Original Paper
Landslides (2016) 13:18 Baolin Wang I Bhuwani Paudel I Haoqiang Li
DOI 10.1007/s10346-014-0549-y
Received: 4 September 2014
Accepted: 18 December 2014 Behaviour of retrogressive thaw slumps in northern
Published online: 9 January 2015
Her Majesty the Queen in Right
Canadathree-year monitoring results from 18 sites
of Canada as represented by:
Author(s): Wang, B. 2015
Abstract Eighteen (18) retrogressive thaw slumps (typical land-
slides in ice-rich permafrost) in northern Canada were monitored
for 3years to investigate the characteristics of their retrogressive
behaviour. The head scarp retreat distances and scarp wall heights
were measured annually. The first year monitoring results from 13
of the sites were released earlier, which demonstrated a correlation
between thaw retrogression rate and scarp wall height. More data
were obtained from a subsequent 2 years of monitoring and with
five monitoring locations added. The additional data enhanced the
initial findings that the retrogression rate increased with the in-
crease of the scarp wall height. An updated correlation between the
retrogression rate and scarp wall height is presented in this paper.
The effect of slope orientation on thaw slump retrogression was
also investigated. The data provided evidence that the occurrence
of the retrogressive thaw slumps had no preference over slope
orientation. The retrogression rates were also not affected by the
slope facing direction.
Keywords Landslide . Retrogressive thaw slump . Ice-rich
permafrost
Introduction
A retrogressive thaw slump is a common type of slope failure of
ice-rich permafrost (Everdingen 2002). Ice-rich permafrost is typ-
ically associated with fine-grained soils. The retrogressive failure
usually starts from an area where the permafrost is exposed to the
atmosphere. The material melts and flows away, leaving behind a
near vertical scarp wall. The melting continues throughout the
summer as long as the ice-rich permafrost is exposed. The debris
is formed by the materials ablated from the scarp face. It is often
fluid due to high water content hence flows easily even at a shallow
angle. Such thaw slumps may pose serious threat to nearby infra-
structure or facilities, e.g. highways and pipelines (Ma et al. 2006).
Such slope failures are commonly observed in the Mackenzie River
valley (or Mackenzie valley), Northwest Territories (NWT),
Canada. Wang et al. (2005) reported 97 active landslides that were
counted along a proposed pipeline route in the northern part of
the Mackenzie valley in June 2005. Most of those slides were
retrogressive thaw slumps. Understanding the behaviour of such
thaw slumps is important for mitigation of the planned
infrastructure.
Several researchers have discussed the behaviour of retrogressive
thaw slumps. Lamothe and St-Onge (1961) reported that the retreat
rates of a thaw flow in Ellef Ringes Island, NWT averaged about 7 m,
with a maximum of 10 m in a summer season. McRoberts and
Morgenstern (1974) investigated ablation problems of different thaw
flows along the Mackenzie valley, NWT. They reported that the
ablation rate was 1 to 20 cm per day with an average of 10 cm per
day. Burn and Friele (1989) studied retrogressive thaw flows near
Mayo, Yukon Territory, Canada. They reported short-term mean
retreat rates of two slumps to be 7.4 and 11.1 cm per day, respectively,
from June 30 to July 9, 1982. The mean (year 1949 to 1987 and 1961 to
1987) annual head scarp retreat rates were measured to be approx-
imately 12 and 14 m, respectively, for the two slumps. Burn and
Lewkowicz (1990) reported that the head scarp ablation continued
at a rate of 14 to 16 m per year for many years at Stewart River near
Mayo. Burn (2000) reported that ablation rate of a retrogressive thaw
flow near Mayo was as high as 16 m per year, with an average ablation
rate of 10 m per year.
Kerfoot (1969) observed slope failures on Garry Island (NWT,
Canada) and indicated that the movement of head scarps did not
have preference on south-facing or north-facing slopes. Aylsworth
et al. (2000) indicated that landslide inception showed no partic-
ular preferred orientation in a survey of 3400 landslides in the
Mackenzie valley region. McRoberts and Morgenstern (1974) indi-
cated that the lack of orientation preference may have been due to
the 24-h daylight during the summer months. However,
Lewkowicz (1988), who studied short-term rates of ablation of
massive ground ice in Mackenzie Delta, NWT, reported that ice
face orientation affected the timing and total amount of ablation
during clear skies, but was insignificant during overcast
conditions.
Pufahl and Morgenstern (1980) discussed the behaviours of
retrogressive thaw flows based on energy balance calculations.
The observed average rate of ablation for a north-facing large
slump, near Fort Simpson, NWT, was 14 cm per day from June
21 to July 21, 1974. It was concluded that there was no correlation
between retreat rates and insolation. There was no reduction in
retreat rates for the slopes facing north.
Thus far, the reported behaviours of the retrogressive thaw
slumps have been based on studies of limited sites or duration.
There is also lack of strong evidence or even consensus whether
the retrogressive thaw rates have any preference over slope orien-
tation. A field monitoring programme was therefore launched at
the Geological Survey of Canada (GSC) in 2007 to investigate the
retrogressive behaviour of the thaw slumps in the Mackenzie
valley. The programme covered a relatively large number of sites
for a 3-year period. During the first year in 2007, a total of 13 sites
were monitored. The first year data was reported in Wang et al.
(2009). The monitoring programme continued for two more years
in 2008 and 2009 with five more thaw slumps added to a total of 18
sites. This paper presents the 3-year monitoring results. A corre-
lation between retrogression rate and scarp wall height as reported
in Wang et al. (2009) has been updated based on the new data.
Site description
The locations of the 18 retrogressive thaw slumps monitored from
2007 to 2009 are shown in Fig. 1. These slumps were selected as
they were representative of the typical retrogressive thaw slumps
observed in the region. Seven of the slumps are located north of
Inuvik, and 11 southeast of Inuvik. The northern sites are in
Inuvialuit Settlement Region and are denoted as I1 to I5. The
Landslides 13 & (2016) 1
 Original Paper

Fig. 1 Site location map (dots with notation I1 to I5 and G1 to G12 indicate study sites. Weather stations are located in Inuvik as marked on the map)

southern sites are in Gwichin Settlement Region and are denoted wooded areas, indicating fire activity prior to the establishment of
as G1 to G12. The permafrost in both regions extends to a depth of the current forest. The soil materials at the southern site are
the order of hundred metres according to Taylor et al. (2000) and
Smith and Burgess (2002).
The terrain of the study region is gently rolling with abundant
lakes. The vegetation varies considerably between the northern
and the southern regions. The vegetation at the northern sites (I1
to I5) is typical tundra with occasional shrubs and few or no trees
(see Fig. 2 for example). The ground in this region is hummocky
and typically covered by an organic layer of about 15 to 30 cm thick
(locally variable) that consists of roots and moss. The subsurface
materials exposed on the scarp face are clayey silt or silty clay with
trace sand and gravel and occasional cobbles (glacial till deposit).
The active layer is about 0.5 m thick. The sediments below the
permafrost table are ice-rich.
The southern sites (G series) are in forest areas (Figs. 3 and 4).
However, most of the slumps are located in areas burned by wild
fires that occurred in or before 1998. Trees were burned (Fig. 4)
and mineral soils were often exposed due to destruction of the
organic cover. Vegetation was still recovering at those sites at the
time of this investigation. The exceptions are G4 and G8 where Fig. 2 Typical retrogressive thaw slump in Inuvialuit region north of Inuvik (aerial
mature trees were observed (Fig. 3) and there was no sign of recent photo of site I1, Sept. 23, 2004). Approximate dimension: length 95 m, width 115
fire activity. Some burned trees were occasionally observed in the m, outlet flowing tongue 45 m long

2 Landslides 13 & (2016)


Fig. 3 A typical retrogressive thaw slump in wooded area southeast of Inuvik. It is
about 300 m long and about 300 m wide. The head scarp is about 15 m high facing
north (aerial photo of site G8, June 20, 2007; see Fig. 1 for location). Note: Behind
the scarp wall is a winter road. A telecommunication line used to run along this
route but was abandoned. The road was cut off by the slump later in 2010. Also
note two new slumps developing in the background, indicated by white arrows
mostly silty clay with high ice content as observed from the scarp
faces. The active layer is typically about 1 m thick in the burned
areas and thinner (about 0.5 m) in the unburned areas where
undisturbed organic cover provides better insulation to the
permafrost.
Weather condition
Weather stations were available in Inuvik, which were about
30 km from the nearest northern site and about 90 km from
the nearest southern site (Fig. 1). Historical weather data re-
corded in Inuvik from 1958 to 2009 were obtained from
Environment Canada (2011). The minimum and maximum
mean annual air temperature (MAAT) and air thawing indices
(Iat) are provided in Table 1. The MAAT and Iat for the study
years of 2007, 2008 and 2009 are given in Table 2. The weekly
Fig. 4 A typical retrogressive thaw slump in burned area southeast of Inuvik. Note
dead trees in background (aerial photo of G1, Sept. 23, 2004)
Table 1 Historical mean annual air temperatures (MAAT) and air thawing indices
(Iat) in Inuvik (19582009)
Description MAAT (C) Iat (Cday)
Minimum 11.8 939
Average 8.8 1309
Maximum 4.5 1822
average air temperatures of 2007, 2008 and 2009 are plotted in
Fig. 5 for comparison purpose.
As noted in Tables 1 and 2, the mean air temperatures in 2007,
2008 and 2009 were close to the average historical conditions. It
was slightly cooler in 2008 and 2009 than in 2007 with the MAAT
lower by only 0.1 C. The Iat was also slightly lower in 2008 and
2009. However, the differences are marginal compared to the
historical variations shown in Table 1.
The ground surface was usually covered by snow from late
September or early October to the end of May or early June during
the study period. Freezing usually starts in late September (Fig. 5).
The field trips were scheduled during the spring break up and
right before freezing in 2007, 2008 and 2009.
Geotechnical properties
The scarp walls of the thaw slumps provided ideal access to the
freshly exposed subsurface soils. More than 280 soil samples were
taken from the monitored slumps during the field visits. Most of
the samples were taken for water content measurements for a
separate study. The sampling locations were selected so that they
are representative of the typical conditions. Grab samples were
taken with hand tools from the scarp wall face after removing the
surficial materials that were believed to have been affected by
exposure to the atmosphere. The samples were taken every
10 cm from ground surface to a maximum depth of 2.8 m, which
was well below the permafrost table.
Figure 6 shows the grain size data of the soil samples from six
sites (seven locations). The materials were mostly silty clay or
clayey silt. The samples from the southern sites (G-series) typically
contain more fines than those from the northern sites (I-series).
The total water contents (including pore ice melt) of the soil
samples were measured in the laboratory at the Geological Survey
of Canada. Typical soil moisture profiles for the northern and
southern sites are shown in Fig. 7. More detailed geotechnical data
from those other samples are available from Wang et al. (2008). As
seen in Fig. 7, the total water contents typically increased with
depth and ranged from about 20 % in the active layer to greater
than 100 % below the permafrost table. The fine-grained soils were
usually ice-rich below permafrost table.
Table 2 Mean annual air temperatures (MAAT) and air thawing indices (Iat) in
Inuvik of study period (2007 to 2009)
Year MAAT (C) Iat (Cday)
2007 7.6 1376
2008 7.7 1212
2009 7.7 1306
Landslides 13 & (2016) 3
 Original Paper
Fig. 5 Weekly average air
temperature in Inuvik (2007 to 2009)
(source: Environment Canada 2011)

Fig. 6 Typical grain size distribution

of soil samples from the monitored
sites

Fig. 7 Select soil moisture-depth

profiles for a northern site (a) and a
southern site (b)

(a) (b)

4 Landslides 13 & (2016)

Table 3 Field measurements of scarp wall heights and retrogression rates
Site-location Scarp 2007 2008 2009
facing Scarp Scarp Scarp Scarp Scarp Scarp
direction height (m) retrogression height (m) retrogression height (m) retrogression
(m/year) (m/year) (m/year)
I1-STK1 N 50 E 9.1 7.0 2.2 4.0
I1-STK2 N 50 E 9.1 6.0 2.2 3.0 1.7 0.0
I1-STK4 N 52 E 1.7 1.0
I2-STK4 N 164 E 2.5 4.3 1.7 0.4
I2A-STK1/STK5 N 196 E 6.0 10.0 6.0 8.8
I3-STK1A N 0 E 2.5 1.8 2.5 0.8 1.5 0.3
I3A-STK1 N 60 E 2.5 1.2 2.0 0.1
I4-STK1 N 80 E 13.0 8.0 13.0 14.3 10.0 11.7
I5-STK1 N 100 E 6.6 5.6 8.1 8.4 5.5 4.1
G1-STK1/STK2 N 253 E 7.0 9.3 8.0 15.1 6.0 14.1
G1-STK3 N 180 E 6.0 9.0 6.0 11.0
G1-TP2 N 343 E 5.0 6.7 5.0 6.0
G2-STK1 N 0 E 5.8 9.0 10.6 11.0
G3-STK1 N 241 E 4.0 6.5 4.0 12.2
G4-STK1/STK2 N 30 E 15.0 8.8 15.0 9.4 14.0 8.8
G5-STK1/STK2 N 290 E 8.2 7.0 4.7 11.4 4.8 10.0
G6-STK1/STK2 N 18 E 6.4 6.5 6.0 9.1 5.0 8.1
G7-STK2 N 56 E 2.2 4.7 1.9 3.7 2.5 0.0
G8-STK1/STK2 N 0 E 14.5 9.7 14.5 10.8 12.0 8.6
G9-STK1/STK2 N 190 E 7.0 7.6 4.4 11.3 4.0 7.2
G10-STK1 N 22 E 7.5 9.8 7.7 10.1
G12-STK1 N 117 E 4.6 4.9

Liquid limits measured for some samples are also shown in moisture contents below permafrost table were usually greater
Fig. 7. The soils above and below the permafrost table were usually than the liquid limit. This was almost always evident by the high
observed as the same type of materials. However, the total ice content observed in the permafrost zone. When such soils are

Fig. 8 Thaw slump retrogression

rate vs. height of scarp wall

Landslides 13 & (2016) 5

 Original Paper
Table 4 Constants for correlation between retrogression rate and scarp wall height
(Eq. 1)
Constant Lower bound Average Upper bound
A 8.5 12.0 15.5
B 22.0 28.5 35.0
C 3.4 2.6 1.7

exposed to the atmosphere and thaws, they behave like a slurry

and flow easily.

Retrogression rates
The retrogressive thaw slumps were monitored with survey stakes
installed at the back of the scarp walls at the beginning of the first
summer thaw season. The retreat distances of the scarp walls were
measured from the survey stakes with a measuring tape at the end
of the thaw season in 2007, 2008 and 2009. Most of the survey
stakes were installed at the higher grounds above the scarp walls,
while some were installed at lateral locations towards which the
scarps were retreating. Also measured were the heights of the scarp
walls. The slope orientation or the scarp facing direction at each
monitored location was measured as well. Table 3 provides the 3-
year monitoring data.
Retrogression correlation
An initial correlation between the thaw slump retrogression rate
and the height of the scarp wall was identified by Wang et al.
(2009), which was from the first year (2007) monitoring data.
Continued monitoring during the subsequent 2 years (2008 and
2009) enhanced the data. The annual retrogression rates for all
3 years were plotted against the height of the scarps as shown in
Fig. 8. As noted from this figure, the distribution of the 20082009
data followed the same pattern as established by the 2007 data.
However, the upper bound curve moved higher with the addition
of the 2008 and 2009 data. The lower bound curve did not change.
The 3-year data are more scattered than the first year data alone.
As noted in Wang et al. (2009), the increased scattering was
expected with more data becoming available.
Fig. 10 Illustration of thaw slump process
As discussed earlier, the average air temperatures for 2008 and
2009 were very close to that of 2007. It is therefore unlikely that
the air temperature contributed to the increase of the upper limit
of the data. Note that, based on the slightly lower air thawing
indices in 2008 and 2009 (Table 2), the retrogression rates should
have dropped slightly, which was actually not the case. In fact, as
indicated in Table 3, amongst the 16 locations at the 13 sites
monitored in 2007, the retrogression rate dropped at five locations
and increased at seven locations in both 2008 and 2009, while the
other four locations experienced a drop in 1 year and an increase
in another year. Also note from Table 3 that the six monitoring
points at five slumps added in 2008 and 2009 indicated near-
average retrogression rates. Therefore, the increase of the upper
bound limit would not have been attributed to the newly added
sites.
Nevertheless, given the 3-year duration of measurements and
the near normal average air temperature conditions, the
retrogression rates collected are believed to be representative of
the normal rates in the region. Based on this set of data, the

Fig. 9 Extended monitoring data

indicating no relationship between
retrogression rate and slope angle

6 Landslides 13 & (2016)

correlation as reported in Wang et al. (2009) has been updated as
follows:
H=C
R AB  e
Where R denotes scarp wall retrogression rate (m/year); H is
height of scarp wall (m) and A, B and C are fitting constants with
values given in Table 4.
Another correlation between thaw retrogression and overall
slope angle was presented in Wang et al. (2009). However, a point
was made in the paper that the correlation with slope angle was
less meaningful than that with scarp wall height. The new data
confirmed this point. Further correlation of retrogression rate
versus slope angle was not found from the extended monitoring
data as shown in Fig. 9.
The measured retrogression behaviour can be explained by
examining the processes of the thaw slumps. The head scarp
retreat is associated with thawing of ice-rich permafrost.
Ablation continues along the scarp wall in summer as long as
there is enough ice-rich material in the wall. Depending on the
water content, the ablated materials build up at the toe of the wall
and cover part of the ice-rich wall as illustrated in Fig. 10. Lower
scarps tend to retreat more slowly as shown in Fig. 8. One of the
reasons could be attributed to the debris covering the ice-rich wall
from time-to-time. The materials in the active layer have a lower
moisture content compared to those of the ice-rich permafrost.
The ice-rich soil behaves like slurry when thawed and flows away
relatively quickly from the scarp (Wang et al. 2005). When the
slurry is mixed with the slumped materials from the active layer,
its fluidity decreases, resulting in a higher and steeper debris
buildup at the toe. The debris buildup reduces the area of ice-
rich permafrost being exposed and therefore a reduced amount of
slurry is generated. This further decreases the fluidity of the
slurry mix and hence more sediment accumulation at the toe.
Fig. 11 Distribution of scarp retrogression rate and scarp wall facing direction
N
1
Survey stake G1-STK3/G2-STK1
G2
G1
Fig. 12 Two retrogressive thaw slumps (G1 and G2) retreating towards each other
(photo taken: Sept. 8, 2007). A survey stake (labelled as G1-STK3 and G2-STK1) was
installed between the two slumps. The south-facing scarp wall of G1 and the north-
facing scarp wall of G2 retreated at the same rate in 2007 and 2008. The two
slumps merged together by September 2008 (see Fig. 13)
However, as the debris builds up, its weight increases. When the
buildup is high enough, the mass slides away driven by its own
weight. Thus, the covered ice-rich permafrost along the face of the
wall is re-exposed, and thawing is accelerated until it is covered
again. This process is illustrated in Fig. 10. The cycle repeats
throughout the summer.
In summary, the maximum scarp retrogression rate ranged
from about 8 m/year to about 15 m/year. The retrogression rate
was found to increase with increased scarp wall height. However,
once the scarp reaches a certain height, say about 6 to 8 m (Fig. 8),
the increase of the retrogression rate starts to taper off.
Effect of slope orientation
The relationship between thaw slump retrogression and slope
orientation for the 3-year monitoring period is plotted in Fig. 11.
A vector drawn from the coordinate origin to any data point
G2
G1
Fig. 13 Two retrogressive thaw slumps (G1 and G2) merged together (photo
taken: Sept. 10, 2008)
Landslides 13 & (2016) 7
 Original Paper
indicates the scarp facing direction and distance of retreat at that
point. As indicated by Wang et al. (2009), the retrogression was not
influenced by the slope orientation observed from the first year
(2007) data. This conclusion was supported by the extended moni-
toring data from 2008 and 2009. The thaw slump retrogression rates
are fairly evenly distributed in all directions where data are available.
Although it happened that more data are available in the northeast
and southwest quadrants, it was observed in the field that most of the
slumps are in near circular shape. It means that the scarp walls have
been retreating relatively evenly around the slump. A portion of the
wall would inevitably face southeast and northwest. In other words,
the thaw slump occurrence and their retrogression rates have no
preference over any slope facing direction. To further support this
argument, Fig. 12 demonstrates two separate thaw slumps retreating
towards each other. A survey stake (labelled as G1-STK3 and G2-
STK1) was installed at a midpoint between the two slumps. As seen in
Table 3, the south facing scarp wall of G1 and the north-facing scarp
wall of G2 retreated at exactly the same rate of 9 m/year in 2007 and
11 m/year in 2008. The two slumps merged together in September
2008 as shown in Fig. 13.
Conclusions
Retrogressive behaviour of 18 thaw slumps in the Mackenzie valley
was monitored for 3 years. The data enhanced the initial findings
reported by Wang et al. (2009). The results indicated that the thaw
slump retrogression rate increases with the increase of scarp wall
height. The increase of the retrogression rate starts to taper off after
the scarp height reaches about 6 to 8 m. The maximum retrogression
rates may range from about 8 to 15 m/year in the northern Mackenzie
valley region. The average air temperatures during the three moni-
toring years were very close to the historical average. The 3-year
monitoring data are believed to be representative of the average
conditions. The correlation between the retrogression rate and scarp
wall height is therefore believed to be representative of the typical
behaviour of the retrogressive thaw slumps in the region.
The monitoring data provided evidence that the thaw slump
occurrence and retrogression rate have no preference over slope
orientation. In other words, the north-facing slumps do not nec-
essarily retrogress slower than those facing south.
Acknowledgments
The authors would like to thank Wendy Sladen for her careful
review and comments on the manuscript that helped improve the
quality of the paper. This paper was released under an internal ESS
Contribution Number 20140195.
8 Landslides 13 & (2016)
References
Aylsworth JM, Duk-Rodkin A, Robertson T, Traynor JA (2000) Landslides of the Mackenzie
valley and adjacent mountainous and coastal regions. Geological Survey of Canada
Bulletin 547:167176
Burn CR (2000) The thermal regime of a retrogressive thaw slump near Mayo, Yukon
Territory. Can J Earth Sci 37:967981
Burn CR, Friele PA (1989) Geomorphology, vegetation succession, soils characteristics
and permafrost in retrogressive thaw slumps near Mayo, Yukon Territory. Arctic
42:3140
Burn CR, Lewkowicz AG (1990) Retrogressive thaw slumps. Can Geogr
34:273276
Environment Canada (2011) http://climate.weatheroffice.ec.gc.ca/climateData/
canada_e.html. Accessed 11 Feb 2011
Everdingen RV (2002) Multi-language glossary of permafrost and related ground-
ice terms. National Snow and Ice Data Center/World Data Center for
Glaciology, Boulder
Kerfoot DE (1969) The geomorphology and permafrost conditions of Garry Island NWT.
Ph.D Thesis, University of British Columbia, Vancouver, B.C.
Lamothe C, St-Onge D (1961) A note on a periglacial erosional process in the Isachsen
Area NWT. Geographical Bulletin 16:104113
Lewkowicz AG (1988) Ablation of massive ground ice, Mackenzie Delta. Proceedings, 5th
International Conference on Permafrost vol. 1, pp 605610
Ma W, Niu F, Akagawa S, Jin D (2006) Slope instability phenomena in permafrost regions
of Qinghai-Tibet Plateau, China. Landslides 3:260264
McRoberts EC, Morgenstern NR (1974) The stability of thawing slopes. Can Geotech J
11:447469
Pufahl DE, Morgenstern NR (1980) The energetics of an ablating head scarp in
permafrost. Can Geotech J 17:487497
Smith SL, Burgess MM (2002) A digital database of permafrost thickness in Canada.
Geological Survey of Canada, Open File 4173; 38 pages, 1 diskette
Taylor AE, Burgess MM, Judge AS, Allen VS (2000) Deep ground temperatures. The
physical environment of the Mackenzie Valley, Northwest Territories: a base line for
the assessment of environmental change. Dyke LD, Brooks G (eds) Geological Survey
of Canada, Bulletin 547:105109
Wang B, Nichol S, Su X (2005) Geotechnical field observations of landslides in fine-
grained permafrost soils in the Mackenzie Valley, Canada. Proceedings of the First
General Assembly of the International Consortium on Landslides. Sassa K, Fukuoka H,
Wang F, Wang G (eds) Washington, DC, pp 203212
Wang B, Li H, Paudel B (2008) The transient layer and slope stability in fine-grained
permafrost soils. Proceedings of the 61st Canadian Geotechnical Conference, Sept.
2124, 2008, Edmonton, Alberta: 919924
Wang B, Paudel B, Li H (2009) Retrogression characteristics of landslides in fine-grained
permafrost soils, Mackenzie Valley, Canada. Landslides 6:121127
B. Wang ()) : B. Paudel : H. Li
Natural Resources Canada,
Ottawa, Ontario K1A 0E8, Canada
e-mail: bwang@NRCan.gc.ca
 Original Paper
Landslides (2016) 13:923 Yueping Yin I Yuliang Cheng I Jingtao Liang I Wenpei Wang
DOI 10.1007/s10346-015-0554-9
Received: 16 April 2014
Accepted: 5 January 2015 Heavy-rainfall-induced catastrophic rockslide-debris
Published online: 14 January 2015
Springer-Verlag Berlin Heidelberg 2015
flow at Sanxicun, Dujiangyan, after the Wenchuan
Ms 8.0 earthquake
Abstract This paper uses the catastrophic rockslide at Sanxicun
village in Dujianyan city as an example to investigate the forma-
tion mechanism of a rapid and long run-out rockslide-debris flow
of fractured/cracked slope, under the application of a rare heavy
rainfall in July 2013. The slope site could be affected by the
Wenchuan Ms 8.0 Earthquake in 2008. The sliding involved the
thick fractured and layered rockmass with a gentle dip plane at
Sanxicun. It had the following formation process: (1) toppling due
to shear failure at a high-level position, (2) shoveling the accumu-
lative layer below, (3) forming of debris flow of the highly weath-
ered bottom rockmass, and (4) flooding downward along valley.
The debris flow destroyed 11 houses and killed 166 people. The
run-out distance was about 1200m, and the accumulative volume
was 1.9106m3. The rockslide can be divided into sliding source,
shear-shoveling, and flow accumulative regions. The stability of
this fractured rock slope and the sliding processes are discussed at
four stages of cracking, creeping, separating, and residual accu-
mulating, under the applications of hydrostatic pressure and uplift
pressure. This research also investigates the safety factors under
different situations. The double rheological model (F-V model) of
the DAN-W software is utilized to simulate the kinematic and
dynamic processes of the shear-shoveling region and debris flow.
After the shear failure occurred at a high-level position of rock, the
rockslide moved for approximately 47s downward along the valley
with a maximum velocity of 35m/s. This is a typical rapid and long
run-out rockslide. Finally, this paper concludes that the identifi-
cation of the potential geological hazards at the Wenchuan moun-
tain area is crucial to prevent catastrophic rockslide triggered by
heavy rainfall. The identified geological hazards should be prop-
erly considered in the town planning of the reconstruction works.
Keywords Wenchuan Ms 8.0 earthquake .
Heavy rainfall . Rockslide . Debris flow
Introduction
During July 712 of 2013, a heavy rainfall occurred at the
Wenchuan Ms 8.0 earthquake area. The 5 days accumulative pre-
cipitation in the Dujiangyan area of Sichuan province was up to
1129 mm, which was rare in previous years and exceeded the mean
annual precipitation of 1110 mm over the 25 years from 1987 to
2012. The heavy rainfall triggered a rapid and long run-out
rockslide in Sanxicun village. Its debris had a volume of 1.9 million
m3. It caused a sliding distance up to 1200 m and killed 166 people
in the village. The Sanxicun landslide was located in Zhongxing
town, Dujiangyan city of Sichuan province. The site is located at
(E1033348, N305455), 68 km away from Chengdu city to the areas that should be subject to monitoring and prevention.
east and 16 km away from the Dujiangyan urban district (Fig. 1).
This area was famous for its agreeable climate and beautiful
natural mountain scenery, lakes, and abundant vegetation. The
typical natural conditions to resorts and vocations often concealed
the potentialities of rockslide, rockfall, and debris flow. This re-
sulted in a misunderstanding in lower risk of geological hazards.
This area enjoyed a high reputation stemming partly from the
Qincheng MountainDujianyan Scenic Resort and attracted a
great many tourists every summer. Therefore, research on
rainfall-induced post-earthquake geological hazards in these areas
has significance to the reconstruction of towns and villages and the
tourism safety after the earthquake. The rockslide dynamic re-
sponse characteristics triggered by the Wenchuan Ms 8.0 earth-
quake and post-earthquake debris flow disaster have received
much attention from researchers (Yin et al. 2009; Tang et al.
2011). Tang et al. (2009) reported a heavy-rainfall-induced debris
flow disaster on September 24 of 2008 at the earthquake-ruined
Beichuan town in the central part of the area affected by the
Wenchuan Ms 8.0 earthquake and presented the possibility that,
due to the ground shaking, the critical amount of accumulated
precipitation and the hourly rainfall intensity necessary to initiate
debris flow were reduced compared with values before the
earthquake. Recently, Zhou and Tang (2013) summarized 11 rainfall
events that induced debris flows between 2008 and 2012 after the
Wenchuan earthquake and discussed the rainfall thresholds for
debris flow occurrence. However, the researches on rainfall-
induced rockslides after the earthquake and its transferring pro-
cess into a long run-out debris flow have made little significant
progress.
The parent rock of the Sanxicun rockslide is a slightly inclined,
nearly horizontal strata sandstone formation with intact rock
constitutes. This rockslide was located at southwest of Wenchuan
Ms 8.0 earthquake fault and was only 12 km away from the
epicentral region. The seismic intensity of the Sanxicun area was
up to the IX degree. According to the emergency investigation
1 week after the Wenchuan Ms 8.0 earthquake, three small scale
rockfalls were induced. Two rockfalls were at the frontier of the
slope of the rockslide source area in the moving direction of
NNE10, and the other was at the southeastern corner in the
moving direction of NWW288 (Fig. 2, also see Fig. 8). They
presented a disturbance, or even damage on the slope from the
earthquake. Research on this rockslide is significant for prevention
on post-earthquake geological hazard prevention. Therefore, this
paper uses the Sanxicun landslide as an example to study the
stability of fractured slope after the Wenchuan Ms 8.0 earthquake
and the formation mechanism of rapid and long run-out
rockslides triggered by heavy rainfall. The adverse conditions of
the geological features prior to the rockslide are investigated,
which provide a scientific consideration for classifying hazard
Environmental geological condition
In terms of geomorphology, the Sanxicun rockslide was located at
the transition zone from the Chengdu Plain to the QinghaiTibet
Landslides 13 & (2016) 9
 Original Paper
Fig. 1 Location of catastrophic rockslide-debris flow at Sanxicun, Dujiangyan due
to heavy rainfall on July 10, 2013(also indicated map of density of landslides
triggered by Wenchuan Ms 8.0 earthquake on May 12, 2008)
Plateau, which features erosion tectonics with low mountains and
hills. Mountains generally run in the NESW direction with rela-
tively flat peaks. The mountains are subject to the control of joint
cracks in NESW and EW directions. The northwest, northeast,
and north of the mountains are free faces.
In terms of regional geology, the Sanxicun rockslide was located
at the Longmenshan tectonic belt. It was on the SE side of the
Erwangmiao secondary fault and the SE side of the Shenxianqiao
Fig. 2 3D images showing catastrophic rockslide-debris flow at Sanxicun (red
dashed line indicates the small rockfall due to Ms 8.0 earthquake)
10 Landslides 13 & (2016)
rear slope. The slope is a monoclinal structure mountain with a
Cretaceous Guankou Formation (Kg) whose rock consists of
brown-red and purplish red thick blanket sandstone, siltstone,
and sandy mudstone.
With regards to precipitation, this area belongs to the middle
semi-tropical moisture monsoon climate of the Sichuan Basin with
abundant rainfall which is concentrated from May to September of
every year. According to the 25-year precipitation statistics from
1987 to 2012, the mean annual precipitation is approximately
1100 mm. The precipitation from May to September accounts for
80 % of the annual rainfall, among which August has the maxi-
mum mean monthly precipitation of up to 243 mm (Fig. 3).
Field surveys on rockmass structure indicated that the parent
rock of the Sanxicun rockslide was thick blanket sandstone and
siltstone of the Cretaceous Guankou Formation (Kg) doped with
small amounts of sandy mudstone. The bedding of the parent rock
is 11016, indicating that it has a flat dipping bedrock slope.
Remote sensing image presented that the rockslide body was cut
into block structure by the three sets of preferred joints and a soft
layer. The north and west sides of the rockslide body were free
faces since its east one was obstructed by mountain. This resulted
in a deflection in the principle slide direction from 110 (true dip
angle) to the northeast direction (apparent dip angle). Figures 4
and 5 show the three sets of preferred joints in the rockslide and
adjacent areas: (1) joint crack belt in NESW direction, (2) joint
crack belt adjacent to EW direction, and (3) joint crack belt
adjacent to SN direction. Combining joint crack direction and
geomorphologic features, the first set of joint surfaces can be
judged as inclined tension crack surfaces with a crack occurrence
of 3247580. The second set of joint surfaces can be judged as
transverse tension crack surfaces, which are parallel to the rear
tension surface and are known as killing ditch big cracks in
residents with a crack occurrence of 386. The third set of joint
crack surfaces can be judged as longitudinal tension crack surfaces
with a crack occurrence of 2887080. The length of joint cracks
in rockslide and adjacent areas is between 10 and 60 m. The
numbers of the above three sets of joint cracks are more than
70 % in total, which reflects the preferred controlling characteris-
tics on rockslide body, and the lower microrelief also reflected the
consistency along the joint cracks.
Basic characteristics of rockslide
The Sanxicun rockslide was a gently inclined laminated rockslide.
It is hard to know previously the failure of rockslide in the area
due to the heavy vegetation coverage. The remote sensing images
before and after the sliding are compared to analyze the extent of
the slope failure effectively. Sheared at a high-level position, the
rockmass was transferred to a rockfall and shoveled the bottom
accumulative layer as well as severely weathered rock. The slide
movement direction deflected 67 from 57 NE to 10 NW due to
the obstruction of the front slope. Under the application of rainfall
runoff, the rockfall crushed the valley downstream in the form of a
flow rockslide, destroyed 11 houses located in the valley and adja-
cent tourism area, and killed/buried 166 people.
The elevations of rear edge and front accumulative area are 1140
and 720 m, respectively, which indicates an elevation difference of
up to 420 m between the rear edge and front edge. The rockslide
width had a maximum value of 420 m and a minimum value of
30 m. The rockslide had a longitudinal length of approximately
Fig. 3 Diagram of mean month
precipitation at Dujiangyan from
1987 to 2012
1200 m, and the rockslide area was 18.1103 m2. The maximum
thickness of the rockslide accumulation was 30 m with a mean
thickness of 10 m and a volume of 1.915106 m3. The movement
and accumulative characteristics of the rockslide can be classified
into rockslide source area, impact-shoveling area, and flow accu-
mulative area (Figs. 6 and 7) that will be described in detail in the
following three paragraphs.
Rockslide source area
Rockslide source area was located approximately 100 m northwest
from the downside slope of the Shangqing Palace Building in the
Qingcheng Mountain Scenery Area. The terrain of this area is steep
with a slope of approximately 30. This area mainly consists of
sandstone and siltstone. The surface is comprised of clay gravel
soil covered by a Quaternary loose residual layer with a thickness
of <2 m.
The boundaries of the east and the west and the rear edge of the
rockslide were controlled by joints, the frontier was a free face
(Fig. 8). The plane configuration of this area is an approximate
parallelogram whose sections exhibit trapezoidal shapes with a
larger thickness of rear part and a smaller thickness of front part.
The elevation difference between the rockslide rear edge and the
shear outlet is 96 m with an elevation of 1140 m for the former and
1044 m for the latter. The thickness of this area ranges between 30
Fig. 4 Joint length and percentage
in and around the Sanxicun rockslide
and 54 m with a mean thickness of 40 m. The longitudinal length
of this area ranges between 50 and 200 m with a mean value of
100 m. The plane of this area exhibits an irregular trapezoidal
shape whose width is between 250 and 420 m with a mean value
of 335 m. The mean area is 33.5103 m2, and the estimated volume
is 1.47106 m3.
After the sliding, the parent rock of rockslide only partially sepa-
rated and retained the structural characteristics of rock. To estimate
the volume of the rockslide accumulation, the enlarged index (Ie) is
applied, i.e., the Ie is suggested that the volume of the accumulation is
about 1.21.4 times the volume of the original rockmass of the
rockslide (Yin et al. 2011). Therefore, by multiplying with a rockmass
loose factor of 1.2, the accumulative volume of the rockslide source
area can be computed as 1.21.47106 m3=1.77106 m3.
According to the estimate from remote sensing images prior
and post-sliding, the entirety of the rockslide body moved down-
ward about 50 to 80 m. A sliding mass with a volume of approx-
imately 300103 m3 was sheared out at a high-level position and
moved along two directions, forming a major slide area and a
branch slide area. The major slide moved along the N57 E direc-
tion with an estimated volume of 275103 m3 and transferred to
debris flow with a run-out distance of >1000 m from the exit of
rockslide (also see Figs. 6 and 7). The major slide resulted in severe
disaster and is the focus of this paper. The branch slide moved
along the 10 NW direction with a flow distance of 380 m, a mean
Landslides 13 & (2016) 11
 Original Paper
Flow-accumulative area
Debris flow with an approximate volume of 335103 m3 moved
down towards the valley in a NW direction. The length of the
valley is about 650 m. The upper edge elevation of the valley was
approximately 840 m, and the front toe edge elevation was about
740 m (an elevation difference of 100 m). The slope angle of valley
had a maximum of 15, and the corresponding terrain turned
relatively flat. The valley features a seasonal water flow creek. Its
water supplement is mainly from atmospheric precipitation, which
controls the amount of water in the valley. In general, the valley
has the water flow characteristics of a creek with intensive seasonal
rise and fall. During July 810 of 2013, a heavy rainfall occurred in
the valley and formed mountain torrents. It is estimated that the
flow rate of the torrents was up to 5 m3/s. Therefore, after flooding,
soil and mud joined the rock flow, and the rockslide turned into
debris flow. The resultant debris flow accumulated along valley
beds. According to site exploration report for the emergency
prevention on debris flow (Figs. 10 and 11), the accumulation
Fig. 5 Joint preferred direction in and around the Sanxicun rockslide mainly consisted of huge rocks, stones, boulders, and silt clay.
The content of huge rocks and boulders comprised approximately
25 % with a dominant size of 0.5 m and a maximum size up to
width of 65 m, an accumulative thickness of 1 m, and an estimated several meters. The accumulation had a maximum thickness of
volume of 25103 m3. The branch slide area did not result in
disaster, which is not the focus of this paper.
The residual accumulation volume of the rockslide was approx-
imately 1.47106 m3. Two evident dips were formed at the rear edge
and in the middle of the rockslide, which exhibited multiple
imbricate slide features of rock. The rear edge dip was created by
propagation and shearing application of the original cracks. The
cracks were extended into a depth of 30 m, width of 64 m, and
length of 350 m along the NWSE direction (Fig. 9, also see Fig. 8)
from a width of 48 m, depth of 8 m, and length of 180 m. The
middle dip of the slope had a depth of 58 m, a width of 1020 m,
and an arch extension with a length of 160 m. Although the middle
rockmass remained better rock structure, multiple tension cracks
were developed with evident sinking features.

Impact-shoveling area
After being sheared at a high-level position, the major ditch area
with a volume of 275103 m3 moved down along the N57 E
direction and was accelerated by passing a 20-m high fault scarp.
The rockslide was accordingly impacted to break and shovel the
top 2-m thick surface rockmass and vegetation. This resulted in a
triangle shoveling slope surface along the east side of the valley
with a length of 380 m, an upper edge width of 280 m, and a front
toe edge of only 60 m. The shoveled volume was about 119103 m3
and increased the rockslide volume (also see Fig. 8). The remain-
ing residual accumulation with a volume of 60103 m3 at the
impact-shoveling area moved downward forming a debris flow
with a volume of approximately 335103 m3. The upper edge
elevation of the impact-shoveling area was approximately
1024 m, and the front toe edge elevation was about 840 m (an
84 m difference).
The slope angle of the impact-shoveling area was 27 in average,
up to 47 or steeper near the exiting zone of rockslide, which was
beneficial to rapid sliding. Subjected to direction turning from the
front valley, the debris flow of rockslide directly impacted the right
side of the valley, which forced the rockslide to move with a 20
deflection and in an approximately northward direction. Fig. 6 Map of Sanxicun rockslide-debris flow

12 Landslides 13 & (2016)

Fig. 7 Longitudinal profile of rockslide-debris flow at Sanxicun
8 m with a mean value of 6 m, a width of 60110 m with mean
value of 90 m, an approximate area of 60103 m2, and an approx-
imate volume of 360103 m3.
Initiation mechanism of rockslide
The front of the rockslide source area had a scarp with an eleva-
tion difference more than 20 m (NNE direction). The west side of
the rockslide source area was a deep cut valley. The upper bound-
ary of the rockslide source area was a tension fracture created by
preferred joints along the NWWSEE direction. It was recorded in
the literature that this crack had been formed 400 years ago and
had a maximum width of 10 m and a visible depth of up to 25 m
(Fig. 12a) (Li 1998). After sliding, the maximum continuous exten-
sion length of this fracture was 250 m, and the width was 64 m
(Fig. 12b). There were several tension cracks parallel to the killing
ditch in the residues of the rockslide. These cracks had lengths of
up to 100 m, splitting rocks into multiple blocks. The east bound-
ary of the rockslide extended to the NNENE direction joints
along the valley (also see Figs. 6 and 8). The rockslide was
Fig. 8 RS images showing initial and final border of rockslide at Sanxicun
obviously separated by two sets of dominate steep joints with
discontinued interface and soft layer in bedrock that had an
attitude of 11016. One was tense joint nearly vertical to the
strike of the bedrock with smooth-jointed surface with an attitude
of 386. The other was a compressive joint nearly parallel to
the strike of bedrock with an attitude varied between 2887080
and 3247580. Due to the barrier of stable bedrock at the east
boundary, a compressional area was formed at the southeast
corner where the east boundary interconnected with rear edge
cracks. This compressional area blocked the movement of
rockslide along the direction of the true dip angle and forced a
deflection of sliding direction from 110 SE to 57 NE, which
formed a rockslide with an inclined dip azimuth. In other word,
the rockslide was forced to change the movement direction into an
apparent dip of NE from the true dip of SEE (also see Fig. 8 and
15). This kind of slide was typically studied taking a case of the
Jiweishan rockslide at Wulong, Chongqing city, which occurred on
June 5, 2009, illustrating a typical failure pattern of massive rock
slope that rockmass rapidly slides along apparent dip, and then
Landslides 13 & (2016) 13
 Original Paper
Fig. 9 Previous crack at back border of rockslide at Sanxicun (also see Fig. 8)
transforms into a long run-out rock avalanche (fragment flow).
Before the Jiweishan rockslide, the failure pattern of an oblique
inclined bedding slope was conventionally recognized as a lateral
rockfall/topple and then a transformation into a rockfall accumu-
lation landslide (Yin 2011; Feng et al. 2012). There were several sets
of bedding silt lamellar shatter zones in the rockmass (with occur-
rence of 11016), which cut and separated the sliding rockmass
from the parent rock at the slope bottom to form the bottom
sliding face. At the initial stage of rockslide deformation, the weak
layer and surrounding rock layers were relatively complete due to
no obvious dislocation, which was resisting to groundwater
(Fig. 13a). During the creeping or sliding of rockslide, that brittle
weak layer and surrounding rock layers were fractured or cracked,
which provided a pass way to flow and an aquifer to storage of
groundwater (Fig. 13b). The uplift pressure on the bottom of
rockslide was created because the zones provided beneficial con-
ditions for groundwater preservation and runoffs.
Starting at 2000 hours, July 8th 2013, a regional heavy rainfall
occurred in Dujiangyan area. This strong rain featured long dura-
tion, a wide scope of influence, and severe hazard. The heaviest
rains occurred between 2000 hours, July 8th, and 2000 hours, July
10th. During this time period (2 days long), the rainfall at 35
Fig. 10 Cross-section of debris
deposit at Sanxicun (CC) (also see
Fig. 6 for the section layout)
14 Landslides 13 & (2016)
monitoring stations of Dujiangyan all reached above 250 mm,
the monitored rainfall at the other 12 stations exceeded 500 mm,
and the accumulated maximum precipitation was 1059 mm. This
heavy rainfall achieved a historic record. It has been the heaviest
rainfall since 1954 when Dujiangyan began to record meteorolog-
ical data (Fig. 14).
Apparently, during this long-term heavy rainfall, a great deal of
rain infiltrated into the rear edge tension fractures along the
rockslide and generated hydrostatic pressure. Meanwhile, ground-
water also infiltrated into rockmass along the gentle dip fractured
rockmass to produce water head pressure. These processes dete-
riorated the deformation of rockmass along the dip angle and
resulted in the separation of rockmass and the shear failure at a
high-level position. Many investigators have studied the rockslide
trigger mechanism of gentle dip-horizontal rockmass and pro-
posed a sliding mechanism jointly driven by hydrostatic pressure
at the rear edge inclined fractures and uplift pressure at the
bottom surface as a result of groundwater infiltration (Ji et al.
2000; Fan et al. 2009). Of course, some scholars did not agree with
this view for that the idea of a significant fluid pressure in a
tension crack is likely not very real. First of all, it is hard to
imagine that a tension crack could hold water for any period of
time. Secondly, the volume of water in a narrow tension crack
could be very small, so any slight lateral movement of the sliding
mass would result in an immediately disappearance of the water
and the associated lateral force. Thus, considerable thought has to
be given to whether a tension crack filled with water can actually
exist, and if it can exist, would the lateral hydraulic force remain
when there is some as movement (GEO-SLOPE International Ltd.
2008).
The Sanxicun rockslide met the two conditions mentioned
above. First, it must be noted that the section cut along a true
dip angle of 110 of rockmass (Fig. 15, also see Fig. 6) that was
basically parallel to the rear edge tension fracture. The highest
ground point prior to sliding was located at the west boundary at
an elevation of 1125 m, and the lowest ground point was at the east
boundary at an elevation of 1050 m. There was an elevation
difference of 75 m between these two ground points. The east
boundary was not only the slide-blocking boundary but also the
water barrier for groundwater due to the compression of rockslide
movement along true dip at the early stage. This configuration
Fig. 11 Cross-section of debris
deposit at Sanxicun (DD)(also see
Fig. 6 for the section layout)
assured that the groundwater in rear edge tension fracture could
steadily reach the needed height before sliding and at the early
stage of sliding. It could be speculated from Fig. 15 that the
groundwater in the tension fracture could keep a relatively steady
depth due to a closed hydrogeological condition not easily yielding
to discharge formed by the oblique apparent dip layered rockmass.
Secondly, the heavy rainfall before and during the rockslide not
only exceeded the record of recent decades but also created a
duration record. This kind of continuous several-days-long ex-
treme heavy rainfall was enough to provide the necessary ground-
water lastly recharge and seepage into the crack during the
rockmass sliding. Thus, the slide of the rockmass could be gener-
alized into four stages of cracking, creeping, separating, and re-
sidual accumulation (Fig. 16).
Cracking stage
This stage corresponded to the early process of rock crack defor-
mation. Due to the incised valley, a terrain with two free faces was
formed surrounding the east and north sides of the slope. Multiple
sets of joints resulted in unloading cracks in the rock. These
unloading cracks on the track of large joints in the NWWSEE
direction of the rear edge then formed preferred tension cracks.
Meanwhile, there was no evident slide in the weak layer consisting
of siltstone and shall sandstone at the bottom of the rock. During
the rainfall, surface water infiltrated into the rockmass along steep
inclined cracks at the rear edge, which produced hydrostatic
Fig. 12 Photos showing the width of back crack of rockslide before and after sliding
pressure on the rear edge of rockmass. However, the gently in-
clined cracks along the bottom of the rockmass were not well
developed and had poor water permeability; therefore, penetrating
uplift pressure could not occur (Fig. 16a).
Due to a lack of groundwater and runoff information in the
rockslide, it is difficult to set up the seepage field in the rockmass.
The 2D analysis on typical longitudinal section is preferred instead
of 3D numerical simulation. This paper utilizes the Geo-Studio
2007 commercial software to simulate the stability of rockmass
under the application of hydrostatic pressure in the relaxed crack
stage (Swedish method of slices). Rockslide of the Cretaceous
Guankou Formation has a wide distribution in the Sichuan area,
which makes it possible to select parameters in an analogy meth-
od. This paper selects a typical strength testing index of saturated
rockslide soil recommended by the Rockslide Investigation Report
at Fenghuangshan, Chengdu (Zhao et al. 2009) which includes:
1. Shear strength for sliding surface(clay): cohesion c=13.54 kPa
and internal friction angle =8.74
2. Rockslide mass(mud rock): cohesion c=70100 kPa; internal
friction angle =2042; density =21 kN/m3.
A calculated model is conceptualized according to the
hydrogeological condition and rockmass structure, in which the size
of model has the same size as the actual rockslide with a longitudinal
length of about 140.6 m and a height varying from 55.6 to 34.7 m
Landslides 13 & (2016) 15
 Original Paper
Fig. 13 Fractured rockmass showing a rich water-bearing layer
(Fig. 17, also see Fig. 7). The model has three sub-blocks at the stages
of cracking and creeping, and then front block sliding away accord-
ing to the remote sensing images. Since the water level that filled in
tension cracks of rockmass during the rainfall has not been actually
tested, an approximate method using the percentage ratio of filling
level to crack thickness is used for trend analysis.
When cracks are not filled with groundwater, the factor of safety
(FOS) equaled 1.105 (Table 1), which indicates that rockmass is
generally stable. However, the rear edge of rockmass is subject to
hydrostatic pressure with the increase in the filling depth of ground-
water. The corresponding safety factor decreases. When the filling
depth (h) of groundwater reaches 60 % of the tension crack thickness
at the rear edge of rock, the hydrostatic pressure (Ps) applied to the
rear edge of rockmass can be determined by the equation:
Ps h2 =2 10  32:42 =2 5248:8 kN=m
Under this situation, the FOS equaled 1.007, and the rockmass
is at the critical slide status.
Creeping stage
This stage corresponded to the evolution process that the steep
inclined joints separated from adjacent parent rock and cracked
rockmass transformed into sliding mass. A weak zone with good
property of water penetration was formed due to the long-term
Fig. 14 Diagram of heavy rainfall at
Dujiangyan from June 15 to July 19
of 2013
16 Landslides 13 & (2016)
movement of the weak bottom layers along the inclined direction.
The penetration of rain along the steep inclined cracks produced
hydrostatic pressure, which applied to the rear edge of sliding
mass and also formed uniform uplift pressure. This uplift pressure
resulted from the evident slide at the bottom weak zone. The
pressure was applied to the free face area from the rear edge to
the front edge of sliding mass, where there was a certain thickness
of gentle inclined crack zone and good water permeability
(Fig. 16b, also see Fig. 12b).
The sliding mass was simultaneously subject to the application
of rear edge hydrostatic pressure (Ps) and bottom uplift pressure
(Pu) (Fig. 18 and Table 1).
When the filling depth (h) of groundwater reached 60 % of the
rear edge tension crack thickness (H), the sliding mass was subject
to not only hydrostatic pressure of 5248.8 kN/m at rear edge but
also to uplift pressure at the bottom with a value determined
according to the following equation:
Pu hl 10  32:4  150 48; 600 kN=m
This situation corresponds to a FOS equal to 0.819 and a sliding
status. This superposition effect results in an evidently lower FOS.
Actually, under joint application of hydrostatic pressure and uplift,
if the filling depth of groundwater exceeds 20 % of the crack depth
Fig. 15 Dip-direction profile section of rockslide at Sanxicun
(H) of the rear edge, i.e., 11 m, FOS would be equal to 1.0 and
sliding would occur.
Separating stage
This stage corresponds to the evolution process of an evident slide,
gradual failure of rock, separation from parent rock, and rockslide
formation. Multiple steep inclined joints developed, continued to
cut the slide mass into multiple blocks, and resulted in separation
of the slide mass. Especially at the steep location with free faces at
the front edges, the rockmass was readily sheared to rockslide
failure at a high-level position, the front rockslide changed to fall,
topple, or rotate (Fig. 16c). Rainfall infiltrated into rockmass
through a variety of steep inclined cracks and produced multiple
sections of hydrostatic pressure. Meanwhile, the uniform uplift
pressure applied in the free face area from the rear edge to the
front edge at the bottom gentle dip laminated crushing zone
produced multiple sections of uplift pressure. Many investigators
studied the characteristics of groundwater pressure in the multiple
cracks. With the physical simulation, Xu et al. (2012) conducted a
research on the deformation and failure evolution of the transla-
tional rockslide, which was caused by the cooperation of the
hydrostatic pressure and the uplift pressure. The clear relationship
Fig. 16 Sliding pattern of cracked
layered rockmass under water
pressure
between the critical water head and the dip angle of the sliding
plane was revealed in the results: The larger the dip angle of the
slide plane was, the smaller the critical head would be under the
same physico-mechanical parameters of rock and soil in the
rockslide. Wang et al. (2012) presented a similar result that the
wedge-shaped cracking expansive force of groundwater played a
key role in the course of deformation and destruction of slope for
the near horizontal layered rock slope.
In the early stage of rock separation, the cracks in sliding mass
were not filled by groundwater. Therefore, the bottom uplift pres-
sure generally exhibited a triangle distribution, i.e., P u=hl
(Fig. 19a). With the increase in movement, the rockslide was
gradually separated, and the wedging expansion from the ground-
water in cracks correspondingly increased and expanded. This
resulted in the bottom uplift pressure changing overall and
exhibiting a composition distribution of a trapezoid and triangle
(Fig. 19b, also see Fig. 16c), i.e.
Pu h1 h2 l1 =2 hll1 
Under the condition of rockmass cracking, a special situation of
h2=h1 might occur in which the uplift pressure applied on the rear
Landslides 13 & (2016) 17
 Original Paper
Fig. 17 Seepage analysis with
hydrostatic pressure (cracking stage)
block varies from trapezoid to rectangular, and the safety factor
decreased accordingly (Table 2).
It must be noted that during sliding, separation of rockmass
structure was accelerated, and the discharge of groundwater was
accomplished more readily. Accordingly, the groundwater hydro-
static pressure at the tension cracks of the rear edge and the uplift
pressure at the bottom cracked rockmass decreased significantly.
Residual accumulating stage
This stage corresponds to the process that the decrease in ground-
water hydrostatic pressure after the shear failure of rockmass
resulted in a slow slide of material on a gentle dip surface
(Fig. 16d). In the process of a large slide, rockmass further sepa-
rated to stone blocks. The major application of groundwater
during rainfall was represented by pore water pressure. However,
in relatively large rocks, hydrostatic water pressure and uplift
pressure were still jointly applied. The resultant process of a
laminated rockslide would be an accumulation slide.
Analysis of rapid and long run-out slide process
In Southern China, it is a very typical disaster mode that a
rockslide triggered by a heavy rainfall is transformed into a rapid
and long run-out debris flow (Yin 2011). Xu et al. (2012) reported a
catastrophic avalanchedebris flow that was triggered by the heavy
rainfall on July 27, 2010 in Sichuan. A mass of strongly weathered
basalts with a volume of 480,000 m3 was initiated from a valley
side slope and then moved downstream along the valley,
entraining a large amount of unconsolidated substrate and
Table 1 Calculating results of factor of safety (FOS) with various depths of groundwater
Groundwater
Depth h (m)
1 0.0
2 10.8
3 21.6
4 32.4
a
Ratio of depth between ground (h) and crack (H)
b
Hydrostatic pressure in rear crack
c
hydrostatic pressure in rear crack+bottom uplift pressure along sliding plane
18 Landslides 13 & (2016)
bilateral materials and colluviums. The entrainment increased
the volume of slide to 1.0 million m3 and may also enhance the
mobility of the rockslide. Zhang et al. (2011) conducted a study on
the transform mechanism from rockslide to debris flow with ring
shear test that indicated the so-called soft base effect. Because of
shearing shrinkage, grain crushing, and layering, the sliding body
volume of soil reduces in the course of shearing, which makes
excess pore water pressure increase, and thus its shear strength
decreases. Simultaneously, a thick liquefied layer is formed at the
bottom of sliding body, of which thickness increases with shearing
displacement increasing. The liquefied layer carries the sliding
body moving fast and reduces its stability greatly. There is a
different slide-flow transferring process between the Sanxicun
rockslide and above cases. The Sanxicun rockslide had no
obvious liquefied layer. After sliding, the rockslide at front edge
was subject to shear failure, impacted, and then shoveled the
surface rock, soil, and vegetation. With the increase in sliding
volume, the rockslide flooded into the valley and merged into a
mountain flood to form a debris flow. Hungr (2010) proposed a
variety of rheological relations and developed the DAN-W
rockslide dynamic analysis software to simulate this process.
DAN-W is an MS Windows-based program used to model the
post-failure motion of rapid rockslides. The basic premise of the
analysis is that, as a result of sliding or other failure, a predefined
volume of soil or rock (the source volume) changes into a fluid
and runs downslope, following a path of a defined direction and
width. The mass can entrain additional material from the path and
eventually deposits, when it reaches slopes that are sufficiently flat
FOS
h/Ha State 1b State2c
0.0 1.105 1.105
0.2 1.079 1.015
0.4 1.033 0.907
0.6 1.007 0.819
Fig. 18 Seepage analysis with
hydrostatic and uplift pressure
(creeping stage)
(the deposition area). The model implements a one-dimensional
Lagrangian solution of the equations of motion and is capable of
using several alternative rheological relationships. Hungrs
method simplifies the complicated sliding structure into a kind
of equivalent fluid. The method assumes that the movement of
rockslides are a shallow flow approximately similar to a parallel
sliding bed with smaller sliding thickness. The sliding mass is
divided into a certain number of mutually contacted blocks
under curvilinear coordinates. These blocks can freely transform
but have constant volume during movement. Through the
Lagrange method, continuous equations and momentum
equations are resolved on the basis of curvilinear coordinates
and moving meshes. In the DAN method, Hungr (1995) took into
Fig. 19 Seepage analysis with hydrostatic and uplift pressure (separating stage)
account factors such as terrain limits on both sides of the sliding
approach and the internal strength of the sliding mass.
Establishment of model
The Double rheology model (F-V model for short) is based on a
friction criterion and Voellmy criterion. It is established per the
difference in ground rocks and divisions of sliding approach. In
the F-V model, the rheological relation is based on the friction
criterion for the impact-shoveling area but on the Voellmy criteri-
on for the debris flow section (Hungr 2008; Hungr and Evans
2004), of which:
a. The bottom rockmass from rockslide shear failure at a high-
level position in an impact-shoveling area is simulated using
the friction criterion. The friction criterion is a rheological
criterion with a single variable and its shear-resistant stress
can be determined through the equation:
1r u tan
where stands for shear stress at the bottom of the sliding mass,
stands for the overall stress in the vertical direction of sliding
approach, ru stands for the ratio of pore water pressure to normal
stress and is normally assumed to be constant, and stands for
friction angle. While using the friction criterion and taking into
account the fractured rockmass after earthquake, the value of
must tend to be lower than the normal value.
b. After sliding into the valley, the rockslide is transformed into a
debris flow. Therefore, the Voellmy criterion is used to simu-
late the corresponding process. The shear stress resistance of
the Voellmy criterion can be determined by the equation:
2
f g
where f stands for the friction coefficient, stands for the density
of the sliding mass, g stands for gravitational acceleration, stands
for the mean velocity of the sliding mass, and stands for
turbulivity (its value is the square of the Chzy coefficient). The f
Landslides 13 & (2016) 19
 Original Paper
Table 2 Calculating results of sactor of safety (FOS) vs. different shapes of bottom uplift pressure with various depths of groundwater
Groundwater Factor of Safety
h/Ha (%) h2 =(ll1)h1/lb h2 =h1c
1 0.0 1.079 1.079
2 0.2 1.013 1.000
3 0.4 0.969 0.938
4 0.6 0.950 0.916
a
Ratio of depth between ground (h) and crack (H)
b
Triangle-shaped distribution of uplift pressure, h2=water pressure head at l1
c
Rectangular-shaped distribution of uplift pressure, h2 at l1=h1 at rear crack

and the are the two parameters to be determined when using this it continued to move forward for 40 s and stopped 1250 m away
criterion. from the original location.
The research and analysis on the dynamic and movement The rockslide with a volume of 276,000 m3 was subject to shear
processes of rockslides are still in the early exploration stage. failure in a high-level position, moved forward 400 m, and then
Hungr and Evans (2004) indicated that the results of the calcula- shoveled down the slope surface to a depth of 2 m with a total
tions were entirely dependent upon the data provided by the user.
Therefore, persons using the program to make run-out estimates
should be geoscience professionals thoroughly familiar with
rockslides, soil and rock material behavior, and rheology, who
have studied recent research publications on rockslide dynamics,
including the relevant references listed at the end of this manual.
The properties entered into the program should always be checked
by back-analysis of real rockslide case histories, similar to the
existing or potential rockslide being studied. The results of the
analysis should never be relied on exclusively, but should be
interpreted carefully by a qualified person in the light of field
observations, empirical estimates, other analyses, and
appropriate judgment and experience. Hungr and Evans (2004)
held that its analysis via computation results was completely
dependent on the data provided. Therefore, this paper adopts a
trial-and-error method and references the research outcomes from
Hungr and McDougall (2009) and Qi et al. (2012) to comprehen-
sively investigate the movement process of the Sanxicun rockslide.
Based on this and compared with the analysis results from para-
graphs of Basic characteristics of rockslide, we propose the
rheology parameters of the F-V model as follows:

Internal friction angle of single variable for the friction crite-

rion: =20
Friction coefficient: f=0.21 and turbulivity: =200 m/s2 in the
Voellmy criterion

It is noteworthy that the friction criterion is applied, and the

internal friction angle =25 at the area where houses were
destroyed by the debris flow.

Analysis of simulation results

Run-out distance and accumulation

Figure 20 shows the movement status of the sliding mass under
different simulations by the F-V model. After a shear failure 300 m
away from original location (also see Fig. 7), the front edge of the
sliding mass took 47 s to arrive at a location 1100 m away from
original location. Due to the application of water flow in the valley, Fig. 20 Scenario of 2D depth profile of rockslide-debris run-out at Sanxicun

20 Landslides 13 & (2016)

Fig. 21 A height difference of 15 m
between the level and accumulation
of debris
volume of 296,500 m3 (Fig. 20a). When moving forward 500 m, the
rockslide continued to shovel down the slope surface to a depth of
23 m with a resultant volume of 319,800 m3 (Fig. 20b). When
moving forward 600 m, a shoveling zone with a length of 250 m
and depth of 13 m was formed, and the resultant volume of the
rockslide increased to 338,400 m3 (Fig. 20c).
The rockslide converged with the flood in the valley at 600 m to
form a debris flow and continued to move forward. When moving
forward 700 m, the mud level in the confluence was as high as 18 m
with a volume of 356,700 m3 (Fig. 20d). When it arrived at 800 m, the
mud level in confluence was as high as 22 m with a volume of
371,100 m3 (Fig. 20e). When it arrived at 8001000 m, houses were
destroyed, and the volume increased to 388,800 m3 (Fig. 20f, g).
Subsequently, the debris flow continued to move downward and
the mud height in the confluence decreased from 17 to 12 m with a
resultant volume of 398,400 m3 (Fig. 20h, k). The difference of the
debris flow between the estimated volume of 360,000 m3 from the
field observation and the calculated volume of 398,400 m3 based on
the computer simulation is 38,400 m3. It should be noted that, when
the water in debris flow was exhausted, the thickness of the solid
accumulation was generally <8 m (Fig. 21, also see Figs. 10 and 11).
In general, after shear failure in a high-level position, the
rockslide moved downward about 950 m in 47 s. When the
rockslide moved forward 600 m, a shovel zone with a length of
250 m and depth of 13 m was created. Its volume increased to
338,400 m3, a debris flow was formed by its confluence with the
flood of the valley, and the maximum height of the debris flow
reached 22 m. When the rockslide arrived 8001000 m away,
Fig. 22 Velocity of slide mass vs. distance of run-out of rockslide debris at
Sanxicun
houses were destroyed with a resultant volume of up to
398,400 m3.
Movement velocity
If the initial velocity at the shear failure of the rockslide is set to
zero, the sliding velocity kept increasing between 300 and 600 m
(Fig. 22, also see Fig. 20). This sliding velocity reached a maximum
value of 35 m/s at the location of 600 m with a movement time
t=13.6 s. After entering the valley, the rockslide was transformed
into a debris flow, and the corresponding velocity decreased to
28 m/s. At the distance of 800 m, the velocity before destroying
houses was 22 m/s with a corresponding moving time of t=21.2 s
and a time interval of t=21.213.6=7.6 s. Between the distances of
800 and 1000 m, 14 houses were destroyed, and the movement
velocity decreased to 14.2 m/s with a corresponding moving time
of t=31.3 s and a time interval of t=31.321.2=10.1 s. After that, the
rockslide continued to move forward approximately 100 m with a
corresponding moving time of t=46.7 s and a time interval of
t=46.731.3=15.4 s. Finally, the rockslide tended to stabilize at
1250 m with a corresponding moving time of t=46.7 s and a time
interval of t=83.446.7=36.7 s.
After shear failure occurred in a high-level position, the
rockslide moved downward for approximately 950 m in 47 s with
a maximum velocity of 35 m/s. This was a typical rapid and long
run-out rockslide.
Discussions
Theoretically, the prediction and early warning on debris flows
after earthquake is possible through the establishment on the
critical precipitation to initiate debris flow after the earthquake.
Quan et al. (2013) developed a model that could give a rough
spatial and temporal prediction of expected change in landslide
hazard in an area following an earthquake. The model is able to
describe the reduced impact of earthquakes with distance from
epicenter as well as how the soil gradually regains its strength with
time. Yu et al. (2014) studied the ratio of rainfall thresholds after
and before earthquakes in relation to the peak ground acceleration
(PGA) of earthquakes and the number of rainy seasons after the
Landslides 13 & (2016) 21
 Original Paper
earthquake. The Wenchuan earthquake area is a mountainous area
with height difference over several thousands of meters. The rain-
fall observation stations are very inadequate. Especially, there is
almost no station in the initiation area to debris on the upper of
the tributary. Existing researches on the initial threshold of rainfall
is merely replaced by in the deposit area of debris at the month of
the tributary. The local effect on rainfall is very obvious. The
rainfall is much different in the various zones and the various
elevations. That is why, so far, there is none of successful or
reliable case of early warning to be reported on the above theory,
ideal and method in the Wenchuan earthquake area. The method
on critical precipitation to prediction/warning is not satisfied. A
lot of fractured/cracked slopes that were resulted or intensified
from the Wenchuan Ms 8.0 earthquake have similar geological
conditions and structural characteristics as those associated with
the Sanxicun rockslide. The early detection conditions of
rockslide-debris flow disaster can be concluded as follows:
Geology and geomorphology: The upper mountain is cracked
by earthquake. A great deal of loose accumulation is in the
middle valley. The lower valley contains villages in a relative
flat terrain.
Triggering factors: During a heavy rainfall, if the upper rockfall
occurs, it would flood and erode the saturated accumulation in
the valley and trigger the flow sliding to shovel loose accumu-
lation. It can result in an increase in volume of sliding mass
and creates a rapid and long run-out rockslide in a flow state.
Hazard objects: The hazard objects include villages, roads, and
bridges located in lower and relatively flat terrain.
The visible spectral remote sensing technique for early detec-
tion of landslides is difficult in this area due to the heavy vegeta-
tion coverage. The integration of field geological investigation with
RS and InSAR is applied as an early detection tool (Zhao et al.
2013) to investigate the large-scale oblique dip apparent rockslide
in the sliding source area. An outline of the detection method can
be given as follows: (1) The tributary and cliff provide a free space
for the apparent dip movement; (2) the block-fracture bedding
structure is intercalated with weak layer as sliding plane; (3) the
rockmass is split into obvious smaller, distinct blocks with several
groups of large joints; (4) the sliding rockmass (i.e., the rockslide
structure before movement) exists along a dip angle and is barri-
caded by an inclined stable bedrock area. The subsequent sliding
direction is deflected from a true dip angle to an apparent dip
angle. (5) The macrodeformation characteristics are obvious in
and near the rockslide, such as small rockslide/fall and large
crack/collapse, etc.
These indices can serve as a scientific consideration on which
hazard regions can be classified and monitoring prevention can be
established. Therefore, the early identification of adverse geologi-
cal features in the potential landslide hazardous regions and haz-
ard prevention (or mitigation) should be established on the basis
of this disaster formulation as follows: (1) A high-resolution re-
mote sensing and detail investigation must be applied to recognize
the potential hazardous slope, (2) rainfall-based real-time moni-
toring stations are established on the source area of the rockslide-
debris flow, (3) a safe siting work on the base of long run-out risk
assessment is necessary to relocate and reconstruct, and (4) the
22 Landslides 13 & (2016)
function for land-use should be changed. The strategic plan of the
reconstruction of towns and villages in the Wenchuan Ms 8.0
earthquake area should address a limit on the scale of urban
construction and plan appropriate areas, limited areas, and for-
bidden areas for construction. In the future, 510 years or even
over a longer period, some ecological protection zones could be
established to prevent damage to properties and reduce the num-
ber of casualties.
Conclusions
This paper uses the catastrophic rockslide at Sanxicun village,
Dujianyan city as an example to investigate the characteristics of
rapid and long run-out rockslidedebris flow from the cracked
slopes intensified by the Wenchuan Ms 8.0 earthquake. This paper
analyzes the stability of the cracked slopes after the Wenchuan Ms
8.0 earthquake, especially the instability mode under the condition
of heavy rainfall. Four sliding stages of tension crack, creep sliding,
separation, and residual accumulation under the application of
hydrostatic water pressure and uplift pressure are discussed. Ki-
netic and dynamic processes of impact-shoveling area and debris
flow are simulated. This paper summarizes the early detection
signs of hazard objects and the geologic/geomorphic disaster char-
acteristics, based on the rockslide-debris flow formation mecha-
nisms at the Wenchuan Ms 8.0 earthquake area and heavy rainfall.
These indices can serve as a scientific consideration for classifying
hazard areas and monitoring for prevention of heavy rainfall type
rockslides in earthquake areas.
Acknowledgments
The authors express a gratitude to Professor Zhongqi Yue from
Hong Kong University for his time on English editing of this paper.
References
Fan XM, Xu Q, Zhang ZY (2009) The genetic mechanism of a translational rockslide. Bull
Eng Geol Environ 68:231244
Feng Z, Yin YP, Li B (2012) Centrifuge modeling of apparent dip slide from oblique thick
bedding rockslide. Chin J Rock Mech Eng 31(5):890897 (in Chinese)
GEO-SLOPE International Ltd (2008) Stability modeling with SEEP/W 2007an engi-
neering methodology, 3rd edn. Calgary, Alberta
Hungr O (1995) A model for the run out analysis of rapid flow slides, debris flows, and
avalanches. Can Geotech J 32:610623
Hungr O (2008) Simplified models of spreading flow of dry granular material. Can
Geotech J 45(8):11561168
Hungr O (2010) Users manual for DAN-W Dynamic analysis of landslides. 4195 Almondel
Rd., West Vancouver, B.C., Canada, V7V 3L6
Hungr O, Evans SG (2004) Entrainment of debris in rock avalanches: an analysis of a
long-runout mechanism. Geol Soc Am Bull 116:12401252
Hungr O, McDougall S (2009) Two numerical models for rockslide dynamic analysis.
Comput Geosci 35:978992
Ji SW, Zhang ZY, Liu HC (2000) The mechanism of deformation and failure for the slope
composed of nearly horizontal competent and incompetent intercalated rock mass
strata. Chin J Geol Hazard Control 11(9):4953 (in Chinese)
Li YQ (1998) Master Zhang Daqian and Mt. Qiangcheng. Sichuan Today 11(4):3234
Qi C, Xing AG, Yin YP (2012) Numerical simulation of Dynamic behavior of Donghekou
rockslidedebris avalanche. J Eng Geol 20(3):334339
Quan LB, Vangelsten BV, Liu ZQ, Eidsvig U and Nadim F (2013) Landslides induced by the
interaction of an earthquake and subsequent rainfall. A spatial and temporal model.
Proceedings of the 18th International Conference on Soil Mechanics and Geotechnical
Engineering, Paris, pp 22372240
Tang C, Zhu J, Li WL, Liang JT (2009) Rainfall-triggered debris flows following the
Wenchuan earthquake. Bull Eng Geol Environ 68:187194
Tang C, Zhu J, Ding J (2011) Catastrophic debris flows triggered by a 14 August 2010 Zhao X, Kang JW, Jiang J (2009) Strength index determine methods of the sliding zone
rainfall at the epicenter of the Wenchuan Ms8.0 earthquake. Landslides 8:485497 soil in rockslide investigation. J Geol Hazards Environ Preserv 20(1):4349 (in Chinese)
Wang ZJ, Wang CG, Wang K (2012) Research on mechanism of rockslide of nearly Zhao CY, Zhang Q, Yin YP, Zhong L (2013) Pre-, co-, and post-rockslide analysis with
horizontal soft rock slope caused by wedge-shape cracking expansionary force. J ALOS/PALSAR imagery data: a case study of Jiweishan rockslide, China. Nat Hazards
Railw Eng Soc No 12:3641 (in Chinese) Earth Syst Sci 13:28512861
Xu Q, Fan XM, Dong XJ (2012) Characteristics and formation mechanism of a catastrophic Zhou W, Tang C (2013) Rainfall thresholds for debris flow initiation in the Wenchuan
rainfall-induced rock avalanchemud flow in Sichuan, China. Landslides 9:143154 earthquake-stricken area, southwestern China. Landslides. doi:10.1007/s10346-013-0421-5
Yin YP (2011) Recent catastrophic landslides and mitigation in China. J Rock Mech
Geotech Eng 3(1):1018
Yin YP, Wang FW, Sun P (2009) Landslide hazards triggered by the 2008 Wenchuan Y. Yin ()) : W. Wang
Ms8.0 earthquake, Sichuan, China. Landslides 6(2):139152 China Institute of Geo-Environmental Monitoring,
Yin YP, Sun P, Zhang M, Li B (2011) Mechanism on apparent dip sliding of oblique Beijing, China
inclined bedding rockslide at Jiweishan, Chongqing, China. Landslides 8(1):4965 e-mail: yyueping@mail.cgs.gov.cn
Yu B, Wu YF, Chu SM (2014) Preliminary study on the effect of earthquake to the rainfall
threshold of debris flows. Eng Geol. doi:10.1016/j.enggeo.2014.04.007 Y. Cheng : J. Liang
Zhang M, Yin Y, Hu R, Wu S, Zhang Y (2011) Ring shear test for transform mechanism of Sichuan Geological Survey,
slide-debris flow. Eng Geol 118:5662 Chengdu, China

Landslides 13 & (2016) 23

 Original Paper
Landslides (2016) 13:2537 Xiaojun Guo I Peng Cui I Yong Li I Qiang Zou I Yingde Kong
DOI 10.1007/s10346-014-0541-6
Received: 23 December 2013
Accepted: 17 November 2014 The formation and development of debris flows in
Published online: 4 December 2014
Springer-Verlag Berlin Heidelberg 2014
large watersheds after the 2008 Wenchuan Earthquake
Abstract The Wenchuan earthquake has caused abundance of
loose materials supplies for debris flows. Many debris flows have
occurred in watersheds in area beyond 20km2, presenting charac-
teristics differing from those in small watersheds. The debris flows
yearly frequency decreases exponentially, and the average debris
flow magnitude increases linearly with watershed size. The rainfall
thresholds for debris flows in large watersheds were expressed as
I=14.7 D0.79 (2h<D<56h), which is considerably higher than those
in small watersheds as I=4.4 D0.70 (2h<D<37h). A case study is
conducted in Ergou, 39.4km2 in area, to illustrate the formation
and development processes of debris flows in large watersheds. A
debris flow develops in a large watershed only when the rainfall
was high enough to trigger the wide-spread failures and erosions
on slope and realize the confluence in the watershed. The debris
flow was supplied by the widely distributed failures dominated by
rill erosions (14 in 22 sources in this case). The intermittent
supplying increased the size and duration of debris flow. While
the landslide dam failures provided most amounts for debris flows
(57% of the total amount), and amplified the discharge suddenly.
During these processes, the debris flow velocity and density in-
creased as well. The similar processes were observed in other large
watersheds, indicating this case is representative.
Keywords Debris flows . Triggering rainfall . Formation . Large
watershed . Wenchuan earthquake
Introduction
The region severely affected by the Wenchuan Earthquake (May 12,
2008, in Sichuan Province, China) has been particularly vulnerable
to debris flows (Cui et al. 2011; Tang et al. 2012). Debris flows
occurred frequently even in watersheds larger than 20 km2, while
in other regions most debris flows occur in watersheds smaller
than 10 km2 (Li et al. 2002).
Debris flows are generally initiated in response to failure of a
discrete landslide mass on a hillslope or failure of channel bed
material (Innes 1983). Various studies have investigated the initi-
ation mechanisms and established some critical thresholds (e.g.,
Takahashi 1978; Cui 1992; Iverson 1997; Iverson et al. 1997; Huang
et al. 2009). Comparing with the case of small watershed where
debris flow originates from single or several sources, debris flow in
large watershed originates in numerous sources, which requires
more intense rainfall to trigger the slope failures and of which the
confluence depends on complex structures of the full watershed
(Cui et al. 2013). Therefore, the critical rainfall or runoff thresholds
that derived from the aforementioned models that based on indi-
vidual slope or in laboratorial channel and the previous proposed
linear debris flow discharge calculation methods (Kang and Zhang
1991; Hu et al. 2010) should be limited applied on a watershed
scale, only with full acquaintance of its formation and develop-
ment of debris flows.
To reveal the formation conditions and development processes
in large watersheds, the debris flow event in Ergou, a 39.4-km2
watershed, was taken as a case. The rainfall and materials supply-
ing conditions, the developmental processes of its velocity, dis-
charge, and density were analyzed. It is hoped that this study wll
be helpful in revealing the debris flow characteristics in large
watersheds in the Wenchuan Earthquake area.
Study area
The study area is located upstream on the Minjiang River, between
Yingxiu Town on the south and Wenchuan County on the north
(Fig. 1). The physiography is characterized by mountain ranges
trending in an approximately north-northeasterly direction which
are dissected by the Minjiang River and its tributaries. The topog-
raphy is typical of that formed by tectonic uplifting and fluvial
erosion. The channel gradients range from 5 to 30, and the slope
gradients range from 25 to 50. The geology conditions are shown
in Fig. 2.
Most of watersheds (D1D16) have a wet climate, with annual
average precipitation of 1,200 mm, and D17 and D18 are located in
relative dry climate with annual average precipitation of 800 mm.
After the earthquake, rainfall stations were established in selected
watersheds primarily for the purpose of supporting early warning
systems for debris flows. Compared with other regions in south-
western China, rainfall measurements in this region are conducted
relatively intensively (Fig. 1). Four of these stations were
established by authors, in Ergou (Fig. 1), from source region to
the outlet (Fig. 3).
Ergou, with an area of 39.4 km2, is a tributary of Minjiang River,
and located about 6 km from the epicenter (Fig. 1). The slope
gradient averages 40. Elevations range from 930 to 4120 m. The
gully length is about 12 km, and the mean channel gradient is
about 12, and ranges from 5 to 12 in the debris flow transporta-
tion region. Landslide deposits, with depth between 1 and 30 m, are
spatially distributed over about 24 % of the watershed, amounting
to 1.0108 m3. Debris flows occurred in low-order gullies in Ergou
before, according to the field investigation, but only the one on 10
July 2013 flowed out and was recorded at the outlet.
Debris flows in watersheds with different size
Debris flow activities
Debris flows behaviors vary in watersheds with different size. We
collected the information of landslide area ratio in a watershed,
slope gradient, and debris flow numbers in Ergou and its neigh-
boring watersheds (Table 1). Here, the landslide area ratio, the
landslide area divided by the total watershed area, is defined to
represent the materials supplying level, which is negatively pro-
portional to the area (P=0.410.0007A). Debris flow yearly fre-
quency ranges from 0.17 to 2.67, which decreases exponentially
with watershed size (Fig. 4a), while the average magnitude of
debris flow increases linearly with watershed size (Fig. 4b).
Area, landslide area, landslide proportion, and channel gradi-
ent were obtained from the topography map and aerial photos; the
Landslides 13 & (2016) 25
 Original Paper
Fig. 1 Debris flow and rainfall gauge distributions and the geomorphology map in the study area
landslide proportion is calculated as dividing the area of landslides
by the total watershed area. Debris flow times were obtained by
field investigation. The debris flow yearly frequency was deter-
mined as dividing debris flow times by the number of years
(generally 6 years from 2008 to 2013, with exception of watershed
no. 13 with debris flow prevention engineering since 2012). The
average debris flow magnitude was obtained by field investiga-
tions. Debris flow only occurred on 10 July 2013 in the watersheds
with asterisks.
The larger landslide ratio and steeper channel slope in small
watersheds or slope gullies lead the more frequent debris flow
activities. Confluence processes, indicating the control effort
of a fluvial system, significantly influence debris flow activities
in larger watersheds. Debris flows in lower-order valleys did
not necessarily move downward to the outlet. On the other
hand, due to the wide-spatially distributed landsliding and
loose materials accumulation by sediment supply from low-
26 Landslides 13 & (2016)
order valleys, the magnitudes in large watersheds are gener-
ally tremendous.
It is noted that 10 out of the recorded 18 gullies have an area bigger
than 10 km2, which is contrary to the usual cases that less than 30 %
debris flow occur in gullies bigger than 10 km2 according to statistics
in the southwest China (Li et al. 2002), showing the debris flow
susceptibility in large area. Also, debris flow yearly frequency ranges
12.67 and 0.171 in areas in watersheds with area smaller and larger
than 20 km2, respectively. Therefore, we take 20 km2 as the boundary
of large watershed to analyze the rainfall thresholds.
Rainfall conditions for debris flows
As shown in Table 2, the mean minimum triggering rainfall inten-
sity and inducing rainfall for debris flows are 14.8 and 43.5 mm/h.
The minimum triggering rainfall intensity and inducing rainfall
for debris flows is as low as 3.7 mm/h and 5.3 mm, on 17 May, 2008,
quite short following the earthquake. Maximum triggering rainfall
intensity and inducing rainfall for debris flows is 38.9 mm/h and
140.8 mm. The mean and minimum rainfall condition required for
debris flows are significantly lower than the 3050 mm/h intensity
and 80100 mm/day daily rainfall required before the earthquake
(Tan and Han 1992). Major debris flow events group-occurring in
both slope gullies and large watersheds were recorded on rain-
storms such as 24 Sep 2008, 13 Aug 2010, 3 July 2011, and 18 Aug
2012. These events destroyed roads and vehicles, blocked rivers,
dammed lakes, and submerged villages, resulting in significant
loss of life and property damage (Guo et al. 2012; Su et al. 2012).
Debris flows also occurred during some small rainfalls on single
slopes or source slope gullies, for instance, rainfall on 16 July 2009,
22 Aug 2009, and so on.
We studied 22 debris flow events during the period from
May 2008 to 2011. Only 15 events occurred in watersheds with
areas larger than 20 km2. The methodology for identifying
rainfall intensity (I) and duration (D) thresholds follows those
of previous studies for areas with widely different climates
and geological settings (Caine 1980; Guzzetti et al. 2007, 2008;
Cannon et al. 2008; Saito et al. 2010). These methodologies
have been used to represent the critical rainfall thresholds for
triggering debris flows in these regions. Based on rainfall data
collected, we identified the rainfall duration (D) and intensity
(I) necessary for debris flow formation. The rainfall I-D pat-
tern based on available rainfall records is presented in Fig. 5
according to the following formulas:

Thresholds for watersheds with areas larger than 20 km2 : I 14:7 D0:79 2 h < D < 56 h 1

Thresholds for watersheds with areas smaller than 20 km2 : I 4:4 D0:70 2 h < D < 37 h 2

This suggests that debris flows on slopes were induced by is quite lower than ID thresholds in other regions in Wenchuan
rainfall with an average intensity of 4.4 mm/h, while for large earthquake hit area (Tang et al. 2012; Zhou and Tang 2013), but it is
watersheds the average intensity was greater than 14.7 mm/h. This reasonable. as for shortly after the earthquake, especially in May

Table 1 Basic characteristics and debris flow behaviors of 18 watersheds in Ergou and neighboring watersheds
No. Area Landslide Landslide Channel DF DF yearly Average magnitude
(km2) area ratio gradient () times frequency (103 m3)
1 12.7 2.03 0.16 15 7 1.17 100
2 2.34 0.78 0.33 23 8 1.33 50
3 0.46 0.22 0.48 32 14 2.33 4
4 2.67 1.26 0.47 31 12 2.00 20
5 2.5 0.98 0.39 28 16 2.67 5
6 0.26 0.11 0.42 30 12 2.00 4
a
7 39.4 8.34 0.21 15 1 0.17 500
8 27.31 5.11 0.19 14 2 0.33
9 7.46 3.59 0.48 27 10 1.67 200
10 2.94 1.16 0.39 29 8 1.33
11 24.26 8.76 0.36 17 5 0.83 300
a
12 28.65 4.18 0.15 14 1 0.17 200
13 16.97 3.82 0.23 17 7 1.40 100
14 32.62 4.55 0.14 14 2 0.33 200
15a 49.9 5.33 0.11 12 1 0.17 500
16 0.225 0.08 0.36 31 12 2.00 4
a
17 21.4 5.76 0.27 15 1 0.17 500
18a 52.4 7.44 0.14 10 1 0.17 800
Area, landslide area, landslide proportion, and channel gradient were obtained from the topography map and aerial photos; the landslide proportion is calculated as dividing the area
of landslides by the total watershed area. Debris flow times were obtained by field investigation. The debris flow yearly frequency was determined as dividing debris flow times by
the number of years (generally 6 years from 2008 to 2013, with exception of watershed no. 13 with debris flow prevention engineering since 2012). The average debris flow
magnitude was obtained by field investigations
a
Debris flow only occurred on 10 July 2013 in the watersheds

Landslides 13 & (2016) 27

 Original Paper
2008, the loose materials accumulated on the unstable slopes as
the critical condition, therefore debris flows can be initiated by
rainfall under 10 mm (on 17 May 2008) .
To illustrate the triggering conditions, materials supplying and
debris flow formation, development processes, and its damage in
large watersheds, we take debris flows in Ergou Gully (Fig. 3) on 10
July 2013 as a case. Considering that there was debris flow prevention
engineering prior the debris flows in D15, and the climate conditions
of D18 is different from most of the debris flows in this region as
shown in Fig. 1, although Ergou is not the largest one in area (smaller
than D15 and D18), it was the most representative one.

Debris flow in Ergou on 10 July 2013

Methodologies applied in the field investigation

Four rainfall gauges were established to obtain rainfall informa-
tion related to debris flows.
Field reconnaissance and measurements using hand-held GPS
and laser Rangefinders, were conducted in order to (1) investigate
the sediment supply of the debris flows; (2) evaluate the formation
and development processes of the debris flows; (3) determine the
volume of the debris flows. Natural color aerial photographs
(spatial resolution=0.6 m, acquisition date: 17 April 2013; and
spatial resolution=0.5 m, acquisition date: 21 July 2013) were also
used to identify the potential distribution of landslides, and to
evaluate the debris flow run-out area.

1. Investigation of the sediment supply

Fig. 4 a The regression relation between average debris flow yearly frequency
post-earthquake and the watershed areas. b The regression relation between
average debris flow magnitude post-earthquake and the watershed areas
Field reconnaissance before and shortly following the debris
flow had been carried out in May 2013 and July 2013 to
determine the location of seismic landslides in this watershed
and to analyze the sediment supply. The measurement results
and photographs were compared to evaluate the supplying vol-
ume for debris flow on 10 July 2013.
2. Evaluation of debris flow properties
Five sections in the transportation zones were assessed along
the channel (Fig. 3), and their geometries and the gradients of
neighboring channel beds were measured. Debris flow velocity
was calculated by the Manning formula, which is widely used

Table 2 Responsible rainfall information for some debris flows

Date R (mm) I (mm/h) Date R (mm) I (mm/h)
a
13 May 2008 47.3 9.1 22 August 2009 25.5 5.1
17 May 2008 5.3 3.7 26 August 2009 39.6 19.4
6 August 2008 40 15 14 September 2009 46.6 15.4
a
11 August 2008 24.7 7.2 1314 August 2010 220 32.2
2224 September 2008a 94.7 10.3 1819 August 2010a 63.6 21
8 July 2009 29.7 9.3 13 July 2011a 144.6 34.5
1617 July 2009 21.4 4.2 20 August 2011 91.4 31
a
22 July 2009 36.5 9.6 1718 August 2012 158.4 38.9
25 July 2009 27.9 10.1
R event rainfall volume (in millimeter), I triggering rainfall intensity (in millimeter per hour)
a
Group debris flows occurred in this rainfall event (debris flows occurred in majority of watersheds in this region); and debris flows occurred in small number of watersheds in other
events

28 Landslides 13 & (2016)


Fig. 5 Rainfall I-D thresholds for triggering debris flows in watersheds with
different areas
and proved to be reasonable in Western China (Kang 1987; Cui
et al. 2013), and discharge was calculated by debris flow velocity
and the responding section area, according to field investigation.
Debris flow samples were collected from three sections in chan-
nel and the depositional fan (Fig. 3). On each section, samples were
collected from several points. The particle-size distribution was
determined following the conventional approach (Godt and Coe
2007), that is by visually estimating the relative percentage of clasts
and matrix in the field and from photographs and by mechanical
sieving and hydrometer tests of matrix samples in the laboratory.
3. Determination of the debris flow volume
Debris flow deposition depth was measured in situ and combin-
ing the deposit length and lateral width measured from aerial pho-
tographs, the deposition volumes on the fans were calculated.
Rainfall triggering debris flows
Field survey results indicated that the debris flow in Ergou oc-
curred at 3:00 am on 10 July 2013. Observations from the first
rainfall station (Fig. 3) indicated that the antecedent rainfall vol-
ume from July 1 through July 4 was 47 mm, and there was no
rainfall from July 5 to 7. The rainfall event responsible for the
debris flow began on July 8 and ended on July 11, with a total
rainfall volume of 328.5 mm, most concentrated on July 9 and 10.
The volume of rainfall, directly inducing the debris flow, lasted
from July 8 to the time of occurrence of the debris flow. It was
measured as 191.5 mm, and the triggering rainfall intensity was
26.5 mm/h (Fig. 6a). Measurements from the other three gauges in
the watershed and the nearby Yinxing station (2 km from the
outlet of Ergou) were comparable, with triggering rainfall intensity
ranging from 19.0 to 23.0 mm/h, and triggering rainfall amounts
from 188.0 to 210.5 mm (Fig. 6b). While the rainfall information
from Yingxiu station, which is 6 km from the outlet of Ergou was
even larger, with 292.2 mm triggering rainfall and 22.4 mm/h
triggering rainfall intensity (Fig. 6c).
Notably in Fig. 6a, the rainfall intensity at 4:00 am on July 9,
which at 22 mm/h approached the triggering rainfall intensity at
3:00 am on July 10, did not trigger debris flows. This demonstrates
the significance of accumulated rainfall on debris flow initiation in
Ergou.
Material supplies
Field reconnaissance after the debris flow revealed three processes
of sediment initiation and supplies: (1) soil failures along rills on
slopes, (2) landslide dam failures along the stream channel, and (3)
bed erosion of loose materials in the channels. Twenty-two source
areas from the outlet to the junction A were identified as the
supply sources and their geometries were measured (Fig. 3). The
volume of a slope failure and/or slope rill is typically calculated by
multiplying the failure area by an estimated average depth (Martin
et al. 2002), and their topographic parameters were listed in
Table 3.
1. Slope debris flows initiated by rills
A small rill on slope can form sizeable debris flows in a short
period and this phenomena was verified through field investigations.
For an individual rill, the length, average width of the erosion region
was measured and the average depth was estimated by comparing
two investigation results before and after the debris flow. Then the
volume and its residues on slope were calculated as shown in Fig. 7.
The volume of debris flow fans remained along the flanks of the
channel was also measured as the potential supply for future debris
flows.
Rills on slopes ranged from 10 to 20 m in width and 3 to 6 m in depth,
and were observed along the flanks of the channel (Fig. 3). Fourteen of the
22 source regions investigated as suppliers of materials for debris flows
formed in rills, with magnitudes ranging from 1.4103 m3 to 24103 m3,
with a total supply volume of about 130103 m3. However, a volume of
more than 1,500103 m3 of material was deposited in the form of residues
on slope and debris flow fans along the river channel, indicating the
likelihood of future debris flows.
2. Failure of landslide dams along the river channel
Failure occurred from the bottom to the top of landslide dams due
to the unconsolidated nature of the sediments, as a result of the
dredging and erosion effect of flood and/or debris flows. Similarly
with measurement for a slope rill, the height and average width of
the failure region were measured and the average depth were esti-
mated, then the volume and its residues were calculated as shown in
Fig. 8.
Six landslide dams failed from the outlet to the junction A
(Fig. 3). The volume of materials supplied for these debris flows
ranged from 10103 m3 to 72103 m3. The largest landslide deposit,
lying about 250 m upstream of the deposit area with dimensions of
150 m (width) and 40 m (height), failed at a depth of about 12 m
(Fig. 8). Six deposits provided materials with an approximate vol-
ume of 180103 m3. This represents 20 % of the total volume of
landslide dams. In other words, a volume of more than 1,000103 m3
of loose materials remains on the slopes and may supply future
debris flows.
3. Bed erosion of loose materials in channels
Observations revealed that one of the eroded channel sections has
an approximate length and depth of 100 and 3.5 m, respectively, and
it was widened by 20 m.
Two branch gullies formed deposits with volumes of 100103 m3
and 300103 m3 and are present as debris flow fans. In actuality,
debris flows in branch gullies formed by all of the three aforemen-
tioned modes.
Landslides 13 & (2016) 29
 Original Paper
Fig. 6 a Rainfall during the period
July 8July 11 measured by rainfall
station 1 in Ergou. b Rainfall during
the period July 8July 11measured
by rainfall station 3 in Ergou. c
Rainfall during the period July 8July
11 measured by Yingxiu rainfall
station

30 Landslides 13 & (2016)

Table 3 Supply modes and their parameters
No. Supply mode Failure width (m) Failure height/length (m) Average failure Amount Deposit residue
depth (m) (103 m3) (103 m3)
1 2 150 40 12 72 300
2 1 20 150 6 18 100
3 2 60 80 5 24 100
4 1 6 80 3 1.44 10
5 1 12 110 4 5.28 80
6 1 10 100 4 4 10
7 2 50 80 8 32 160
8 1 10 100 4 4 10
9 1 20 200 4 16 200
10 1 8 80 4 2.56 15
11 2 50 40 5 10 150
12 1 6 80 4 1.92 10
13 1 20 100 4 8 20
14 2 40 50 5 10 50
15 1 12 100 4 4.8 100
16 2 100 80 4 32 300
17 1 10 100 4 4 250
18 1 20 200 6 24 400
19 1 20 150 8 24 200
20 1 18 120 4 8.64 100
21 3 20 100 3.5 7
22 3 18 90 2 3.24
The supply modes (13) are as follows: 1 Slope debris flows formed by rill erosion; 2 failure of landslide dams along the river channel; 3 bed erosion of loose materials in the channel.
The failure heights correspond primarily to debris flow dam failures, while failure lengths are for rills formed on loose deposits and/or slopes. Deposit residue refers to the amount of
deposit that did not supply the current debris flow but may potentially supply a future one

evaluation of Mannings roughness coefficient for debris flows, while

Development of debris flow
1. Flow velocity and evolution of material discharge
The debris flow was initially dominated by fast moving surges of
materials, after which it was dominated by flooding that washed
away some of the deposits into the Minjiang River. Debris flow
deposit residues were detected at junction A. Based on the section
dimensions measured and the gradient of neighboring channel beds
(Fig. 3), the velocity V is calculated by the Manning formula:
 
V H 2=3 J 1=2 =Rn
Then discharge Q was calculated by debris flow velocity and the
responding section area, according to field investigation:
Q AU
where H is the debris-flow depth in meter; J is the channel bed
gradient, and Rn is the roughness coefficient for viscous debris
flows, and A is the cross-sectional area in square meter. The
critical to calculate the velocity, is very complex as it includes internal
resistances owing to solids collision and contact friction and exterior
resistances triggered by the rough boundaries. Therefore, it is approxi-
mately evaluated empirically (Xu and Feng 1979). The debris flow was
viscous and flowed down a meandering channel with large stones, sand
and gravel present and forming obstacles, therefore the roughness coeffi-
cient for all the debris flows in Ergou was estimated to be 0.1 (Xu and Feng
1979). Note this method is also highly recommended by Chien and Wan
(1999) and used by Cui et al. (2013).
3 Although the debris flow sediments were found on the channel
where it widens, on the whole, as indicated in Fig. 9, the velocity of
the debris flow increased from 7.2 to 10.1 m/s over a long transpor-
tation distance. The phase ABCD represented the gradual scale
increase of the debris flow from about 400 m3/s to about 500 m3/s,
4 owing to continuous supply of material. In addition, channel bed
degradation with armoring was observed, with an estimated erosion
depth of 34 m between sections C and D. With particular regards to
sections D and E, the paroxysmal variation of debris flow discharge
is attributed to the cascading failure of landslide dam, as shown in

Landslides 13 & (2016) 31

 Original Paper

Fig. 7 Appearance of a debris flow formed by slope rill, and schematic diagram of rill erosion measurement. Photo (a) is a slope rill, and photo (b) is a close shot of photo
(a). In the figures, W1, W2, and W3 are erosion width on different sections, D1, D2, and D3 are erosion depth on different sections. Combining rill length, the erosion
volume can be calculated. W1, W2, and W3 are width on different sections of the loose slope; D1, D2 are depth on different sections of the loose slope. Combining
the slope lengths, the potential volume of loose materials for future erosion is calculated

Fig. 8. Ultimately, the peak discharge near the deposition region was
710 m3/s.
2. Constituents of debris flow deposits and variations in densities
In all cases, laboratory analysis revealed that the debris flow deposits
were dominated by sand and gravel-sized particles and the percentage of
fine-grained material (smaller than 0.075 mm) ranged from 5.7 to 7.3 %.
The particle size distributions are indicated in Fig. 10.
The cumulative particle-size curve P(D) (percentage of particles
with diameters exceeding D) was plotted using the following distri-
bution (Li et al. 2013):
PD CD expD=Dc
Where, P (D) is the percentage of particles with diameters greater
than D (in millimeter); the power exponent , can be associated with
and estimated by the criterion of shear flow; D c defines a
characteristic size for the total aggregation of particles that can be
linked to the flow density by Eq. (6) (Li et al. 2013):
Dc 0:04718:1446 6
For these samples, the parameters C, , and Dc ranged from 75.15
to 83.56, 0.047 to 0.073, and 5.93 to 22.2, respectively (Table 4).
The overall increase in Dc during debris flow transport is evident,
suggesting that greater proportions of coarse particles were assimi-
lated into the matrix and the kinetic energy and transportation
capacity increased. Meanwhile, debris flow density increased as well
5 because of more solid materials supplied along the channel. The
value of more than 2.0 g/cm3 showed the debris flow was evidently
viscous. On the debris flow fan, Dc and were smaller because of the
lower velocity and due to the increased amounts of coarse particles
in subsequent floods.

32 Landslides 13 & (2016)

Fig. 8 Appearance of the largest landslide dam before and after the debris flow, and schematic diagram of slope failure measurement

Deposit characteristics
Five sections can be identified in the deposit according to the
topography (Fig. 11). The deposit depths, varying with channel
width, slope gradient, and building distribution, were measured
by marks left on constructions (e.g., Fig. 12). Section A, at the tail
of the depositional region, is about 200 m long with depths of
about 5 m, while sections B and C are relatively wide and with
depths of about 510 m. These deposits are present in the key area
of destruction. Section D is a narrow region and is shaped like a

Fig. 9 Evolution of the debris flow along Ergou. In the table, H represents the height of debris flow, according to the residues or marks visually (in meter); I is the slope
gradient; D is the width of the measured section (in meter)

Landslides 13 & (2016) 33

 Original Paper

Fig. 10 Debris flow particle size distribution in Ergou. The figure on the top left corner is the fitted P(D) curve according to Li et al. (2013)

channel; therefore it was relatively shallow (4 m). Section E is at
the head of the region as the debris flow fan with a deposit depth
of about 34 m. The deposit parameters are listed in the table in
Fig. 11. The total volume of residue in the channel and debris flow
fan was estimated to be 500103 m3. However the debris flow
lasted for 4 h, exhibited typical surge characteristics, and
transported a great volume of debris flow material into the
Minjiang River. Therefore, the total volume of this debris flow
event was significantly greater than 500103 m3.
Figure 12 Buildings indicating deposit height. On the left, photo
(a) (taken in May 2013) was Ergou power station prior in section C to
the debris flow, and photo (b) (taken in July 2013) was the same
building after the debris flow. H1, H1 and H2, H2 represent the
vertical distances from the same white object in red circles and the
same line to the ground before and after the debris flow, indicating
the depth of debris flow deposit was about 5.5 m. The red line in
photo (c) shows debris flow deposit on against another building in
Section D, indicating the derbsi flow deposit height was about 4 m.
Discussion
Debris flow formation and development features in large watersheds
The debris flow in Ergou is typical for debris flow in large water-
sheds. Similar processes are observed in other large watersheds
(D11, D15, D17, and D18 in Fig. 1, the debris flow magnitude are also
listed in Table 1) (Yan et al. 2014; Zeng et al. 2014), indicating that
debris flows in a large watershed are potentially initiated and
supplied by various processes than small gully or single slope,
such as new landslides and collapses, slope rill erosion, landslide
dam failures, and bed erosion of loose materials in channels. The
processes are common:
1. Material supplies from diverse slope processes:
The failures and rill erosions on loose, coarse-grained colluviums
were widely distributed as the most common supplying mode for
debris flows (14 of 22 in this study), and their intermittent supplying
lead the flood to behave like surges and lasting for hours, which is
also different from the common debris flows on slopes.
2. Sediment assimilation:
The mobilization of granular material allows the gradual transformation
of a flood into a debris flow and enlarges the discharge of debris flows,
velocity of the flow body also possibly decrease according to the local
slope gradient, leading the debris flow discharge vary along the channel,
which is also verified by previous studies (Coe et al. 1997; Godt and Coe
2007; Cui et al. 2013).
3. Landslide dam failures:
In addition, the corresponding sediments of certain landslides
and/or collapses are very common in drainages in this region due

Table 4 GSD parameters and debris flow density for the debris flow in Ergou
Collected section no. Coefficient Power index Characteristic size R2 Debris flow density
C Dc (mm) (g/cm3)
S1 75.15 0.060 5.93 0.9923 1.81
S2 72.06 0.073 16.6 0.9827 2.05
S3 80.16 0.047 22.2 0.9917 2.13
Debris flow fan (S4) 83.56 0.037 13.7 0.9978 2.01

34 Landslides 13 & (2016)

Fig. 11 The debris flow deposition
area and the geometry parameters of
deposit sections. The debris flow
depth means the average depths in
sections

Fig. 12 Some local buildings indicate the debris flow deposit height. On the left, photo (a) (taken in May 2013) was Ergou station prior in section C to the debris flow,
and photo (b) (taken in July 2013) was the same building after the debris flow. H1, H1 and H2, H2 represent the vertical distances from the same white object in red
circles and the same line to the ground before and after the debris flow, indicating the depth of debris flow deposit was about 5.5 m. The red line in photo (c) shows
debris flow deposit on against another building in section D, indicating the debris flow deposit height was about 4 m

Landslides 13 & (2016) 35

 Original Paper
to the effect of earthquake, and most of them partially blocked the
channel. The failures of these dams, supplied debris flows by great
proportion (57 % in this case), and increasing the velocity and
discharge of debris flows suddenly, ultimately leading to the giant
magnitude. This is very common in Wenchuan earthquake hit area
because great amount of co-seismic landslides deposit exists along
the channel, and very likely to fail from the toe during the strong
storm.
Critical rainfall for debris flows in large watersheds
Even though the rainfall exceeds the threshold as shown as Eq. 2,
some but limited source materials supplied debris flows in form of
these kinds of processes, the debris flows can hardly form as a
watershed behavior. With respect to this, the critical rainfall or
runoff thresholds that derived from laboratory experiments, which
were aiming at slope failures at some specific location, can hardly
be used for forecasting debris flows in large watersheds. Experi-
ments show that the magnitude of supplying materials exponen-
tially increases with the rainfall intensity (Guo et al. 2013),
indicating only under certain rainfall conditions, such as the
rainstorm on 10 July 2013, do these processes occur together,
resulting in the formation of large debris flows.
During the period of 20082013, in 22 recorded debris flows
events, 15 of them happened in some large watersheds, while, only
7 among them (including the event on 10 July 2013) happened in
most of large watersheds. Eliminating the one that happened on 13
May 2008, shortly after the Wenchuan earthquake, the event rain-
fall thresholds for grouping-debris flows occurrence can be
regarded as 64 mm, as shown in Table 2. For Ergou, the inducing
rainfall volume on 10 July 2013 was significantly greater than any
recorded following the earthquake, indicating that the rainfall
threshold for flood-like debris flows in this watershed is greater
than 160 mm (Table 2), but less than 190 mm.
It is also worth mentioning that on one hand, the development
of debris flows in big watershed with complex drainage topogra-
phy can enlarge the debris flow magnitude, and the density, and
transportation capacity as well. On the other hand, deposition
areas for large watersheds tend to be relatively wide, which is a
significant factor to consider in mountainous areas in southwest-
ern China. Buildings and infrastructure are densely distributed
across these outlets and adjoining areas which increases the po-
tential for catastrophic damage from debris flows. For example,
the major factories were destroyed in D7 (Ergou) and D15, and
most residences were ruined in D17 and D18 (Fig. 1), in this rainfall
event. The characteristics of rainfall-triggered debris flows de-
scribed in this study can be used as indicators for future debris
flows in large watersheds following rainstorms.
Conclusion
Because of the widely distributed loose materials induced by the
Wenchuan Earthquake, debris flows occurred frequently in not
only slopes but also in large watersheds. The rainfall thresholds
and debris flow behaviors such as debris flow frequencies and
magnitudes are different from those on small slopes in lower-
order valleys, The latter are more easily to be triggered as
I=4.4D0.70 (2 h<D <37 h) and the former as I=14.7D0.79 (2 h<D
<56 h). The debris flow frequency followed an exponentially de-
creasing tendency with watershed area as F=2.05 exp (0.051A),
36 Landslides 13 & (2016)
and the average magnitudes linearly increased with watershed area
as M=11.7A13.6.
Study of a debris flow in Ergou shows that the triggering rainfall
threshold is between 160 and 190 mm. Slope failures along rills due
to overland flow erosion, failures of landslide dams, and channel-
bed erosion in gullies are the main processes responsible for
instigating and supplying debris flows. Among these processes, rill
erosion was the most common type, while landslide dam failures
contributed the largest volume of material. Velocity accelerated
from 7.2 m3/s to more than 10 m3/s over the long transportation
distance, and discharge volumes were gradually supplied by
channel-bed erosion and slope rill erosion along the channel.
Landslide dam failure suddenly amplified the discharge from 504
to 710 m3/s, and increasing the resulting damage. The formation
and development processes described in this watershed can be
used as indicators for future debris flows in large watersheds
following rainstorms.
Acknowledgments
The authors acknowledge the Meteorological Bureau of Sichuan
Province, for providing the rainfall data. This study was supported
by the National Natural Science Foundation (grant no. 41301008),
the Key Deployment Project of Chinese Academy of Sciences
(grant no. KZZD-EW-05-01) and the National Key Technologies
R&D Program of China (grant no. 2012BAK10B04). We also ac-
knowledge Dr. William B. Mahoney for improving the English.
References
Caine N (1980) The rainfall intensityduration control of shallow landslides and debris
flows. Geografiska Annaler Series A Phys Geogr 62:2327
Cannon S, Gartner J, Wilson R, Bowers J, Laber J (2008) Storm rainfall conditions for
floods and debris flows from recently burned areas in southwestern Colorado and
southern California. Geomorphology 96:250269
Chien N, Wan ZH (1999) Mechanics of sediment transport. ASCE Press, Reston
Coe JA, Glancy PA, Whitney JW (1997) Volumetric analysis and hydrologic characteriza-
tion of a modern debris flow near Yucca Mountain Nevada. Geomorphology 20:1128
Cui P (1992) Studies on condition and mechanism of debris flow initiation by means of
experiment. Chin Sci Bull 37(9):759763
Cui P, Chen XQ, Zhu YY, Su FH, Wei FQ, Han YS, Liu HJ, Zhuang JQ (2011) The Wenchuan
Earthquake (May 12, 2008), Sichuan Province, China, and resulting geohazards. Nat
Hazards 56(1):1936
Cui P, Zhou GGD, Zhu XH, Zhang JQ (2013) Scale amplification of natural debris flows
caused by cascading landslide dam failures. Geomorphology 182:173189
Godt JW, Coe JA (2007) Alpine debris flows triggered by a 28 July 1999 thunderstorm in
the central Front Range, Colorado. Geomorphology 84:8097
Guo XJ, Xiang LZ, Zhou XJ, Yang W (2012) Research on the debris flow hazards in Gaojia
Gully and Shenxi Gully in 2011. J Catastrophology 27(3):8185 (in Chinese)
Guo XJ, Li Y, Cui P (2013) Exponential amplification effect of rainfall on slope failures in
debris flow source region. J Mt Sci 31(4):406412
Guzzetti F, Peruccacci S, Rossi M, Stark C (2007) Rainfall thresholds for the initiation of
landslides in central and southern Europe. Meteorog Atmos Phys 98:239267
Guzzetti F, Peruccacci S, Rossi M, Stark C (2008) The rainfall intensityduration control of
shallow landslides and debris flows: an update. Landslides 5:317
Hu KH, Cui P, You Y, Zhuang JQ, Chen XQ (2010) Nonlinear modified flood method for
calculating the debris-flow peak discharge in Wenchuan Earthquake region. J Sichuan
Univ (Eng Sci Ed) 42(5):5256 (in Chinese)
Huang CC, Ju YJ, Hwu LK, Lee JL (2009) Internal soil moisture and piezometric responses
to rainfall-induced shallow slope failures. J Hydrol 370:3951
Innes JL (1983) Debris flows. Prog Phys Geogr 7:469501
Iverson RM (1997) The physics of debris flows. Rev Geophys 35(3):245296
Iverson RM, Reid ME, LaHusen RG (1997) Debris-flow mobilization from landslides. Annu
Rev Earth Planet Sci 25:85138
Kang ZC (1987) A velocity research of debris flow and its calculating method in China. Mt Yan Y, Ge YG, Zhang JQ, Zeng C (2014) Research on the debris flow hazards in Cutou
Res 5(4):247259 (in Chinese) Gully, Wenchuan County on July 10, 2013. J Catastrophology 29(3):229234 (in
Kang ZC, Zhang J (1991) Study and calculation of debris-flow peak discharge. Soil Water Chinese)
Conserv China 2:1518 (in Chinese) Zeng C, Cui P, Ge YG, Zhang JQ, Lei Y, Yan Y (2014) Characteristics and mechanism of
Li Y, Hu KH, Cui P, Chen XQ, Yue ZQ (2002) Morphology of basin of debris flow. J Mt Sci buildings damaged by debris flows on 11 July, 2013 in Qipangou of Wenchuan,
20(1):111 Sichuan. J Earth Sci Environ 36(2):8191 (in Chinese)
Li Y, Zhou XJ, Su PC, Kong YD, Liu JJ (2013) A scaling distribution for grain composition Zhou W, Tang C (2013) Rainfall thresholds for debris flow initiation in the Wenchuan
of debris flow. Geomorphology 192:3042 earthquake-stricken area, southwestern China. Landslides. doi:10.1007/s10346-013-
Martin Y, Rood K, Schwab JW, Church M (2002) Sediment transfer by shallow landsliding 042
in the Queen Charlotte Islands, British Columbia. Can J Earth Sci 39:189205
Saito S, Daichi N, Hiroshi M (2010) Relationship between the initiation of a shallow
X. Guo : P. Cui ()) : Y. Li : Q. Zou : Y. Kong
landslide and rainfall intensity-duration thresholds in Japan. Geomorphology
118:167175
Su PC, Wei FQ, Cheng ZL (2012) Debris flow activity of Mozi Gully after Wenchuan Key Laboratory of Mountain Surface Process and Hazards/Institute of Mountain Hazards
Earthquake on May 12. J Yangtze River Sci Res Inst 29(2):1622 (in Chinese) and Environment,
Takahashi T (1978) Mechanical characteristics of debris flow. J Hydraul Div ASCE Chinese Academy of Sciences,
104(8):11531169 Chengdu, 610041, China
Tan WP, Han QY (1992) Study on regional critical rainfall indices of debris flow in Sichuan e-mail: pengcui@imde.ac.cn
Province. J Catastrophology 7(2):3742 (in Chinese with English abstract) X. Guo : P. Cui : Y. Li
Tang C, Van Asch TWJ, Chang M, Chen GQ, Zhao XH, Huang XC (2012) Catastrophic Asian Network on Debris Flow,
debris flows on 13 August 2010 in the Qingping area, southwestern China: the Chengdu, 610041, China
combined effects of a strong earthquake and subsequent rainstorms. Geomorphology
139140:559576 X. Guo : Q. Zou : Y. Kong
Xu MD, Feng QH (1979) Roughness of debris flows. Proceeding of the First Conference of Graduate University of Chinese Academy of Sciences,
Chinese Research of Debris Flows, pp 5152 (in Chinese) 100049, Beijing, China

Landslides 13 & (2016) 37

 Original Paper
Landslides (2016) 13:3954 Jia-wen Zhou I Peng Cui I Ming-hui Hao
DOI 10.1007/s10346-014-0553-2
Received: 22 August 2014
Accepted: 30 December 2014 Comprehensive analyses of the initiation and entrainment
Published online: 14 January 2015
Springer-Verlag Berlin Heidelberg 2015
processes of the 2000 Yigong catastrophic landslide in Tibet,
China
Abstract The 2000 Yigong landslide was one of the most cata-
strophic landslides worldwide, resulting in huge casualties and
property losses. The dynamic process of the Yigong landslide
was very complicated, especially for the initiation and entrainment
mechanism during the landslide movement process. The topogra-
phy, geological condition, traces left by the landslide, and distri-
bution characteristics of the landslide deposits were determined by
field investigations, combined with several years of monitoring the
temperature and rainfall data in this region. The initiation mech-
anism of the Yigong landslide is presented. The main reasons for
the landslide initiation are as follows: the strength reduction of
rock masses (especially for the weak structural surface), the impact
from years of freeze-thaw cycles, the superposition of glacier
melting and heavy rainfall on the slope, and a slope that was
almost at the limit state before the landslide. Laboratory tests
and physical modeling experiments were carried out to study the
entrainment process of this landslide. Combined with the topo-
graphic survey data and theoretical analyses, the entrainment
mechanism during the movement process of the Yigong landslide
is presented. The old landslide deposits on the lower slope collided
with and were scraped by the high-speed debris avalanche, which
resulted in the volume amplification of the landslide. The exis-
tence of water plays a key role during the landslide initiation and
movement processes.
Keywords Yigong landslide . Dynamic process . Initiation
mechanism . Entrainment . Water
Introduction
Rapid catastrophic landslides, often defined as having a minimum
volume of 106 m3, can cause huge property loss and human death
because of their large volume and high speed (Korup et al. 2007;
Kuo et al. 2013; Pudasaini and Miller 2013; Weidinger et al. 2014).
Previous studies have shown that in recent years, the occurrence
frequency of rapid and long-runout catastrophic landslides has
significantly increased because of the changes of global climate,
such as rising temperature, heavy rainfall, and retreating snowline
(Crosta et al. 2009). Catastrophic landslides are widely distributed
worldwide, especially in the mountainous regions, such as the
southwest of China (Wang et al. 2013). On 12 May 2008, the great
Wenchuan earthquakes resulted in many catastrophic landslides
and provided rich data to study the dynamic problem of rapid
giant landslides (Huang and Li 2009; Tang et al. 2010; Dai et al.
2011; Yin et al. 2011; Zhou et al. 2013a). The initiation mechanism
and motion process are very important to understand the dynam-
ical process of giant landslides, which can provide possible pre-
ventive measures for potential giant landslides.
Landslides and other slope failures can be triggered by different
factors, such as earthquakes, heavy rainfall, freeze-thaw cycles,
glacier melting, and human activities (Evans et al. 2001). The
initiation mechanisms of landslides under different triggering
factors are not the same and are also influenced by the geological
condition of the slope (Acharya et al. 2011). Triggering factors for
the landslide can be divided into two main types: long-term
factors, such as freeze-thaw cycles, weathering, and unloading,
and short-term factors, such as strong earthquakes and heavy
rainfall (Barth 2014). Most of the catastrophic landslides are trig-
gered by the strong earthquake, such as the Tangjiashan landslide,
the Donghekou landslides, the Wenjiagou landslide, and the
Daguangbao landslide triggered by the 2008 Wenchuan earth-
quake (Yin et al. 2011) and the Tsaoling landslide triggered by the
1999 Chi-Chi earthquake (Tang et al. 2009). Because strong earth-
quake poses huge energy on the slopes in the earthquake hit area,
some of the catastrophic landslides are triggered by heavy rainfall,
for example, on 9 August 2009, a catastrophic landslide (Hsiaolin
landslide) is triggered by the accompanying heavy rainfall from
Typhoon Morakot in Kaohsiung County, Taiwan (Lo et al. 2011).
Previous studies have pointed out that water content within the
sliding mass and the sliding bed can accelerate rock mass dilation
and fragmentation and a successive increase in mobility (Crosta
et al. 2009). Landslides may be initiated by one triggering factor or
several coupling factors. Furthermore, the failure of the slope is
controlled by internal factors, such as terrain condition, mechan-
ical properties of the rock mass, and distribution of joints and
structural surface; the failure mode and probability for different
slopes are not the same (Mather et al. 2014).
During the motion process of rapid giant landslides, the rapid
sliding masses will interact with the bed materials along the trav-
eling paths, especially for the saturated accumulations (Legros
2002; Dufresne 2012). The high-speed sliding mass can entrain
large volumes of sediments both in a dry state and a saturated
state (Angeli et al. 1998; Dufresne and Davies 2009). Entrained dry
material is generally the consequence to reduce the mobility of
sliding mass (Crosta et al. 2009). However, according to the
Xiejiadianzi catastrophic landslide triggered by the 2008
Wenchuan earthquake, Wang et al. (2013) have pointed out that
an earthquake can simultaneously trigger liquefaction of runout
path materials before the arrival of the landslide mass, thus greatly
increasing the mobility of an overriding landslide. Entrainment of
almost saturated material can have a more complex consequence,
in which most can increase the runout distance, because of sudden
undrained loading and impact on loose deposits, and entrainment
of a saturated substrate can cause enhanced mobility (Crosta et al.
2009; Sassa 2000; Zhu et al. 2000). The bed materials are scoured
and eroded by the sliding mass and mixed with the sliding mass
and water, accreted to the main flow front or entrained, resulting
in the change of the landslide form (Dufresne and Davies 2009; Lo
et al. 2011). Some studies for rainfall-induced landslides have
pointed out that high-speed sliding motion leads to grain crushing
along the sliding surface, and the liquidated bed materials are
Landslides 13 & (2016) 39
 Original Paper
entrained by the upper sliding masses and finally resulted in the
long-runout distance and increasing volume of sliding mass (Sassa
2000). The entrainment of sufficiently large volumes of bed mate-
rials and surface water in the traveling path can transform a dry
rock avalanche into a debris avalanche/flow, which has greater
mobility (Lv et al. 2003; McDougall and Hungr 2005). The volume
of avalanches can be significantly increased through entrainment
of a substantial amount of bed material during the movement
process (Niemi et al. 2005). Here, field investigation, laboratory
test, and theoretical analysis are used to study the initiation mech-
anism and entrainment process of the 2000 Yigong landslide.
Background
The 2000 Yigong catastrophic landslide occurred in the mountains
of Tibet at 94 58 03 E, 30 12 11 N (Xu et al. 2012). The initiation
zone of the 2000 Yigong landslide was located in Zhamu Creek, at
the north side of the Yigong River, a tributary of the Yarlu River. In
this section, the geomorphology and geological condition in the
study area is briefly introduced, and then the main features of the
2000 Yigong catastrophic landslide are described.
Geomorphology
The 2000 Yigong landslide occurred in the steep high mountain-
ous region, which is located on the southern flank of the
Nyainqentanglha Shan Mountains in southeast Tibet (Shang
et al. 2003). Figure 1a shows the location of the Yigong landslide
area; the distance to Yigong Town is approximately 26 km, and the
location is 50 km from Lulang Town and 95 km from Bomi City.
An old dammed lake was formed on the Yigong River in 1900, and
after 100 years, a new catastrophic landslide occurred in Zhamu
Creek and also formed a huge landslide dam. Figure 1b shows the
geomorphology condition of the Yigong landslide region.
As shown in Fig. 1b, the length of Zhamu Creek is approximate-
ly 9.7 km, with thick ancient glaciers and snow on the top slope
when the slope elevation is greater than 4500 m. Bedrock is bared
at an elevation of approximately 31004200 m, and large volumes
of landslide deposits of outwash, glacial moraines, and others are
accumulated at the lower slope and along Zhamu Creek. The total
length of Zhamu Creek is approximately 10.6 km, the maximum
elevation at the top creek is approximately 5515 m, and the mini-
mum elevation at the lower creek of the Yigong River bed is
approximately 2185 m, with a total elevation difference of 3330 m
and a mean longitudinal inclination of 31.4 % (Xu et al. 2012). The
2000 Yigong landslide occurred at the top slope of Zhamu Creek;
the elevation range of the initiation zone is approximately 4000
5515 m. Zhamu Creek is perpendicular to the Yigong River. When
the landslide deposits rushed into the river, the water flow was
blocked by the mixture of rocks and soils and formed a dammed
lake.
Figure 2 shows the three-dimensional visualization of the
Yigong landslide region. At the upper creek, the elevation is ap-
proximately 38005515 m, with the slope angle larger than 50 and
side slope angles approximately 4050 (Hu et al. 2009a). The
upper slope is mainly covered by thick glaciers. Slope failures
occur easily under the special geomorphology conditions.
Figure 3a shows the steep terrain of the upper slope at the
Zhamu Creek, and multiple glaciers cover the mountain top. At
the middle creek, the elevation is approximately 29003800 m,
40 Landslides 13 & (2016)
with the slope angle approximately 1518 and side slope angles
approximately 4580. Large volumes of loose landslide deposits
are accumulated in this section due to the effect of old slope
failures, strong physical weathering, glacier melting, and other
processes. At the lower slope, the elevation is approximately
21852900 m with a gentle slope inclination about 1015. The
outlet of the Zhamu Creek is the junction position with the Yigong
River. Large amounts of landslide deposits are accumulated at this
section due to the 1990 Yigong landslide in the same region (Yin
2011). Here, we select four terrain sections along the Zhamu Creek
to analyze the terrain changes across the channel (the locations of
the terrain sections are shown in Fig. 2).
As shown in Fig. 3a, erosions were strongly intensive at the
upper slope; that means several landslides and collapses occurred
in the same area in the past. According to the terrain data at
different sections along the creek (Fig. 3b), we can find that the
inclination for the slopes on both sides of the Zhamu Creek is very
steep, while the width of the creek is relatively large at the upper
section and gradually narrows with decreasing elevation. At the
lower section of the creek, the terrain is flat and the width of the
creek is enlarged. This special terrain condition will affect the
motion of the landslide; part of the boulders may be blocked at
the back of the narrowest location (Yin and Xing 2012).
Geology
In the study area, the rock masses at the slope surface are strongly
weathered. Joints and fractures are well developed in the rock
masses. The occurrences of the main joints in the study area
include the following: (1) the dip direction of 328 and dip of
46.5, (2) the dip direction of 120 and dip of 74, (3) the dip
direction of 64 and dip of 32, (4) the dip direction of 180 and
dip of 48, and (5) the dip direction of 284 and dip of 72. The
layers of rock masses are also most consistent with the strike of the
slope. The slope stability is controlled by the structural surfaces
and rock layers. Wedge instability controlled by different structur-
al surfaces is the typical slope failure mode in this region.
Under the effect of tectonic movements and geomorphic evo-
lutions, the distribution of lithology in the Yigong landslide area is
very complicated. The study area is located near the Gangdise
batholiths (Xu et al. 2012). There are four main types of rocks in
this area: granite, marble, sandstone or slate, and limestone, with
different degrees of weathering and variability. Figure 4 shows four
images of the different types of rocks in the Yigong landslide
region.
At the upper section of Zamu Creek, the rock masses are mainly
composed of granite and steep slate (Fig. 4a, c). Here, we empha-
size that dolomite existed in the landslide initiation zone and
deposition zone, with different weathering degrees (Fig. 4d). At
the middle section, there are epimetamorphic rocks of the upper
Pondo Group (C pn ) of the Carboniferous, including marble
(Fig. 4b), sandstone, and slates interbedded with marbles (Shang
et al. 2005). Large volume of loose landslide deposits also existed
at the creek bottom in the middle section, mainly composed of
fragmented gravels, sandy silt, and other debris. At the lower
section, thick deposits from old rock avalanches or debris flows
are accumulated.
During the field investigation process, a small number of the
dolomites were weakly weathered; most of the dolomites were
strongly weathered and transformed into white powder (Fig. 5a),
 (a)
(b)
Fig. 1 a Location of the study area and b geomorphology condition of the Yigong landslide region (modified from Google Earth, 2014)
which is similar to lime but has more coarse particles and rough-
ness. The powder of strongly weathered dolomite existed at dif-
ferent deposition locations, mainly at the middle section of
landslide depositions. Figure 5b shows the X-ray diffraction anal-
ysis results for the sample of dolomite powder, and the mineral
composition of dolomite is relative to a single of CaMg(CO3)2. The
dolomite can be weathered more easily and significantly
influenced by the freeze-thaw cycle, resulting in the decreasing of
shear strength (Angeli et al. 1998; Borgatti and Soldati 2010).
The 2000 Yigong landslide
On 9 April 2000, a catastrophic landslide took place at Yigong,
Southeastern Tibet, one of the largest non-seismic landslides in the
world. Interpretation of remote sensing images shows that the
Landslides 13 & (2016) 41
 Original Paper
N (a)
Section 1-1
Section 2-2
Section 3-3
Section 4-4

(b)

Fig. 2 Three-dimensional visualization of the Yigong landslide region (landslide

region in Fig. 1)
(c)
total volume of the Yigong landslide deposits is approximately
3108 m3, while the accumulation area is approximately 6106 m2,
and the average depth of depositions is approximately 50 m
(Shang et al. 2003). The landslide was initiated at an elevation of

(a) (d)

Powder Block

Fig. 4 Images of the different types of rocks in the Yigong landslide region: a
granite, b marble, c slate, and d dolomite

approximately 5000 m (the volume of the landslide initiation

region is estimated to be larger than 1108 m3) and then acceler-
ated after a downward path of 1500 m because of the steep
(b)
inclination of the slope. Rock fragmentations occurred in the rock
5200
mass because of the high motion speed and were then transformed
into rock avalanche. The old deposits at the creek were impacted
and scoured by the high-speed rock avalanche (which was mixed
4400 with ice blocks and water) and finally transformed into a super
Meter (a.s.l.)

1-1 high-speed debris avalanche, eroding both banks of the creek (Xu
et al. 2012). Finally, the debris avalanche was flushed into the
Yigong River and formed a huge landslide dam. The landslide
3600 \2-2
deposits are widely distributed at Zhamu Creek.
3-3 Figure 6 shows the geological investigation results of the 2000
Yigong landslide. As shown in Fig. 6a, the 2000 Yigong landslide
4-4
2800 can be divided into three zones: the initiation zone, movement
amplification zone, and deposition zone. The landslide was initi-
ated at an elevation between 3800 and 5515 m and then ran into the
amplification zone (which corresponds to the middle creek). The
2000
0 1 2 3 4 moving mass involved the scouring of loose debris on both banks
Kilometer and the bottom at a high speed. Figure 6e shows the main geolog-
ical profile for the location of AB; we can see clearly that the slope
Fig. 3 Geomorphology features of the Zhamu Creek: a site photo and b terrain gradually decreased during the motion. When the moving mass
data at different sections along the creek (locations of the sections are shown in attained the narrow location of the creek, the moving mass became
Fig. 2) thicker and some boulders were blocked here.

42 Landslides 13 & (2016)

 (a)
Strongly weathered
Dolomite powder
(b)
6000
5000
Intensity(Counts)
4000
3000
2000
1000
0
36-0426> Dolomite - CaMg(CO3)2
10 20 30 40 50 60 70
2-Theta(
Fig. 5 a Powder of strongly weathered dolomite and b X-ray diffraction analysis
for the sample of dolomite powder
From the field investigation results, we can determine the
material compositions at different locations: (a) at the tail of
deposition zone, the deposition height is approximately 410 m
(Fig. 6b) and the landslide deposits are mainly composed of
boulders with a proportion of approximately 7090 %; (b) at the
middle of the deposition zone, the deposition height was approx-
imately 3565 m (Fig. 6c) and the landslide deposits are mainly
composed of rock debris, sandy silt, and solitary stone; the pro-
portion of sandy silt is approximately 4060 %; and (c) at the
lower of deposition zone, the deposition height is approximately
5080 m (Fig. 6d) and the landslide deposits are mainly composed
of rock debris and sandy silt; the proportion of sandy silt is
approximately 5070 %. At the middle and lower deposition
zones, the new 2000 landslide deposits covered the old 1900
landslide deposits. Figure 7 shows the distribution of the landslide
deposits for the 2000 Yigong landslide.
As shown in Fig. 7(a), the deposition of the 2000 Yigong
landslide can be divided into three zones: (1) the boulder deposi-
tion zone, (2) the debris and sandy silt deposition zone, and (3) the
impact and scouring zone. As shown in Fig. 7(b), at deposition
zone 2, the landslide deposits are composed of two main types of
material: rock debris and soils. Three-layered structural character-
istics existed in the lower deposition zone (from one eroded
section) because of the high mobility of the debris avalanche.
The top layer is the new 2000 landslide deposit, and the lower
layer is the old 1900 landslide deposit interbedded with an over-
consolidated layer (which is mainly composed of fine particles). At
the tail of the landslide deposits region, boulders accumulated
(Fig. 7(c)) and several giant stones exist in this region, which
belong to the landslide deposition zone 1. During the motion
process of debris avalanche, the old depositions were scoured
and entrained by the moving mass, due to the high speed of the
moving mass and steep terrain of Zhamu Creek.
Initiation mechanism of the Yigong landslide
In this section, combined with the geomechanical analysis and
long-term monitoring data of temperature and rainfall, the
initiation mechanism of the 2000 Yigong landslide is
presented.
Geomechanical analysis of slope
At the initiation zone of the 2000 Yigong landslide, steep slate and
granite are the main rock masses. The spacing of the slate is
approximately 1.05.0 m, and under the effect of the upper weight
of the rock mass and thick glacier, toppling and bending failure
occurred to the slate during a long geological history. The bending
failure surface can provide a potential sliding surface for the
initiation of landslides. The slope stability of the initiation zone
is controlled by the structural surfaces. Figure 8 shows the typical
geological condition of the initiation zone for the 2000 Yigong
80
landslide.
As shown in Fig. 8(a), there are five joint sets in the landslide
initiation zone; the occurrence parameters of those joints are as
follows: (1) in joint set 1, the dip is approximately 4050 and the
spacing is approximately 1.53.0 m; (2) in joint set 2, the dip is
approximately 7080 and the spacing is approximately 0.51.5 m;
(3) in joint set 3, the dip is approximately 2535 and the spacing is
approximately 0.51.5 m; (4) in joint set 4, the dip is approximately
7080 and the spacing is approximately 2.05.0 m; and (5) in joint
set 5, the dip is approximately 3055 and the spacing is approxi-
mately 1.54.5 m. The landslide initiation zone can be divided into
three subzones: (a) a tensile failure surface at the tail of the slope
(zone 1), in which a clearly fractured surface can be seen; (b) a
dislocation layer with several rock layers along the direction of
slate (zone 2); and (c) a compound weak layer (zone 3). The upper
part is a weak layer of bending failure surface of slate (rough), and
the lower part is a rock layer (smooth). To the left of subzone 1,
clearly bending failure surface existed and an old landslide oc-
curred in this area (Fig. 8(b)). The 2000 landslide initiation zone is
a typical wedge failure mode, which is controlled by two structural
surfaces and a tensile failure surface, and a free surface at the slope
front provided the potential initiation of the landslide in this
region.
Furthermore, thick glaciers covered the slope rock masses and
could have resulted in three adverse impacts on the landslide
initiation zone: (1) an increase in the weight of the sliding mass,
(2) the melting and freezing of the glacier effect on the rock masses
or structural surfaces such as the unloading-loading cycle for
rocks, and (3) the melting water could decrease the shear strength
of the structural surface and increase the pore water pressure in
the slope rock masses. The typical geological condition on the top
slope provides favorable conditions for the initiation of landslides.
The variation of the external environment condition on the slope
can be summarized in two aspects: decreasing shear strength and
increasing stress.
Landslides 13 & (2016) 43
 Original Paper
(e)
Fig. 6 Geological investigation results of the 2000 Yigong landslide: a overview of the landslide motion and deposition zone, b cross section aa, c cross section bb, d
cross section cc, and e main geological profile for the location of AB
Effect of temperature changes
In the study area, the 19512011 data recorded by the Yigong
Meteorological Station shows that the average temperature was
approximately 810 C. Since the 1970s, the temperature in this
area has had a slowly rising tendency with a total ratio of 0.3 C/
10 years (Shang et al. 2003). Figure 9 shows the monitoring results
of temperature in the study area.
As shown in Fig. 9a, the average temperature gradually
increased during 1 March to 4 May, which means that the
melting of glacier and snow gradually increased during this
season. However, there was no significant difference between
the years 1998, 1999, and 2000. Along with the rises of tem-
perature from April to May, glaciers and snow cover melt and
water flow was generated in Zhamu Creek. The water flow not
44 Landslides 13 & (2016)
(a)
(b)
(c)
(d)
only affects the loose landslide deposits located at the creek
but also influences the slope stability in this region. As shown
in Fig. 9b, the daily temperature difference is the difference
between maximum temperature and minimum temperature in
1 day. The daily temperature difference in the study area was
relatively large; the average value was approximately 12.5 C
from 1 January 2000 to 1 January 2001, and the maximum
value attained was 26 C. The temperature difference from 1
April to 15 April 2000 was relatively large, which had a signif-
icant effect on the initiation of landslide on 9 April 2000.
Temperature changes can provide two aspects for the initiation
of landslides: (a) a freeze-thaw cycle on structural surfaces and
(b) glacier melting resulted in the decreasing of shear strength
and increasing pore water pressure.
 (a)
(b)
2000 landslide deposits
Over-consolidate layer
1900 landslide deposits
Fig. 7 Distribution of the landslide deposits for the 2000 Yigong landslide: a partition of deposition zone (Xu et al. 2012, modified), b site photo taken at the lower
deposition zone, c site photo taken at the upper deposition zone, and d giant stone at the tail of landslide deposits region
There are two types of freeze-thaw cycles for the rock mass in
this area: the seasonal freeze-thaw cycle (with a period of 1 year)
and the daily freeze-thaw cycle (with a period of 1 day). Figure 10a
shows the temperature monitoring result from 1 January 1996 to 30
December 2003, and Fig. 10b shows the temperature monitoring
results from 15 March to 16 April 2000.
As shown in Fig. 10a, the annual temperature variation period-
ically changes; the minimum temperature (about 5 C) occurred
in December or January and then gradually rose to the maximum
temperature (approximately 20 C) in June or July. This long-term
freeze-thaw cycle resulted in the gradual decreasing of shear
strength of rock masses and structural surfaces. As shown in
Fig. 10b, the freeze-thaw cycle property of daily temperature is
exhibited. However, strong cycles did not occur because the
daily temperature variation was disordered. The mechanical
properties of rock masses were also influenced by the daily
freeze-thaw cycle.
Previous studies have shown that freeze-thaw cycles have a
significant influence on the mechanical properties of rocks, in-
cluding porosity, compressive strength, elastic modulus, and shear
(d)
(c)
strength (Ma et al. 1999). Microcracks and micro-defects occur in
the rocks, and water will run into the rocks. The water in the rocks
begins to freeze when the temperature is lower than 0 C.
Volumetric expansion occurs during the freezing process and leads
to microcrack propagations in the rock. Local damage occurs in
the rocks and results in the decreasing of compressive strength
and shear strength (Draebing et al. 2014). However, the ice in rocks
melts when the temperature is greater than 0 C, while pore water
pressure is formed in the rocks, and also results in the decreasing
of shear strength (Zhu et al. 2011; Krautblatter et al. 2013). With the
increasing numbers of freeze-thaw cycles, the shear strength of
rock masses or structural surface is gradually decreased (Lai et al.
2010). The freeze-thaw cycle not only decreases the shear strength
of rock masses but also results in the increasing stress distribution
in the slope and finally causes the initiation of landslides in high
mountainous regions (Bayard et al. 2005).
Effect of rainfall
According to the 19512011 data from the Yigong Meteorological
Station, the mean annual rainfall in the Yigong area was
Landslides 13 & (2016) 45
 Original Paper
(a)
(b)
Fig. 8 Typical geological condition of the initiation zone for the 2000 Yigong landslide at the top of Zhamu Creek
approximately 8101250 mm. The rainfall from April to October
accounts for 90 % of the annual total rainfall, and November to
February is the dry season (Shang et al. 2003). Figure 11 shows the
rainfall monitoring results in the Yigong region.
As shown in Fig. 11a, most of the daily rainfall is approximately
040 mm, but some heavy rainfall occurred on special days. The
maximum daily rainfall occurred on 20 October 1998, approxi-
mately 111.7 mm. The daily rainfall was not distributed uniformly
during the whole year; the daily rainfall during the rainy season
was greater than during the dry season. The rainfall in different
seasons has similar characteristics. The rainfall data are monitored
in the valley region, and the rainfall level at the top of the moun-
tain is greater. Thus, in the initiation zone of Yigong landslide, the
rainfall is relatively abundant. Figure 11b shows the comparison
results for the monthly rainfall of 1998, 1999, and 2000. We can see
that rainfall occurred mainly during April to October. On April,
the monthly rainfall in 2000 was approximately 175 mm, more
than the years 1998 (38 mm) and 1999 (80 mm). Increasing rainfall
on April 2000 played some key effects on the initiation of the 2000
Yigong landslide. Figure 12 shows the daily rainfall during 1 March
to 4 May 2000 at the study area.
As shown in Fig. 12, in 2000, when the recent Yigong Landslide
occurred, the rainfall from April 2 to April 9 was more than 60 mm
(Hu et al. 2009b) and the total rainfall in April was approximately
175 mm, 90135 % more than the mean value from the same period
in southeast Tibet (Liu 2002). The daily rainfall on 9 April 2000
was approximately 25 mm. The abundant rainfall resulted in the
46 Landslides 13 & (2016)
saturation of loose landslide deposits at Zhamu Creek and increas-
ing pore water pressure in the upper slope, especially in the
structural surfaces. Furthermore, relatively more water flow
formed in Zhamu Creek, which could have affected the motion
of debris avalanche.
Comprehensive analysis
Through the above analyses of the geological and environmental
conditions at the Yigong area, we found that the main reasons for
the initiation of the 2000 Yigong landslide were typical slope
geological conditions, temperature changes, and heavy rainfall
on April 2000.
Table 1 summarized the annual environment conditions at the
Yigong area from 1991 to 2001, including temperature and rainfall.
Extreme maximum temperatures always occurred from June to
August, with values of approximately 29 to 31 C. However, the
extreme minimum temperatures always occurred from December
to February, with values of about 10 to 13 C. The maximum
rainfall for the whole year of 2000 occurred on 9 April.
The reasons for the initiation of the 2000 Yigong landslide are
the coupling of long-term effects and short-term impact.
Regarding the long-term effects, there are two aspects, as men-
tioned above: (a) the loading-unloading cycle on the rock masses
or structural surfaces due to the melting and freezing of glaciers
and (b) the continuous seasonal and daily freeze-thaw cycles on
the rock masses or structural surfaces. These two aspects resulted
in the gradual decrease of mechanical properties of rock masses or
 (a)
20
2000
1999
1998

Average temperature (C)

15

10

0
3/1 3/9 3/17 3/25 4/2 4/10 4/18 4/26 5/4
Time (mm/dd)

(b)
30 Larger value Average value is 12.15 C
Temperature difference (C)

25

20

15

10

0
01/01/00 03/01/00 05/01/00 07/01/00 09/01/00 11/01/00 01/01/01
Time (mm/dd/yy)

Fig. 9 Temperature monitoring results in the study area: a comparison analysis for the average temperature during 1 March to 4 May in 1998, 1999, and 2000, and b
daily temperature difference from 1 January 2000 to 1 January 2001

structural surfaces, including compressive strength, porosity, and
elastic modulus, especially for the shear strength of rock masses or
structural surfaces. Furthermore, the tensile fracture has gradually
been increasing over a long period. Other long-term factors in-
cluding weathering and unloading can also have adverse impacts
on the slope stability.
Regarding the short-term effects, the rising temperature will
accelerate the melting of glacier and snow; by coupling with heavy
rainfall, the water content of the structure surface is increased
sharply and results in a larger decreasing of shear strength. The
heavy rainfall and melting of glacier and snow also resulted in the
increasing of pore water pressure on the structural surfaces.
Another key factor for the initiation of the landslide is the special
geological conditions of the upper slope at Zhamu Creek. The
potential sliding wedge is controlled by two structural surfaces
and a tensile failure surface, which are sensitive to temperature
change and water content.
Regarding the initiation mechanism of the 2000 Yigong land-
slide, special geological conditions at the top slope provide neces-
sary conditions for the initiation of the landslide. Long-term
freeze-thaw and loading-unloading cycles resulted in the gradually
deterioration of the slope stability. The direct triggering factors are
short-term heavy rainfall and rising temperatures on April 2000.
The increasing water content and pore water pressure in the
structural surface finally resulted in the catastrophic landslide in
the Yigong area in 2000.
Entrainment effect during motion process
For the 2000 Yigong landslide, the volume of landslide deposits
was about three times the volume of initiation the zone. The
entrainment effect during the motion process obviously occurred
because of the high motion speed and water flow.
Motion of high-speed landslide
The entrainment of path material is an important feature of
many rapid landslides (Hs 1978; Evans et al. 2001; Dufresne
et al. 2010). During the motion process of rock avalanches,
the landslide paths are typically covered by thick surficial
deposits. These deposits may be loose and have high water
content. The rapid loading due to the weight and momentum
of the moving masses may cause failure and mobilization of
these bed materials and resulted in the volume amplification
of the landslide (Scott et al. 2005; Cepeda et al. 2010).
Figure 13 shows the topography changes for the main geolog-
ical section of Zhamu Creek before and after 2000 Yigong
landslide, which is extracted for the high-resolution digital
elevation model of the study area.

Landslides 13 & (2016) 47

 Original Paper
(a)
20
15
Temperature (C)
10
-5
01/01/96 12/31/96 12/31/97 12/31/98 12/31/99 12/30/00 12/30/01 12/30/02 12/30/03
(b)
25
Temperature (C)
15
-5
3/15
Fig. 10 Different freeze-thaw cycles for the rock mass in the study area: a seasonal type and b daily type
As shown in Fig. 13, many loose surficial deposits were covered
on Zhamu Creek during the motion process of the high-speed rock
avalanche, and these bed materials were scoured and eroded by
the sliding masses, which finally resulted in the volume amplifica-
tion of the landslide. The serious scouring of the bed materials at
the middle section of Zhamu Creek by sliding masses can clearly
be seen.
Laboratory tests
Regarding the entrainment problem of high-speed landslides,
a simple laboratory test was carried out to study the bed
materials impact and scouring by the sliding masses.
Figure 14 shows the experimental equipment for the laborato-
ry tests. The horizontal length of the flume is 2.0 m, the
height is 1.5 m, and the width is 0.3 m. Loose bed materials
are accumulated at the lower of flume, and the sliding mass
(composed by rock blocks) are placed in the topper of flume
(Fig. 14b). These experiments are solely conceptual since no
scaling calculations were done. Table 2 shows the experimen-
tal design for the impact and scouring effect on the bed
materials.
As shown in Table 2, 11 tests under different material composi-
tions and weight ratios, with and without water, were carried out.
The weight ratio is a key factor affecting the motion distance of
bed materials. Figure 15 shows the movement distance of bed
materials influenced by the weight ratio between sliding mass
and bed depositions.
48 Landslides 13 & (2016)
Time (mm/dd/yy)
3/23 3/31 4/8 4/16
Time (month/day)
As shown in Fig. 15, these test results were obtained under dry
conditions, without water flow. The movement distance of bed
materials increased with the weight ratio between sliding mass
and bed depositions, an approximately linear relationship.
During the landslide motion process, the bed depositions are
impacted and scoured by the sliding masses (Van Wyk de Vries
et al. 2001; Crosta et al. 2005; Dufresne 2012). A larger weight ratio
between the sliding mass and bed deposition means the difference
of impact energy and friction resistance is increased, resulting in
the longer movement distance of bed materials. Figure 16 shows
the experimental results for the impact and scouring of sliding
masses.
As shown in Fig. 16a, rock blocks were used to simulate the rock
avalanche and placed at the top of the flume and the loose deposits
were covered on the lower flume with a uniform thickness.
Figure 16b shows the final deposited situation of test No. 6 under
dry condition, and the WR is 1.0. The bed depositions were im-
pacted and scoured by the moving mass, and the total motion
distance of the bed materials was 16.0 cm. Most of the rock blocks
were blocked at the tail of landslide deposits, and some blocks
moved to the front of the flume. When water flow was added in the
experimental process, the bed depositions were saturated, pore
water pressure formed in the soils, and the friction resistance of
the bed materials decreased sharply. When the sliding mass im-
pacted on the bed materials, the huge impact energy and excess
pore water pressure resulted in the failure and scouring of bed
materials. Then, the bed loose deposits were mixed in the sliding
 (a)
50 111.7 mm

Daily rainfall (mm)

40

30

20

10

0
01/01/96 12/31/96 12/31/97 12/31/98 12/31/99 12/30/00 12/30/01 12/30/02 12/30/03
Time (mm/dd/yy)
(b)
250
1998
200
1999
Rainfall (mm)

2000
150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12
Month

Fig. 11 Rainfall monitoring results in the Yigong region: a daily rainfall during the years 19962003 and b monthly rainfall for the years 1998, 1999, and 2000

mass and moved down along the flume. The final deposited Movement amplification effect
situation of test No. 4 is shown in Fig. 16c. The motion distance From the field investigation and above analysis results, the dy-
of the bed materials is larger than the dry condition, approximate- namical process of the 2000 landslide can be summarized as
ly 43.8 cm. Compared with the actual deposited situation in Fig. 7a, follows: landslide initiation (wedge failure)acceleration and
the experimental deposited situation has the same distribution fragmentationrock avalanchescouring and
characteristics as the actual landslide. The impact and scouring entrainmentdebris avalanchelandslide deposition, as shown
of bed materials resulted in the volume amplification of the sliding in Fig. 17a. The landslide was initiated under heavy rainfall and
mass. rising temperature at the top slope of Zhamu Creek, with a wedge
failure type. Then, the sliding mass moved down the steep terrain
and attained a high speed. The sliding mass was fragmented by the
30
2000
effect of strike, collision, and other impact loading, and then, the
1999
1998
sliding mass transformed into a rock avalanche. During the mo-
25
tion process of the rock avalanche, much bed deposit was impact-
Daily rainfall (mm)

20 ed and scoured by the sliding mass. This situation became more

15
serious because the water flow, the decrease in the shear strength
of bed material, and the excess pore water pressure facilitated the
10 failure. The failed bed material was mixed in the rock avalanche
5 and moved with the sliding mass and resulted in the rock ava-
lanche transforming into debris avalanche. When the sliding mass
0
3/1 3/9 3/17 3/25 4/2 4/10 4/18 4/26 5/4
moved to the lower section of Zhamu Creek, a thin layer was
Time (mm/dd) formed between the new avalanche debris and the 1900 landslide
deposits. The material composition of this layer is mainly water
Fig. 12 Daily rainfall during 1 March to 4 May 2000 at the study area and fine particles, so that we can clearly see the three-layered

Landslides 13 & (2016) 49

 Original Paper
Table 1 Annual environmental conditions in the Yigong area from 1991 to 2011
Year Average Extreme maximum Extreme minimum Annual Maximum daily rainfall
temperature temperature temperature rainfall
(C) Value (C) Time Value (C) Time (mm) Value (mm) Time
(dd-mm) (dd-mm) (dd-mm)
1991 8.6 28.8 22-Jun 12.0 17-Jan 1128.5 43.8 2-Jun
1992 8.4 28.8 18-Jun 13.7 24-Dec 714.2 43.1 24-Jun
1993 8.7 28.4 7-Jul 12.6 29-Jan 810.0 26.5 28-Sep
1994 8.9 31.0 12-Jul 12.1 23-Dec 847.3 36.9 10-Oct
1995 9.2 31.0 6-Aug 13.0 28-Jan 1130.7 48.1 14-Jun
1996 8.9 28.3 21-Jul 12.5 3-Jan 1068.4 44.1 29-Oct
1997 8.2 28.7 8-Aug 12.7 19-Jan 930.6 44.1 9-Aug
1998 9.3 28.0 12-Jul 10.8 9-Feb 1108.8 111.7 20 Oct
1999 9.4 30.2 1-Aug 12.1 12-Jan 932.2 31.8 22-May
2000 8.8 29.2 25-Jul 12.2 31-Jan 808.9 25.0 9-Apr
2001 9.1 31.2 8-Jul 11.4 24-Dec 801.0 25.1 24-Mar
2002 9.1 30.2 9-Jun 12.6 2-Jan 862.5 36.8 10-Apr
2003 9.1 30.0 7-Aug 12.4 16-Jan 823.7 22.8 8-May
2004 9.0 29.3 6-Aug 11.5 28-Dec 880.9 46.0 15-Apr
2005 9.6 28.6 15-Jun 11.5 13-Jan 777.2 32.9 18-Apr
2006 9.8 31.2 17-Jul 9.6 24-Dec 864.1 34.9 12-Sep
2007 9.7 29.4 13-Jul 10.6 24-Jan 855.9 45.0 16-May
2008 9.1 29.1 12-Jun 10.3 15-Feb 800.1 86.0 27-Oct
2009 10.2 30.9 20-Jul 10.0 28-Dec 536.2 22.1 22-Feb
2010 9.6 30.7 15-Aug 10.2 3-Jan 1080.6 39.3 11-May
2011 9.2 30.1 31-Aug 9.2 16-Jan 838.7 44.8 25-Mar

structural characteristics of the landslide deposits (Fig. 7b). The

middle layer is composed of over-consolidated materials and has a (a)
higher compression strength and lower porosity.
Figure 17 shows the whole dynamic process of the movement of
the 2000 Yigong landslide. There are three types of entrainment
modes for the high-speed landslide: impact, scouring, and erosion.
When the rock avalanche attained the tail of the bed deposits, the
entrainment of the sliding mass occurred through impact and
scouring. The high-speed rock avalanche led to great impact en-
ergy on the bed materials and resulted in the failure of old

Before landslide 6000

After landslide (b)
4900 150
Meters (a.s.l.)

Sliding mass
3800
100
Height (cm)

2700
50
Bed materials
1600
12 10 8 6 4 2 0
Kilometers 0
0 50 100 150 200
Length (cm)
Fig. 13 Topography changes for the main geological section of the Zhamu Creek
before and after the 2000 Yigong landslide Fig. 14 Experimental equipment: a actual image and b size design

50 Landslides 13 & (2016)

Table 2 Experimental design for the impact and scouring effect on the bed materials
No. Depositions (kg) Rock blocks (kg) Water WRa Motion distance (cm)
<1 mm 25 mm 2040 mm
1 4.225 13.040 Y 3.1 72.3
2 6.235 11.560 Y 1.9 54.3
3 10.000 8.000 Y 0.8 11.6
4 3.000 5.000 12.000 Y 1.5 43.8
5 4.000 6.000 10.000 Y 1.0 23.0
6 4.000 6.000 10.000 N 1.0 16.0
7 4.000 6.000 4.000 N 0.4 2.5
8 4.000 6.000 8.000 N 0.8 11.0
9 4.000 6.000 12.000 N 1.2 28.0
10 4.000 6.000 15.000 N 1.5 46.0
11 4.000 6.000 20.000 N 2.0 58.0
Motion distance referred to the bed depositions
a
The weight ratio between sliding rock blocks and bed depositions

deposits. Furthermore, during the motion process of the rock
avalanche, the bed material was scoured by the sliding mass.
This former entrainment process is shown in Fig. 17b. After the
landslide initiation, the form and material composition of land-
slide varied during the whole motion process, and the cumulative
volume of the landslide changed enormously along the path of
movement. The path is described as a series of reaches that have a
distinct morphology, width, and slope inclination, and the struc-
tural characteristics, material compositions, and mechanical prop-
erties are also not the same. These changes made the entrainment
process of sliding mass very complicated.
As shown in Fig. 17c, the bed deposits were eroded by the high-
speed sliding mass, especially when water flow occurred. The bed
erosion at the base of the landslide is similar to bedload transport
in fluvial hydraulics. The failure of bed material can be caused by
the impact force and scouring effect and the generation of excess
pore water pressures due to rapid undrained loading with possible
liquefaction. The existence of water plays a key role during the
motion process of landslides, especially for the entrainment of
landslides.
In the Yigong area, the special terrain condition allowed the
water flow to be easily collected in the Zhamu Creek. Heavy
rainfall and glacier on April 2000 resulted in the saturation of
60
y = 37.058x - 14.533
Movement distance (cm)
bed deposits and larger water flow. During the motion process of
the landslide, the landslide volume gradually increased through
the entrainment effect and finally resulted in the volume amplifi-
cation of the landslide deposits.
Discussion and conclusion
Discussion
The initiation mechanism and motion process of landslide are
very complicated for the dynamical problem of high-speed land-
slides (Zhou et al. 2010). The initiation of landslides is not only
influenced by the geological condition but also affected by the
short-term environmental conditions, such as earthquake, heavy
rainfall, or rising temperature (Wang et al. 2003; Tsou et al. 2011).
Previous studies show that landslides may be initiated by some
sudden events, such as strong earthquake, heavy rainfall, or rising
temperature in a short time (Evans et al. 2001; Dai et al. 2011; Kuo
et al. 2013; Barth 2014). However, heavy rainfall is not the single
triggering factor for the 2000 Yigong landslides, and the rising
temperature on April 2000 is another short-term triggering factor.
Furthermore, the rock masses in a slope generally experienced a
long-term geological evolution process, such as the tectonic effect,
weathering and unloading, freeze-thaw cycles, and loading-
unloading cycle. These long-term geological effects can result in
the change of the slope stability state, including the decreasing of
shear strength of rock masses and increasing of pore water pres-

R2 = 0.9759
sure in slope (Zhou et al. 2013b). These reasons make the initiation
40
mechanism of landslide very complicated under different condi-
tions. However, one key issue should be emphasized here: the
20 initiation of most landslides is a result of the coupling of long-
term evolution and short-term effects on a slope.
Mass movement of the rock avalanche or debris avalanche is
0 mainly influenced by the Lithology and topographical conditions.
0.0 0.5 1.0 1.5 2.0 2.5
Landslide/Deposits Fragmentation of rock masses and mobility of sliding masses are
influenced by the lithology and its structural properties. For the
Fig. 15 Movement distance of bed materials influenced by the weight ratio 2009 Hsiaolin landslide, Lo et al. (2011) found that huge amounts
between sliding mass and bed depositions of fragmented rock materials moved quickly downward and

Landslides 13 & (2016) 51

 Original Paper
(a) (b) (c)

Fig. 16 Experimental results for the impact and scouring of sliding masses: a experimental preparation, b final deposited situation of test No. 6, and c final deposited
situation of test No. 4

transformed into a debris avalanche. Rock fragmentation can
easily be happened during the motion process of high-speed land-
slide, part of the debris rushed over a high-level terrace and spread
along the hillslope (Hungr et al. 2005). Special terrain condition
plays a key role during the mass movement process. Regarding to
the 2000 Yigong landslide, the sliding mass moved down the steep
terrain with a high speed and then encountered the bed deposits.
Huge impact energy is posed on the bed deposits, and a large
volume of bed deposit was impacted and scoured by the sliding
mass, resulting in the rock avalanche transformed into debris ava-
lanche. Meantime, a part of the sliding masses are stopped and
accumulated at the foot of the steep slope, because of the energy loss
and the narrow terrain. The flat and straight topographical condi-
tions along the runout path provide a favorable condition for the
mass movement. The presence of water not only increases the
mobility of sliding masses but also poses a great impact on the
entrainment process during mass movement process.
For the entrainment problem during the movement process of
high-speed landslide, well-documented landslide events have ver-
ified this phenomenon (Staron 2008). The landslide entrainment
includes three main types: impact failure, scouring, and erosion
(Pudasaini and Miller 2013). Using a small-scale physical models
(linear scale of 1:104), Dufresne (2012) found that substrates are
failing by the effect of upper sliding masses due to low internal and
basal frictional resistance and then turn into the active basal
sliding layer, increasing basal boundary roughness of erodible,
deformable substrates resulting in shorter runout and longer de-
posits (decreased mobility of sliding mass). Using laboratory tests,
Shea and van Wyk de Vries (2008) found that, in the presence of a
substrate, the granular flow was slowed and contraction-

(a)

Top-down
High speed rock avalanche Debris + soil
Water + fine particle
Old depositions
Evolution
Impact and scouring
Motion direction

Loose depositions Motion

(b) Impact and scouring (c) Movement erosion

Fig. 17 Dynamic process of the movement of the 2000 Yigong landslide: a scheme of whole landslide process, b impact and scouring of bed materials by the high-speed
rock avalanche, and c erosion of bed materials erosion the motion of debris avalanche

52 Landslides 13 & (2016)

generated surface ridges formed due to a lack of flow spreading,
while the substrate itself was bulldozed at the flow front.
According to the round top (RT) rock avalanche deposit, using
field investigations and laboratory tests, Dufresne et al. (2010)
found that the volume increase from the initial source volume to
that of the final deposit existed because of the substrate entrain-
ment during mass movement process, and the substrate entrain-
ment increases the mobility of the avalanche mass by introducing
significant quantities of water. Entrainment of bed deposits by
surface water or sliding masses can lead to the change of motion
behavior, such that the bed deposits on the steep slope entrain-
ment by surface water can transform a dry rock/debris avalanche
into debris flows, leading to greater areas affected and longer-
runout distances covered (e.g., Kerle and vanWyk de Vries 2001;
Evans et al. 2001). For the 2000 Yigong landslide, bed deposit
entrainment by surface water and sliding masses transforms the
rock avalanche into more mobile debris avalanche. However, the
entrainment of landslide is a dynamical process, and its mechan-
ical mechanism is very complicated and involves multiple areas of
expertise: geotechnical engineering, hydromechanics, kinematics,
and others. It is hard to present a theoretical model to describe the
entrainment problem during the landslide motion process, and
also, numerical simulations cannot be true representations of the
entrainment process of landslides. However, the existence of water
flow during the landslide process can aggravate the entrainment of
landslides.
Conclusion
In this paper, the 2000 Yigong catastrophic landslide is taken as
the study example, and the initiation mechanism and entrainment
process of landslide are analyzed through field investigation, the-
oretical analysis, and laboratory tests. Some useful conclusions are
presented here.
For the initiation mechanism of the 2000 Yigong landslide, the
special geological conditions at the top slope provide precedent
conditions for the initiation of landslide. The reasons for the
initiation of the 2000 Yigong landslide are the coupling of long-
term effects and short-term impact. For the long-tern effects, there
are two aspects as mentioned above: (a) the loading-unloading
cycle on the rock masses or structural surfaces due to the melting
and freezing of glaciers and (b) the continuous seasonal and daily
freeze-thaw cycles on the rock masses or structural surfaces. For
the short-tern effects, the rising temperature on April 2000 accel-
erated the melting of glacier and snow, by coupling with the heavy
rainfall, resulting in the decrease of shear strength and the increase
of pore water pressure on the structural surfaces.
For the entrainment process of the 2000 Yigong landslide, the
dynamical process can be summarized as follows: landslide initi-
ation (wedge failure)acceleration and fragmentationrock
avalanchescouring and entrainmentdebris
avalanchelandslide deposition. There are three types of entrain-
ment modes for the high-speed landslide: impact, scouring, and
erosion. During the motion process of rock avalanche, the shear
strength of bed material is decreased, excess pore water pressure is
formed, and many bed deposits are impacted and scoured by the
sliding mass. The landslide volume is gradually increased by the
entrainment effect and resulted in the volume amplification of
landslide. The existence of water flow plays a key role during the
initiation and motion process of landslides.
Acknowledgments
We gratefully acknowledge the support of the National Natural
Science Foundation of China (41472272, 41030742, and 41102194)
and the Science Foundation for Excellent Youth Scholars of Si-
chuan University (2013SCU04A07). Dr. Gong-dan Zhou and Dr.
Zhi-man Su provided valuable discussion on the initiation mech-
anism of the Yigong landslide. Critical comments by the reviewers
greatly improved the initial manuscript.
References
Acharya G, Cochrane T, Davies T, Bowman E (2011) Quantifying and modeling post-
failure sediment yields from laboratory-scale soil erosion and shallow landslide
experiments with silty loess. Geomorphology 129:4958
Angeli MG, Bunma J, Gasparetto P, Pasuto A (1998) A combined hillslope hydrology/
stability model for low-gradient clay slopes in the Italian Dolomites. Eng Geol 49:113
Barth NC (2014) The Cascade rock avalanche: implications of a very large Alpine Fault-
triggered failure, New Zealand. Landslides 11:327341
Bayard D, Sthli M, Parriaux A, Flhler H (2005) The influence of seasonally frozen soil on
the snowmelt runoff at two Alpine sites in southern Switzerland. J Hydrol 309:6684
Borgatti L, Soldati M (2010) Landslides as a geomorphological proxy for climate change:
a record from the Dolomites (northern Italy). Geomorphology 120:5664
Cepeda J, Chvez JA, Martnez CC (2010) Procedure for the selection of runout model
parameters from landslide back-analyses: application to the Metropolitan Area of San
Salvador, El Salvador. Landslides 7:105116
Crosta GB, Imposimato S, Roddeman D, Chiesa S, Moia (2005) Small fast-moving flow-
like landslides in volcanic deposits: the 2001 Las Colinas Landslide (El Salvador). Eng
Geol 79:185214
Crosta GB, Imposimato S, Roddeman D (2009) Numerical modelling of entrainment/
deposition in rock and debris-avalanches. Eng Geol 109:135145
Dai FC, Tu XB, Xu C, Gong QM, Yao X (2011) Rock avalanches triggered by oblique-
thrusting during the 12 May 2008 Ms 8.0 Wenchuan earthquake, China. Geomor-
phology 132:300318
Draebing D, Krautblatter M, Diakau R (2014) Interaction of thermal and mechanical
processes in steep permafrost rock walls: a conceptual approach. Geomorphology
226:226235
Dufresne A (2012) Granular flow experiments on the interaction with stationary runout
path materials and comparison to rock avalanche events. Earth Surf Process Landf
37:15271541
Dufresne A, Davies TR (2009) Longitudinal ridges in mass movement deposits. Geomor-
phology 105:171181
Dufresne A, Davies TR, McSaveney MJ (2010) Influence of runout-path material on
emplacement of the Round Top rock avalanche, New Zealand. Earth Surf Process
Landf 35:190201
Evans SG, Hungr O, Clague JJ (2001) Dynamics of the 1984 rock avalanche and associated
distal debris flow on Mount Cayley, British Columbia, Canada; implications for
landslide hazard assessment on dissected volcanoes. Eng Geol 61:2951
Hs KJ (1978) Albert Heim: observations on landslides. In: Voight B (ed) Rockslides and
avalanches, vol 1. Elsevier, Amsterdam, pp 7093
Hu M, Chen Q, Wang F (2009a) Experimental study on formation of Yigong long-distance
high-speed landslide. Chin J Rock Mech Eng 28:138143 (in Chinese)
Hu M, Wang F, Chen Q (2009b) Formation of tremendous Yigong landslide based on
high-speed shear test. Chin J Geotech Eng 31:16021606 (in Chinese)
Huang RQ, Li WL (2009) Analysis of the geo-hazards triggered by the 12 May 2008
Wenchuan earthquake, China. Bull Eng Geol Environ 68:363371
Hungr O, Corominas J, Eberhardt E (2005) Estimating landslide motion mechanism, travel
distance and velocity. In: Proceedings of the International Conference on Landslide
Risk Management, pp. 99128
Kerle N, van Wyk de Vries B (2001) The 1998 debris avalanche at Casita volcano,
Nicaraguainvestigation of structural deformation as the cause of slope instability
using remote sensing. J Volcanol Geotherm Res 105:4963
Korup O, Clague JJ, Hermanns RL, Hewitt K, Strom A, Weidinger JT (2007) Giant
landslides, topography, and erosion. Earth Planet Sci Lett 261:578589
Krautblatter M, Funk D, Gnzel FK (2013) Why permafrost rocks become unstable:
a rockice-mechanical model in time and space. Earth Surf Process Landf
38:876887
Landslides 13 & (2016) 53
 Original Paper
Kuo Y, Tsai Y, Chen Y, Shieh C, Miyamoto K, Itoh T (2013) Movement of deep-seated
rainfall-induced landslide at Hsiaolin Village during Typhoon Morakot. Landslides
10:191202
Lai Y, Yang Y, Chang X, Li S (2010) Strength criterion and elastoplastic constitutive model
of frozen silt in generalized plastic mechanics. Int J Plast 26:14611484
Legros F (2002) The mobility of long-runout landslides. Eng Geol 63:301331
Liu W (2002) Study on the characteristics of huge scale-super high speed-long distance
landslide chain in Yigong, Tibet. Chin J Geol Hazard Control 13:918 (in Chinese)
Lo C, Lin M, Tang C, Hu J (2011) A kinematic model of the Hsiaolin landslide calibrated to
the morphology of the landslide deposit. Eng Geol 123:2239
Lv J, Wang Z, Zhou C (2003) Discussion on the occurrence of Yigong landslide in Tibet.
Earth Sci J China Univ Geosci 28:107110 (in Chinese)
Ma W, Wu Z, Zhang L, Chang X (1999) Analyses of process on the strength decrease in
frozen soils under high confining pressures. Cold Reg Sci Technol 29:17
Mather AE, Hartley AJ, Griffiths (2014) The giant coastal landslides of Northern Chile:
tectonic and climate interactions on a classic convergent plate margin. Earth Planet
Sci Lett 388:249256
McDougall S, Hungr O (2005) Dynamic modelling of entrainment in rapid landslides. Can
Geotech J 42:14371448
Niemi NA, Oskin M, Burbank DW, Heimsath AM, Gabet EJ (2005) Effects of bedrock
landslides on cosmogenically determined erosion rates. Earth Planet Sci Lett 237:480
498
Pudasaini SP, Miller SA (2013) The hypermobility of huge landslides and avalanches. Eng
Geol 157:124132
Sassa K (2000) Mechanism of flows in granular soils. In: Proceedings of GeoEng2000,
Melbourne, 1924 November, vol. 1. Technomic Publishing Company, USA, pp 1671
1702
Scott KM, Vallance JW, Kerle N, Macas JL, Strauch W, Devoli G (2005) Catastrophic
precipitation-triggered lahar at Casita volcano, Nicaragua: occurrence, bulking and
transformation. Earth Surf Process Landf 30:5979
Shang Y, Yang Z, Li L, Liu D, Liao Q, Wang Y (2003) A super-large landslide in Tibet in
2000: background, occurrence, disaster, and origin. Geomorphology 54:225243
Shang Y, Park H, Yang Z, Yang J (2005) Distribution of landslides adjacent to the
northern side of the Yarlu Tsangpo Grand Canyon in Tibet, China. Environ Geol
48:721741
Shea T, van Wyk de Vries B (2008) Structural analysis and analogue modelling of the
kinetics and dynamics of rockslide avalanches. Geosphere 4:657686
Staron L (2008) Mobility of long-runout rock flows: a discrete numerical investigation.
Geophys J Int 172:455463
Tang CL, Hu JC, Lin ML, Angelier J, Lu CY, Chan YC, Chu HT (2009) The Tsaoling landslide
triggered by the Chi-Chi earthquake, Taiwan: insights from a discrete element
simulation. Eng Geol 106:119
Tang H, Jia G, Hu X, Li D, Xiong C (2010) Characteristics of landslides induced by the
great Wenchuan earthquake. J Earth Sci 21:104113
Tsou C, Feng Z, Chigira M (2011) Catastrophic landslide induced by Typhoon Morakot,
Shiaolin, Taiwan. Geomorphology 127:166178
Van Wyk de Vries B, Self S, Francis PW, Keszthelyi L (2001) A gravitational spreading
origin for the Socompa debris avalanche. J Volcanol Geotherm Res 105:225247
54 Landslides 13 & (2016)
Wang G, Sassa K, Fukuoka H (2003) Downslope volume enlargement of a debris slide
debris flow in the 1999 Hiroshima, Japan, rainstorm. Eng Geol 69:309330
Wang G, Huang R, Chigira M, Wu X, Loureno SDN (2013) Landslide amplification by
liquefaction of runout-path material after the 2008 Wenchuan (M 8.0) earthquake,
China. Earth Surf Process Landf 38:265274
Weidinger JT, Korup O, Munack H, Altenberger U, Dunning SA, Tippelt G, Lottermoser W
(2014) Giant rock slides from the inside. Earth Planet Sci Lett 389:6273
Xu Q, Shang Y, van Asch T, Wang S, Zhang Z, Dong X (2012) Observations from the large,
rapid Yigong rock slidedebris avalanche, southeast Tibet. Can Geotech J 49:589
606
Yin Y (2011) Recent catastrophic landslides and mitigation in China. J Rock Mech Geotech
Eng 3:1018
Yin Y, Xing A (2012) Aerodynamic modeling of the Yigong gigantic rock slide-debris
avalanche, Tibet, China. Bull Eng Geol Environ 71:149160
Yin Y, Zheng W, Li X, Sun P, Li B (2011) Catastrophic landslides associated with the M8.0
Wenchuan earthquake. Bull Eng Geol Environ 70:1532
Zhou JW, Xu WY, Yang XG, Shi C, Yang ZH (2010) The 28 October 1996 landslide and
analysis of the stability of the current Huashiban slope at the Liangjiaren Hydropower
Station, Southwest China. Eng Geol 114:4556
Zhou JW, Cui P, Fang H (2013a) Dynamic process analysis for the formation of
Yangjiagou landslide-dammed lake triggered by the Wenchuan earthquake, China.
Landslides 10:331342
Zhou JW, Cui P, Yang XG (2013b) Dynamic process analysis for the initiation and
movement of the Donghekou landslide-debris flow triggered by the Wenchuan
earthquake. J Asian Earth Sci 76:7084
Zhu P, Wang C, Tang B (2000) The deposition characteristic of supper debris flow in
Tibet. J Mt Sci 18:453456 (in Chinese)
Zhu Z, Ling X, Wang Z, Lu Q, Chen S, Zou Z, Guo Z (2011) Experimental investigation of
the dynamic behavior of frozen clay from the Beiluhe subgrade along the QTR. Cold
Reg Sci Technol 69:9197
J.-w. Zhou ())
State Key Laboratory of Hydraulics and Mountain River Engineering,
Sichuan University,
Chengdu, Sichuan 610065, Peoples Republic of China
e-mail: jwzhou@scu.edu.cn
P. Cui
Key Laboratory of Mountain Hazards and Earth Surface Processes,
Chinese Academy of Science,
Chengdu, Sichuan 610044, Peoples Republic of China
M.-h. Hao
College of Water Resources and Hydropower,
Sichuan University,
Chengdu, Sichuan 610065, Peoples Republic of China
 Original Paper
Landslides (2016) 13:5565 Fujun Niu I Jing Luo I Zhanju Lin I Jianhong Fang I Minghao Liu
DOI 10.1007/s10346-014-0545-2
Received: 30 April 2014
Accepted: 16 December 2014 Thaw-induced slope failures and stability analyses
Published online: 9 January 2015
Springer-Verlag Berlin Heidelberg 2015
in permafrost regions of the Qinghai-Tibet Plateau,
China
Abstract The distribution of permafrost-related slope failures
along the Qinghai-Tibet Highway from Wuddaoliang to
Fenghuoshan correlates with ice content, slope gradient, and
ground temperature. Slope failures are of two types. (1) Retrogres-
sive thaw slumps result from icy permafrost being exposed by
either man-induced excavation or fluvial-thermal erosion and
undercutting of basal slopes. (2) Active-layer-detachment failures
are caused by thaw of icy permafrost at the active layer-permafrost
interface. After initial failure, active-layer-detachment failures can
lead to retrogressive thaw-slumping and localized surficial land-
slide. Common trigger mechanisms for failure include high sum-
mer air temperatures and heavy summer precipitation. A third
possible trigger mechanism for slope failure is earthquake occur-
rence. A geotechnical slope stability analysis was undertaken for
an active-layer-detachment failure that had progressed into a
retrogressive thaw slump. A safety factor (Fs) of 1.24 for the natural
slope was determined using in situ tested strength parameters.
However, the slope would lose stability when either the ground-
water level over the permafrost table exceeded 1.42m or seismic
acceleration reached, or exceeded, 0.03g.
Keywords Permafrost . Active-layer-detachment
failure . Retrogressive thaw slumping . Stability
assessment . Qinghai-Tibet Plateau
Introduction
Most permafrost on the Qinghai-Tibet Plateau (QTP) is continu-
ous and characterized by warm ground temperatures (1 to 0 C)
(Zhou et al. 2000). The region has experienced noticeable warming
during the past 4050 years (Cheng and Wu 2007); for example,
both mean annual ground temperature (MAGT) and active-layer
thickness have increased in the last two decades (Jin et al. 2006,
2008). These changes can lead to slope instability and failure. Such
failures generally occur in the active layer in summer. They are
triggered by the formation of a shear zone near the base of the
active layer that results from thaw of the ice-rich zone at the active
layer-permafrost interface. Thawing is usually attributed to either
high summer air temperatures or extreme precipitation events.
The QTP is also an earthquake-prone region and seismic activity
can accelerate the possibility of slope instability. For example, the
Kunlun Mountain earthquake (magnitude 8.1) on November 14,
2001 attracted significant attention from engineers and scientists.
It follows therefore that an assessment of slope stability is essential
for engineering planning and implementation.
Here, we describe thaw-induced slope failures that have oc-
curred within a 10-km-wide zone along the Qinghai-Tibet Highway
(QTH), from Wudaoliang (35 13 N, 93 4 E) in the north to
Fenghuo Mountain (34 43 N, 92 53 E) in the south (Fig. 1). The
zone includes the Hoh Xil Hill region, the Beiluhe Basin region,
and the Fenghuo Mountain region. Approximately 85 % of the
study area is higher than 4500 m above sea level (ASL). The QTH,
the Qinghai-Tibet Railway (QTR), and other linear engineering
facilities are located along this corridor (Fig. 2). The objectives
are to (1) summarize the distribution and morphological charac-
teristics of thaw-induced slope failures, (2) investigate the thermal
regime and failure development of a typical failure, (3) determine
shear strength parameters in the basal zone of the active layer
using in situ direct shear tests, and (4) assess slope stability in both
its natural state and under the influence of an earthquake.
Study area
Data from the meteorological station in Wudaoliang town (Fig. 2)
shows that both mean annual air temperature (MAAT) and pre-
cipitation have increased during the past 3050 years (Fig. 3). The
MAAT in the study area is lower than 4 C, while the extreme low
is about 30 C at the end of January and highest is approximately
25 C in July. The annual precipitation, about 300 mm, mostly falls
between May and August.
The active-layer thickness ranges between 1.5 and 3.0 m and
near-surface permafrost may contain ice that exceeds 20 % by
volume; this characterizes about 80 % of the permafrost in the
study area (Liu and Wu 2000; Wu et al. 2002, 2004). The MAGT
monitored at a depth of 15 m varies between 3.5 and 0.5 C, with
lowest values in Fenghuoshan and highest in the Beiluhe Basin.
Soil in the study area is mainly loose alpine steppe and alpine
meadow; these support a sparse vegetation cover that is dominat-
ed by monotone community physiognomy, characterized by low
stature (10 to 15 cm), short growth period, and low biomass (Niu
et al. 2011).
Thaw-induced slope failures
Types and morphological characteristics
A total of 42 slope failures were recognized using SPOT-5 imagery
and field investigation within the 10-km lateral zone along the
QTH from Wudaoliang to Fenghuo Mountain pass. The failures
are classified as being either active-layer-detachment slides
(failures) or retrogressive thaw slumps (see French 2007). They
are well known in the North American literature and have been
described from forested, tundra, and high Arctic environments
(Burn and Lewkowicz 1990; Lewkowicz and Harris 2005a, b).
Retrogressive thaw slumps are triggered by the thaw of perma-
frost and the melt of exposed ground ice. This then leads to the
retreat of a steep frozen back scarp, or headwall. Thaw slumps
along the Qinghai-Tibet Engineering Corridor (QTEC) are usually
initiated following excavation of material for use in construction
of local infrastructures (Ma et al. 2006; Niu et al. 2012). They
generally have a width of 20 to 90 m and a length of 40 to 150 m
and occur at concave slope segments. Most are arc-shaped and
usually consist of a series of slumped arcuate blocks. Compression
Landslides 13 & (2016) 55
 Original Paper
Fig. 1 Location of the study area on the Qinghai-Tibet Plateau. Permafrost underlies 75 % of the total area of the plateau. The study site is in the continuous permafrost
zone. (Source: Environmental and Ecological Science Data Center for West China, National Natural Science Foundation of China. http://westdc.westgis.ac.cn/data)
ridges in their lateral and toe zones cannot be observed, and the
sliding distances are often only a few meters. For example,
Fig. 4a, b shows a typical thaw slump on a NW facing 7 slope
developed at K3035 mile of the QTH. The actively retreating
headwall is shown in Fig. 4c. Behind the headwall, a series of
fissures can be observed. Once slumping has occurred, newly
collapsed material pushes earlier collapsed blocks and forms a
draped and superposed structure in the failure zone (Fig. 4d).
The thaw depth in the failure area is about 2.2 m, slightly higher
than the 2.0 m measured in undisturbed natural ground.
In the case of active-layer-detachment failures, the active layer
may slide over the permafrost table for distances as much as
several tens of meters; they constitute localized surficial landslides.
As such, these can be larger than thaw slumps, with widths of 80 to
160 m and lengths of 120 to 280 m. The mean slope angle at failure
sites was 7.6, much lower than that for landslides in non-
permafrost regions of China (Huang 2007). These types of slope
failure are usually arc-shaped or bell-shaped, and often with ver-
tical headwalls and compression ridges across the toe zone. After
failure, and due to thaw of the icy layers at the active layer-
permafrost interface, the headwall may develop into a retrogres-
sive thaw slump.
Geometric measurement
Morphometric variables including horizontal width and length,
slope angle, and length-to-width ratio were calculated for each
slope failure using satellite imagery validated with field observa-
tions. Both the widths and lengths of slope failures showed con-
siderable variability. The mean width and length of the thaw-
induced landslides were 1.52.0 times larger than those of the thaw
slumps. The length-to-width ratios of the features were commonly
between 1 and 3, which demonstrates that the slope failures in the
study area were dominantly compact in morphology rather than
elongated (Fig. 5a).
The long axis orientation was defined as the direction toward
the head scarp of the feature. The slope failures are commonly
56 Landslides 13 & (2016)
oriented in the NW direction (Fig. 5b), indicating that they
developed on SE facing slopes. Since ground ice content does
not vary significantly with aspect, the main reason for the
observed distribution is probably the greater amount of solar
radiation received by SE facing slopes. However, other studies
(Lewkowicz and Harris 2005b; Wang et al. 2009) from the
Mackenzie Valley, Canada, have shown that the majority of
slope failures are concentrated on NE facing slopes where
higher ice contents are found, indicating that in this environ-
ment, solar insolation was not a primary factor influencing
the occurrence of slope failures.
Spatial distribution of thaw-induced slope failures
The spatial distribution of slope failures appears to be closely
related to ice content, slope gradient, and ground temperature
(Fig. 6). As regards to ice content, statistical analysis shows
that the vast majority (94 %) of slope failures occur in areas of
icy permafrost or where massive ground ice occurs. As regards
to slope gradient, failures occur predominantly on slopes of
moderate angle (Fig. 6); for example, about 88 % of the
failures occur on slopes with gradients of between 6 and 10.
However, the occurrence of slope failures in permafrost-free
regions is positively correlated with slope gradient (Akgun
2012; Gnther et al. 2013). As regards to ground temperature,
about 40 % of failures occur in areas where the MAGT is above
0.5 C, 32 % occur in areas with MAGT of between 1.0 and
0.5 C, and 28 % occur in areas where the MAGT is lower
than 1.0 C.
The slope failure at K3035W
A typical active-layer-detachment failure that developed into a
localized surficial landslide was selected for detailed study. It
was first examined on October 12, 2010. It is located on the
west side of K3035 mile of QTH (34 59 N, 92 58 E). To
distinguish it from the K3035 thaw slump, we refer to it as the
K3035W landslide. An aerial view, taken on March 27, 2010
Fig. 2 Topographic map of the study area
(Fig. 4f), shows tensional fissures (red dash line). However,
the landslide occurred sometime before October 12, 2010,
resulting in wheel ruts visible in the landslide mass that had
moved downslope 28 m (Fig. 4g). Therefore, the landslide
probably occurred at the end of September 2010 when the
thaw depth reached its maximum. The size of this failure is
about 230 m in length from the back scarp to the edge of the
compressed area, and 90 m wide in its middle part. A 1.8-m-
high headwall formed after failure occurred; this initiated
retrogressive thaw slumping; retreat of about 9 m between
2011 and 2013 was monitored. The total volume of material
moved was 22,500 m3, several shear fissures formed on the
surface (Fig. 4e), and compression ridge of about 3050 cm
high formed at the toe zone (Fig. 4h). According to borehole
data obtained in September 2012, the thaw depth was 1.8 m in
the undisturbed natural conditions, 1.0 m in the failure plane,
and 2.1 m in the main displaced mass.
Thermal regime
In August 2012, thermistors were installed within three 15-m-deep
boreholes (No. 1, 3, and 4) and one 10-m-deep borehole (No. 2)
along the major axis of the K3035W landslide (Fig. 7). Borehole No.
1 is located in undisturbed terrain, borehole No. 2 in recently
disturbed terrain close to the foot of back scarp, borehole No. 3
in the area between the scarp and the sliding mass at a distance of
30 m from the back scarp, and borehole No. 4 was drilled in the
main sliding mass. Each thermistor string consisted of 20 therm-
istors at 0.5-m interval from the surface down to 10 m and 5
thermistors at 1.0-m interval below 10-m depth. The temperature
measurement accuracy is 0.02 C. The data acquisition system is
made up of a CR3000 data logger and a solar panel recharging a
12-V battery (Fig. 7).
Figure 8 plots MAGT against depth from September 2012 to
September 2013 for the four boreholes. Due to the thermal distur-
bance caused by the landslide, the MAGT in the sliding mass area
Landslides 13 & (2016) 57
 Original Paper
Fig. 3 Climate changes in the study area in the past 50 years. a MAAT in the past 50 years; b mean annual precipitation in the past 30 years
was slightly higher than that in the undisturbed area, and this
difference mainly existed between the 26-m depth. The maximum
differences in MAGT were about 1.3 C (between the scar area and
undisturbed terrain), and 0.5 C (between recently disturbed ter-
rain and undisturbed terrain). Such differences were caused by the
thinner active layer in the failure area than that in the undisturbed
area during the initial period of the landslide occurrence.
In order to further investigate the thermal regime of the land-
slide, ground temperature isotherms as a function of depth and
time, from September 2012 to October 2013, are plotted for the
undisturbed terrain (Fig. 9a), the recently disturbed terrain
(Fig. 9b), the scar area (Fig. 9c), and the sliding mass (Fig. 9d).
As might be expected, the depth to the permafrost table in the
recently disturbed area and the scar area was significantly less than
that in undisturbed terrain; in 2012, the thaw penetration depth
reached only about 100 cm in recently disturbed terrain and the
scar area. However, the thickness of the active layer increased in
2013 and reached 150 cm in depth. With time, it is probable that the
thickness of the active layer will eventually reach, or even exceed,
the active-layer thickness in the adjacent undisturbed terrain.
Failure process
In order to investigate the retrogressive nature of the initial active-
layer-detachment failure (K3035W landslide), 3D laser images of
the area were acquired for October 2011, June 2012, October 2012,
June 2013, and October 2013. Based on these images, morpholog-
ical changes between 2011 and 2013 are shown in Fig. 10. The total
collapsed area was 670 m2 from October 2011 to October 2012,
while it was only 65 m2 from October 2012 to October 2013. This
indicates that the rate of headwall retreat decreased in 2013. This
was confirmed by a number of manual transects that monitored
58 Landslides 13 & (2016)
retreat of the headwall from October 2011 to October 2013. Table 1
shows that the maximal headwall retreat from October 2011 to
October 2012 was 7.4 m at transect No. 4, while maximal retreat
from October 2012 to October 2013 was only 1.6 m, also at the same
transect. In general, both the laser imagery and the transect mon-
itoring indicate that the rate of headwall retreat was slower in the
second year. However, more data over a longer time period are
required before any average rate of headwall retreat can be stated.
Geotechnical properties of the active layer
Grain size analyses conducted on material from within the active
layer indicated that the clay content varied significantly with
depth. The highest amount, 48.7 %, was present in the layer closest
to the permafrost table (Fig. 11). This is consistent with studies of
active-layer-detachment failures in northern Canada (Lewkowicz
and Harris 2005a). Therefore, high clay content near the perma-
frost table may be another important condition facilitating land-
slide initiation in permafrost regions.
Atterberg limits and plasticity indices of soil samples were also
measured (Table 2). Results indicate that liquid limits are consis-
tent with plasticity indices. The maximal liquid limit and plasticity
index was also found in soil closest to the permafrost table;
specific values are 32.12 % and 12.25, respectively. In addition, the
peak shear strengths of undisturbed soil samples collected from
depths of 0.5, 1.0, and 1.7 m and the strength parameters tested by
the direct shear test are given in Table 2.
Shear strength of the basal zone
Active-layer-detachment failures are generally caused by low shear
strength in the basal zone of active layer. Knowledge of the
strength parameters of the material that comprises the active layer
Fig. 4 Photographs of slope failures in the study area. a A typical thaw slump developed at K3035 mileage of the QTH; b satellite image of the thaw slump on March 27,
2010 (acquired from Google Earth: http://www.google.com/earth); c the head scarp of the thaw slump; d the draped and superposed structure in the slumped material; e
a typical active-layer-detachment slide located at the west side of K3035 mileage of the QTH; f satellite image of the failure site on March 27, 2010 (acquired from Google
Earth); g satellite image of the failure on October 12, 2012 (acquired from Google Earth); h compression ridges developed at the toe zone of the failure. The red arrow
represents the direction of mass movement

is essential for accurate stability assessment. Therefore, in situ K3035W landslide. The apparatus used, shown in Fig. 12, consisted
shear tests were conducted at a location near the headwall of the of vertical loading devices, horizontal loading devices, and

Fig. 5 The morphological characteristics of the slope failures in the study area. a Histograms of the length-to-width ratios, b long axis orientation (0=north, 90=east,
180=south, 270=west)

Landslides 13 & (2016) 59

 Original Paper

Fig. 6 Statistical relations between the distribution of slope failures and the ice content, slope gradient, and MAGT (measured at a depth of 15 m from the year 2005 to
2013). Permafrost is classified as ice-poor when ice content is less than 20 %, ice-rich when ice content is between 20 and 50 %, and massive ground ice when ice content
is higher than 50 % (Niu et al. 2002)

Fig. 7 Plan and profile views of the K3035W landslide. a Plan view, delineated landslide regions, and thermistor locations, and b slope profile and borehole stratigraphy

60 Landslides 13 & (2016)


Fig. 8 Mean annual ground temperature profiles from boreholes at the K3035W
landslide site, September 2012 to September 2013
displacement measurement devices. The vertical and horizontal
loads were applied by two lifting jacks with a maximal output force
of 10 kN. The strength levels were measured by two pressure
gauges, which were calibrated in the Lanzhou Jiaoda Engineering
Fig. 9 Two-dimensional ground temperature fields for the boreholes at the K3035W landslide. a Undisturbed area away from the slide, b recently disturbed area, c scar
area, and d displaced mass area. The thick red line represents the 0 C isotherm
Measurement Co., Ltd. before the testing. The vertical and hori-
zontal displacements were measured by four dial indicators with a
precision of 0.01 mm.
Two sets of tests, including eight samples, were undertaken.
The dimensions of each sample were 5050 cm in length and
30 cm in height. In order to reduce disturbance, each sample
was manually exposed by shovel after removal of the top soil
layers. The exposed frozen surface around each sample (i.e.,
the permafrost table) was covered by a 2-cm-thick layer of
sand to reduce quick thawing and to prevent the shear box
from plugging into the icy surface. The small inter-space be-
tween the sample and the shear box was also filled with fine
sand. In addition, the ice content at the active layer-permafrost
interface beneath each sample was measured after testing. The
volumetric ice contents from beneath of all eight samples
varied from 75 to 80 %; this guaranteed little influence of ice
content on the test results.
During testing, five levels were loaded at 5-min time intervals.
Vertical displacement was measured every 5 min. Horizontal stress
was applied when vertical displacement did not exceed 0.01 mm in
5 min. Before horizontal loading started, the maximal shearing
strength was evaluated and averaged. The peak horizontal stress
was recorded when the sample slipped, and residual stress was
acquired at the time when horizontal displacement kept a constant
value.
The strength parameters in the basal zone of active layer were
obtained based on the Mohr-Coulomb theory (Fig. 13) and are
listed in Table 3. In order to guarantee the test result accuracy, we
used the mean values of the shear strength parameters for subse-
quent slope stability calculation.
Landslides 13 & (2016) 61
 Original Paper

Fig. 10 The location of the head scarp at different times. The transects (No. 18) for monitoring the head scarp retreat are also shown. The green numbers refer to the
distance from the scanner to each transect

Table 1 Head scarp retreat distances along eight transects (see Fig. 10 for the location of each transect)
Date (year-month) Retrogressive distance (m)
No.1 No.2 No.3 No.4 No.5 No.6 No.7 No.8
2011-10 0 0 0 0 0 0 0 0
2012-06 0.5 1.0 0.9 0.5 0.4 0 1.3 0
2012-10 1.3 3.1 2.5 7.4 2.3 3.0 3.2 3.5
2013-06 1.3 3.1 2.5 7.9 2.5 3.0 3.2 3.5
2013-10 1.5 3.1 2.5 9.0 3.5 3.1 3.2 3.5

Fig. 11 a the stratigraphy of the active layer and top of permafrost, and b grain size distribution of soils from the test sites

62 Landslides 13 & (2016)

Table 2 Geotechnical properties of the active layer at the K3035W landslide site
Depth (m) Moisture content (%) Liquid limit (%) Plasticity index Shear strength
c
0.5 17.02 28.2 0
1.0 18.18 23.54 6.98 24.1 20.6
1.5 23.65 27.32 8.29
1.7 36.65 32.12 12.25 33.5 28.7

Fig. 12 Pictures of the in situ strength tests. a A sample on the surface of the massive ground ice, and b test setup

Slope stability in natural conditions
Most thaw instability studies in permafrost regions adopt an
infinite slope model because the thaw of icy permafrost will lead
to high pore water pressure at the base of active layer (McRoberts
and Morgenstern 1974). This is generally considered to be the main
trigger for slope failure in permafrost regions (Savigny et al. 1995).
Therefore, slope stability should be assessed using effective stress
analysis. Slope stability can be estimated from the following equa-
tion, assuming that seepage direction is parallel to the ground ice
surface (Niu et al. 2005):
c0 f1m m sat w gzcos2 tan0
Fs
1m m sat zsincos
where Fs is the safety factor, c is effective cohesion, is effective
frictional angle in the basal zone of active layer, is the angle of
slope, sat is unit weight of saturated soil, is unit weight of the
soil above groundwater level, w is unit weight of water, z is
vertical depth of the slip surface, and m is the ratio of the height
of the water table above the slip surface to the depth of the slip
surface (z).
In undisturbed terrain at the site of the K3035W failure, the
angle of slope is 7.5, the unit weight of saturated soil and soil
above the groundwater level are 21.66 and 18.84 kN/m3, respective-
ly, the depth to the permafrost table is 1.8 m, and the level of the
water table above the permafrost table was 0.8 m at the end of
September 2013. Using the strength parameters of the basal zone
obtained from the in situ shear tests (Table 3), the calculated safety
factor (Fs) of the slope was 1.24 when thaw depth reached its
maximum, indicating that the slope is stable in its natural status.
Three factors trigger the instability of slopes in permafrost
1
regions. First, extreme high air temperature in summer, or forest
fires that reduce the thickness of insulating layer, can lead to an
increase in pore water pressure near the permafrost table (Savigny
et al. 1995; Lewkowicz and Harris 2005a). Second, persistent rain-
fall can elevate the water table in the warm season. The calculated
result indicates that the safety factor would reach its critical value
when the water table level increased to 1.42 m. Third, the

Table 3 In situ shear strength parameters in the basal zone of active layer ( and r are effective peak and residual angle of friction, c and cr are peak and residual
cohesion)
() c (kPa) r () cr (kPa)
Test 1 10.8 0.51 9.3 0
Test 2 11.2 0.55 9.8 0
Mean 11.0 0.53 9.6 0

Landslides 13 & (2016) 63

 Original Paper
Fig. 13 Horizontal stress-displacement curves for a test 1 and b test 2
occurrence of an earthquake will result in an instantaneous in-
crease in the sliding force and pore water pressure. This last trigger
factor is discussed below.
Slope instability under the influence of earthquake
In order to assess slope instability on the QTP under an earth-
quake influence, the Fs in the conditions of static equilibrium is
given as follows:
c0 f1m m sat cos2 a1m msat sincosm w cos2 gztan Development Program of Basic Research of China (973 Plan,
Fs
1m m sat zsincos a1m msat zcos2 2012CB026101), and the Foundation for Innovative Research
where the Fs, c, , , sat, , w, z, and m have been mentioned
above, a is the seismic acceleration, and sat and are the densities
of, respectively, saturated soil and soil above groundwater level
(2.21 and 1.92 g/cm3 in the K3035W landslide site).
Based on Eq. 2, the critical value of seismic acceleration for Fs=1
is 0.03g (g=9.8 m/s2) in the K3035W failure site. According to the
Seismic Ground Motion Parameter Zonation Map of China (GB
18306-2001), the seismic peak ground acceleration is 0.1g in the
study area; this is sufficient to promote slope instability of a site
with the same conditions as the K3035W failure site. Furthermore,
the above calculation does not consider the influence of pore water
pressure; this will increase when an earthquake occurs and will
result in more unstable conditions. Therefore, the slope may reach
its failure threshold even when seismic acceleration is less than
0.03g.
Conclusions
Thaw-induced slope failures are widespread on the QTP, espe-
cially along the QTEC. The distribution of failures is closely
related to the occurrence of icy permafrost, slope gradients of
between 6 and 10, and mean annual ground temperatures
(MAGT) that are higher than 0.5 C. Slope failures in the study
area are classified as being either thaw slumps or active-layer-
detachment failures that incorporate retrogressive thaw
slumping of the headwall.
Monitoring of a typical failure indicates that the MAGT and
depth to the permafrost table in the failure area are lower than
similar values in undisturbed terrain. Headwall retreat is highest in
the initial years after failure. The shear strength parameters in the
basal zone of the active layer were determined by in situ shear
tests. Slope stability analysis for the K3035W failure site was
64 Landslides 13 & (2016)
undertaken using in situ tested strength parameters. The slope at
the K3035W site appears to be stable in its undisturbed state but
could reach failure threshold when either the groundwater level in
the active layer is higher than 1.42 m or earthquake acceleration
reaches, or exceeds, 0.03g.
Acknowledgments
This work was supported by the Western Project Program of the
Chinese Academy of Sciences (KZCX2-XB3-19), the State Key
2 Groups of the National Natural Science Foundation of China
(Grant No. 41121061). The authors are indebted to Professor
Fuchu Dai, Institute of Geology and Geophysics, Chinese Acad-
emy of Sciences, and Dr. Sergey Marchenko, University of Alas-
ka Fairbanks, for constructive comments. Professor Hugh
French, Emeritus Professor, University of Ottawa, undertook
final editing.
References
Akgun A (2012) A comparison of landslide susceptibility maps produced by logistic
regression, multi-criteria decision, and likelihood ratio methods: a case study at zmir,
Turkey. Landslides 9(1):93106
Burn CR, Lewkowicz AG (1990) Retrogressive thaw slumps. Can Geogr 34(3):273276
Cheng GD, Wu TH (2007) Responses of permafrost to climate change and their
environmental significance, Qinghai-Tibet Plateau. J Geophys Res 112:F02S03.
doi:10.1029/2006JF00063
French HM (2007) The periglacial environment, 3rd edn. Wiley, UK, p 458
Gnther A, Reichenbach P, Malet JP, Eeckhaut MVD, Hervs D, Dashwood C, Guzzetti F
(2013) Tier-based approaches for landslide susceptibility assessment in Europe.
Landslides 10(5):529546
Huang RQ (2007) Large-scale landslides and their sliding mechanisms in China since the
20th century. Chin J Rock Mech Eng 26(3):433454 (in Chinese)
Jin HJ, Zhao L, Wang SL, Jin R (2006) Modes of degradation and thermal regimes of
permafrost along the Qinghai-Tibet Highway. Sci China D 49(11):11701183, in
Chinese
Jin HJ, Yu QH, Wang SL (2008) Changes in permafrost environments along the Qinghai-
Tibet engineering corridor induced by anthropogenic activities and climate warming.
Cold Reg Sci Technol 53(3):317333
Lewkowicz AG, Harris C (2005a) Morphology and geotechnique of active-layer detach-
ment failures in discontinuous and continuous permafrost, northern Canada. Geo-
morphology 69(1):275297
Lewkowicz AG, Harris C (2005b) Frequency and magnitude of active-layer detachment
failures in discontinuous and continuous permafrost, northern Canada. Permafrost
Periglac 16(1):115130
Liu YZ, Wu QB (2000) Study on ground temperature field in permafrost regions of Zhou YW, Guo DX, Qiu GQ (2000) Permafrost in China. Science Press, Beijing, pp 106404
Qinghai-Tibet Plateau. Highway 2(1):48 (in Chinese) (in Chinese)
Ma W, Niu FJ, Satoshi A, Jin DW (2006) Slope instability phenomena in permafrost
regions of Qinghai-Tibet Plateau, China. Landslides 3(3):260264
McRoberts EC, Morgenstern NR (1974) Stability of slopes in frozen soil, Mackenzie Valley,
N.W.T. Can Geotech J 11(4):554573 F. Niu : J. Luo : Z. Lin : M. Liu
Niu FJ, Zhang JM, Zhang JZ (2002) Engineering geological characteristics and evaluations State Key Laboratory of Frozen Soil Engineering, Cold and Arid Regions Environmental
of permafrost in Beiluhe testing field of Qinghai- Tibetan Railway. J Glaciol Geocryol and Engineering Research Institute,
24(3):264269 (in Chinese) Chinese Academy of Sciences,
Niu FJ, Cheng GD, Ni WK, Jin DW (2005) Engineering-related slope failure in permafrost No. 320, West Donggang Road, Lanzhou, Gansu 730000, China
regions of the Qinghai-Tibet Plateau. Cold Reg Sci Technol 42(3):215225
Niu FJ, Lin ZJ, Liu H, Lu JH (2011) Characteristics of thermokarst lakes and their influence F. Niu
on permafrost in Qinghai-Tibet Plateau. Geomorphology 132(3):222233 Key Laboratory of Highway Construction & Maintenance Technology in Permafrost
Niu FJ, Luo J, Lin ZJ, Ma W, Lu JH (2012) Development and thermal regime of a thaw Region, Ministry of Transport,
slump in the Qinghai-Tibet plateau. Cold Reg Sci Technol 8384:131138 CCCC First Highway Consultants Co., Ltd,
Savigny W, Logue C, MacInnes K (1995) Forest fire effects on slopes formed in ice-rich Xian, 710075, China
permafrost soils Mackenzie Valley, Northwest Territories. 48th Canadian Geotechnical J. Luo ()) : M. Liu
Conference, Vancouver. Canadian Geotechnical Society, Vancouver, pp 989998 University of Chinese Academy of Sciences,
Wang BL, Paudel B, Li HQ (2009) Retrogression characteristics of landslides in fine- Beijing, 100049, China
grained permafrost soils, Mackenzie Valley, Canada. Landslides 6(2):121127 e-mail: luojinabcd@163.com
Wu QB, Liu YZ, Tong CJ (2002) Interactions between the permafrost and engineering
environments in the cold regions. J Eng Geol 8(3):281287 (in Chinese) J. Fang
Wu QB, Cheng GD, Ma W (2004) The impact of climate warming on Qinghai-Tibetan Qinghai Research Institute of Transportation,
Railroad. Sci China D 47(Supp I):122130, in Chinese Xining, 810001, China

Landslides 13 & (2016) 65

 Original Paper
Landslides (2016) 13:6783 Vanessa Wirz I Marten Geertsema I Stephan Gruber I Ross S. Purves
DOI 10.1007/s10346-014-0544-3
Received: 12 March 2014
Accepted: 10 December 2014 Temporal variability of diverse mountain permafrost
Published online: 16 January 2015
Springer-Verlag Berlin Heidelberg 2015
slope movements derived from multi-year daily GPS
data, Mattertal, Switzerland
Abstract In this study, high resolution surface measurements of
diverse slope movements are compared to environmental factors
such as ground surface temperature (GST) and snow cover, in order
to reveal and compare velocity fluctuations caused by changing
environmental conditions. The data cover 2years (20112013) of
Global Positioning System (GPS) and GST measurements at 18
locations on various slope movement types within an alpine study
site in permafrost (Mattertal, Switzerland). Velocities have been
estimated based on accurate daily GPS solutions. The mean annual
velocities (MAV) observed at all GPS stations varied between 0.006
and 6.3ma1. MAV were higher in the period 2013 compared to
2012 at all stations. The acceleration in 2013 was accompanied by a
longer duration of the snow cover and zero curtain and slightly lower
GST. The amplitude (0600%) and the timing of the intra-annual
variability were generally similar in both periods. At most stations,
an annual cycle in the movement signal was observed, with a phase
lag of 14months to GST. Maximum velocity typically occurred in
late summer and autumn, and minimum velocity in late winter and
beginning of spring. The onset of acceleration always started in
spring during the snowmelt period. At two stations located on steep
rock glacier tongues, overprinted on the annual cycle, short-term
peaks of velocity increase, occurred during the snowmelt period,
indicating a strong influence of meltwater.
Keywords GPS . Slope movements . Permafrost .
Rock glacier . Landslide
Introduction
Many mountain ranges, such as the Alps and the Himalayan
Hindu-Kush, have relatively dense populations at risk from natu-
ral hazards related to slope instabilities (e.g. (Heim 1932; Govi
1989)). Since thawing processes reduce the strength of ice-rich
sediments (Morgenstern and Nixon 1971; Harris and Davies 1998)
and frozen jointed bedrock (Gruber and Haeberli 2007; Hasler
et al. 2011), permafrost degradation can develop or accelerate slope
instabilities, such as deep-seated bedrock failures, creep-related
processes, rockfalls, active layer detachment failures or increased
debris flow activity (Harris et al. 2001). Further, if average and
extreme precipitation increase, as predicted for some areas due to
climate change (IPCC 2013), the frequency of rainfall-triggered
slope instabilities may rise (Beniston et al. 1996). Thus, while some
factors controlling slope stability (e.g. topography or lithology)
remain rather constant overtime, others are sensitive to climate
forcing. Melting of surface or subsurface ice may respond rapidly
to climate forcing and may cause unexpected types of slope move-
ments (Gruber 2013). Therefore, anticipated global climate change
is likely to increase instability of rock slopes and movement of ice-
rich debris (Haeberli et al. 1997).
Although, there is still no unambiguous evidence that the fre-
quency or the magnitude of landslides has changed over recent
decades (Huggel et al. 2012), some studies from the Alps point to
an increasing number of rockfalls (Ravanel and Deline 2010),
debris flows (e.g. (Hirschmugl 2003)) and large rock slope failures
(>105 m3, (Fischer 2010)) that initiated in permafrost areas over the
past two decades. Further, for many rock glaciers, an acceleration
(Roer et al. 2008; Delaloye et al. 2013) and for one even a total
collapse (Krysiecki et al. 2008) has been observed and hypothe-
sized to be caused by increasing air temperature (e.g. (Roer 2005;
Delaloye et al. 2010)). Increased thawing at rock glacier tongues,
debris loading of gullies and the development of process chains
could further increase the potential for debris flows and floods
(Haeberli et al. 1990; Zimmermann and Haeberli 1992; Harris et al.
2009; Lugon and Stoffel 2010; Graf et al. 2013). Geertsema et al.
(2006) also noted decadal increases in large landslides in British
Columbia between 1973 and 2003 with most rock slides in their
study initiating in permafrost areas.
Detecting spatial and temporal variation in the movement of
individual slopes is therefore central to our understanding of
underlying slope processes, and ultimately for application in risk
reduction. The timing of acceleration or deceleration of slope
movements has the potential to allow us to identify governing
controls, for example, snowmelt infiltration and ground tempera-
ture. One simple and effective way to study the dynamics of
potentially unstable areas is the observation of surface displace-
ment. To date, most displacement measurements focus on either
spatial resolution and coverage (e.g. (Kb et al. 1997; Strozzi et al.
2012)) with limited time steps or higher temporal resolution,
typically for a small number of punctual measurements (e.g.
(Malet et al. 2002)). To analyse short-term velocity fluctuations,
measurements with a high temporal resolution (e.g. daily) are
required. In addition, effective methods for their interpretation
are needed because many slopes have low rates of displacement (a
few cm per year). Continuous Global Positioning System (GPS)
measurements have proven to be particularly suitable for moni-
toring slope movements with high accuracy over long time periods
and have been utilized to monitor landslides (e.g. (Malet et al.
2002; Baldi et al. 2008)) and rock glaciers (e.g. (Limpach and
Grimm 2009)).
Alpine slope movement types range from slow continuous
displacement of entire slopes, where the main process is creep,
to rapid, catastrophic landslides, with varied movement types
often coexisting on the same slope or in close proximity. We use
the term cryosphere-related mountain slope movements (CM-
movements, adapted from (Wirz et al. 2013)) to refer to slope
movements that (a) occur on moderate to steep mountain slopes
(angle: 1550) and (b) are related to cryosphere, i.e. strongly
influenced by occurrence or changes to permafrost, seasonal frost,
glacier debuttressing and/or snow. Studies on short-term variabil-
ity of CM-movements are few and mainly focused on one type,
such as rock glaciers (e.g. (Haeberli 1985; Krainer and He 2006;
Landslides 13 & (2016) 67
 Original Paper
Perruchoud and Delaloye 2007)), rock slides (e.g. (Nordvik et al.
2010)), active layer detachment slides (e.g. (Lewkowicz 2007)) or
deep-seated landslides (e.g. (Coe et al. 2003)). The few available
observations of short-term (intra-annual) variability of slope
movements have shown that velocities are typically either constant
(e.g. (Krainer and He 2006)) or have a seasonal variability resem-
bling a sine function (e.g. (Coe et al. 2003; Nishii and Matsuoka
2010; Delaloye et al. 2010)), though short periods with increased
velocities might exist (e.g. (Buchli et al. 2013)). Due to the range of
processes involved in downslope movement, the temporal velocity
characteristics of individual landforms may differ, even where they
are classified as the same geomorphological landform (e.g. rock
glacier, (Delaloye 2010)).
Our aim in this study was to explore the relationship between
multiple, high resolution GPS time series of permafrost slope
movements to a range of environmental factors including near-
surface ground temperature (GST), air temperature, rain and snow
cover duration. We calculate velocities from daily GPS positions,
using a method developed to allow extraction of velocity signals
from high resolution GPS even for very small displacements (Wirz
et al. 2014), measured at 18 stations located on eight landforms
including rock glaciers and composite landslides.
In this paper, we addressed the following research questions:
& What are the main patterns of short-term temporal variability
of movement observed within the study site?
& Do measured slope movements differ from those expected
based on the geomorphological setting?
& How do changes in environmental conditions affect the differ-
ent types of slope movements?
Study site
General description
The study site is located above the villages Herbriggen and Randa
on the orographic right side of the Mattertal in Canton Valais,
Switzerland (Fig. 1). It spans 6 km north-south and 1.5 km east-
west and includes slopes of the peaks Breithorn, Gugla, Dirruhorn
and Grabenhorn that are mainly west-facing and have average
slope angles of approximately 30. Elevation ranges from 2600 to
3200 m a.s.l. Permafrost is likely to exist beneath much of the
study area, especially above 2850 m a.s.l. (Boeckli et al. 2012). The
main lithology is gneiss belonging to the crystalline Mischabel unit
(200150 mya, (Labhart 1995)). In most places, bedrock is debris
covered, originating from in situ weathered bedrock, morainal
deposits and colluvium. Vegetation is sparse to nonexistent in
the study area. Due to various slope movements in the study site
and the large amount of unconsolidated material, the amount of
debris deposited in gullies is increasing; thus, the related risk of
down valley debris flows (Dikau 1996; Graf et al. 2013).
In the Mattertal region, the reconstruction of prehistoric glacier
stages is difficult because of the steep valley walls and frequent
mass movements (Hantke 1980). Based on field visits and discus-
sions with an expert (M. Maisch), two interpretations of the ice
level during the Last Glacial Maximum (LGM, 2619 ky BP) are
possible: the ice elevation either reached an elevation of approxi-
mately (a) 3100 m a.s.l. or (b) 2600 m a.s.l. Case (a) seems
plausible, because the upper slopes (e.g. close to Dirruhorn) are
68 Landslides 13 & (2016)
much more precipitous and heavily gullied than the slopes found
in the study area. Case (b), however, is in line with the findings by
Kelly (2003) who assumed that in this area, during LGM, the ice
elevation reached 26002800 m a.s.l. Further, a valley shoulder
at about 2600 m a.s.l. can be identified in the entire study area.
Concerning more recent glaciations, several moraines in the study
area, e.g. adjacent to the Dirru and Steintlli rock glacier (Fig. 1),
indicate that those valley bottoms were probably ice-covered until
the Younger Dryas (YD, approximately from 12.8 to 11.5 ka BP;
(Leith 2012)). Further, during the Little Ice Age (LIA, 1850), the
forefield of Dirru glacier and the upper part of Steintlli was
glacier-covered down to an elevation of approximately
3000 m a.s.l. (Maisch 1999).
The recent climate in the study area can be described as inner-
alpine continental. High mountains on both sides shield the
Mattertal, leading to lower precipitation and higher temperatures
compared to other inner-alpine meteorological stations located in
Switzerland (e.g. Andermatt, Davos, Montana). The closest weath-
er station is located at Grchen (1550 m a.s.l., 9 km away from the
study site). The mean annual sum of precipitation in Grchen is
only 523 mm (19611990), the mean annual air temperature
(MAAT) is 5.2 C (19611990). Over the measurement period of
this study (20.08.201119.08.2013), the MAAT in Grchen was ap-
proximately 1 warmer and the mean annual sum of precipitation
was approximately 100 mm higher compared to the long-term
climatic mean (19611990). The MAAT measured at the study area
is about 1.2 C (at 2697 m a.s.l., mean over the measurement period
of this study). The hydrological year 20121 was warmer (+1.2 C)
and dryer (43 mm), and the duration of the snow cover was
shorter (16 days) compared to the hydrological year 2013 (data
derived from the MeteoSwiss station in Grchen).
Measurement locations and geomorphological setting
The GPS stations were placed on various landforms, e.g. rock
glaciers with different morphologies as well as deep-seated com-
posite landslides. Eight units (Fig. 1 and Table 1) have been selected
based on mapping in the field and on aerial photographs, consul-
tation with federal authorities (personal communication of Dr.
Hugo Raetzo, Federal Office for the Environment), as well as
previous observations (Winterhalter 1959; Delaloye et al. 2008b;
Strozzi et al. 2009a, b; Delaloye 2010). For the selection of the
measurement locations, two criteria were considered: Firstly, min-
imal estimated displacements must be of the order of a decimetre
per year (Fig. 1), in order to obtain reliable displacements with
GPS. Secondly, clear geomorphological signs of movement must be
visible in the field (Table S1 in supplementary material). Within
each unit, one to three well-grounded boulders were chosen as
measurement locations. For each unit, the main characteristics are
summarized in Tables 1 and S1 (in supplementary material). The
abbreviation RG in unit names denotes rock glacier.
The unit Breithorn (L1) includes a huge terraced slope (1545,
Fig. 2) with evidence of downslope extension (Table S1): Cracks
and voids, which are not filled with sediments, double-ridges
toward the top (antiscarp gull), some lobate tongues and exposed
planar bedrock rupture surfaces. The unit can be described as a
deep-seated complex landslide. We assume that the rupture sur-
face (or surfaces) dips more or less parallel to the terrain surface.
1
01.10.201130.09.2012
Fig. 1 Overview of study site and main instrumentation. Estimated velocity range are low few decimetres per year, medium several decimetres per year and high more
than a metre per year. This was estimated based on field observations, expert knowledge (Dr. Hugo Raetzo, Federal Office for the Environment), and published
observations (Winterhalter 1959; Delaloye et al. 2008b; Strozzi et al. 2009a, b; Delaloye 2010). Background: orthophoto from the year 2009 (reproduced by permission of
swisstopo BA14032)

Water flow is expected to occur in the subsurface, where perma-
frost and bedrock may act as flow barriers. The slope of Breithorn
supplies four main debris flow initiation zones with debris (Bhler
and Graf 2013). Within the unit Breithorn, three permanent GPS
stations were installed: L1a is located on a double-ridge (height:
1050 m, length: 100 m) close to the top of Breithorn at an
elevation of 3140 m a.s.l. L1b is mounted on a huge block
(250 m3, at 2973 m a.s.l.) in a less steep zone. L1c is located at
2870 m a.s.l. in a zone with comparably finer surface material and
more vegetation.
The unit RG Breithorn (R2) includes several distinct lobes with
characteristic flow patterns of a rock glacier. The lower part of the rock
glacier has a convex profile curvature: approximately 100 m above the
front, the slope suddenly increases from about 20 to 35. In this area,

Landslides 13 & (2016) 69

 Original Paper
Table 1 Characteristics of GPS stations by terrain unit (cf. Fig. 1)
Unit Name GPS stations Elevation range (m a.s.l.) Area (m2) Mean slope ()
L1 Breithorn L1a, L1b, L1c 26003178 600000 30 (L1a: 24, L1b: 21, L1c: 22)
R2 RG Breithorn R2a, R2b 25502950 105000 24 (R2a: 25, R2b: 34)
L3 Lngenschnee L3a, L3b 25702980 249100 30 (L3a: 30, L3b: 32)
R4 RG Lngenschnee R4a 27503170 63600 32 (R4a: 16)
L5 Gugla L5c, L5b, L5a 27503200 220000 32 (L5a: 18, L5b; L5c: 30,)
R6 RG Steintlli R6b, R6a 29503100 75600 21 (R6a: 20, R6b: 11)
R7 RG Dirru R7a, R7b, R7c 23702970 92950 25 (R7a: 17, R7b: 23, R7c: 33)
L8 Grabengufer L8a, L8b 28502980 19800 30 (L8a: 15, L8b: 25)
Mean slope angle is calculated from a digital elevation model with a cell size of 25 m
RG rock glacier

Fig. 2 Overview and local context of instruments placed. The crack (at 2660 m a.s.l.) on the orographic left side of unit RG Breithorn is approximately 20 m long and
0.5 m wide

70 Landslides 13 & (2016)

transverse fissures (Fig. 2) and depressions indicate extension. Based
on annually repeated GPS measurements, Delaloye et al. (2013) ob-
served that the annual surface velocities at the tongue of the rock
glacier had increased from 2.8 ma1 in 2007/08 to more about 7 ma1
in 2011/12. This fast phase likely started before 1995 and persisted until
the time of writing (Delaloye et al. 2013). The tongue of the rock glacier
reaches a gully with a high debris flow potential (Fig. 2). There is no
stream visible at the surface of this unit, but in spring and summer, a
water source forms at the front of the rock glacier. GPS station R2a is
located on the more gentle part of the rock glacier (2704 m a.s.l.), R2b
on the steep tongue of RG Breithorn (2650 m a.s.l.).
The unit Lngenschnee (L3) includes a terraced slope and can
roughly be divided into three zones: (A) The uppermost zone,
dominated by finer materials, has sparse vegetation cover and
several solifluction lobes. (B) The middle zone has three open
transverse fractures (Fig. 2) as well as several voids (pockets).
The fractures are approximately 1 to 2 m wide, 720 m long and
up to several metres deep. All fractures are either in bedrock or
reach bedrock, mostly filled with debris. (C) The lowest zone
forms a depression with large angular blocks (several m) and few
fines exist. Toward the upper and lower limit of this zone, the slope
angle drastically increases to more than 45. Both rotational sliding
and toppling occur here. We assume that water flow in the unit
Lngenschnee occurs in the subsurface. Water emerges at the
surface further down at an elevation between 2400 and
2600 m a.s.l. GPS station L3a was installed directly below the
lowest fracture (zone B, 2794 m a.s.l.) on rather solid rock, L3b
in zone C (2610 m a.s.l.).
The unit RG Lngenschnee (R4) is located in a gully between the
units Lngenschnee and Gugla (Fig. 2). The feature has the typical
form of a rock glacier with a steep front and lateral margins. Above
the rock glacier, a head scarp with a height of about 50 m extends
over its entire width. It was potentially caused by a rotational slide,
which predates 1955 (Winterhalter 1959). The rock glacier surface
has several elongated ridges and depressions, indicating extension.
We assume that those depressions and ridges are caused by the
slope movement, but the depressions could also be related to
bedrock topography (e.g. at 2850 m a.s.l.) and the ridges might
result from rockfall deposits. The GPS station R4a is mounted on
the tongue of the rock glacier (2792 m a.s.l.). The surface next to
the GPS station is gentle (slope angle of 15), but 20 m uphill, the
slope angle is steeper than 35, similar to the front 30 m downslope.
The unit Gugla (L5) includes a huge west-exposed slope of
Gugla (3380 m a.s.l.). Slope movements here can be roughly divid-
ed into three parts: (a) The movement at the top of the unit seems
to be a rockslide (strongly fractured rock mass). Arcuate fractures
as well as fresh extension zones in dark weathered coarse rubble
are visible in a strongly fractured rock mass. The surface shows
signs of shallow movement, such as frost-boil features and angled-
up stones that may add to the deeper underlying movement. This
might also indicate a sorting (based on texture) of the near sub-
surface. (b) In the middle part of the slope, a well-defined scarp of
about 50 m height and 250 m width exists (Fig. 2). Here, the main
processes are rockfall and toppling. As no clear deposit is visible,
we assume that the rockfall from this scarp is an ongoing contin-
uous process rather than one single large event. (c) In the lower
part of the slope, movement can be described as a mixture of
sliding and flowing mainly in rubble. Transverse fractures and
voids not infilled by surrounding fines indicate longitudinal
extension. GPS station L5a was located about 100 m above the
scarp (zone A). Stations L5b and L5c (Fig. 2) were installed in the
lower part of zone B on ridges, which mainly consists of blocks
and/or strongly rugged and weathered bedrock with infill of fines.
The unit RG Steintlli (R6) includes two active superimposed
rock glaciers (Fig. 2). Both features have typical steep fronts and a
ridge-furrow structure indicating compressive flow. Both rock gla-
ciers may originate from ice-cored moraines. The unit was glacier-
covered in the YD (Leith 2012). The top feature blends into a trimline
(on the orographic right side at 3170 m a.s.l.) that most probably
originates from the YD. GPS station R6a (3020 m a.s.l.) was located
on the upper, R6b (2991 m a.s.l.) on the lower rock glacier.
The unit RG Dirru (R7) has a length of more than 3.3 km and
includes various rock glacier lobes and fronts (Fig. 2), originating
from different generations (Delaloye et al. 2013). Potentially, some
of those nested lobes were originally lateral moraines that formed
during the YD. Today, inactive as well as active features with high
movement rates exist. The lobes on the orographic left side (with
more debris input and less solar radiation) are still active. Their
slope has a convex profile and slope angle increases toward the
lower part of the unit to more than 30. The lobe in the steep part
on the orographic left side currently seems destabilised and has
mean annual velocities of about 5 to 7 ma1 (max. activity in 2008/
09, (Delaloye 2010; Delaloye et al. 2013)). Its front already col-
lapsed in some parts in the recent past. At this front, water
emerges in spring and summer. Based on former photographs, it
was found that the actual acceleration phase of its frontal part
started progressively during the 1970s and 1980s and that the
origin of the destabilization of the entire rock glacier seems to be
older (back to LIA, (Delaloye et al. 2013)). GPS station R7a was
mounted on a longitudinal ridge (at 2772 m a.s.l.) in the upper flat
and shadowed part of the rock glacier. R7b is on a steeper but
rather inactive lobe (at 2706 m a.s.l.). R7c was installed on a small
longitudinal ridge at 2673 m a.s.l. more or less in the middle of the
steep and fast moving tongue.
The unit Grabengufer (R8) is located on the orographic left side
at the west-exposed slope of Grabenhorn (3193 m a.s.l., Fig. 2).
Within the unit, various ridges and depressions indicate a rota-
tional slide. Rock surfaces (light, unweathered) at the lower limit
of the unit imply recent rockfall activity (toppling). In September
2010, a rockfall (5000 m3) occurred at the lower limit of this unit
(Bhler and Graf 2013). L8a is located in a depression zone (at at
2882 m a.s.l.) in the middle of the unit. GPS station L8b was
mounted 70 m below L8a on a ridge (at at 2868 m) at the lower
limit of the unit. Directly below L8b, the slope angle increases to
more than 40.
Instrumentation
The deployed GPS stations (e.g. Fig. 2) were designed for high-
mountain environments (Beutel et al. 2011; Buchli et al. 2012) and
comprise of a low-cost single-frequency GPS receiver (ublox LEA-
6T) and a two-axis inclinometer (SCA830-D07, VTI Technologies
(2010)). Energy was provided by a photovoltaic system and a
battery. The stations were installed on large boulders assumed to
be carried along with the displacement of the entire slope. None-
theless, an inference has to be made from these point measure-
ments at the surface to extrapolate behaviour at depth and across a
larger area around the measurement. For continuous monitoring,
GPS antennae must be positioned higher than the expected snow
Landslides 13 & (2016) 71
 Original Paper
depth to prevent signal loss. Therefore, GPS antennas were mounted
on a mast (hmast =0.51.5 m). The set-up allowed continuous mea-
surements of position and mast inclination with high temporal
resolution, temporal coverage of several years and high accuracy
(sub-cm accuracy, (Buchli et al. 2012; Wirz et al. 2013)).
GST was measured at each station using five miniature tempera-
ture loggers (iButtons DS1922L, c.f., (Gubler et al. 2011)). The iButtons
were distributed within a radius of about 7 m around the GPS station,
covering a range of aspects. The devices were buried a few
centimetres below the surface to prevent exposure to direct sunlight.
If the surface material consisted of large boulders with no fine grains,
the iButtons were distributed in small voids between the boulders.
Further, a weather station (WXT520, (Vaisala 2012)), mounted next
to the GPS reference station, provided information about air tem-
perature, liquid precipitation and wind.
Data and methods
GPS and inclinometer data
GPS data were collected between summer 2011 and summer 2013
(Table 2). Due to energy loss, some stations have data gaps
(Table 2), lasting around 30 days, with few exceptions (e.g. L8a:
data gaps >100 days). For one station (R7a), only data until
summer 2012 exists. Daily GPS solutions are provided in the Swiss
coordinate system CH1903. Solutions were calculated with a static
approach at the Geodesy and Geodynamics Lab of ETH Zurich.
Therefore, all daily measurements are used to compute a single
and highly accurate daily solution based on a single-frequency
differential carrier-phase technique applying the software Bernese
(Limpach and Grimm 2009; Dach et al. 2007; Buchli et al. 2012).
The standard deviation of the GPS solutions and their covariances
were estimated based on the precision of the daily GPS solutions
(Wirz et al. 2014). The standard deviations of the three compo-
nents of the GPS positions (E , N , mah ) were typically about
1.5 mm in the horizontal and about 3.5 mm in the vertical, and
covariances were small (< 0:01 mm).
Having the GPS antenna mounted on a mast makes the mea-
surement system more sensitive to local rotations, which can be
misinterpreted as translations of the slope. Measuring mast incli-
nation in combination with the GPS allowed separation of trans-
lation and rotation and, thus, correction of GPS positions for mast
tilt (Wirz et al. 2014). However, these corrections could only be
made where tilting is greater than the uncertainty of the inclinom-
eter measurements (Wirz et al. 2014). Inclinometer measurements
have the same temporal resolution (daily) and coordinate system
(CH1903) as the GPS data. They were calculated from raw incli-
nometer measurements with a temporal resolution of 5 min (see
(Wirz et al. 2014), for further details). Standard deviations in
inclination () were typically around 0.4, and around 2.5 for
the orientation (az) of the mast tilt. The corresponding covari-
ances were relative small (10
2
10
5 ). Note that for R7a and R7c,
only inclinometer data until summer 2012 exist.
The data period 2012 includes data from 20.08.2011 to 19.08.2012;
the period 2013 includes the data from 20.08.2012 to 19.08.2013.
Auxiliary data
GST measurements have a temporal resolution of 3 h and were
aggregated to daily means. The accuracy of the measurements is
specified by the manufacturer to be 0.5 C, but found to be
72 Landslides 13 & (2016)
smaller near 0 C (0.125 C, (Gubler et al. 2011)). Based on GST
data, the duration of the temporal pattern of the snowmelt was
quantified by calculating the duration of the zero curtain, which
starts with the basal-ripening date (RD) and ends with the melt-
out date (MD), applying the approach of Schmid et al. (2012). In
this study, the term zero curtain refers to the duration of the zero
curtain effect, which is the effect of latent heat in maintaining
temperatures near 0 C over extended periods in freezing or
thawing soils (e.g. (Outcalt et al. 1990)). Measurements of iButtons
that were found at the surface are excluded from all analysis. Mean
annual GST (MAGST2012, MAGST2013) was only calculated if at
least 360 measurements exist within a year. Both, MAGST and zero
curtain period are calculated for the individual stations (average
over the iButtons that are located next to the stations) and for the
entire study site (including all iButtons within the study site).
Air temperature and rain intensity measurements were aggre-
gated to daily mean values based on measurements with a tempo-
ral resolution of 2 min. Accuracy was specified by the
manufacturer to be 0.3 C (around 0 C) for air temperature
and 5 % for rain intensity measurements (Vaisala 2012).
Velocity estimations
Movement parameters, such as velocity, must be estimated using a
set of GPS coordinates and are thus strongly dependent on the
selected time window (number of measurement points, (Jerde and
Visscher 2005; Laube and Purves 2011)). GPS measurements of
slope displacements with high temporal resolutions (e.g. daily)
typically have a low signal-to-noise-ratio (SNR) (e.g. (Vieli et al.
2004; Dunse et al. 2012; Massey et al. 2013)). To calculate mean-
ingful movement parameters, a method is required that ensures
sufficient movement between consecutive estimates (Jerde and
Visscher 2005) since data with a low SNR cannot support reliable
estimates (Laube and Purves 2011). In this study, the method
signal-to-noise thresholding (SNRT, (Wirz et al. 2014)) was applied
to calculate the magnitude of the velocity (v, in this paper referred
to as velocity) and the direction of movement (aziv ). Compared to
other methods that have been applied to estimate velocities based
on GPS data, e.g. smoothing over several days (Dunse et al. 2012),
Spline (e.g. (Copland et al. 2003)) or Kernel Regression Smoothing
(KRS, e.g. (Vieli et al. 2004)), SNRT is suitable especially for data
with a high noise level and variable SNR (Wirz et al. 2014): The
advantage of the SNRT method is that a displacement signal is
only detected where it actually exists (without oversmoothing
large displacements), while with other methods, the temporal
variability during periods of constant linear displacement is often
overestimated. In addition, an important quality of SNRT com-
pared to the other methods is that calculated velocities stay within
the range of the data and derived velocities represent average
velocities representative for a given period. If velocities and, thus,
the SNR are high, obtained velocities have a high temporal reso-
lution and peaks are not smoothed out. For periods with small
velocities (and low SNR), the smoothing window is larger, which
allows separation of the signal from the noise and thus enhances
the reliability of the estimated velocity. This is important when
velocity variations are used to infer processes.
SNRT adaptively calculates a smoothing window based on the
SNR of the position data. For each velocity period, the SNR must
be higher than a predefined threshold (t). The SNR of the position
data is estimated using Monte Carlo Simulation (MCS, maximum
 Table 2 Characteristics of individual GPS stations
Unit pos Numberh From To disptot MAV1213 MAV2012 MAV2013 VAR12, 13 MAGST2012 MAGST2013 Snow2012 Snow2013 zc2012 zc2013
(m) (ma1) (ma1) (ma1) (%) (C) (C) (d) (d) (d) (d)
L1 L1a 653 20.08.11 18.08.13 0.354 0.177 0.169 0.230 36 0.71 0.60 199 199 8 9
a a a
L1 L1b 584 20.08.11 25.06.13 0.401 0.217 0.186 0.322 73 0.44 0.52 221 216 64 46
L1 L1c 482 23.02.12 25.06.13 0.237b 0.177b 0.167b 0.183b NA NA 0.66 0 186 0 12
R2 R2a 535 23.02.12 19.08.13 2.319c 1.559c 1.355c 1.663 NA NA 0.93 0 218 0 48
g g
R2 R2b 669 20.08.11 19.08.13 12.623 6.525 5.881 7.209 23 1.38 1.09 180 189 32 25
L3 L3a 663 20.08.11 19.08.13 0.062 0.031 0.030 0.043 43 1.74 0.39 166 206 46 46
L3 L3b 524 25.02.12 19.08.13 0.273c 0.184c 0.129c 0.212 NA NA 0.20 0 191 0 32
R4 R4a 664 20.08.11 19.08.13 0.063 0.032 0.028 0.046 67 0.09 0.67 0 0 0 0
L5 L5a 649 20.08.11 19.08.13 0.011 0.006 0.005 0.010 98 0.61 0.55 216 238 17 54
L5 L5b 438 20.08.11 19.08.13 0.078 0.039 0.019 0.073 293 0.18 1.15 200 237 16 13
L5 L5c 587 20.08.11 19.08.13 0.084 0.042 0.038 0.061 62 0.66 0.74 210 223 11 33
R6 R6a 607 20.08.11 19.08.13 1.033 0.516 0.437 0.738 69 0.26 0.31 214 230 7 67
R6 R6b 664 20.08.11 19.08.13 0.419 0.209 0.195 0.275 41 0.53 0.37 217 244 21 49
d d d
R7 R7a 277 20.08.11 16.07.12 1.229 1.355 1.355 NA NA 0.31 0.45 0 0 0 0
R7 R7b 434 20.08.11 06.08.13 0.009 0.004 0.003 0.010 276 0.71 1.20 205 223 18 57
R7 R7c 458 20.08.11 19.08.13 8.541 4.270 4.235 5.975 41 0.38 0.15 206 225 23 57
e e e
L8 L8a 630 29.09.11 19.08.13 0.578 0.306 0.274 0.335 22 0.75 0.96 230 250 22 47
f f f f
L8 L8b 570 29.09.11 25.06.13 0.698 0.401 0.352 0.454 29 0.46 0.66 195 224 6 36
Zero curtain and snow cover duration refer to the median of all iButtons near a GPS station. Only for R2b, measurements are corrected for mast tilt. For all other stations, total rotation is smaller than its uncertainty.
From/To measurement period, disptot total displacement, MAV1213 mean velocity for the entire measurement period; MAV2012/MAV2013 mean annual velocities, VAR1213 inter-annual variability, MAGST2012/MAGST2013 mean annual
GST, Snow2012/Snow2013 duration of an insulating snow cover, zc2012/zc2013 duration of the zero curtain
a
Refers to measurements from 20.08.2011 to 25.06.2013
b
Refers to measurements from 23.02.2012 to 25.06.2013
c
Refers to measurements from 23.02.2012 to 19.08.2013
d
Refers to measurements from 20.08.2011 to 16.07.2012
e
Refers to measurements from 29.09.2011 to 19.08.2013
f
Refers to measurements from 29.09.2011 to 25.06.2013
g
At the antenna: MAV12
13 6.312; VAR12,13=39
h
Number of daily GPS fixes obtained

Landslides 13 & (2016)

73
 Original Paper
2000 realizations). The stability of the results is tested after every
additional 250 MCS realizations. If the standard deviation of the
SNR (SNR ) over the last 250 realizations is smaller than 0.08, the
MCS is stopped. Velocities are calculated based on linear fits
through daily positions as a function of time.
The suitable threshold t that affects the duration of the smooth-
ing windows is chosen empirically. Velocity estimations with dif-
ferent thresholds are (a) summarized in box plots, and (b) the sum
of the differences of aziv to the median of aziv is calculated.
Thresholds t of 5, 10, 15, 20, 30, 40, 50 or 60 are applied. We then
argue similar to (Laube and Purves 2011) that for suitable thresh-
olds, between the results calculated with different t, no significant
difference in (a) the calculated velocities and (b) variations in aziv
exists, because the velocity estimations are no longer affected by
the uncertainty in the data. Besides, for suitable thresholds, the
variations in aziv should be minimal.
Mean annual velocities (MAV12 , MAV13 ) are given as total
displacement divided by time and were only calculated if within
the first and last 10 days of the period at least one data point
existed. MAV12
13 refers to the mean velocity over the entire 2-
year measurement period. Inter-annual variability between the
periods 2012 and 2013 was calculated with respect to the mean
annual velocity in 2012 (MAV12 ), which allowed us to compare
inter-annual variability of all stations including a wide range of
velocities (Mean annual velocities). The intra-annual variability
in velocity of each GPS station was calculated individually for the
two periods 2012 and 2013 and is given as relative deviation to the
mean annual velocity of this period.
Results and interpretation
This section starts with a brief paragraph on parameters applied
for the velocity calculations. In the following sections, we then
describe the main findings concerning the MAV (Mean annual
velocities), as well as the inter-annual (Inter-annual variability
between 2012 and 2013) and intra-annual (Intra-annual
variability) velocity variations.
Between 500 and 2000, Monte Carlo realizations were needed
to obtain stable results (SNR 0.08) for velocity and SNR. For a
few velocity periods after 2000 realizations, SNR was still above
0.08, but always below 0.046, which is still very small compared to
a threshold of 40 and thus a SNRv above 40. With a threshold of
40, estimated velocities for all GPS stations seem to be no longer
affected by uncertainties in position data (Fig. S2S20 in supple-
mentary material) and this value was used for all results, unless
otherwise indicated. To correct velocities for a tilt of the mast is
only possible where tilting is greater than the uncertainty of the
inclinometer measurements (Wirz et al. 2014). This is only the case
for the stations R2b2; hence, MAV of R2b were corrected for the tilt
of the mast. For velocities with higher temporal resolution, veloc-
ities derived from the GPS positions of the antenna and those
corrected for the mast tilt have a nearly equal temporal variability
and differences are very small ( 1.4 %, q95 1.3 %3). Hence, the accord with the weather station data (MAAT and snow height)
2
Also for R6c, the tilt is grater than its uncertainty, but inclinom-
eter data are only available until July 2012.
3
Mean and 95-quantile of the absolute relative difference between
the horizontal velocity of the antenna and the foot (in comparison
to the mean velocity of the antenna), (v f
va )(100/vantenna ) for
R2b with the largest mast tilt (incmax 33 ).
74 Landslides 13 & (2016)
tilt of the mast has negligible influence on the results of this study.
Therefore, for better comparison between stations, velocities
refer to horizontal velocities at the GPS antenna, unless oth-
erwise noted.
The main movement patterns visible in the velocity data are
summarized in Figs. 3 and 4.
Mean annual velocities
The mean annual velocities over the entire data period (MAV12
13 )
of the GPS stations varied between 0.004 and 6.3 ma1 over the
entire data period (Table 2). MAV12
13 4.3 ma1 was measured at
two GPS stations (R7c, R2b), located on rock glacier tongues
within the units RG Breithorn and RG Dirru. MAV12
13
0.04 ma1 was observed at five GPS stations, located within the
units Gugla, Lngenschnee, RG Lngenschnee and RG Dirru. MA
V12
13 between 0.1 and 0.6 ma1 was measured at eight stations
within the units Breithorn, Lngenschnee, RG Steintlli and
Grabengufer.
The variability of MAV12
13 between stations within one unit
was the largest for the units RG Breithorn (4.8 ma1) and RG Dirru
(4.3 ma1. Within the unit RG Steintlli, the upper station (R6a)
was 0.31 ma1 faster than the lower one (R6b). GPS stations within
the units Breithorn and Grabengufer have mean velocities between
0.2 and 0.4 ma1, and variations between the stations are relative
small ( 0.1 ma1). In comparison to the low MAV12
13 ( 0.18
ma1) within the units Gugla and Lngenschnee, the relative var-
iability of MAV12
13 between the stations within one unit are
comparably high ( 0.02 ma1).
Inter-annual variability between 2012 and 2013
Inter-annual variability ranged from 20 to 290 % (Table 2). At all
stations (with data for both periods) velocities increased in 2013
compared to 2012. The greatest acceleration from 2012 to 2013
occurred at L5b (290 %) and R7b (280 %). However, absolute
velocities were very small at both stations (MAV12
13 0.04 ma1)
and thus even small changes in MAV have a large influence. The
lowest ( 40 %) inter-annual variability was observed at L1a and
R2b even though the absolute increase in mean annual velocity in
2013 was largest at R2b (2.1 ma1). An insulating snow cover and a
zero curtain period could be detected based on GST measurements
at 13 stations for the period 2012, and at 16 stations for the period
2013. For R7a and R4a, GST data suggest that in both periods, the
snow cover was not thick enough for insulation. For the period
2012, insufficient GST measurements were available at three sta-
tions (L1c, R2a, L3b). Compared to 2012, 2013 was characterized by
a slightly lower MAGST (0.3 C, averaged over all iButtons with
data for both periods) and a longer snow cover duration (8 days,
averaged over all iButtons in the study site with an insulating snow
cover in both periods). In addition, the zero curtain lasted about
4 weeks longer in 2013 than 2012. These in situ measurements thus
described in General description). For those stations with GPS
and GST data in both periods (N=11), at seven stations, MAGST
was lower, and at four stations, it was higher in 2013 than in 2012
(Table 2). However, at nearly all stations (except for L1a), the
duration of an insulating snow cover was longer in 2013, and for
seven stations, the zero curtain period was correspondingly
longer in 2013.
 Inter-annual vari-

Intra-annual vari-

Acceleration dur-

than
Number of full

spring accelera-

Additional short
peak

during

Apparent influ-
ence of strong
during snowmelt
ing snowmelt
Annual cycle

snowmelt
smoother
Positions

Decrease

velocity
ability

ability
MAV
years

Max
tion

rain
R2b 2 4 1 4 4 4 4 4 4 3
R7c 2 4 2 4 4 4 4 3 3 3
R7a 1 4 0 3 4 41 4 1 2 0
R2a 1.5 3 0 2 4 4 3 3 2 3
R6b 2 3 2 3 4 3 3 2 1 0
R6a 2 4 3 2 4 3 2 2 2 0
L8a 2 3 0 4 4 3 2 2 1 0
L8b 2 3 0 4 4 3 2 1 1 0
L1b 2 3 3 2 3 3 0 3 3 0
L1a 2 2 1 1 3 3 0 2 2 0
L3a 2 1 2 3 3 3 0 0 3 0
L3b 1.5 2 0 4 2 3 0 3 2 0
L1c 1.5 2 0 1 2 31 0 0 2 0
R4a 2 1 3 1 2 2 0 0 2 0
L5b 2 1 4 2 2 0 0 0 0 0
L5f 2 2 3 2 0 0 0 0 0 0
L5a 2 1 4 1 0 0 0 0 0 0
R7b 2 1 4 1 0 0 0 0 0 0
1: snowmelt inferred not based on GST, but on webcam images

Fig. 3 Summary of subjectively classified movement characteristics. MAV, inter-annual variability and intra-annual variability are classified based on quantiles: 1 <25 %-
quantile, 2 <50 %-quantile, 3 <75 %-quantile, and 4 >75 %-quantile. For the other main movement patterns, the following classification is used: 1 clearly absent (absence
of movement pattern is clearly visible), 2 likely absent (absence of movement pattern is visible with some doubt in entire period or visible in only part of the period), 3
likely present (movement pattern is visible with some doubt in entire period or visible in only part of the period), 4 clearly present (movement pattern is clearly visible)
and 0 no statement possible (velocity data not available or do not allow statement). In addition, for each GPS station, the number of full years (data period) is given. The
horizontal velocities in comparison to environmental factors that have been used for this classification are shown in Figs. 7, 8 and 9 and S32S49 (in supplementary
material)

Fig. 4 Intra-annual variability of all stations, expressed as deviation from the mean annual velocity within each period. Stations within the same unit have equal colour.
Horizontal grey dashed lines indicate 50 % and +50 deviations. Insert plot shows absolute horizontal velocities. The vertical solid grey line shows the
separation between period 2012 and period 2013

Landslides 13 & (2016) 75

 Original Paper
Intra-annual variability
On average, the amplitude of the intra-annual variability of the
horizontal velocities is around 30 % (median of all stations, Fig. 4).
The greatest intra-annual variability was measured at R7c (2012
597 % and 2013 591 %). High intra-annual variability was also
observed at R2b (2012169 % and 2013 116 %) and L3b (only data
for 2013148 %). At R6b, L8a and L8b the intra-annual variability
was between 50 and 80 % in both periods. No intra-annual vari-
ability could be detected at the slow-moving R7b and L5a. For all
other stations, the intra-annual variability was rather small (7
37 %) in both periods.
Differences in the amplitude of the intra-annual variability
between 2012 and 2013 are generally small. In general, for all
stations with data for both periods and measurable intra-
annual variability, the timing of maximum and minimum veloc-
ities was rather similar in both periods (e.g. R2b and R2a in
Fig. 5). For those stations, where the temporal resolution of the
velocity estimates (duration of velocity periods) allows a state-
ment about the timing of acceleration and deceleration in the
two periods (R7c, R6a, L8a, L8b, R2b and R2a), observed spring
acceleration and autumn/winter deceleration were generally later
in 2013 than in 2012 (e.g. R2b and R2a in Fig. 5).
An annual cycle in the movement signal can be detected in the
horizontal velocity data of 11 GPS stations (Figs. 3 and 4). For eight
stations, this annual cycle is pronounced (R2b, R7c, R7a, R2a, R6b,
R6a, L8a, L8b, L1b). The lowest velocities tend to occur in late
winter and the beginning of spring (Mar.Apr.), and the highest
velocities at the end of summer or autumn (Sept.Dec., Figs. 6, 7
and 9). The phase lag of the minimum and maximum velocities,
compared to GST, is between 1 and 4 months (Fig. S21S31 in
supplementary material). For four stations (L3b, L1c, R4a, L5b)
an annual cycle was not observed, and for the remaining three
stations (R7b, L5a, L5c), intra-annual variability was too small to
detect any cyclical behaviour (Fig. 3). At all stations with a visible
annual cycle, spring acceleration commenced during the snowmelt
period (Figs. 3, 7, 8 and 9). For two further stations (L1c and L3b)
with no visible annual cycle, spring acceleration also appears to
start during snowmelt. For five stations (R2b, R7c, R7a, R2a, R6b)
with a visible annual cycle, winter deceleration appears smoother
than the spring acceleration (Figs. 3, 7 and 8). For six stations, this
can either not be confirmed (N=3) or was not the case (N=3, Fig 7).
At two stations (R7c, R2b), two differing movement regimes are
observed (Figs. 3 and 8): As well as an annual cycle, an additional
Fig. 5 Comparison of the intra-annual variability of the periods 2012 and 2013 for
R2b and R2a located within unit RG Breithorn (deviation to mean annual velocity
of the respective period)
76 Landslides 13 & (2016)
short peak in velocity occurred during the snowmelt period. The
highest velocities were measured during those short peaks, but
within the snowmelt period velocities then decreased. At three
other stations, an additional peak in velocity occurred during the
snowmelt period, but with maxima still in late autumn (Figs. 3
and 9). The absence of an additional velocity peak during the
snowmelt period is visible for six stations. At three stations,
peaks in velocity likely occurred during/after periods with strong
rain (e.g. beginning of July Figs. 3 and 7). However, for most
stations (N=15), the temporal resolution of the velocity estimates
is insufficient to allow a definitive statement about a link be-
tween typically short-lived precipitation events and velocity.
Discussion
This section starts with a discussion of the approach taken. Then,
results are discussed in view of the three research questions
outlined in Introduction.
Advantages and limitations of our approach
The temporal resolution of the velocity estimations has a strong
influence on observed short-term velocity variations. Certainty
about observed movement patterns increases with higher temporal
resolutions in velocity estimations (cf. Fig. 3). Coarse resolutions
of estimated velocities make the detection of movement patterns
difficult or impossible. Importantly, by applying SNRT, we could
adapt the size of the smoothing window and thus temporal
resolution of velocity estimations to the SNR of the data (Ve-
locity estimations). Overall, the high temporal resolution of our
measurements over a period of 2 years, allowed us to both
confirm existing observations and insight on the short-term
variability of slope movements, such as higher velocities in years
with a thicker snow cover (Ikeda et al. 2008) and on the other
hand to detect patterns that would not be visible based on
measurements with a coarser resolution (e.g. (Kb 2005;
Perruchoud and Delaloye 2007)) or over shorter time periods
(e.g. (Krainer and He 2006)).
Difference of GPS velocities from expectations based on morphology
For most stations, velocities derived from GPS are similar to
expectations based on surface morphology and previous measure-
ments (e.g. (Delaloye et al. 2008b; Strozzi et al. 2009a; Strozzi et al.
2009b; Delaloye 2010; Delaloye et al. 2013)). However, at the sta-
tions in units RG Lngenschnee, Lngenschnee and Gugla (except
at L3b) observed velocities are much lower than expected
(MAV12
13 0.04 ma1). The tongue of RG Lngenschnee, for ex-
ample, was located approximately 100 m further upslope in 1959
than it is today (at 2860 m a.s.l., (Winterhalter 1959)), implying
that the rock glacier had much faster phases then than today
(0.03 ma1). Further, and in contrast to R7b (RG Dirru,
0.004 ma1), RG Lngenschnee lacks typical signs for inactive
rock glaciers such as vegetation cover on its front (Wahrhaftig and
Cox 1959; Haeberli 1985; Ikeda and Matsuoka 2002). The unit
Gugla also appears to have experience more rapid movement in
the recent past than today: a hut, built in 1959 in the middle of the
unit at 2860 m a.s.l. (Winterhalter 1959), has been destroyed,
apparently by ground movement (open voids in the rubble below
the hut and fractures in the floor of the hut are visible). On the
other hand, observed velocities in unit RG Breithorn are nearly
Fig. 6 Intra-annual variability of selected stations, expressed as deviation from the mean velocity of the entire data period (20.08.1119.08.13). Stations within the same
unit are plotted in the same colour. Horizontal grey dashed lines indicate 50 % and +50 %- deviations. A sine function with a period of a year, visually fitted to
match the minima and maxima in velocity, is plotted in black. The vertical solid grey line shows the separation between period 2012 and period 2013

1 ma1 (at R2a), higher than determined in a previous study measurements on rock glaciers in the European Alps, where typ-
(Strozzi et al. 2009b). Nevertheless, the observed velocity at R2a, ical movement rates of 0.1 to 2 ma1 have been reported (for active
as well as at R6b, R6a and R7a, agree well with older rock glaciers, (Barsch 1996; Haeberli 1985)). By contrast, unusually

Fig. 7 Comparison of the horizontal velocities of L8b (unit Grabengufer) to environmental factors. Top horizontal velocity, periods with a SNR below t (t=40) are
indicated in light blue. Middle GST measured with iButtons near GPS station (grey individual, red average) and air temperature measured at the study site (green).
Zero curtain period (orange) and insulating snow cover (light blue) also estimated from individual iButtons. These point measurements are not necessarily
representative for the larger area around the station. Bottom rain intensity (black, measured at the study site) and snow height (blue, measured at Grchen). Insulating
snow cover (blue, earliest in grey) zero curtain periods (dark orange, latest in grey) as derived from all iButtons in the study area (N=90, median value)

Landslides 13 & (2016) 77

 Original Paper

Fig. 8 Comparison of the horizontal velocities of R2b (unit RG Breithorn) to environmental factors. Top horizontal velocity, periods with a SNR below t (t=40) are
indicated in light blue. Middle GST measured with iButtons near GPS station (grey individual, red average) and air temperature measured at the study site (green).
Zero curtain period (orange) and insulating snow cover (light blue) also estimated from individual iButtons. These point measurements are not necessarily
representative for the larger area around the station. Bottom rain intensity (black, measured at the study site) and snow height (blue, measured at Grchen). Insulating
snow cover (blue, earliest in grey) zero curtain periods (dark orange, latest in grey) as derived from all iButtons in the study area (N=90, median value)

high velocities (MAV12
13 4.3 ma1) were observed at the steep
fronts of RG Breithorn and RG Dirru. Both rock glaciers have
previously been identified as exceptional rapidly moving
(MAV 5 ma1, (Delaloye et al. 2013)). Cracks (Fig. 2) and fast
changes at their steep fronts further indicate a destabilization of
those landforms (cf., (Delaloye et al. 2010)). Currently, 15 such
rapidly moving (destabilized) rock glaciers have been identified
in the European Alps (Delaloye et al. 2010). As expected, measured
velocities on slow-moving landslides (0.0070.401 ma1) are gen-
erally lower than on rock glaciers, but are within the wide range of
measured surface velocities in previous studies (e.g. (Strozzi et al.
2005; Coe et al. 2003; Baldi et al. 2008; Brueckl et al. 2006; Strozzi
et al. 2010)). However, velocities measured within the units
Grabengufer (3140 cma1) and Breithorn (1822 cma1 are high
when compared to previously observed velocities of a few to
several cma1 on a deep-seated landslides in permafrost, located
in a neighbouring valley (Saas valley, CH; (Strozzi et al. 2005)).
Generally, no correlation between observed slope movements
and slope angle was found, although at the fastest two sta-
tions (R2b and R6c), slope angles are highest (R2b 33 and
R6c 34, Table 1).
Patterns of temporal variability
In this section, the main movement patterns of the inter-annual
and intra-annual variabilities found in this study (Results and
interpretation) are discussed and compared to previous studies.
We also discuss similarities and differences between landslides and
rock glaciers.
At all sites with data for both periods, MAV13 were higher than
MAV12 (3969 %). This observed change accords well with the
measured acceleration (448 %) in the hydrological year 2013
compared to 2012 at seven rock glaciers in the Valais and
Bernese Alps made by the Swiss permafrost monitoring net-
work (PERMOS 2014).
We detected three main intra-annual patterns of movement
in rock glaciers and landslides: (a) an annual cycle, (b) a
decrease in winter that is smoother than the rapid acceleration
in spring and (c) additional short and strong peak(s) in spring
(during snowmelt period, Fig. 3). By contrast to previous
studies (e.g. (Krainer and He 2006)), no continuous displace-
ment (without intra-annual variability) was observed at sta-
tions with sufficient temporal resolution to derive intra-annual
variations.

78 Landslides 13 & (2016)

Fig. 9 Comparison of the horizontal velocities of R2a (unit RG Breithorn) to environmental factors. Top horizontal velocity, periods with a SNR below t (t=40) are
indicated in light blue. Middle GST measured with iButtons near GPS station (grey individual, red average) and air temperature measured at the study site (green).
Zero curtain period (orange) and insulating snow cover (light blue) also estimated from individual iButtons. These point measurements are not necessarily
representative for the larger area around the station. Bottom rain intensity (black, measured at the study site) and snow height (blue, measured at Grchen). Insulating
snow cover (blue, earliest in grey) zero curtain periods (dark orange, latest in grey) as derived from all iButtons in the study area (N=90, median value)

1. Annual cycle. An annual cycle was observed at various stations,
including both rock glaciers and landslides (Fig. 3). The timing of
the minima (Apr.Mar.) was similar for stations on landslides
and rock glaciers and agrees well with previous studies on rock
glaciers (e.g. (Lambiel et al. 2005; Perruchoud and Delaloye 2007;
Buchli et al. 2013)), as well as on landslides in permafrost (Shan
et al. 2013) and seasonal frost (Coe et al. 2003). However, the
observed timing maxima in velocity at stations on landslides
(Nov.Dec.) differ from previous observations on landslides in
seasonal frost (Shan et al. 2013) or permafrost (Coe et al. 2003),
where the maximal velocity was found in spring or early summer.
By contrast, the timing of maxima in velocity on rock glaciers
(Sept.Oct.) that occurs 1 to 3 months earlier than on landslides
agrees well with previous observations on rock glaciers (Haeberli
1985; Roer 2005; Kb et al. 2005; Perruchoud and Delaloye 2007;
Buchli et al. 2013), although differences between individual rock
glaciers and from year to year have been observed (maxima in
velocity between late summer and beginning of winter have been
observed in the Alps, (Delaloye et al. 2010)).
2. Smooth winter decrease/rapid spring acceleration. For stations
on rock glaciers (except for R6a), the decrease in winter was
smoother compared to the acceleration in spring (Figs. 3, 8 and
9). Such an asymmetric annual cycle seems typical for rock
glaciers (Krummenacher et al. 2008; Perruchoud and Delaloye
2007; Delaloye et al. 2010; Buchli et al. 2013), but was not
always detected in previous studies (e.g. on Muragl rock gla-
cier, (Arenson et al. 2002; Kb 2005)) most likely due to
coarser temporal resolutions (monthly).
3. Short-term peaks. At two stations located on steep fast-flowing
rock glacier tongues (R7c and R2b with MAV4.3 ma1), short
velocity peaks occurred in spring, in addition to the annual
cycle. The highest velocity during those peaks was more than
three times higher than the maximum velocity in late summer
(cf. maxima of annual cycle). These peaks occurred during
snowmelt and related zero curtain period, but mostly termi-
nated (slowdown) before the end of the zero curtain period.
On a rock glacier in a nearby valley, similar short peaks in
velocity during the snowmelt period, with the highest velocities
exceeding 140 % of the summer maximum, were detected
based on borehole-inclinometer data with a temporal resolu-
tion of 6 h (Turtmann valley, CH; (Buchli et al. 2013)). This
rock glacier has high velocities (> 3ma
1 ) and visible signs of

Landslides 13 & (2016) 79

 Original Paper
destabilizations (Buchli et al. 2013) similar to the tongues or
RG Dirru and RG Breithorn.
The observed annual cycle of slope movements has previously
been described as sinusoidal (cf. (Arenson et al. 2002; Kb 2005;
Buchli et al. 2013)) and approximated with a sine function (cf.
(Kb et al. 2007)). In this study, we found that for all stations, the
inter-annual variability differs significantly from a smooth sine
function with a period of a year (Fig. 6), because either the time lag
between the minima and maxima in velocity is more than
6 months, the deceleration in winter is smoother than the accel-
eration or, in addition to the annual cycle, peaks in velocity occur
during the snowmelt period.
Effects of environmental conditions
In this section, the influence of environmental factors is discussed,
first for the inter-annual and then for the intra-annual variability
of slope movements.
Acceleration in 2013 compared to 2012 was accompanied by a
slightly lower MAGST and a longer duration of the snow cover and
zero curtain period (median over all iButtons in the study area
with GST data for both periods). The longer duration of the zero
curtain in 2013 occurred, because in 2013, snowmelt started early in
April, but in May, the weather was cold and wet (with several
snowfall events), followed by intense snowmelt due to warm
weather in June. Previous studies on rock glaciers have also re-
ported higher velocities in years with a thicker snow cover and
thus more meltwater in spring (e.g. (Ikeda et al. 2008; Bodin et al.
2009)). Further, a positive correlation between annual velocities
and MAGST was observed at various rock glaciers (Hoelzle et al.
1998; Ikeda et al. 2003; Lambiel et al. 2005; Perruchoud and
Delaloye 2007; Delaloye et al. 2010; Buck and Kaufmann 2010;
PERMOS 2013), but with a phase lag of between 1 year
(Perruchoud and Delaloye 2007; Delaloye et al. 2010; Buck and
Kaufmann 2010) and several months (Delaloye et al. 2010). In this
study, a possible phase lag between changes in MAGST and the
inter-annual velocity changes cannot be investigated, because we
only have 2 years of data. Further, GST refer to point measure-
ments close to the stations and, thus, are not necessarily represen-
tative for the entire landform, as spatial variability in GST may be
high within short distances (Gubler et al. 2011). In addition, the
small difference in MAGST (0.3 C) between the two periods is
close to the accuracy of the GST measurements and at individual
stations both higher and lower MAGST were measured (mean over
iButtons located next to the station). Today, it is generally assumed
that the inter-annual changes in rock glacier velocities relate to
external climatic factors (e.g. MAGST) rather than internal ones
(e.g. slope), because a rather homogenous and synchronous be-
haviour of the inter-annual variability over several years was
observed at 16 rock glaciers of variable size and morphology
located in different regions of the European Alps (Delaloye
et al. 2008a).
Similarities in the intra-annual variability, both in the timing
and the amplitude, of deep-seated landslides (unit Grabengufer
and Breithorn) and rock glaciers (RG Breithorn, RG Steintlli and
RG Dirru, Fig. 6) suggest that intra-annual variability is indeed
driven by external factors (e.g. meteorological forcing) rather than
by internal characteristics (composition, depth of shear horizon,
cf. (Delaloye et al. 2008a; Buck and Kaufmann 2010)). Further, it
80 Landslides 13 & (2016)
seems that both rock glaciers and deep-seated landslides respond
similarly to external forcing, for example GST or water infiltration.
Nonetheless, we also observed clear differences, both in the timing
and magnitude of intra-annual movement variability, between
stations on rock glaciers and landslides.
For most rock glaciers, we observed a rapid and strong accel-
eration in spring compared to the more or less smooth gradual
deceleration in winter (Figs. 3, 8 and 9). Those observations indi-
cate that for rock glaciers, the response time of movement to
external factors is longer in winter than in spring. In winter, a
decrease of the ground temperature occurs, sometimes influenced
by a snow cover. In spring, in addition to the increase in ground
temperature, meltwater infiltration and, thus, an increase in the
pore water pressure might play an important role. By contrast, the
deceleration in winter was similar to the acceleration in spring at
stations on landslides. Potentially, this is because the meltwater
does not immediately reach the rupture surfaces of the landslides,
which are assumed to sit at greater depths, because the phase lag
between the annual cycle of velocity and GST for landslides was
mostly 1 to 2 months longer (34 months) than for rock glaciers
(12 months). A greater depth of the rupture surface leads to
a comparably longer reaction time to changes at the surface
(e.g. GST or water infiltration). Further, if the rupture surface
lies below the active layer depth, the permafrost might act as
a flow barrier exerting control on permeability and water
pressure, and ground ice-melt might delay the warming of
the ground. This might also explain why the observed maxima
in velocity for stations on landslides in the study site was
about 6 months later compared to a landslide situated in
seasonal frost (no permafrost, (Coe et al. 2003)) or a shallow
landslide in permafrost (Shan et al. 2013). Nonetheless, at
most stations (N=13), including landslides, spring acceleration
started during the snowmelt period, which would rather point
to an influence of meltwater.
The influence of snowmelt infiltration on velocity variations is
clearly visible for two stations (R7c and R2b, Fig. 8) located on
steep rock glacier tongues, as short-lived, high velocity, peaks
occur during the snowmelt period. Such an immediate and sudden
increase in velocity might indicate a shear horizon at shallow
depths, as was observed for the rapidly moving Furggwanghorn
rock glacier (Buchli et al. 2013) and was also assumed to be the case
for Hinteres Langtalkar rock glacier (Avian et al. 2009). Buchli
et al. (2013) found based on borehole measurements that sharp
peaks in surface velocity in spring are caused by a strong increase
in displacement rates within the active layer (4-m depth), possibly
caused by melt water infiltration forming a water table on top of
the permafrost.
The influence of heavy rain on velocity variations is uncertain
and only seemed apparent for three stations (Figs. 3 and 8) with
relatively high velocities and thus a sufficiently high temporal
resolution in velocity estimations. Correlation of velocity varia-
tions in deep-seated movements with precipitation data is gener-
ally difficult without knowing the general hydrological condition
an entire catchment and the resulting fluctuation in pore water
pressure at the shear surface (Terlien 1998; Baldi et al. 2008).
Furthermore, velocities of deep-seated landslides are most likely
driven by long-lived precipitation events (with moderate intensity)
rather than intense short-lived rainfall events (with moderate
intensity, (Crosta 2004)) and the time lag between rain events
and increased velocities might be several months (Baldi et al.
2008). This leads to difficulties in linking rain events to deep-
seated movements as the separation of other influencing fac-
tors (e.g. GST) becomes difficult. Nonetheless, although a
relation between intense rainfall events and increased veloci-
ties was observed for only three stations, rainfall events can-
not be excluded as an influencing factor on the short-term
variability of other slopes.
Conclusions
We compared high resolution velocity estimations over 2 years
(20112013) for 18 GPS stations, situated on various slope move-
ment types to environmental factors (GST, snow (-melt), and rain).
Velocities are estimated based on accurate daily GPS solutions
(hor 1.5 mm) using the SNRT method (Wirz et al. 2014), which
adapts the smoothing window to the SNR of the data.
The main findings of this study are the following:
& Mean annual velocities during the measurement period ob-
served at all GPS stations varied between 0.006 and 6.3 ma1,
with the highest and lowest velocities measured on rock
glaciers.
& Measured velocities mostly agree well with expectations
based on the geomorphological setting and previous
measurements.
& The observed acceleration in 2013 compared to 2012 at all
stations was accompanied by a longer duration of the snow
cover and zero curtain period and a slightly lower MAGST.
& The amplitude (0600 %) and the timing of the intra-annual
velocity were mostly similar for the periods 2012 and 2013.
& An annual cycle in the movement signal was observed at
various stations both on landslides and on rock glaciers, with
maxima between September and December and minima in
March or April.
& Where an annual cycle existed, the acceleration always started
during the snowmelt period, pointing to an influence of melt-
water infiltration.
& The phase lag between GST and velocity was generally longer at
stations on deep-seated landslides (24 months) than on rock
glaciers (12 months).
& For active rock glaciers, we observed a sudden and strong
acceleration in spring compared to the more or less smooth
continuous decrease in winter.
& At two steep rock glacier tongues, the highest velocities were
measured during short peaks in velocity during the snowmelt
period, indicating a strong and immediate influence of
meltwater.
Overall, through the use of well-sited GPS on a number of
landforms, we were able to observe short-term variations in ve-
locities for both deep-seated landslides and rock glaciers. Our
measurements revealed that both similarities, but also clear differ-
ences in the temporal variability, exist not only between different
landform units, but also between stations located on the same
landform. Further, the high resolution and spatial proximity of
our measurements allowed us to demonstrate links between envi-
ronmental processes and patterns of movement. As the length of
time series describing such movements increases, the possibility to
significantly improve our understanding of the physical processes
linked to such landforms through long-term observations is
considerable.
Acknowledgments
This project was funded through nanotera.ch, project X-Sense.
This work was also supported by the Grid Computing Competence
Center (GC3, www.gc3.uzh.ch) with computational infrastructure
and support, including customized libraries (gc_gps and GC3Pie)
and user support. This study would not have been possible with-
out the collaboration with colleagues from the project X-Sense,
notably Jan Beutel (Computer Engineering and Networks Lab of
ETH Zurich) who has led the design and deployment of the
measurement infrastructure, Dr. P. Limpach (Geodesy and
Geodynamics Lab of ETH Zurich) who processed the daily GPS
solutions and Hugo Raetzo (Federal Office for the Environment)
who had provided valuable insight and discussion concerning
slope movements and placement of GPS stations. We thank
Reynald Delaloye (Physical Geography at the University of Fri-
bourg) and Andreas Vieli (3G, Department of Geography of the
University of Zurich) for valuable comments on the manuscript.
The software iAssist (Keller et al. 2010) was used to program and
read-out the iButtons.
References
Arenson L, Hoelzle M, Springman S (2002) Borehole deformation measurements and
internal structure of some rock glaciers in Switzerland. Permafr Periglac Process
13:117135. doi:10.1002/ppp.414
Avian M, Kellerer-Pirklbauer A, Bauer A (2009) Lidar for monitoring mass
movements in permafrost environments at the cirque Hinteres Langtal, Austria,
between 2000 and 2008. Nat Hazards Earth Syst Sci 9:10871094. doi:10.5194/
nhess-9-1087-2009
Baldi P, Cenni N, Fabris M, Zanutta A (2008) Kinematics of a landslide derived
from archival photogrammetry and GPS data. Geomorphology 102(3
4):435444
Barsch D (1996) Rock glaciers. Indicators for the present and former Geoecology in high
mountain environments. Springer Verlag, Berlin
Beniston M, Fox D, Adhikary S, Andressen R, Guisan A, Holten J, Innes J, Maitima J, Price
M, Tessier L (1996) Impacts of climate change on mountain regions. Second
assessment report of the Intergovernmental Panel on Climate Change (IPCC) pp
191213
Beutel J, Buchli B, Ferrari F, Keller M, Thiele L, Zimmerling M (2011) X-sense: Sensing in
extreme environments. In: Proceedings of design, automation and test in Europe,
2011, IEEE, pp 16
Bodin X, Thibert E, Fabre D, Ribolini A, Schoeneich P, Francou B, Reynaud L, Fort
M (2009) Two decades of responses (19862006) to climate by the Laurichard
rock glacier, French Alps. Permafr Periglac Process 20(4):331344. doi:10.1002/
ppp.665
Boeckli L, Brenning A, Gruber S, Noetzli J (2012) Permafrost distribution in the European
Alps: calculation and evaluation of an index map and summary statistics. Cryosphere
6(4):807820. doi:10.5194/tc-6-807-2012
Brueckl E, Brunner F, Kraus K (2006) Kinematics of a deep-seated landslide derived from
photogrammetric, GPS and geophysical data. Eng Geol 88(34):149159.
doi:10.1016/j.enggeo.2006.09.004
Buchli B, Sutton F, Beutel J (2012) GPS-equipped wireless sensor network node
for high-accuracy positioning applications. In: Wireless Sensor Networks,
Springer, pp 179195
Buchli T, Merz K, Zhou X, Kinzelbach W, Springman S (2013) Characterization and
monitoring of the Furggwanghorn rock glacier, Turtmann Valley, Switzerland: results
from 2010 to 2012. Vadose Zone J 12(1):115. doi:10.2136/vzj2012.0067
Buck S, Kaufmann V (2010) The influence of air temperature on the creep behaviour of
three rockglaciers in the Hohe Tauern. In: Grazer Schriften der Geographie und
Landslides 13 & (2016) 81
 Original Paper
Raumforschung.10th International Symposium on High Mountain Remote Sensing
Cartography, vol 45, pp 159170
Bhler Y, Graf C (2013) Sediment transfer mapping in a high-alpine catchment using
airborne LiDAR. In: Graf C (ed) Mattertal - ein Tal in Bewegung, vol 29. Publikation zur
Jahrestagung der Schweizerischen Geomorphologischen Gesellschaft, St. Niklaus, pp
113124
Coe J, Ellis W, Godt J, Savage W, Savage J, Michael J, Kibler J, Powers P, Lidke D, Debray S
(2003) Seasonal movement of the Slumgullion landslide determined from Global
Positioning System surveys and field instrumentation. Eng Geol 68(1):67101.
doi:10.1016/S0013-7952(02)00199-0
Copland L, Sharp M, Nienow P (2003) Links between short-term velocity variations and
the subglacial hydrology of a predominantly cold polythermal glacier. J Glaciol
49(166):337348. doi:10.3189/172756503781830656
Crosta G (2004) Introduction to the special issue on rainfall-triggered landslides and
debris flows. Eng Geol 73:191192. doi:10.1016/j.enggeo.2004.01.004
Dach R, Hugentobler U, Fridez P, Meindl M (2007) Bernese GPS Software. Version 5.0.
Astronomical Institute, University of Bern
Delaloye R (2010) GPS Messungen Mattertal 2010. Departement of Geosciences, Univer-
sity of Fribourg, Fribourg, Switzerland, unpublished report
Delaloye R, Kaufmann M, Bodin X, Hausmann H, Ikeda A, Kb A, Kellerer-
Pirklbauer A, Krainer K, Lambiel C, Mihajlovic D, Roer I, Thibert E (2008a)
Recent interannual variations of rock glacier creep in the European Alps. In:
Proceedings of the 9th International Conference on Permafrost, Fairbanks,
Alaska, USA, pp 343348
Delaloye R, Strozzi T, Lambiel C, Perruchoud E, Raetzo H (2008b) Landslide-like devel-
opment of rockglaciers detected with ERS-1/2 SAR interferometry. In: Proceedings of
the 8th International Conference on Permafrost, 2125 July, Zrich, Switzerland, pp
2630
Delaloye R, Lambiel C, Grtner-Roer I (2010) Overview of rock glacier kinematics research
in the Swiss Alps. Geogr Helv 65:135145. doi:10.5194/gh-65-135-201
Delaloye R, Barboux C, Morard S, Abbet D, Gruber V (2013) Rapidly moving rock glaciers in
Mattertal. In: Graf C (ed) Mattertal - ein Tal in Bewegung. Publikation zur Jahrestagung
der Schweizerischen Geomorphologischen Gesellschaft, St. Niklaus, pp 2131
Dikau R (1996) Untersuchungen zur Murgangaktivitt im Mattertal. In: International
Symposium. Interpreavent 1996 - Garmisch-Partenkirchen. Tagungspublikation, Band
1, pp 397408
Dunse T, Schuler T, Hagen J, Reijmer C (2012) Seasonal speed-up of two outlet glaciers of
Austfonna, Svalbard, inferred from continuous GPS measurements. Cryosphere
6(2):453466. doi:10.5194/tc-6-453-2012
Fischer L (2010) Slope instabilities on perennially frozen and glacierized rock walls: multi-
scale observations, analysis and modelling. PhD thesis, University of Zrich, Zrich,
Switzerland
Geertsema M, Clague J, Schwab J, Evans S (2006) An overview of recent large
catastrophic landslides in northern British Columbia, Canada. Eng Geol 83(1
3):120143. doi:10.1016/j.enggeo.2005.06.028
Govi M (1989) The 1987 landslide on Mount Zandila in the Valtellina, Northern Italy.
Landslide News 3:13
Graf C, Deubelbeiss Y, Bhler Y, Meier L, McArdell B, Christen M, Bartelt P (2013)
Gefahrenkartierung Mattertal: Grundlagenbeschaffung und numerische Modellierung
von Murgngen. In: Graf C (ed) Mattertal - ein Tal in Bewegung, vol 29. Publikation
zur Jahrestagung der Schweizerischen Geomorphologischen Gesellschaft, St. Niklaus,
pp 85112
Gruber S (2013) Landsides in cold regions: making a science that can be put into
practice. In: Margottini Cea (ed) Landslide Science and Practice, Springer, vol 4, pp
329333
Gruber S, Haeberli W (2007) Permafrost in steep bedrock slopes and its temperature-
related destabilization following climate change. J Geophys Res 112:110.
doi:10.1029/2006JF000547
Gubler S, Fiddes J, Gruber S, Keller M (2011) Scale-dependent measurement and analysis
of ground surface temperature variability in alpine terrain. Cryosphere 5(2):431443.
doi:10.5194/tc-5-431-2011
Haeberli W (1985) Creep of mountain permafrost: internal structure and flow of
alpine rock glaciers. Mitt Versuchsanstalt Wasserbau, Hydrol Glaziologie ETH
Zrich 77:5142
Haeberli W, Rickenmann D, Zimmermann M, Rsli U (1990) Investigation of 1987 debris
flows in the Swiss Alps: general concept and geophysical soundings. IAHS Publ
194:303310
Haeberli W, Wegmann M, Vonder Mhll D (1997) Slope stability problems related to
glacier shrinkage and permafrost degradation in the Alps. Eclogae Geol Helv
90(3):407414. doi:10.5169/seals-168172
82 Landslides 13 & (2016)
Hantke R (1980) Eiszeitalter 2: Kalt-/Warmzeit-Zyklen und Eistransport im alpinen und
voralpinen Raum. Ott Verlag, Thun
Harris C, Davies M (1998) Pressures recorded during laboratory freezing and thawing of a
natural silt-rich soil. In: Proceedings of the 7th International Conference on Permafrost
Conference, 2327 June, Nordicana, Yellowknife, Canada, pp 2327
Harris C, Davies M, Etzelmller B (2001) The assessment of potential geotechnical
hazards associated with mountain permafrost in a warming global climate. Permafr
Periglac Process 12(1):145156. doi:10.1002/ppp 376
Harris C, Arenson L, Christiansen H, Etzelmller B, Frauenfelder R, Gruber S, Haeberli W,
Hauck C, Hlzle M, Humlum O et al (2009) Permafrost and climate in Europe:
monitoring and modelling thermal, geomorphological and geotechnical responses.
Earth Sci Rev 92(3):117171. doi:10.1016/j.earscirev.2008.12.002
Hasler A, Gruber S, Haeberli W (2011) Kinematics of steep bedrock permafrost. Journal of
Geophysical Research: Earth Surface (20032012) 117(F1). doi: 10.1029/
2011JF001981
Heim A (1932) Bergsturz und Menschenleben. Fretz & Wasmuth, Zurich
Hirschmugl M (2003) Debris flows in the mountain permafrost zone: Hohe Tauern
National Park (Austria). In: Phillips M, Springman S, Arenson L (eds) Proceedings of
the 8th International Conference on Permafrost, 2125 July, Zrich, Switzerland, pp
413418
Hoelzle M, Wagner S, Kb A, Vonder Mhll D (1998) Surface movement and internal
deformation of ice-rock mixtures within rock glaciers at Pontresina-Schafberg, Upper
Engadin, Switzerland. In: Proceedings of the Seventh International Conference on
Permafrost, pp 465471
Huggel C, Clague J, Korup O (2012) Is climate change responsible for changing landslide
activity in high mountains? Earth Surf Process Landf 37(1):7791. doi:10.1002/
esp.2223
Ikeda A, Matsuoka N (2002) Degradation of talus-derived rock glaciers in the
Upper Engadin, Swiss Alps. Permafr Periglac Process 13(2):145161.
doi:10.1002/ppp.413
Ikeda A, Matsuoka N, Kb A (2003) A rapidly moving small rock glacier at the lower
limit of the mountain permafrost belt in the Swiss Alps. In: Proceedings of the 8th
International Conference on Permafrost, 2125 July, Zrich, Switzerland, 1: 455460
Ikeda A, Matsuoka N, Kb A (2008) Fast deformation of perennially frozen debris in a
warm rock-glacier in the Swiss Alps: an effect of liquid water. J Geophys Res
113(F01021):112. doi:10.1029/2007JF000859
IPCC (2013) Climate change 2013: the physical science basis. In Contribution of Working
Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate
Change. Cambridge University Press, Cambridge, United Kingdom and New
York, NY, USA
Jerde C, Visscher D (2005) GPS measurement error influences on movement model
parameterization. Ecol Appl 15(3):806810. doi:10.1890/04-0895
Kb, A (2005) Remote sensing of mountain glaciers and permafrost creep. PhD thesis,
Geographisches Institut der Universitt Zrich
Kb A, Haeberli W, Gudmundsson G (1997) Analysing the creep of mountain permafrost
using high precision aerial photogrammetry: 25years of monitoring Gruben rock
glacier, Swiss Alps. Permafr Periglac Process 8:409426. doi:10.1002/(SICI)1099-
1530(199710/12)8:4<409::AID-PPP267>3.0.CO;2-C
Kb A, Huggel C, Fischer L, Guex S, Paul F, Roer I, Salzmann N, Schferli S, Schmutz K,
Schneider D, Strozzi T, Weidmann Y (2005) Remote sensing of glacier and permafrost-
related hazards in high mountains: an overview. Nat Hazards Earth Syst Sci 5:527
554. doi:10.5194/nhess-5-527-2005
Kb A, Frauenfelder R, Roer I (2007) On the response of rockglacier creep to surface
temperature increase. Glob Planet Chang 56(12):172187. doi:10.1016/
j.gloplacha.2006.07.005
Keller M, Hungerbuehler G, Knecht O, Sheikh S, Beutel J, Gubler S, Fiddes J,
Gruber S (2010) iAssist: rapid deployment and maintenance of tiny sensing
systems. In: Proceedings of the 8th ACM Conference on Embedded Networked
Sensor Systems, ACM, New York, USA, SenSys10, pp 401402, doi:10.1145/
1869983.1870043
Kelly M (2003) The late Wrmian Age in the western Swiss Alpslast glacial maximum
(LGM) ice-surface reconstruction and 10Be dating of late-glacial features. Ph.d. thesis,
Philosophisch-naturwissenschaftlichen Fakultt der Universitt Bern, Bern,
Switzerland
Krainer K, He X (2006) Flow velocities of active rock glaciers in the Austrian Alps. Geogr
Ann Ser A Phys Geogr Ser A Phys Geogr 88(4):267280. doi:10.1111/j.0435-
3676.2006.00300.x
Krummenacher B, Mihajlovic D, Nussbaum A, Staub B (eds) (2008) 20 Jahre Furggentlti.
Permafrostuntersuchungen auf der Gemmi. Geographica Bernensia G80,
Geographisches Institut der Universitt Bern
Krysiecki J, Bodin X, Schoeneich P (2008) Collapse of the Brard rock glacier (Southern
French Alps). In: Proceedings of the 9th International Conference on Permafrost,
Fairbanks, Alaska, USA, pp 153154
Labhart T (1995) Geologie der Schweiz, 3rd edn. Ott Verlag, Thun
Lambiel C, Delaloye R, Perruchoud E (2005) Yearly and seasonally variations of surface
velocities on creeping permafrost bodies. Cases studies in the Valais Alps. In: 3rd
Swiss Geoscience Meeting, Zrich
Laube P, Purves RS (2011) How fast is a cow? Cross-scale analysis of movement data.
Trans GIS 15(3):401418. doi:10.1111/j.1467-9671.2011.01256.x
Leith K (2012) Stress development and geomechanical controls on the geomorphic
evolution of alpine valleys. PhD thesis, Department of Earth Sciences, ETH Zurich,
Switzerland
Lewkowicz A (2007) Dynamics of active-layer detachment failures, Fosheim Peninsula,
Ellesmere Island, Nunavut Canada. Permafr Periglac Process 18:93103. doi:10.1002/
ppp.578
Limpach P, Grimm D (2009) Rock glacier monitoring with low-cost GPS receivers. In:
Abstract Volume 7th Swiss Geoscience Meeting, November 2009, Neuchatel, Switzer-
land, pp 247248
Lugon R, Stoffel M (2010) Rock-glacier dynamics and magnitude-frequency relations of
debris flows in a high-elevation watershed: Ritigraben, Swiss Alps. Glob Planet Chang
73(34):202210. doi:10.1016/j.gloplacha.2010.06.004
Maisch M (1999) Die Gletscher der Schweizer Alpen: Gletscherhochstand 1850, aktuelle
Vergletscherung, Gletscherschwund-Szenarien. Schlussbericht NFP31, Vdf,
Hochschulverlag AG an der ETH
Malet J, Maquaire O, Calais E (2002) The use of Global Positioning System techniques for
the continuous monitoring of landslides: application to the Super-Sauze earthflow
(Alpes-de-Haute-Provence, France). Geomorphology 43(12):3354. doi:10.1016/
S0169-555X(01)00098-8
Massey C, Petley D, McSaveney M (2013) Patterns of movement in reactivated landslides.
Eng Geol 159:119. doi:10.1016/j.enggeo.2013.03.011
Morgenstern N, Nixon J (1971) One-dimensional consolidation of thawing soils. Can
Geotech J 8(4):558565. doi:10.1139/t71-057
Nishii R, Matsuoka N (2010) Monitoring rapid head scarp movement in an alpine
rockslide. Eng Geol 115(1):4957. doi:10.1016/j.enggeo.2010.06.014
Nordvik T, Blikra L, Nyrnes E, Derron M (2010) Statistical analysis of seasonal displace-
ments at the Nordnes rockslide, northern Norway. Eng Geol 114(3):228237.
doi:10.1016/j.enggeo.2010.04.019
Outcalt S, Nelson F, Hinkel K (1990) The zero-curtain effect: heat and mass transfer across
an isothermal region in freezing soil. Water Resour Res 26(7):15091516.
doi:10.1029/WR026i007p01509
PERMOS (2013) Permafrost in Switzerland 2008/2009 and 2009/2010. Noetzli, J. (ed.),
Glaciological Report (Permafrost) No. 10/11 of the Cryospheric Commission of the
Swiss Academy of Sciences
PERMOS (2014) Key messages on permafrost in Switzerland 2012/2013. unpublished
report by the PERMOS Scientific Committee p 7 pp
Perruchoud E, Delaloye R (2007) Short-term changes in surface velocities on the Becs-de-
Bosson rock glacier (western Swiss Alps). Grazer Schriften Geogr Raumforschung
43:131136
Ravanel L, Deline P (2010) Climate influence on rockfalls in high-Alpine steep
rockwalls: the north side of the Aiguilles de Chamonix (Mont Blanc massif)
since the end of the Little Ice Age. The Holocene 21(2):357365. doi:10.1177/
0959683610374887
Roer I (2005) Rockglacier kinematics in a high mountain geosystem. PhD thesis,
Mathematisch-Naturwissenschaftlichen Fakultt der Rheinischen Friedrich-Wilhelms-
Universitt Bonn
Roer I, Haeberli W, Avian M, Kaufmann V, Delaloye R, Lambiel C, Kb A (2008)
Observations and considerations on destabilizing active rock glaciers in the European
Alps. In: Proceedings of the 9th International Conference on Permafrost, Fairbanks,
Alaska, vol 2, pp 15051510
Schmid M, Gubler S, Fiddes J, Gruber S (2012) Inferring snow pack ripening and melt out
from distributed ground surface temperature measurements. Cryosphere 6(5):1127
1139. doi:10.5194/tc-6-1127-2012
Shan W, Hu Z, Jiang H, Guo Y, Wang C (2013) Mechanism of permafrost landslide based
on GPS and resistivity surveying. In: Progress of Geo-Disaster Mitigation Technology in
Asia, Springer, pp 349361, doi:10.1007/978-3-642-29107-4-18
Strozzi T, Farina P, Crosini A, Ambrosi C, Thring A, Zilger J, Wiesmann A, Wegmller U,
Werner C (2005) Survey and monitoring of landslide displacements by means of L-band
satellite SAR interferometry. Landslides 2:193201. doi:10.1007/s10346-005-0003-2
Strozzi T, Delaloye R, Raetzo H, Wegmller U (2009a) RADAR interferometric observations
of destabilized rockglaciers. In: Proceedings of the Fringe 2009 Workshop, Frascati,
Italy, 30 November 4 December 2009 (ESA SP-677, March 2010), pp 15
Strozzi T, Wegmuller U, Werner C, Wiesmann A, Delaloye R, Raetzo H (2009b) Survey of
landslide activity and rockglaciers movement in the Swiss Alps with TerraSAR-X. In:
Geoscience and Remote Sensing Symposium, 2009 I.E. International, IGARSS, 1217
July 2009, vol 3, pp III53III56, DOI 10.1109/IGARSS.2009.5418175
Strozzi T, Delaloye R, Kaeaeb A, Ambrosi C, Perruchoud E, Wegmller U (2010) Combined
observations of rock mass movements using satellite SAR interferometry, differential
GPS, airborne digital photogrammetry, and airborne photography interpretation. J
Geophys Res 115(F1):F01,014. doi:10.1029/2009JF001311
Strozzi T, Wegmuller U, Werner C, Kos A (2012) TerraSAR-X interferometry for surface
deformation monitoring on periglacial area. In: Geoscience and Remote Sensing
Symposium (IGARSS), 2012 I.E. International, IEEE, pp 52145217
VTI Technologies (2010) Sca830-d07: 1-axis inclinometer with digital SPI interface. Data-
sheet. Doc.nr. 82 823 00 C. wwwvtifi
Terlien M (1998) The determination of statistical and deterministic hydrological
landslide-triggering thresholds. Environ Geol 35(23):124130. doi:10.1007/
s002540050299
Vaisala (2012) Vaisala Weather Transmitter WXT520Ref. Product specification. Doc.nr.
B210417en-J. http://www.vaisalade
Vieli A, Jania J, Blatter H, Funk M (2004) Short-term velocity variations on Hansbreen, a
tidewater glacier in Spitsbergen. J Glaciol 50(170):389398. doi:10.3189/
172756504781829963
Wahrhaftig C, Cox A (1959) Rock glaciers in the Alaska range. Geol Soc Am Bull
70(4):383436. doi:10.1130/0016-7606(1959)70[383:RGITAR]2.0.CO;2
Winterhalter R (1959) Terrainbewegung am Osthang der Gugla, stlich Herbriggen. Tech.
rep., Baudepartement des Kanton Wallis
Wirz V, Limpach P, Buchli B, Beutel J, Gruber S (2013) Temporal characteristics of
different cryosphere-related slope movements in high mountains. In: Margottini C,
Canuti P, Sassa K (eds) Landslide science and practice. Springer, Berlin, pp 383390
Wirz V, Gruber S, Gubler S, Purves RS (2014) Estimating velocity from noisy GPS data for
investigating the temporal variability of slope movements. Nat Hazards Earth Syst Sci
14:25032520. doi:10.5194/nhess-14-2503-2014
Zimmermann M, Haeberli W (1992) Climatic change and debris flow activity in high-
mountain areasa case study in the Swiss Alps. Catena Suppl 22:5959
Electronic supplementary material The online version of this article (doi:10.1007/
s10346-014-0544-3) contains supplementary material, which is available to authorized
users.
V. Wirz ()) : R. S. Purves
Department of Geography,
University of Zurich,
Zurich, Switzerland
e-mail: vanessa.wirz@geo.uzh.ch
M. Geertsema
Ministry of Forests, Lands and Natural Resource Operations,
Prince George, Canada
S. Gruber
Department of Geography and Environmental Studies,
Carleton University,
Ottawa, Canada
Landslides 13 & (2016) 83
 Original Paper
Landslides (2016) 13:8596 Lanhao Zhao I Jia Mao I Xin Bai I Xiaoqing Liu I Tongchun Li I J. J. R. Williams
DOI 10.1007/s10346-014-0552-3
Received: 17 April 2014
Accepted: 30 December 2014 Finite element simulation of impulse wave generated
Published online: 13 January 2015
Springer-Verlag Berlin Heidelberg 2015
by landslides using a three-phase model
and the conservative level set method
Abstract A numerical model has been developed using the finite
element method for the simulation of impulse waves generated by
landslides. The fluid-like landslide is modeled as a generalized
non-Newtonian visco-plastic fluid. A three-phase flow model
based on the incompressible viscous NavierStokes equations is
solved using the finite element method to describe the motion of
the three types of fluid in landslide. The conservative level set
method is expanded to n-phase flow cases and employed to cap-
ture the interface of the three phases: air, water, and the landslide.
The overall performance of the approach is checked by a number
of validation cases: a RayleighTaylor instability problem to illus-
trate the capability of the proposed method to deal with interface
capturing, a benchmark test of a subaerial landslide generated by
an impulse wave is carried out and compared with the published
experimental data and numerical results, and finally, the 1958
Lituya Bay landslide generated impulse wave, and its results are
compared against a scaled-down experiment and other published
numerical results. It can be noted that the current model has an
excellent ability to capture the complex phenomena that occurs
during the whole process of the landslide-generated impulse wave,
and considering the simplified treatment of the landslide and the
numerical model, fairly good agreement between computed and
experimental results has been observed for all simulation cases.
Keywords Landslide . Impulse wave . Three-phase
flow . Conservative level set . Finite element method
Introduction
Landslides represent one of the most destructive hazards in nature,
and the enormous damage associated with this phenomena is not
only due to the large amount of mud and debris material involved
but also because of the potentially large generated impulse waves
(Serrano-Pacheco et al. 2009), also referred to as landslide
tsunamis(Ward 2001). The impulse waves are characterized by a
short wave length and high amplitude and can be more devastating
than meteorological flood waves (Cremonesi et al. 2011) as they can
overtop dams or run-up along the shoreline destroying villages
located along the shore with catastrophic consequences. Numerous
instances have been observed all over the world, and two of the most
devastating examples include the 1958 Lituya Bay case, where a
landslide tsunami destroyed a forest up to a run-up height of
524 m above mean sea level, and the 1963 Vaiont case in Northern
Italy, where an impulse wave overtopped a dam by more than 70 m
and caused about 2000 casualties (Heller and Spinneken 2013). It is,
therefore, of great importance to understand and estimate the be-
havior and characteristics of landslide-induced impulse waves.
The motion of a subaerial landslide is typically divided into three
processes: (1) after triggering, the slide propagates over some distance,
only interacting with air; (2) the slide penetrates the water, while part
of the slide still interacts with air; and (3) the whole slide becomes
submerged in water and its effect on the free surface decreases with
time (Abadie et al. 2010). There are at least three phases, liquid (water),
gaseous (air), and solid (the landslide material), in landslide wave
generation and propagation phenomena. Therefore, to simulate this
problem correctly, a model capable of representing the interface
between all these phases must be employed.
Over the last decades, large numbers of experimental and
numerical studies have been conducted concerning landslide-
generated waves, and both 2D and 3D experiments (Fritz 2002;
Grilli and Watts 2005; Liu et al. 2005; Enet and Grilli 2005; Enet
and Grilli 2007; Walder et al. 2003; Fritz et al. 2004; Panizzo et al.
2005; Heller and Hager 2010; Ataie-Ashtiani and Nik-Khah 2008;
de Carvalho and do Carmo 2007; Di Risio et al. 2009; Shigihara
et al. 2006; Zweifel et al. 2006; Najafi-Jilani and Ataie-Ashtiani
2008) have been devoted to studying the behavior of landslide-
generated impulse waves, such as landslide motion, effective pa-
rameters, wave characteristics, and associated run-up. A
comprehensive review of these studies can be found in
Mohammed and Fritz (2012).
However, in order to help understand the nature of the pro-
cesses involved and predict the detailed outcome in specific loca-
tions, the development of computational tools for the modeling of
water waves induced by landslide seems to be inevitable
(Cremonesi et al. 2011). If there is no impulse wave problem
involved, both continuum and discontinuous method can be
employed to descript the propagation of the landslide such as
the finite element method (FEM), the discrete element method
(DEM), smoothed particle hydrodynamics (SPH), and discontinu-
ous deformation analysis (DDA). One of the main difficulties in
developing a numerical method for simulating waves generated by
landslide movement is to model the strong interaction and energy
transfer between the moving deformable slide and the surround-
ing water and air. This has received enormous effort over the last
several decades; however, due to the complexity of the phenome-
na, different levels of simplification have had to be made in the
various computational models used. Prior to the development of
computational tools, only analytical or empirical formula was
proposed to give rough estimates of the wave amplitude. With
the development of the powerful numerical techniques and in-
creasing computational capacity, a large number of numerical
methods have been proposed based on a variety of models and
hypotheses (Ataie-Ashtiani and Yavari-Ramshe 2011). According to
the degree of simplification of the fluid, the developed models can
be categorized as the Boussinesq type (Ataie Ashtiani and Najafi
Jilani 2007; Hanes and Inman 1985; Lynett and Liu 2002;Watts et al.
2003; Fuhrman and Madsen 2009; Zhou and Teng 2009), models
based on the shallow water equations (Fernndez-Nieto et al. 2008;
Medina et al. 2008; Saut 2003; Chaudhry et al. 1983; Townson and
Kaya 1988), and the potential flow equations (Grilli et al. 2002;
Grilli and Watts 1999, 2005). It should be noted that these analyses
Landslides 13 & (2016) 85
 Original Paper
often did not use any special surface tracking and capturing algo-
rithms or address the issue of fluidsolid interaction. Obviously, the
full NavierStokes equations should be employed to simulate
the behavior of the near field characteristics of landslide-
generated impulse waves with an appropriate technique to
tracking the interface between the different phases. In recent
years, several attempts based on NavierStokes equations can
be found in the work of Liu et al. and others (Liu et al. 2005;
Ataie-Ashtiani and Nik-Khah 2008; Rzadkiewicz et al. 1997;
Heinrich 1992; Quecedo et al. 2004; Yuk et al. 2006; Abadie
et al. 2010; Montagna et al. 2011; Horrillo et al. 2013).
As regards to the motion of the deformable slide, there have
also been different levels of simplification made by early
research workers. Initially, a rigid motion was often specified
based on a dynamic force balance according to Newtons
Second Law. Essentially, it is simply a general fluid dynamic
problem with moving boundaries. However, in some more
advanced approaches, the fluidsolid interaction between the
water and landslide is taken into account in a coupled way
rather than using pre-specified slide kinematics (Liu et al.
2005; Enet and Grilli 2005; Lynett and Liu 2002; Grilli and
Watts 1999, 2005; Abadie et al. 2010; Pelinovsky and
Poplavsky 1996; Watts et al. 2005; Ghozlani et al. 2013) with
the slide assumed to be a rigid body when entering the water.
This kind of approach has been fully validated experimentally
for rigid slides of idealized shape but is very different from
real physical cases. Up until now, only a few numerical
models have been proposed which can describe the full cou-
pling between the deformable landslide and water, together
with the surrounding air. The basic idea behind those
methods is that the slide is modeled by a fluid-like material
which requires a three-phase model. The appropriate simula-
tion of the dramatic interaction between the air, water, and
landslide is crucial for the numerical results, and it seems
that this point is not clearly stated in the paper. In fact, for a
problem with moderate free surface changes, such as small
amplitude sloshing, the effect of waterair interaction can be
neglected. But as for the problems involving surface merging
or breaking or other dramatic changes, the interaction be-
tween water and air becomes a key issue for the simulation
of these complicated phenomena. For a typical landslide-
generated impulse wave in practice, the whole process does
involve three phases: liquid (water), gaseous (air), and solid
(the landslide material). Moreover, both the experiment and
the numerical results have shown that the strong interaction
does exist during the land sliding. Previous work along these
lines includes Mader and Gittings (2002, 2003) and Gisler
et al. (2006) who solved compressible multi-material Navier
Stokes equations with continuous adaptive mesh refinement.
Quecedo et al. (2004) and Pastor et al. (2005, 2009) solved the
NavierStokes equations with indicator functions to incorpo-
rate the three fluids, or fluid-like, materials involved in the
problem. More recently, Cremonesi et al. (2011) proposed a
pure Lagrangian finite element method to simulate nonlinear
water waves with the assumption that the landslide was de-
scribed as a continuum, non-linear, viscous fluid. Abadie
et al. (2012) employed a 3D multi-fluid NavierStokes model
THETIS to simulate tsunami waves, and Horrillo et al. (2013)
86 Landslides 13 & (2016)
developed a simplified 3D NavierStokes model for the full-
scale landslide scenario in the Gulf of Mexico, the East-Breaks
underwater landslide (Abadie et al. 2010).
When the three-phase model is employed to simulate impulse
waves generated by landslides, the interface capturing between
each phase becomes a key issue. Among the front capturing
methods, the volume of fluid (VOF) method (Hirt and Nichols
1981; Pilliod and Puckett 2004) and the level set (LS) method
(Osher and Sethian 1988; Sussman et al. 1994) have drawn the
most attention. The VOF method is conservative and can deal
with extensive topological changes of the interface. However,
the interface reconstruction can involve considerable compu-
tational effort. An alternative to the VOF method in capturing
fronts is the LS method, and due to its simplicity and ability
to capture interfaces, it has been widely applied in a variety
of fields. However, it has a severe drawback with mass con-
servation in that there can be an unphysical loss/gain of the
fluid at each time step. On the other hand, the conservative
level set (CLS) method proposed by Elin and Olsson (Olsson
and Kreiss 2005; Olsson et al. 2007) improves the mass con-
servation and retains the simplicity of the original method at
the same time. The main idea of the CLS method is to replace
the signed distanced function of the standard method with a
hyperbolic tangent profile. By doing this, the level set func-
tion becomes a smeared Heaviside function which will be zero
in one fluid and one in the other, and it varies smoothly from zero to
one over the interface. The smeared level set function is then
advected in a conservative way. Compared with the standard LS
method, the CLS method exhibits drastically improved mass conser-
vation properties and has been successfully employed in many
applications (Barton et al. 2011; Owkes and Desjardins 2013;
Desjardins et al. 2008). It was originally proposed for two-phase flow
problems, and in this study, it has been expanded to an n-phase flow
that will capture the interface of different phases.
In summary, although a considerable number of general
model studies exist, the numerical prediction of impulse wave
features remains challenging. In the present paper, the finite
element simulation of impulse waves generated by landslides
is studied using a three-phase model in which the landslide
material is schematized as a liquid model and the CLS meth-
od. Some preliminary numerical simulations have highlighted
the fact that simulating a landslide as a liquid may be accept-
able to predict landslidewater interaction (Biscarini 2010).
The general NavierStokes equations and a proper constitu-
tive model are then employed to describe the motion of the
three types of fluid including the landslide, water, and air.
During the simulation, CLS method is adopted to track the
interface of each phase.
This paper is organized as follows. The mathematic model of
the proposed method is given in Section 2 in which the basic idea
of the conservative level set methods is given, followed by their
extension to the three-phase flow problem. The governing equa-
tions and constitutive model for the deformable fluid-like slide
will also be discussed. The finite element discretization is formu-
lated in detail in Section 3. Several challenging verifications and
application are reported to validate the presented technique in
Section 4. Finally, brief concluding remarks are summarized
in Section 5.
Mathematic model signed distance function as in LS method. The material
properties such as density and viscosity are then com-
Conservative level set method for three-phase flow puted as follows:
One of the main problems found in the propagation phase of
landslide inside water is the correct modeling of the interfaces 1 2 1 H 4
which separate the moving mass of soil from the air and water. To
address this, the CLS method is employed in this study to capture 1 2 1 H 5
these interfaces.
The CLS method was first developed by Olsson et al. The total mass for one of the fluids is given by
(Olsson and Kreiss 2005; Olsson et al. 2007) who computed Z
the motion of a two-phase flow in which the NavierStokes M 1 t 1H d 6
equations were solved simultaneously for the whole domain
for the two kinds of fluid using an indicator function to track
the interface. The difficulty of interface reconstruction can be and conservation of the fluid mass implies
overcome perfectly by including the interface conditions in
1 H 
the governing equations. Therefore, it is able to deal with the u1 H  0 7
complex interface and its violent motion such as merge, cross, t
broken up, etc. The most notable difference between the CLS
By introducing the continuity equation for incompressible flow,
method and the standard LS method lies in the choice of the
the level set equation can be written in a conservative way
level set function. Instead of using the signed distance func-
tion , the CLS methods employ the Heaviside function H to H
identify the position of the interface. As shown in Fig. 1a, a uH 0 8
t
domain is separated into two sub-domains 1 2 occupied
by the two fluids with an interface 12. The value of the Due to the existence of inevitable numerical errors or artificial
Heaviside function H is zero in subdomain 1 and one in diffusion together with velocity variations, the shape of H across
2, while the 0.5 contour of H indicates the interface 12: the interface will be distorted when H is advected and, thus, a re-
8 initialization step is necessary to maintain the shape and width of
<0 for 1 the interface:
H 1=2 for 12 1
:
1 for 2 H
H 1H n Hnn 9

The material properties, MP, can be calculated as:
where n is the normal vector at the beginning of the re-
MP MP1 MP2 MP1 H 2 initialization. It is important to mention that the re-initialization
equation should also be in conservative form. In Eq. (9), the flux
where MPi can be density, viscosity, yield strength, etc. of fluid i. term on the left-hand side causes a compression of the interface
The strong discontinuity of the material properties at the inter- profile, while the right-hand-side term is a diffusive flux. When
face will cause instability of numerical solutions, and, in practice, a Eq. (9) is evolved to a steady state with fictitious time , the
hyperbolic tangent function defined as diffusion in the normal direction of the interface is balanced by
the compressive term, and the shape and the width of the
H 1 tanh=2=2 3 interface are well maintained, which is compulsory for the
conservative property.
is often employed as the level set function for CLS meth- The basic idea of CLS method may be extended to the three-
od, where defines a spreading width of H and is the phase or n-phase flow case by defining more indicator functions.

(a) two-phase (b) three-phase (a) four-phase

Fig. 1 Definitions of n-phase flow problems using indicator function

Landslides 13 & (2016) 87

 Original Paper
As indicated in Fig. 1b, two indicator functions, H1, H2, are (Quecedo et al. 2004). Following this approach, the general-
employed to indicate the interface between the three sub- ized visco-plastic fluid model proposed by Chen and Ling
domains 1, 2, 3 for the three-phase flow problem. In this case, (1996) is employed in this study as the constitutive model
the material properties MP may be calculated as for the slide:
 
MP MP1 MP2 MP1 H 1 MP3 MP2 H 2 10 y0 cos psin 1 1 4 2 1
p 21 I 2D 2 D 2 4I 2D 2 D2 17
I 2D 3
More generally, n1 indicator functions H i (i=1n1) are
defined as follows to indicate the interfaces for n-phase where y0 is the fluid yield stress, is the internal friction
flow, for instance for the four-phase flow problem shown angle, I2D is the second invariant of D, 1 and 2 are viscosity-
in Fig. 1c: like coefficients, and 1 and 2 are flow-behavior exponents.
As suggested by Quecedo et al. (2004), this equation can
Hi 0 for i 1n1  in 1 be further simplified by neglecting the dependency on D2
Hi 1 for i 1 j1 resulting in:
in j2n1 11
Hi 0 for i jn1
Hi 1 i 1n1 in n  
for y0 cos psin 1
p 2I 2D 2 D 18
I 2D
The material properties MP can then be calculated in the same
way as for the two-phase case: The above constitutive models can be specialized to other
rheological models with prescribed parameters:

X
&
n1
Newtonian fluids for water and air: y0=0, =0, =1
MP MP1 MPi1 MPi H i 12
& Bingham fluids: =1
&
i1
Frictional fluid: y0=0, =1

Governing equations and constitutive models

Assuming immiscibility for the three interacting fluids where the In this way, only one constitutive equation is used to describe
landslide is modeled as a fluid-like mass and its motion is the three fluids involved in the problem and can be implemented
governed by the evolution of the NavierStokes equation for easily in the existing code.
viscous incompressible flow:
Numerical scheme
Compared with the finite difference method (FDM) or the finite
u 0 13
volume method (FVM), the finite element method (FEM) can be
easily implemented for a large variety of element types and is thus
u capable of handling complex geometries and boundary conditions.
uu p f b 14
t In this paper, all the equations will be discretized within the
FEM context.
In the above equations, t is time, is the gradient
Early attempts to use the FEM to solve convection-
operator, u is the velocity, is viscous stress tensor, p is
dominated problems suffered from severe spurious oscilla-
the pressure, and fb is the body force. The material proper-
tions. The situation has, however, improved and the tech-
ties, such as density and dynamic viscosity , are deter-
nique has gained great popularity in recent years and
mined by Eq.(12).
some very efficient FEM schemes have been proposed.
The viscous stress tensor is defined as a function of the
Among them, two efficient alternatives are the so-called
symmetric component of the rate of strain tensor D.
TaylorGalerkin (TG) (Donea 1984) and Characteristics-
Based Galerkin (CBG) (Zienkiewicz et al. 1999) formula-
D 15 tions. The TG algorithm can be considered as the FEM
counterpart of the LaxWendroff procedure in the FDM.
where D=(u+Tu)/2.
As with the general procedure, it requires the calculation
These general expressions are applicable to the three fluids
of the derivatives of the flux tensor and source vector
involved in the problem of concern and it will be specialized for
relative to a vector of unknowns which is memory and
each fluid phase. For Newtonian fluid
time consuming. The two-step TaylorGalerkin method
(TSTG), which is of second-order accuracy in both the
2D 16 space and time, was developed from the traditional TG
method by Peraire (1986). It has the advantage of being
For most fast-moving landslides, the behavior of the soil easy to program together with high efficiency and has been
mass can be considered as a single phase so that one consti- successfully employed in many applications and is used in
tutive equation for the whole fluidized soil can be used this paper.

88 Landslides 13 & (2016)

Discretization of the incompressible NavierStokes equations Pressure equation:
In the proposed method, a three-phase model is employed
and the interfaces between different phases are captured im-  
plicitly by CLS method, and the problem can be solved by a 1 tpn2 un 1 u* 23
uniform NavierStokes equation in the whole computation
domain. The governing equations are rewritten in the follow- Velocity correction:
ing time-discretized form:
u**
pn2 24
un1 0 19 t

u Intermediate velocity increment u*

uu pn2 f b 20
tt Defining F=uu, G=/, S=fb, Eq. (22) can then be written
in a general conservative form which is convenient to
where the parameters 1 and 2 introduced to represent the time discretize
stepping can be chosen in the range [0,1].
Following the fractional step procedure proposed by Chorin
u*
(1997), the increment u can be decomposed into two parts: F G S 0 25
t

u u* u** 21 Following the basic idea of the TSTG procedure, Eq. (25) can be
simulated as following two steps:

The discretization of the incompressible NavierStokes equa- 1. Prediction step:

tions can be split into three stages:
Intermediate velocity increment: t
un1=2 un F Sn 26
2
u*
uu f b 22
t which of course allows the evaluation of Fn+1/2 and Sn+1/2.

Fig. 2 RayleighTaylor problem, Atwood number A=0.2, Re=2410, comparison of interface shape at non-dimensional time t between the presented results (the right
side panel for each time) and the results from Pochet et al. (2013) (the left side panel for each time)

Landslides 13 & (2016) 89

 Original Paper

Fig. 3 RayleighTaylor problem, non-dimensional linear growth rate against non-dimensional wave number

2. Correction step: where pn is the nodal pressure increment, K=NTNd,

 
a n d RHS2 tNTp un 1 u* d1 t 2 NTp pn d

u* t Fn1=2 Gn Sn1=2 27 t 2 NT pn2 !
1 p n d , N is the shape function scalar for
p
the pressure field.
The standard Galerkin procedure and the Green theorem can
then be applied while writing the discretization of the variables as Velocity correction
*
u Nu
*
28 The velocity is corrected by adding the effects of pressure
once this has been obtained from the pressure equation by
The overbar in the above denotes the approximate nodal values solving the spatial discretization of Eq. (24) to calculate
and N is the shape functions for the field u. Following the standard u **
Galerkin approximation, it gives **
Mu RHS3 31
*
Mu RHS1 29 where
Z
RHS3 t NT pn 2pn d
where M=NTNd and RHS1=t[NT(Fn+1/
2
+G n )dNT (F n+1/2 +G n )ndN T Sn+1/2 d], n is the outer
normal of boundary .
Discretization of the equations of the conservative level set method
Pressure equation discretization Defining F=uH, G=0, S=0 for Eq. (8) and F=H(1H)n,
Approximating spatially using standard finite element shape func- G=(Hn)n, S=0 for Eq. (8), the same discretization process as
tions and integrating by parts following the Green theorem,
discretization of (23) results in

n
1 2 t 2 Kp RHS2 30

x
-0.5

-1.0

-1.5
-1.0 0.5 1.0 1.5 2.0 2.5 3.0

Fig. 4 The laboratory model setup for water wave generation by a deformable Fig. 5 The finite element model for the water wave generation by a deformable
submarine landslide submarine landslide

90 Landslides 13 & (2016)

 Fig. 8 Photomontage of the 1958 Lituya Bay case showing the boundaries of the
slide area and the maximum wave run-up height of 524 m on the opposite shore
of the bay (from Fritz et al.)

RayleighTaylor problem for two-phase flow

Fig. 6 Comparisons of simulated and measured surface elevation at t=0.4 s When a layer of heavier fluid is placed on top of another lighter
(upper panel) and t=0.8 s (lower panel). Circles: measurements; dashed layer in a gravitational field (gravity pointing downwards), the
line: numerical results by Assier-Rzadkiewicz et al. (1997); light solid line:
initial planar interface is unstable. Any disturbance will grow to
numerical results by Gangfeng Ma et al. (2013); dark solid line: current model
produce spikes as the heavier fluid moves downwards and bubbles
of lighter fluid move upwards. This is the so-called Rayleigh
Taylor instability.
(25) can be performed for the advection and re-initialization of the The density difference is normally represented by the Atwood
indicator function in the CLS method. number A=( h l )/( h + l ), where the subscripts h and l
correspond to the heavier fluids and lighter fluids, respectively.
Z The proposed method will be validated by investigating the same
NT Nd H problem as carried out by Pochet et al. (2013) in which the value of

Z  Z  Z 
t NT Fn1=2 Gn d NT Fn1=2 Gn nd NT Sn1=2 d the Atwood number was A=0.2.
The computational domain is a rectangle of dimensions
32 (1 m4 m) discretized with a fully structured mesh made of
128512 grids. The physical properties are density h=1.5 kg/m3
where Fn+1/2 and Sn+1/2 can be obtained from Hn+1/2: and dynamic viscosity h=0.00195 kg/(m s) for the heavier fluid,
and l=1.0 kg/m3 and l=0.0013 kg/(m s) for the lighter fluid. The
t
H n1=2 H n F Sn 33 ratio of kinetic viscosity is kept to one and surface tension is
2 assumed to be negligible. The amplitude of the disturbance ap-
plied to the interface is d=0.05 m and its initial shape is defined as
y=2+dcos(2x/), where =1.0m denotes the width of the channel.
Verification and application p
A number of numerical experiments are now presented to illus- =g is taken as the time scale to obtain non-dimensional
trate the performance of the new algorithm derived here. These time, where g=9.81m/s2 is the gravitational acceleration. The
include a RayleighTaylor instability problem to illustrate the initial velocity field is zero and the pressure field is set to
capability of interface capturing and a benchmark test of subaerial have hydrostatic equilibrium. The Reynolds number, defined
p
landslide-generated impulse wave. In addition, a further applica- as Re h g=h , then becomes 2410. A no-slip condition
tion has been conducted to numerically reproduce the landslide- is enforced at the bottom and top walls while a free-slip
generated impulse wave of Lituya Bay (1958), and its results are boundary condition is imposed on the two vertical sides. Zero
compared with a scaled experiment. pressure is imposed at the upper wall.

Fig. 7 Snapshots of simulated interface profile at t=0.4 s (left panel) and t=0.8 s (right panel)

Landslides 13 & (2016) 91

 Original Paper
Fig. 9 Simplified geometry of the cross-section of Gilbert Inlet along slide axis, on the basis of the physical and numerical model (from Fritz et al.)
Comparison of the interface shape at selected non-
dimensional times between the presented results (the right-
hand-side panels) and the results from Pochet et al. (2013)
(the left-hand-side panels) are shown in Fig. 2. It can be seen
that the results using the proposed methods match perfectly
with the published data throughout the whole simulation.
Both the large mushroom structure and the small vortex
structures are clearly visible and well resolved. The simulation
is numerically stable at all times and shows good ability in
handling complex flow problems.
The numerical results are now compared to the analytical
solution obtained in Tryggvason (1988). In the early stages
when the amplitude of the perturbation is much smaller than
the wave length, the perturbation of the fluid interface has an
exponential growth
h h0 et
where h is the amplitude at time t, h0 is the initial amplitude,
and is the growth rate of the perturbation. With a kinetic
viscosity ratio of one and negligible surface tension, the
growth rate is a function of both the Atwood number and
wave number k 2=. Chandrasekhar (1961) derived the
following non-dimensional relationship between the growth
rate and the wave number of the initial disturbance:
=g 2 = 1=3 ; k k=g= 2 1=3
where and k is the non-dimensional growth rate and wave
number, respectively. In Fig. 3, the normalized growth rate
and the wave number are plotted and compared with the
analytical solutions of Gretar Tryggvason, and very good
agreement can be observed.
Benchmark test of subaerial landslide-generated impulse wave
In order to understand the capability of the proposed method in
modeling submerged landslides, the experiment of Rzadkiewicz
et al. (1996, 1997) was tested. A sketch of the laboratory model
setup is shown in Fig. 4. The experiment setup consists of a slope
of 45 which allows a mass of sand to slide freely down a friction-
less inclined plane to generate water waves. The channel is 4 m
long, 0.30 m wide, and 2 m high. The sand mass is as wide as a
92 Landslides 13 & (2016)
channel, so that the experiments are equivalent to a two-
dimensional problem in a vertical plane. The dimensions of the
sand mass in cross-section are 0.65 m0.65 m and the mean
apparent density is 1950 kg/m3. Because the fluid yield stress is
unknown, the current simulation assumes a zero yield stress. A
frictional fluid model is adopted with the internal friction angle
=15. The water depth is 1.60 m and the top of the triangular mass
is initially 0.1 m below the water surface. A free-slip condition is
applied along the slope. The computation domain is mesh using
73,628 four nodes quadrilaterals with 74,029 nodes depicted in
Fig. 5. More finer mesh density is employed surrounding the
submarine landslide, and the mesh size varies from 0.008 to
0.04 m. The automatic adjustment of the time step is according
to the following stable conditions C2+2C/Pe1, where C is the
element Courant number and Pe is the element Peclet number.
Figure 6 shows the comparisons of simulated and mea-
sured wave elevations at t=0.4 s (upper panel) and t=0.8 s
34
(lower panel). The results are acceptable but show differences
with the experiments. It can be seen that the model captures
the trend of the interface of the wave generated by the
underwater landslide correctly but slightly over-predicts the
amplitude. Similar numerical results were obtained by other
researchers like Rzadkiewicz et al. (1996) and Gangfeng Ma
et al. (2013) with a much larger discrepancy with the experi-
mental data. The discrepancies are introduced by the simpli-
fied treatment of the landslide in which granular stresses
which could retard the initial failure of the sliding mass have
35 been neglected (Ma et al. 2013). The second reason for the
deviation is that the sediments are modeled by an ideal fluid.
In Fig. 7, the snapshots of the simulated interface profiles are
plotted for each time instance investigated. From that, it can be
seen that the landslide has penetrated deeper into the water
Fig. 10 The finite element model for the Lituya Bay impulse wave
Fig. 11 Numerical evolution of the different phases due to the landslide impact. Time increment is 5 s

channel from t=0.4 s to t=0.8 s and causes a change in the water
air interface. Also, at t=0.8 s, the landslide mass has lost its original
triangular shape due to the shearing effect of the water and several
vortex-like bubbles can be seen to form at the landslide/water
interface which slowly detach from the main mass. This phenom-
enon is similar to the RayleighTaylor instability problem men-
tioned above and provides an insight into deformable landslide
and water interaction.
Based on the above discussion, it can be noted that the pro-
posed method can be used to capture the impulse wave generated
by a subaerial landslide, and an improved match with the exper-
imental data has been obtained over existing numerical models. To
further address the problem, an application to the Lituya Bay
impulse wave generation is presented in the next section.
Application to the Lituya Bay impulse wave
In all the documented cases of landslide-generated impulse waves,
there are only a few experiments with sufficient field details for
conclusive comparison with a simulation. Among those, the 1958
gigantic wave generated in Lituya Bay has received extensive
attention from a number of researchers (Quecedo et al. 2004;
Pastor et al. 2005, 2009; Biscarini 2010; Mader 1999; Fritz 2001;
Weiss et al. 2009).
Lituya Bay is a large inlet on the northeast shore of the Gulf of
Alaska, a seismically active area particularly susceptible to
landslide-induced waves. On July 8, 1958, an 8.3 magnitude earth-
quake along the Fair Weather fault triggered a major subaerial
rockslide into the Gilbert Inlet. The initial slide geometry is as-
sumed to be a prism spanning 730 to 915 m in width with a

Fig. 12 Photo sequence of a granular slide impact experiment at a time increment of 5 s (from Fritz et al.)

Landslides 13 & (2016) 93

 Original Paper
Fig. 13 Time history comparison of wave height at location 885 m between the
presented numerical result and experimental data by Fritz (2001)
thickness of 92 m normal to the slope. The lower extent of the initial
landslide position remains undefined. The slide length was estimated
to be 970 m with a center of gravity at 610 m elevation. An estimated
rockslide volume of about 30106 m3 was released on the northeast
wall of the Gilbert Inlet up to an elevation of 915 m. The rockslide
impacted the water at high speed creating a giant nonlinear wave
with the highest wave run-up in recorded history; the soliton-like
wave ran up to an altitude of 524 m causing forest destruction and
erosion down to bedrock on a spur ridge (see Fig. 8).
Based on generalized Froude similarity, Fritz (2001) built a 2D
physical model of the Gilbert Inlet scaled at 1:675 and the rockslide
was modeled with an artificial granular material (PP-BaSO4)
matching the assumed schist density, 2.7 t/m3. Fritz estimated the
bulk density of this granular material to be 1.61 t/m3, considering a
void fraction of 39 %. The slide granulate, initially contained in the
slide box, is accelerated by a pneumatic landslide generator to
control landslide dynamics and impact characteristics. Laser mea-
surement techniques such as particle image velocimetry (PIV) and
laser distance sensors (LDS) were applied to the decisive initial
phase with rockslide impact and wave generation. PIV measure-
ments of wave run-up on the headland slope were conducted to
complement wave and run-up gage records. In the physical model,
the slide mass was accelerated up to a prototype impact velocity of
110 m/s, which corresponds to an estimated free fall velocity with
the centroid situated at 610 m.
Fig. 14 Effect of the internal friction angle on the final stage of run out: a =0; b =5; c =10; d =15
94 Landslides 13 & (2016)
Figure 9 shows the simplified geometry of the Gilbert Inlet,
which is the basis for the physical experiment and the current
numerical simulations. The computation domain is mesh using
94,758 three nodes triangles with 47,921 nodes depicted in Fig. 10.
The mesh size varies from 5 to 30 m. The automatic adjustment of
the time step is according to the same stable conditions as in
Section 4.2. All the boundaries are considered as frictionless,
impermeable walls. As a bulk density of 1.61 t/m3 was used in the
physical model, this is also used in the present numerical simula-
tion. Because the internal friction angle of this material is un-
known, the current simulation assumes a zero friction angle, the
same approximation made in the work of Quecedo et al. (2004).
Figure 11 presents the density image sequence of the landslide
impact according to the numerical model and Fig. 12 the correspond-
ing from the physical model. Figure 13 gives the comparison of wave
height at location 885 m between the presented numerical result and
experimental data by Fritz et al. It is observed that the interfaces
between the three fluids are sharp in the physical model, i.e., no
mixing occurs during the first moments of the landslide impact
which is also assumed in the numerical model. The simulations
predict the wave height in good qualitative agreement with the
experimental results. In both the physical and numerical models,
according to Fritz (2001), the whole process of a high-speed granular
slide impact may be subdivided into two main stages: (a) rockslide
impact and penetration with flow separation, cavity formation, and
wave generation; and (b) air cavity collapse with rockslide run-out
and debris detrainment causing massive phase mixing.
According to the present simulation, after 9.1 s of sliding, the
slide begins to impact on the water with a velocity of about 72 m/s
while in the physical model it is accelerated to an impact velocity
of 110 m/s, which corresponds to an estimated free fall velocity
with a centroid situated at 610 m (Fig. 9a). The impact continues
and penetration begins to occur followed by the slide bulking
caused by impact on the water surface and deflection at the
channel bottom (Fig. 9b). Flow separation on the rockslide shoul-
der occurs due to fast slide penetration into the water body where
energy is transferred from the landslide to the water (Fig. 9c). A
large air cavity is created on the back of the granular slide where
the slide penetration velocity exceeds the wave propagation veloc-
ity. (Fig. 9df). The air cavity subsequently collapses during the
slide run-out along the bottom causing significant airwater
mixing (Fig. 9g, h). As the wave has already propagated outside
the collapsing air bubble zone, the energy losses due to this
energetic airwater mixing during the air bubble collapse may
have a limited effect on the wave characteristics.
 It is noted, however, that the wave height at a position 885 m
closer to the headland is over-predicted as the numerical result is
232 m compared with the experiment value of 152 m. The main
reason for such a discrepancy is due to the simplified treatment of
the landslide which is the same as the previous validation case and
is still closer to the experimental results compared with other
numerical simulations. In the presented numerical method, the
landslide is simulated as a fluid-like material with a generalized
visco-plastic constitutive model, while the landslide was modeled
with an artificial granular material in the experiment. The property
of landslide in real world may be between those two situations. As
can be seen from Fig. 13, except for the over-predicted wave height,
another aspect of discrepancy between numerical result and ex-
perimental data is that the growth/draw-off rate of the impulse
wave is much more rapid in the presented numerical result. This is
also attributed into the flowable tendency of the landslide when a
three-phase model and a rheological model are employed.
In terms of the maximum run-up at the headland, the maxi-
mum run-up at the headland, 573 m, predicted by the current
simulation is higher than that observed in the experiment, 527 m,
which is attributed to the reduced energy dissipation in the
numerical model. In effect, energy losses due to turbulence and
phase mixing which are not considered in the current numerical
model may further reduce the height elevation close to headland.
Also, as reported by Quecedo et al. (2004), the internal friction
angle of the landslide material which is taken as zero in the current
case may also have influence of the output of the numerical
simulation. Generally, the agreement between the current simula-
tion and the experimental data is still good and captures the basic
process of an impulse wave generated by the landslide. Further
work may involve a thorough study of different constitutive
models of the landslide and its effect on the generated im-
pulse wave.
Figure 14 presents the effect of the internal friction angle on the
final stage of run-out. The wave height at location 885 m reduces
when a non-zero internal friction angle is assumed. With the
growth of the internal friction angle, the phenomenon of stack at
location 1342 m becomes more clear. Besides, the interfaces be-
tween the three fluids are sharper. In spite of these differences, the
main characteristic of results is consistent, as shown in Fig. 14.
Conclusion
The finite element simulation of impulse waves generated by
landslides has been carried out using a three-phase model and
the conservative level set method. The landslide material has been
schematized as a liquid model using the NavierStokes equations
and a realistic constitutive model to describe the motion of the
three types of fluid: the landslide, water, and air. In order to carry
out the simulation, the conservative level set method was expand-
ed to n-phase flow problem and adopted to track the interface
between the different fluids. Following verification of a Rayleigh
Taylor problem and a benchmark test of a laboratory subaerial
landslide-generated impulse wave, the model was used to simulate
the 1958 Lituya Bay impulse wave, which was also experimentally
investigated by Fritz et al. It has been shown that although the
simulation seems to slightly over-predict the surface waves, the
overall agreement between computed and experimental results is
good. From the safety point of view, the proposed model can be
used to model the process of a high-speed fluidized landslide
impact, including landslide impact and penetration with flow
separation, cavity formation, wave generation, air cavity collapse
with landslide run-out, and debris detrainment causing massive
phase mixing. Further research may involve a study of the simu-
lation of impulse waves which are generated by granular material.
And the numerical model can be used as a reference for engineer-
ing applications from the safety point of view.
Acknowledgments
Project supported by the National Natural Science Foundation of
China (No. 51279050), National High-tech R&D Program of China
(863 Program) (No. SS2012AA112507), National Key Technology
R&D Program in 12th Five-Year Plan (2012BAK10B04), and non-
profit Industry Financial Program of MWR (Ministry of Water
Resources of China) (No. 201301058).
References
Abadie S, Morichon D, Grilli S, Glockner S (2010) Numerical simulation of waves
generated by landslides using a multiple-fluid NavierStokes model. Coast Eng
57(9):779794
Abadie S, Harris J, Grilli S, Fabre R (2012) Numerical modeling of tsunami waves
generated by the flank collapse of the Cumbre Vieja Volcano (La Palma, Canary
Islands): tsunami source and near field effects. J Geophys Res 117(C5):C05030
Ataie-Ashtiani B, Nik-Khah A (2008) Impulsive waves caused by subaerial landslides.
Environ Fluid Mech 8(3):263280
Ataie-Ashtiani B, Yavari-Ramshe S (2011) Numerical simulation of wave generated by
landslide incidents in dam reservoirs. Landslides 8(4):417432
AtaieAshtiani B, Najafi Jilani A (2007) A higherorder Boussinesqtype model with
moving bottom boundary: applications to submarine landslide tsunami waves. Int J
Numer Methods Fluids 53(6):10191048
Barton PT, Obadia B, Drikakis D (2011) A conservative level-set based method for
compressible solid/fluid problems on fixed grids. J Comput Phys 230(21):78677890
Biscarini C (2010) Computational fluid dynamics modelling of landslide generated water
waves. Landslides 7(2):117124
Chandrasekhar S (1961) Hydrodynamic and hydromagnetic stability. International Series
of Monographs on Physics, Oxford: Clarendon, 1961, 1
Chaudhry MH, Mercer AG, Cass D (1983) Modeling of slide-generated waves in a
reservoir. J Hydraul Eng 109(11):15051520
Chen C-L, Ling C-H (1996) Granular-flow rheology: role of shear-rate number in transition
regime. J Eng Mech 122(5):469480
Chorin AJ (1997) A numerical method for solving incompressible viscous flow problems.
J Comput Phys 135(2):118125
Cremonesi M, Frangi A, Perego U (2011) A Lagrangian finite element approach for the
simulation of water-waves induced by landslides. Comput Struct 89(1112):1086
1093. doi:10.1016/j.compstruc.2010.12.005
de Carvalho RF, do Carmo JSA (2007) Landslides into reservoirs and their impacts on
banks. Environ Fluid Mech 7(6):481493
Desjardins O, Moureau V, Pitsch H (2008) An accurate conservative level set/ghost
fluid method for simulating turbulent atomization. J Comput Phys
227(18):83958416
Di Risio M, Bellotti G, Panizzo A, De Girolamo P (2009) Three-dimensional experiments
on landslide generated waves at a sloping coast. Coast Eng 56(5):659671
Donea J (1984) A TaylorGalerkin method for convective transport problems. Int J Numer
Methods Eng 20(1):101119
Enet F, Grilli ST (2007) Experimental study of tsunami generation by three-dimensional
rigid underwater landslides. J Waterw Port Coast Ocean Eng 133(6):442454
Enet F, Grilli S (2005) Tsunami landslide generation: modelling and experiments.
In Proc. 5th Intl. on Ocean Wave Measurement and Analysis (WAVES 2005),
Citeseer
Fernndez-Nieto ED, Bouchut F, Bresch D, Castro Daz M, Mangeney A (2008) A new
SavageHutter type model for submarine avalanches and generated tsunami. J
Comput Phys 227(16):77207754
Fritz H, Hager W, Minor H-E (2004) Near field characteristics of landslide generated
impulse waves. J Waterw Port Coast Ocean Eng 130(6):287302
Landslides 13 & (2016) 95
 Original Paper
Fritz HM (2002) Initial phase of landslide generated impulse waves, Mitteilungen der
Versuchsanstalt fur Wasserbau, Hydrologie und Glaziologie an der Eidgenossischen
Technischen Hochschule Zurich
Fritz HM (2001) Lituya Bay case rockslide impact and wave run-up. Sci Tsunami Hazards
19(1):322
Fuhrman DR, Madsen PA (2009) Tsunami generation, propagation, and run-up with a
high-order Boussinesq model. Coast Eng 56(7):747758
Ghozlani B, Hafsia Z, Maalel K (2013) Numerical study of surface water waves generated
by mass movement. Fluid Dyn Res 45(5):055506
Gisler G, Weaver R, Gittings ML (2006) SAGE calculations of the tsunami threat from La
Palma. Sci Tsunami Hazards 24(4):288301
Grilli ST, Watts P (1999) Modeling of waves generated by a moving submerged body.
Appl Underw Landslides Eng Anal Bound Elem 23(8):645656
Grilli ST, Watts P (2005) Tsunami generation by submarine mass failure. I: Modeling,
experimental validation, and sensitivity analyses. J Waterw Port Coast Ocean Eng
131(6):283297
Grilli ST, Vogelmann S, Watts P (2002) Development of a 3D numerical wave tank for
modeling tsunami generation by underwater landslides. Eng Anal Bound Elem
26(4):301313
Hanes DM, Inman DL (1985) Experimental evaluation of a dynamic yield criterion for
granular fluid flows. J Geophys Res: Solid Earth (19782012) 90(B5):36703674
Heinrich P (1992) Nonlinear water waves generated by submarine and aerial landslides. J
Waterw Port Coast Ocean Eng 118(3):249266
Heller V, Hager WH (2010) Impulse product parameter in landslide generated impulse
waves. J Waterw Port Coast Ocean Eng 136(3):145155
Heller V, Spinneken J (2013) Improved landslidetsunami prediction: effects of block
model parameters and slide model. J Geophys Res: Oceans 118:14891507.
doi:10.1002/jgrc.20099
Hirt CW, Nichols BD (1981) Volume of fluid (VOF) method for the dynamics of free
boundaries. J Comput Phys 39(1):201225
Horrillo J, Wood A, Kim GB, Parambath A (2013) A simplified 3D NavierStokes
numerical model for landslidetsunami: application to the Gulf of Mexico. J Geophys
Res: Oceans
Liu P, Wu T, Raichlen F, Synolakis C, Borrero J (2005) Runup and rundown generated by
three-dimensional sliding masses. J Fluid Mech 536(1):107144
Lynett P, Liu PL-F (2002) A numerical study of submarine-landslide-generated waves and
run-up. Proc R Soc Lond A: Math Phys Eng Sci 458(2028):28852910
Ma G, Kirby JT, Shi F (2013) Numerical simulation of tsunami waves generated by
deformable submarine landslides. Ocean Model 69:146165
Mader CL, Gittings ML (2002) Modeling the 1958 Lituya Bay mega-tsunami, II. Sci
Tsunami Hazards 20(5):241250
Mader CL, Gittings ML (2003) Dynamics of water cavity generation. Sci Tsunami Hazards
21(2):91101
Mader CL (1999) Modelling the Lituya Bay mega-tsunami. Sci Tsunami Hazards 17(1):5767
Medina V, Hrlimann M, Bateman A (2008) Application of FLATModel, a 2D finite volume
code, to debris flows in the northeastern part of the Iberian Peninsula. Landslides
5(1):127142
Mohammed F, Fritz HM (2012) Physical modeling of tsunamis generated by three
dimensional deformable granular landslides. J Geophys Res: Oceans (19782012),
117(C11)
Montagna F, Bellotti G, Di Risio M (2011) 3D numerical modeling of landslide-generated
tsunamis around a conical island. Nat Hazards 58(1):591608
Najafi-Jilani A, Ataie-Ashtiani B (2008) Estimation of near-field characteristics of tsunami
generation by submarine landslide. Ocean Eng 35(5):545557
Olsson E, Kreiss G, Zahedi S (2007) A conservative level set method for two phase flow II.
J Comput Phys 225(1):785807
Olsson E, Kreiss G (2005) A conservative level set method for two phase flow. J Comput
Phys 210(1):225246
Osher S, Sethian JA (1988) Fronts propagating with curvature-dependent speed: algo-
rithms based on HamiltonJacobi formulations. J Comput Phys 79(1):1249
Owkes M, Desjardins O (2013) A discontinuous Galerkin conservative level set scheme for
interface capturing in multiphase flows. J Comput Phys 249(0):275302. doi:10.1016/
j.jcp.2013.04.036
Panizzo A, De Girolamo P, Petaccia A (2005) Forecasting impulse waves generated by
subaerial landslides. J Geophys Res: Oceans (19782012), 110(C12)
Pastor M, Herreros I, Fernndez Merodo J, Mira P, Haddad B, Quecedo M, Gonzlez E,
Alvarez-Cedrn C, Drempetic V (2009) Modelling of fast catastrophic landslides and
impulse waves induced by them in fjords, lakes and reservoirs. Eng Geol 109(1):124134
96 Landslides 13 & (2016)
Pastor M, Quecedo M, Herreros M, Gonzlez E, Haddad B, Merodo JF, Mira P (2005)
Modelling of fast landslides and waves induced by them in reservoirs and other water
bodies. Rivista Italiana di Geotecnica 2:4662
Pelinovsky E, Poplavsky A (1996) Simplified model of tsunami generation by submarine
landslides. Phys Chem Earth 21(1):1317
Peraire J (1986) A finite element method for convection dominated flows. PhD thesis,
University of Wales, Swansea
Pilliod JE Jr, Puckett EG (2004) Second-order accurate volume-of-fluid algorithms for
tracking material interfaces. J Comput Phys 199(2):465502
Pochet F, Hillewaert K, Geuzaine P, Remacle J-F, Marchandise E (2013) A 3D strongly
coupled implicit discontinuous Galerkin level set-based method for modeling two-
phase flows. Comput Fluids 87:144155
Quecedo M, Pastor M, Herreros M (2004) Numerical modelling of impulse
w a v e g e n e r a t e d b y fa s t l a n d s l i d e s . I n t J N u m e r Me t h o d s E n g
59(12):16331656
Rzadkiewicz SA, Mariotti C, Heinrich P (1996) Modelling of submarine landslides and
generated water waves. Phys Chem Earth 21(1):712
Rzadkiewicz SA, Mariotti C, Heinrich P (1997) Numerical simulation of submarine
landslides and their hydraulic effects. J Waterw Port Coast Ocean Eng
123(4):149157
Saut O (2003) Determination of dynamic stability characteristics of an underwater vehicle
including free surface effects. MS thesis, Florida Atlantic University
Serrano-Pacheco A, Murillo J, Garca-Navarro P (2009) A finite volume method for the
simulation of the waves generated by landslides. J Hydrol 373(34):273289.
doi:10.1016/j.jhydrol.2009.05.003
Shigihara Y, Goto D, Imamura F, Kitamura Y, Matsubara T, Takaoka K, Ban K (2006)
Hydraulic and numerical study on the generation of a subaqueous landslide-induced
tsunami along the coast. Nat Hazards 39(2):159177
Sussman M, Smereka P, Osher S (1994) A level set approach for computing solutions to
incompressible two-phase flow. J Comput Phys 114(1):146159
Townson J, Kaya Y (1988) Simulations of the waves in Lake Botnen created by the Rissa
Landslide. ICE Proceedings, 145160, Ice Virtual Library
Tryggvason G (1988) Numerical simulations of the RayleighTaylor instability. J Comput
Phys 75(2):253282
Walder JS, Watts P, Sorensen OE, Janssen K (2003) Tsunamis generated by subaerial
mass flows. J Geophys Res 108(B5):2236
Ward SN (2001) Landslide tsunami. J Geophys Res 106(B6):1120111215. doi:10.1029/
2000JB900450
Watts P, Grilli ST, Tappin DR, Fryer GJ (2005) Tsunami generation by submarine mass
failure. II: Predictive equations and case studies. J Waterw Port Coast Ocean Eng
131(6):298310
Watts P, Grilli ST, Kirby JT, Fryer GJ, Tappin DR (2003) Landslide tsunami case studies
using a Boussinesq model and a fully nonlinear tsunami generation model. Nat
Hazards Earth Syst Sci 3:391402
Weiss R, Fritz HM, Wnnemann K (2009) Hybrid modeling of the mega-tsunami runup in
Lituya Bay after half a century. Geophys Res Lett 36(9):L09602
Yuk D, Yim S, Liu P-F (2006) Numerical modeling of submarine mass-movement
generated waves using RANS model. Comput Geosci 32(7):927935
Zhou H, Teng MH (2009) Extended fourth-order depth-integrated model for water
waves and currents generated by submarine landslides. J Eng Mech
136(4):506516
Zienkiewicz O, Nithiarasu P, Codina R, Vazquez M, Ortiz P (1999) The characteristic
basedsplit procedure: an efficient and accurate algorithm for fluid problems. Int J
Numer Methods Fluids 31(1):359392
Zweifel A, Hager WH, Minor H-E (2006) Plane impulse waves in reservoirs. J Waterw Port
Coast Ocean Eng 132(5):358368
L. Zhao : J. Mao : X. Liu ()) : T. Li
College of Water Conservancy and Hydropower,
Hohai University, Xikang Rd., Nanjing, 210098,
Peoples Republic of China
e-mail: zlhhhu@163.com
X. Bai : J. J. R. Williams
School of Engineering & Materials Science,
Queen Mary University of London, Mile End Rd.,
London, E1 4NS, UK
 Original Paper
Landslides (2016) 13:97114 Vahid Moosavi I Yaghoub Niazi
DOI 10.1007/s10346-014-0547-0
Received: 5 October 2014
Accepted: 16 December 2014 Development of hybrid wavelet packet-statistical
Published online: 9 January 2015
Springer-Verlag Berlin Heidelberg 2015
models (WP-SM) for landslide susceptibility mapping
Abstract Landslides are recognized as one of the most important
natural hazards in many areas throughout the world. Producing
landslide susceptibility maps have received particular attention
from a wide range of scientists. The main objective of this study
was to produce landslide susceptibility maps using hybrid wavelet
packet-statistical models (WP-SM). In the first step, landslide
susceptibility maps were produced using single artificial neural
network (ANN), support vector machine (SVM), maximum entro-
py (MaxEnt), and generalized linear model (GLM). In the next
step, the input maps were preprocessed using different mother
wavelets in different levels. Then, the hybrid models were devel-
oped using the wavelet-based preprocessed maps. Results showed
that the wavelet packet transform can be effectively used to pro-
duce precise landslide susceptibility maps. It was shown that
wavelet packet transform significantly enhanced the ability of the
single statistical models. The kappa coefficients were increased
from 0.829 to 0.941, 0.846 to 0.978, 0.744 to 0.829, and 0.735 to
0.817 in hybrid ANN, SVM, MaxEnt, and GLM, respectively. The
best wavelet transform was performed using bior1.5 with a three-
level decomposition. It was also recognized that MaxEnt and GLM
produced approximately poor results. However, SVM performed
better than the other three models both in single and hybrid forms.
ANN also outperformed MaxEnt and GLM models. Spatial distri-
bution of the susceptible area is consistent with the observed
landslide distribution pattern particularly in maps obtained from
the hybrid models. The produced maps showed that the general
pattern of susceptible area intensively followed the pattern of
roads and sensitive geological formations.
Keywords Landslide susceptibility . Wavelet
packet . SVM . ANN . Generalized linear model . Maximum
entropy
Introduction
Landslide is a natural phenomenon that plays a critical role in the
landscape evolution and represents a serious vulnerability in many
areas of the world (Glade et al. 2005). Different phenomena can
cause landslides, including intense or extended rainfall, swift snow
melting, earthquakes (Wilson and Keefer 1985; Ingls et al. 2006),
volcanic activity (Ablay and Hrliman 2000; Moon and Simpson
2002), and human activities (Skempton and Hutchinson 1969).
Landslides involve flowing, sliding, toppling, falling, or a combi-
nation of two or more types of movements at the same time or
during the lifetime of a landslide (Cruden and Varnes 1996; Brckl
2001; Srensen and Bauer 2003). They are the most important,
costly, and damaging natural hazards in mountainous regions
(University of Utah 1984). Consequently, it is valuable to assess
and evaluate the landslide susceptibility and vulnerability in a
region. Landslide susceptibility (LS) can be defined as the proba-
bility of the spatial occurrence of known slope failures given a set
of geological and environmental conditions or the spatial proba-
bility of landslide occurrence (Guzzetti et al. 2006). Therefore, the
main objective of modeling the landslide susceptibility is the
prediction of landslide-prone areas in space and/or time.
Several landslide susceptibility mapping methods such as direct
geomorphological mapping, analysis of landslide inventories, heu-
ristic methods, neural networks, fuzzy logic, expert systems, and
process-based models have been used to assess the vulnerability of
different parts of an area to landslide (Gokceoglu and Sezer 2009;
Bednarik et al. 2010; Mihaela et al. 2011; Pourghasemi et al. 2012a,
b, c, d; Xu et al. 2012, 2013a, b; Tien Bui et al. 2012a, b, c, d). So, two
approaches can be mentioned establishing models for predicting
the geographical distribution of a phenomenon. The first one
attempts to disclose the fundamental mechanisms that the envi-
ronmental factors exert on a certain phenomenon. The second
tries a mathematical function, based on various supervised classi-
fication algorithms, to correlate the spatial distribution of the
phenomenon with the environmental factors (Brenning 2005). A
number of statistical methods (also called niche modeling ap-
proaches), i.e., artificial neural network (ANN), support vector
machine (SVM), maximum entropy (MaxEnt), and generalized
linear model (GLM) which lie in the second category have been
increasingly used in LS mapping (Yao et al. 2008; Yilmaz 2010;
Marjanovi et al. 2011; Xu et al. 2012, 2013a, b, 2014; Pradhan 2013;
Conforti et al. 2014). Neural network is a method that is first
trained from known data and then used to categorize unknown
pixels. SVMs demonstrate a set of theoretically superior machine-
learning algorithms. Development of SVM was initially triggered
by the exploration and formalization of learning machine capacity
control and over-fitting issues (Vapnik 2001). GLM forms a mul-
tivariate regression namely link function, between a generalized
dependent variable and several independent variables (Atkinson
and Massari 1998). The MaxEnt method is a generative approach
which makes predictions using statistical concepts, analyzes the
weights of environmental factors, and finally calculates a contin-
uous probability value for the mentioned parameter distribution
(Phillips et al. 2006). Wavelet transform (WT), as a signal process-
ing approach, can be used for multi-scale analysis of a signal
through dilation and translation that can extract signal features
from both time (or position) domain and frequency domain,
effectively (Peng and Chu 2004). Wavelet transforms offer useful
decompositions of the original signals, so that wavelet-based
transformed data enhance the ability of prediction models by
capturing useful information on various resolution levels (Kisi
2009; Nourani et al. 2009; Adamowski and Sun 2010; Moosavi
et al. 2013b, c). Initial researches indicated that wavelet analysis
appears to be a more effective tool than the Fourier transform
(Adamowski 2007). The wavelet packet transform (WPT), which is
a generalization of wavelet transform, is alternative bases formed
by taking linear combinations of usual wavelet functions (Ogden
1997). The wavelet transform technique has been explored in
numerous studies (Liu and Ngan 2006; Borah et al. 2007; Li
2009; Zhou et al. 2010; Park et al. 2011; Demirhan and Guler 2011;
Watkins 2012; Moosavi et al. 2013a). However, to the best of our
Landslides 13 & (2016) 97
 Original Paper
knowledge, no research has been published that explores coupling
wavelet packet analysis with statistical GIS-based models for land-
slide susceptibility modeling. The main objective of this research is
to evaluate the effect of wavelet packet transform as a preprocess-
ing approach on the GIS-based models in landslide susceptibility
mapping.
Study area
The study area covers a region of nearly 272 km2 around the Ilam
dam, Ilam Province, Iran. The region lies between 46 20 to 46 39 Most of the landslides are shallow rotational with a few transla-
E longitude and 33 23 to 33 38 N latitude (Fig. 1). The maximum tional that occurred in cut slopes or in embankments, alongside
and minimum elevations are 2600 and 900 m above mean sea
level, respectively. Climatic condition in the area is semi-arid
(typical Mediterranean climate) with 595 mm rainfall, and average
temperature is 21.7 C in summer and 4.7 C in winter. Land use/
cover in the area mostly includes cultivated area, barren land,
forest, grass land, garden, and water body. The Ilam dam is one
of the most important dams in the western Iran that supplies
potable water for the Ilam City. The geology of the study area is
very complex, and the lithology units comprise several formations.
In this region, geological units exist from Cretaceous period to
Quaternary sediments. Among various lithological formations in
the area, the marl, shale, and silt deposits are known to be the
most susceptible formations to landslides (Table 1). Tectonic fea-
tures of the study area include folds, thrusts, and strike-slip faults
(Harison and Falcon 1936).
Materials and methods
Figure 2 shows the flowchart of this study. As this flowchart shows,
four general stages have been defined in order to develop hybrid
wavelet packet-statistical models. In the first stage, the most im-
portant conditioning factors were selected according to the litera-
ture review. Then, the base maps were produced using
topographic, geological, and remotely sensed data. In the next
stage, the mentioned base maps were classified and the frequency
ratio (FR) for each attribute class was calculated crossing each
map with the landslide inventory map. Afterward, the FR values
were normalized between 0.1 and 0.9. In the third stage, two
modeling approaches (i.e., single and hybrid) were conducted
separately. In the single modeling part, landslide susceptibility
maps were produced using single ANN, SVM, MaxEnt, and GLM
models. In the hybrid modeling process, first, the base maps were
decomposed to approximation and detail components using dif-
ferent mother wavelets in different levels. Second, their approxi-
mation and detail components were used as inputs (instead of the
original maps) for the abovementioned statistical single models. In
the last stage, TPR, AUC, and kappa were used to evaluate the
performance of both single and hybrid models. Details of these
stages are explained in the following sections.
Landslide inventory map
Producing a landslide inventory map is necessary to describe the
relationship between the landslide distribution and the condition-
ing factors. Detecting the accurate location of landslides is of great
importance in landslide susceptibility analysis. As the detection of
landslide location is often time-consuming and labor-intensive,
remote sensing methods have been employed in addition to field
surveys. The initial landslide inventory map was produced using
98 Landslides 13 & (2016)
the approach proposed by Moosavi et al. (2014). They produced
inventory maps using advanced pixel-based and object-oriented
image classification approaches optimized by Taguchi method
(Taguchi 1990). Their results showed that object-oriented ap-
proaches outperform pixel-based ones. Therefore, we used
Taguchi-based optimized object-oriented approaches in this study.
Then, an intensive field survey has been conducted to validate and
precisely detect and record the landslides geographical coordi-
nates using a GPS. A total number of 102 landslides were recorded.
roads, highways in the mountainous regions of the study area
(Fig. 3). The maximum and minimum size of landslides in the
study area is 0.53 and 0.01 km2, respectively.
Landslide conditioning factors
Determination of the landslide conditioning factors is a key step in
the landslide susceptibility assessment (Ercanoglu and Gokceoglu
2002). Fourteen landslide conditioning factors including elevation,
slope gradient, slope aspect, slope length, topographic wetness
index (TWI), stream power index (SPI), plan curvature, profile
curvature, geology, normalized difference vegetation index
(NDVI), land use, and distance from drainage, fault, and road
have been considered in this research according to the literature
review. Thematic layers were acquired from different resources
using ArcGIS, SAGA GIS, and ENVI packages (Bhner et al.
2006). The descriptions of the used input data, their spatial reso-
lution, and the sources are given in Table 2 and briefly explained
below. As mentioned in this table the resolution of all conditioning
factors is 15 m because we used ASTER satellite imagery (15-m
resolution) to produce several base maps. Total number of pixels is
1,211,275 in the study area from which 14,125 pixels are landslides.
Elevation has been used as a controlling parameter for land-
slides in several research studies (Greenbaum et al. 1995; Jordan
et al. 2000; Dai et al. 2001; Yilmaz 2010). It is well known that
elevation influences a large number of biophysical parameters,
anthropogenic activities, and soil properties (Vivas 1992; van
Beek and van Asch 2004; Kuriakose et al. 2009b). The other
important parameter in landslide susceptibility is slope gradient.
It determines the convergence and divergence of water on a
hillslope and thereby influences soil water content, soil forma-
tion, erosion potential, and slope stability (Lee and Min 2001;
Clerici et al. 2002; Lee et al. 2004; Lee 2005; Vijith and Madhu
2008). Slope aspect which is defined as the compass direction of
the maximum rate of change is also considered by several
researchers as a landslide conditioning factor (Gokceoglu and
Aksoy 1996; Saha et al. 2005; Ercanoglu and Gokceoglu 2004;
Lee et al. 2004). It is reported by Lineback et al. (2001) that
more landslides occur in the wetter north facing aspects than in
drier, south facing ones. Moisture retention and vegetation are
mainly reflected by slope aspect and then influence landslide
initiation to some extent.
Slope length can be considered as another important factor in
landslide activity (Carrara et al. 1995). It is the distance along a
slope subject to uninterrupted overland flow, from of the point at
which overland flow begins to where deposition starts, or where
flow enters a well-defined channel (Wischmeier and Smith 1978),
since longer slope lengths increase the potential of erosive agents
to dislodge and transport materials downslope. Moreover,
 (A)

(B)

(C)

Fig. 1 Study area: a location of Iram Province, b hill-shade of the study area, c geological map of the study area

downslope water velocity is greater on longer slopes (Gomez and
Kavzoglu 2005). The slope length is of paramount importance for
the travel distance of materials. Slope length (LS) factor can be
calculated by the equation proposed by Moore and Burch (1986) as
follows:



As 0:6 sin 1:3
LS 1:4 1 slope angle (in degrees).
22:13 0:0896
where AS is the specific watershed area (m2/m) and is slope angle
(in degrees).
TWI represents a theoretical measure of the accumulation of
flow at any point within a drainage basin and the tendency of
water to move downslope by gravitational forces (Fredlund 1987;
Kuriakose et al. 2009a). The index can be thought of as an abstract
parameter to be used as a basis for estimating the local soil
moisture status (Burt and Butcher 1986; Moore et al. 1991). Soil
moisture plays an important role in slope instability, particularly
for shallow landslides. It is defined as follows (Beven and Kirkby
1979; Moore et al. 1991):


As
TWI ln 2
tan
where As is the specific catchment area (SCA) and is the local
SPI is a measure of the erosive power of water flow based on the
assumption that discharge (q) is proportional to specific catch-
ment area (As) (Eq. 3) (Moore et al. 1991).
SPI AS tan 3
where AS is the specific catchments area (m2/m) and is the slope
gradient (in degrees). Moore et al. (1993) stated that the SPI
controls the potential erosive power of overland flow. Therefore,
these processes can be considered as one of the components of

Landslides 13 & (2016) 99

 Original Paper
Table 1 Types of geological formations in the study area
Geological era Code Formation Lithology
Quaternary Qt Alluvium Alluvium and recent deposits
Miocene Gs Gachsaran Alternation of anhydrite and red to gray marl with intrebedded limestone
Eocene-Oligocene AS Asmari Cream, white well bedded to massive limestone, dolomitic limestone, and marl
Paleocene-Eocene Pd Pabdeh Alternation of gray, white shale, marly limestone, and marl
Upper Cretaceous Lbm Lahbari member Alternation of light brown to gray marl and sandstone
Upper Cretaceous Gu Gurpi Gray, blue shale, and marly limestone with pyrite
Upper Cretaceous Ehm Emamhasan member Gray, thick-bedded limestone with marl
Upper Cretaceous II-Sg Ilam Gray, white thin- to medium-bedded limestone with interbedded of shale
Upper Cretaceous Sg Surgah Gray, dark shale, and interbedded yellow marly limestone
Upper Cretaceous Sv Sarvak Gray, dark thin to massive bitominous limestone with dark shale

landslide occurrence (Lee and Min 2001; Gokceoglu et al. 2005;
Nefeslioglu et al. 2008; Yilmaz 2009; Akgun and Turk 2010).
Distance from drainage, distance from fault, and distance from
road are the other important landslide conditioning factors
(Nagarajan et al. 2000; Dai et al. 2001; Cevik and Topal 2003;
Ercanoglu and Gokceoglu 2004; Pradhan 2010a, b). Drainage net-
works have negative impacts on landslide susceptibility since they
abrade the slope base and saturate the underwater section of the
material forming the slope (Vijith and Madhu 2008). Similar to the
effect of the distance to streams, landslides may occur on the road
and on the side of the slopes affected by roads. This is mainly due
to the fact that the natural condition of the slope is damaged
during the process of road construction when engineering princi-
ples, particularly principles related to drainage network are not
taken into account (Ayalew and Yamagishi 2005). Highly faulted
zones are areas of particularly high incidence of unstable slopes
and the degree of fracturing, and shearing plays an important role
in determining slope stability (Varnes 1984). The plan and profile
curvature which control the concentration, acceleration, and de-
celeration of subsurface flow can affect the occurrence of

Fig. 2 Flowchart of the study

100 Landslides 13 & (2016)

Fig. 3 Landslide inventory map (spatial distribution of landslides in the study area)

landslides (Ercanoglu and Gokceoglu 2002; Troch et al. 2003;
Talebi et al. 2008; Nefeslioglu et al. 2008; Oh and Pradhan
2011). Landslides are also significantly controlled by the lithol-
ogy properties of the land surface (Dai et al. 2001; Duman
et al. 2006). Geological factor is considered as an independent
variable in landslide formation in which lithology and its
varied structures tend to lead to a variation in stone stability
and strength and also to a varied soil texture (Ayalew and
Yamagishi 2005).
The next important factor which affects the occurrence of
landslide is land cover and/or land use (Gokceoglu and Aksoy
1996; van Beek et al. 2005; Kuriakose and van Beek 2011; Devkota
et al. 2012). For example, converting a forested area to grassland or
one where crops are cultivated can increase the moisture in the
soil enough to cause landslide problems. In this study, land cover/
land use map was extracted from ASTER satellite imagery (August
2013) using a pixel-based classification approach with the SVM
algorithm. NDVI is a measure of surface reflectance and gives a
quantitative estimate of the vegetation growth and biomass (Hall
et al. 1995). The NDVI was calculated using the ASTER satellite
imagery. Figure 4 shows the input parameters (conditioning fac-
tors) in the study area.
Normalization of the input dataset
In this step, the frequency ratio (FR) for each individual attribute
class was calculated in each map (Pradhan and Lee 2010). Each

Table 2 Descriptions of the input data, their spatial resolution and sources
Spatial data attributes (notation) Source, scale/resolution Short description
Elevation ASTER image, 15 m DEM of the terrain surface
Slope DEM, 15 m Angle of the slope inclination
Aspect DEM, 15 m Exposition of the slope
Slope length DEM, 15 m Length factor of the slope
Topographic wetness index DEM, 15 m Ratio of contributing area to the tangent of slope
Stream power index DEM, 15 m Multiplication of contributing area and the tangent of slope
Plan curvature DEM, 15 m Index of concavity parallel to the slope
Profile curvature DEM, 15 m Index of concavity perpendicular to slope
Distance from stream DEM, 15 m Buffer of drainage network
Lithology Geo-map, 1:100,000ASTER image, 15 m Rock units
Distance from fault Geo-map, 1:100,000ASTER image, 15 m Buffer of structure
Distance from road Topo-map, 1:25,000ASTER image, 15 m Buffer of road
Landuse/landcover ASTER image, 15 m Image classification using SVM algorithm
NDVI ASTER image, 15 m Interpretation of vegetation, water bodies, and bare soil,
based on NDVI

Landslides 13 & (2016) 101

 Original Paper
Fig. 4 Input parameters (landslide conditioning factors) in the study area: DEM (a), slope (b), aspect (c), plan curvature (d), profile curvature (e), LS factor (f), TWI (g), SPI
(h), lithologic units (i), land cover (j), NDVI (k), distance to road (l), distance to drainage channel (m), distance to fault (n)
attribute class was then assigned a sequence number based on the
FR value. Then, the max-min normalization method was used
to rescale in the range 0.1 to 0.9 using Eq. (4) (Wang and
Huang 2009):
Min
0 UL L
Max Min
where v' is the normalized data matrix, v is the original data
matrix, and U and L are the upper and lower normalization
boundaries. Table 3 shows the values of FR for each class of
conditioning factors.
Landslide susceptibility mapping
Here, two approaches have been used to produce landslide sus-
ceptibility maps. Initially, landslide susceptibility maps were pro-
duced using four single statistical models, i.e., ANN, SVM, GLM,
and MaxEnt. Then, hybrid wavelet-statistical models were
102 Landslides 13 & (2016)
developed using wavelet packet transform. These approaches are
explained below.
Single statistical models
An ANN can be defined as a computational mechanism able to
acquire, represent, and compute a mapping from one multivariate
4
space of information to another, given a set of data representing
that mapping (Garrett 1994). The advantage of neural networks is
related to the high computation rate accomplished by their
inherent parallelism that is a result of a potent arrangement
of interconnections (weights) and simple processors (neu-
rons), which allows processing of very large datasets (Frizzelle
and Moody 2001).
In this part, 102 landslide points were randomly split into two
datasets (75 % of the data for training phase and 25 % for valida-
tion phase). The feed-forward back-propagation network which is
the most frequently used neural network method was used in this
study. This algorithm is a multi-layered neural network, which
Table 3 FR values for different conditioning factors
Domain Class Class pixels (%) Landslide pixels (%) Frequency Normalized
ratio classes
Elevation (m) 9001400 40.38 73.39 1.82 0.90
14001900 47.47 25.80 0.54 0.34
19002400 11.29 0.80 0.07 0.13
>2400 0.86 0.00 0.00 0.10
Slope Angle (degree) 015 37.75 35.32 0.94 0.17
1530 29.80 25.04 0.84 0.10
3045 18.52 17.43 0.94 0.18
4565 10.38 15.43 1.49 0.58
>65 3.55 6.77 1.91 0.90
Slope aspect Flat (1) 0.00 0.00 0.00 0.10
N (022.5; 337.5360) 12.23 16.38 1.34 0.61
NE (22.567.5) 13.93 29.59 2.12 0.90
E (67.5112.5) 8.04 14.26 1.77 0.77
SE (112.5157.5) 8.04 6.44 0.80 0.40
S (157.5202.5) 14.08 6.64 0.47 0.28
SW (202.5247.5) 22.41 4.61 0.21 0.18
W (247.5292.5) 12.50 9.24 0.74 0.38
NW (292.5337.5) 8.77 12.85 1.46 0.65
Slope length 04 36.91 33.32 0.90 0.13
48 23.91 21.78 0.91 0.14
812 16.51 14.47 0.88 0.10
1216 14.59 17.36 1.19 0.43
1620 6.66 10.88 1.64 0.90
>20 1.42 2.19 1.54 0.80
TWI <10 73.07 79.53 1.09 0.90
1015 24.42 19.24 0.79 0.52
1520 2.21 1.09 0.49 0.14
>20 0.31 0.14 0.46 0.10
SPI 025 37.08 33.67 0.91 0.10
2550 18.08 18.39 1.02 0.46
5075 11.24 12.70 1.13 0.84
75100 7.11 8.17 1.15 0.90
>100 26.48 27.07 1.02 0.48
Plan curvature Divergent 11.75 10.18 0.87 0.09
Parallel 70.86 69.80 0.98 0.43
Convergent 17.39 20.03 1.15 0.90
Profile curvature Convex 6.58 11.29 1.72 0.90
Straight 79.42 75.93 0.96 0.15
Concave 14.00 12.78 0.91 0.10
Distance to drainages (m) 0100 4.95 21.46 4.33 0.81
100200 6.49 31.57 4.86 0.90

Landslides 13 & (2016) 103

 Original Paper
Table 3 (continued)
Domain Class Class pixels (%) Landslide pixels (%) Frequency Normalized
ratio classes
200300 15.46 17.51 1.13 0.27
300400 29.81 23.79 0.80 0.21
>400 43.28 5.68 0.13 0.10
Distance to fault (m) 01000 18.34 13.74 1.33 0.74
10002000 22.30 15.50 1.44 0.84
20003000 25.60 16.96 1.51 0.90
>3000 33.75 53.81 0.63 0.10
Geology formation Alluvium 9.99 10.41 0.96 0.38
Asmari 1.08 9.38 0.12 0.13
Emamhasan member 0.20 4.25 0.05 0.11
Gachsaran 0.02 0.87 0.02 0.11
Gurpi 33.70 17.18 1.96 0.67
Ilam 23.43 12.94 1.81 0.62
Lahbari member (old landslides) 19.05 6.89 2.76 0.90
Pabdeh 10.28 26.25 0.39 0.21
Sarvak 0.00 10.00 0.00 0.10
Surgah 2.26 1.84 1.23 0.46
Land use Lake 0.65 0.16 0.25 0.11
Horticulture 0.75 0.17 0.23 0.10
Dry farming 2.74 4.43 2.00 0.71
Rangeland 54.21 54.01 0.98 0.44
Forest 27.92 13.20 0.47 0.21
Baren land 13.74 28.03 2.04 0.90
NDVI 0.270.1 9.93 12.76 1.28 0.90
0.10 71.09 86.61 1.22 0.86
00.1 17.23 0.59 0.03 0.11
0.10.2 1.26 0.03 0.02 0.10
>0.2 0.49 0.01 0.02 0.10
Distance to roads (m) 0100 5.51 19.16 3.48 0.90
100200 5.12 15.02 2.93 0.75
200300 4.92 10.71 2.18 0.53
300400 4.52 4.24 0.94 0.18
>400 79.93 50.86 0.64 0.10

consists of an input layer, hidden layers, and an output layer. The
hidden and output layer neurons process their inputs by multi-
plying each input by a corresponding weight, summing the prod-
uct, and then processing the sum using a nonlinear transfer
function to produce a result. It learns by regulating the weights
between the neurons in response to the errors between the actual
output values and the target output values. In this study, a three-
layered, interconnected neural network consisting of an input
layer, a hidden layer and an output layer was used. In this specific
case, there are 14 input nodes (one each for the elevation, slope
gradient, slope aspect, slope length, TWI, SPI, plan curvature,
profile curvature, distance from drainage, geology, distance to
fault, land use, distance to road, and NDVI). There are several
methods to determine the optimum number of nodes in the
hidden layer including 2n+1 (Kavzoglu and Mather 2003), 2n
(Wong 1991), n (Tang and Fishwick 1993) hidden neurons, where
n is the number of nodes in the input layer. Mishra and Desai
(2006), Moosavi et al. (2013b) and Shirmohammadi et al. (2013)
declared that the trial and error is an appropriate way to deter-
mine the best number of neurons in the hidden layer. As the

104 Landslides 13 & (2016)

abovementioned formulas can be used to obtain an approximate
range in which the best number of neurons in the hidden layer can
be determined, 1430 nodes were tested as the number of nodes in
the hidden layer. In fact, the optimum number of nodes was
determined using a common trial-and-error approach. The
hyperbolic tangent sigmoid, y(x)=[2 / (1+exp(_2x)) _ 1], was
used as transfer function for the hidden layer and the log-
sigmoid, y(x)=[1 / (1+exp(x))], for the output layer. The land-
slide susceptibility value for each pixel is between zero and
one. Therefore, the results of the ANN modeling should be in
this range. As the log-sigmoid function scales its output be-
tween zero and one, all susceptibility values automatically
remain in the range. The same 7525 % data separation
scheme was used for SVM, MaxEnt, and GLM models.
SVM is a supervised learning method based on statistical learn-
ing theory and the structural risk minimization principle (Vapnik
2001). SVM constructs a hyper-plane or set of hyper-planes in a
high- or infinite-dimensional space, which can be used for classi-
fication, regression, or other tasks. Using the training data, SVM
implicitly maps the original input space into a high-dimensional
feature space (Kanevski et al. 2009). Subsequently, in the feature
space, the optimal hyper-plane is determined by maximizing the
margins of class boundaries (Abe 2010). The training points that
are closest to the optimal hyper-plane are called support vectors.
The choice of the kernel function is crucial for successful SVM
training and classification accuracy. Two SVM types, i.e., c-SVC
and nu-SVC, and four types of kernel functions, i.e., linear kernel
(LN), polynomial kernel (PL), radial basis function (RBF) kernel,
and sigmoid kernel (SIG), were used in this study. The LN is
considered to be a specific case of RBF, whereas the SIG behaves
like the RBF for certain parameters (Song et al. 2011). In the
present study, the four mentioned kernel functions were tested
to determine the one which gives the best result.
The principle of maximum entropy states that, subject to pre-
cisely stated prior data (such as a proposition that expresses
testable information), the probability distribution which best rep-
resents the current state of knowledge is the one with the largest
entropy. The MaxEnt model is a niche-based model that suppose
the distribution of the presence data, represents the realized niche
(Heumann et al. 2011). MaxEnt method is a generative approach
which is used to make predictions using statistical concepts
(Phillips et al. 2006). It analyzes individually the weights of
environmental factors and calculates a continuous probability
value for the distribution of a certain phenomenon. As with
other machine learning models, the maximum entropy model
requires some optimization procedures. Thus, before the gen-
eration of the landslide susceptibility map, optimal settings
are first searched by predictive performance measures based
on cross-validation. Using optimal settings, the landslide sus-
ceptibility map over the study area is generated, and the
relationships between the input causal factors and landslide
susceptibility are interpreted.
GLM which allows one to form a multivariate regression
relation between a dependent variable and several indepen-
dent variables is also widely used for landslide susceptibility
mapping (Ayalew and Yamagishi 2005; Brenning 2005). Logis-
tic regression as a generalized linear model with a logit link
function, which is one of the most frequently used method in
landslide susceptibility modeling, was used in this study
(Brenning 2005). This model consists of an additive combina-
tion of single parametric terms, each representing a linear
function of a single predictor.
Hybrid models based on wavelet transform
The wavelet transform as a robust signal processing approach has
been explored in numerous studies (Zou et al. 2008; Wong et al.
2009; Li 2009; Tagluk et al. 2010; Wang et al. 2010; Shirmohammadi
et al. 2013; Moosavi et al. 2013a, b, ). The wavelet transform of a
signal is family coefficients of wavelet, depending on two param-
eters namely scale and translation. According to the analysis
requirements of the signal, continuous wavelet transform (CWT)
or discrete wavelet transform (DWT) can be used. CWT of a signal
x(t) can be defined as Eq. 5.
Z  t 
1
CWT x ; s p xt * dt 5
js j s
where x(t) is the original signal and is function of time or position
(here is function of position), s is the scale parameter, is the
translation parameter, and * represents the complex conjugate
(Cannas et al. 2006). The mother wavelet (t) is the transforming
function. The CWT computation requires a large quantity of cal-
culation time and resources. On the other hand, the DWT scales
and positions are usually based on powers of two (dyadic scales
and positions). It can be introduced as Eq. 6.
!
1 tk 0 s0j
j;k t q
  6
s j  s0j
0
where j and k are integers and s0, which is greater than zero,
is a fixed dilation step (Cannas et al. 2006). The DWT per-
forms two sets of functions namely high-pass and low-pass
filters. A two-dimensional wavelet decomposition can be de-
scribed as a one-dimensional discrete wavelet transform ap-
plied consecutively along the horizontal (rows of the map or
image) and vertical (columns of the map or image) axes
(Mallat 1989).
The wavelet packet decomposition which is a generalization
of the conventional wavelet decomposition that offers a richer
signal analysis (Garcia et al. 2000) was used in this study. In
the first step, the original input map was decomposed into
coarser approximation coefficients (CA) which contains low
frequency horizontal and vertical components of the map,
horizontal detail coefficients (Dhorizontal), vertical detail coeffi-
cients (Dvertical), and diagonal detail coefficients (Ddiagonal).
This procedure was iteratively performed in different levels.
Consequently, the original signal was decomposed to its ap-
proximation and detail components. It finally resulted in a
wavelet decomposition tree. In this study, the input maps
were decomposed using different mother wavelets such as
Daubechies (db), Symlets (Sym), Coiflets (coif), and
BiorSplines (bior) in different levels (1, 2, 3, 4, and 5) using
Matlab software.
Figure 5 shows the decomposition tree obtained from de-
composition of original NDVI map using bior1.5 mother wave-
let in level 3 as an example. In this figure, the first node (0,
0) is the original map (i.e., NDVI) that is decomposed to four
Landslides 13 & (2016) 105
 Original Paper
Fig. 5 An example of wavelet packet decomposition tree (decomposition using bior1.5 in level 3)
components namely CA, D horizontal , D vertical , and Ddiagonal
(which is shown as (1, 0), (1, 1), (1, 2), and (1, 3) nodes,
respectively) in the first level. In the next level, each of these
components is decomposed to its subcomponents (which are
shown as (2, 0), (2, 1), (2, 2), etc.). The same procedure is
carried out in level 3. As it is shown from this figure, a large
number of components are made especially when the decom-
position level increases. It is not possible to use all of these
components in the modeling process. Classical entropy-based
criteria are suitable for this purpose. Entropy is a common
concept in many fields, mainly in signal processing
(Daubechies 1988). Matlab has a function called besttree in
order to compute the optimal subtree from the initial
106 Landslides 13 & (2016)
Level 1
Level 2
Level 3
decomposition tree. This function uses an entropy type crite-
rion. However, as in several cases, the optimal subtree that is
obtained by this function is still large, the five components
that had the least entropy values were selected. Therefore, the
best approximation and detail components (the five compo-
nents with the least entropy values) of each conditioning
factor were used as inputs for all the aforementioned models.
Model performance criteria
Using several statistical evaluation criteria such as true positive
(TP), false positive (FP), true negative (TN), and false negative
(FN), the overall accuracy of the landside susceptibility models
were calculated. The reliability of the landslide susceptibility
models were estimated using Cohens kappa index () (Guzzetti
et al. 2006):
Pobs Pexp
7
1Pexp
Pobs TP TN=n 8 combinations of these, i.e., interaction features (Goldberg and
p and Bruzzone (2005). ANNs also show second best results. Neural
Pexp TP FN TP FP FP TN FN TN = N 9
where n is the proportion of pixels that is correctly classified as
landslide or nonlandslide and N is the number of total training
pixels. Kappa is intended to give us a quantitative measure of the
magnitude of agreement between observers. A kappa of 1 indicates
perfect agreement, whereas a kappa of 0 indicates agreement
equivalent to chance (Viera and Garrett 2005). As another criteri-
on, goodness of fit of each landslide susceptibility model was
evaluated using a receiver operating curve (ROC) and by comput-
ing the area under the ROC curve (AUC). The ROC curve plots the
true-positive rate (TPR) versus the false-positive rate (FPR), where
TPR is the proportion of landslide area that is correctly classified
as susceptible and FPR is the proportion of nonlandslide area
classified as susceptible (Fratinni et al. 2010). The range of values
of the ROC curve area is 0.51 for a good fit, while values below 0.5
represent a random fit (Egan 1975; Fawcett 2006).
Results and discussion
Landslide susceptibility maps were generated using single models
(models without wavelet-based preprocessing on the input factors)
and hybrid models (models with wavelet based preprocessing on
the input factors). Table 4 shows the evaluation criteria for both
single and hybrid models. The boldface figures in this table refer to
the best values of performance criteria for single and hybrid
models. According to this table, the SVM (nu-SVC) model with
the polynomial kernel function outperforms the others in both
single and hybrid models. This may be related to the SVMs
capabilities which are briefly explained below. It solves the convex
optimization problem in which all local solutions (e.g., individual
landslide events) are classified into a global optimum (Bishop
1994). SVMs get flexibility in the selection of the form of the
threshold separating different classes using the kernel function,
which is not necessary to be linear or even have the same func-
tional form for all data, since its function is nonparametric and
operates locally. The other advantage of SVM is that no assump-
tion about the functional form of the transformation, which makes
data linearly separable, is necessary because the kernel implicitly
contains a nonlinear transformation. Also, SVMs provide a good
out-of-sample generalization, if the parameters are appropriately
chosen. This means that, even if the training sample has some bias,
choosing proper parameters, SVMs perform appropriately. SVMs
provide a unique solution, since the optimality problem is convex.
This is an advantage in comparison to the neural networks, which
have multiple solutions related to local minima and therefore they
may not be robust over different samples. The other important
point is that different kernel functions make different results in
SVM modeling. This is related to the nature of the data. For
example here, polynomial function is better fitted to the data in
comparison with other kernel functions particularly sigmoid ker-
nel function that did not fit the data appropriately. Polynomial
kernel function characterizes the similarity of training samples in a
feature space over polynomials of the original variables, allowing
learning of nonlinear models. It looks not only at the given
features of input samples to determine their similarity but also
Elhadad 2008). This result is in accordance with Zhu and
Blumberg (2002), Foody and Mathur (2004), and Camps-Valls
networks are able to model complex nonlinear relationships be-
tween independent and dependent variables (Hinton 1992). They
can detect nonlinearities automatically adjusting the connection
weights in their structure. The predictor variables in a neural
network usually undergo a nonlinear transformation at each hid-
den node and output node; therefore, they can model much more
complex nonlinear relationships (that exist in the natural phenom-
ena) than other models such as GLM and MaxEnt can (Tu 1996). In
the GLM model, it should be mentioned that the absence of strong
collinearity (strong correlation between the predictors) is very
important. While not necessarily reducing the predictive value of
the model, collinearity will inflate the standard errors of the
estimates of the model coefficients and cause unreliable results
(Tabachnick and Fidell 1996). Some variables such as slope, dis-
tance to fault, and geological conditions may show degrees of
collinearity, and the poor performance of GLM may be related to
this fact.
The other notable result is that the performance evaluation
criteria are significantly enhanced when the wavelet packet trans-
form is applied on the input maps in all four models (ANN, SVM,
MaxEnt, and GLM). The bior1.5 was the best mother wavelet which
shows the highest kappa (0.978), TPR (0.962), and AUC (0.977). As
mentioned before, the wavelet transform was performed in differ-
ent levels, i.e., 1, 2, 3, 4, and 5. Figure 6 shows that the performance
of the hybrid models increases by increasing the decomposition
level up to level 3 and after that, decreases dramatically. The
evaluation criteria for the hybrid models shown in Table 4 belong
to the best decomposition level, i.e., level 3.
Wavelet packet transform can significantly improve the ability
of empirical models such as neural network and support vector
machine. It can be related to the ability of wavelet packet to
decompose the original map or image to its components. In a
discrete wavelet packet decomposition of the original map or
image with J levels, both the detail and approximation coefficients
are decomposed at each level, instead of decomposing only the
approximation component in the usual wavelet decomposition.
The main two advantages of the wavelet packet decomposition is
that the texture information belongs to all frequency bands and
not only the approximation one and the coefficients of a wavelet
packet decomposition are less correlated together than the ones
issued from a usual wavelet decomposition. Wavelet transforms
offer a framework in which a signal is decomposed, with each level
corresponding to a coarser resolution or lower frequency band.
For a given wavelet function, a library of bases called wavelet
packet bases is generated. Each of these bases presents a specific
way of coding signals, conserving global energy, and
reconstructing exact features. Wavelet packet components are
Landslides 13 & (2016) 107
 Original Paper
Table 4 Accuracy assessment of the single and hybrid models
Model (single) Evaluation criteria
Kappa index TPR AUC
ANN 0.829 0.904 0.855
SVM(C-SVC)
Linear kernel function (LN) 0.800 0.862 0.818
Radial basis kernel function (RBF) 0.804 0.885 0.839
Polynomial kernel function (PL) 0.823 0.889 0.846
Sigmoid kernel function (SIG) 0.533 0.682 0.608
SVM (nu-SVC)
Linear kernel function (LN) 0.822 0.863 0.834
Radial basis kernel function (RBF) 0.824 0.896 0.85
Polynomial kernel function (PL) 0.846 0.919 0.861
Sigmoid kernel function (SIG) 0.489 0.634 0.572
MaxEnt 0.744 0.805 0.83
GLM 0.735 0.801 0.816
Model (hybrid-db4)
ANN 0.931 0.927 0.936
SVM(C-SVC)
Linear kernel function (LN) 0.844 0.900 0.899
Radial basis kernel function (RBF) 0.909 0.901 0.910
Polynomial kernel function (PL) 0.922 0.908 0.921
Sigmoid kernel function (SIG) 0.756 0.866 0.623
SVM (nu-SVC)
Linear kernel function (LN) 0.889 0.901 0.921
Radial basis kernel function (RBF) 0.930 0.918 0.933
Polynomial kernel function (PL) 0.956 0.951 0.955
Sigmoid kernel function (SIG) 0.539 0.678 0.822
MaxEnt 0.807 0.894 0.920
GLM 0.827 0.890 0.918
Model (hybrid-sym4)
ANN 0.923 0.929 0.948
SVM (C-SVC)
Linear kernel function (LN) 0.814 0.901 0.906
Radial basis kernel function (RBF) 0.889 0.916 0.922
Polynomial kernel function (PL) 0.916 0.922 0.932
Sigmoid kernel function (SIG) 0.800 0.838 0.882
SVM (nu-SVC)
Linear kernel function (LN) 0.809 0.876 0.902
Radial basis kernel function (RBF) 0.919 0.908 0.936
Polynomial kernel function (PL) 0.933 0.949 0.953
Sigmoid kernel function (SIG) 0.539 0.674 0.673
MaxEnt 0.820 0.901 0.918
GLM 0.810 0.890 0.906
Model (hybrid-coeif4)

108 Landslides 13 & (2016)

Table 4 (continued)
Model (single) Evaluation criteria
Kappa index TPR AUC
ANN 0.912 0.931 0.950
SVM(C-SVC)
Linear kernel function (LN) 0.844 0.899 0.895
Radial basis kernel function (RBF) 0.880 0.912 0.930
Polynomial kernel function (PL) 0.910 0.914 0.934
Sigmoid kernel function (SIG) 0.765 0.855 0.809
SVM(nu-SVC)
Linear kernel function (LN) 0.879 0.914 0.868
Radial basis kernel function (RBF) 0.912 0.926 0.947
Polynomial kernel function (PL) 0.932 0.945 0.951
Sigmoid kernel function (SIG) 0.539 0.689 0.661
MaxEnt 0.819 0.906 0.913
LM 0.800 0.904 0.892
Model (hybrid-bior1.5)
ANN 0.941 0.957 0.947
SVM (C-SVC)
Linear kernel function (LN) 0.889 0.932 0.916
Radial basis kernel function (RBF) 0.911 0.943 0.921
Polynomial kernel function (PL) 0.932 0.947 0.923
Sigmoid kernel function (SIG) 0.800 0.867 0.735
SVM(nu-SVC)
Linear kernel function (LN) 0.930 0.907 0.911
Radial basis kernel function (RBF) 0.933 0.954 0.935
Polynomial kernel function (PL) 0.978 0.962 0.977
Sigmoid kernel function (SIG) 0.600 0.671 0.909
MaxEnt 0.839 0.844 0.925
GLM 0.817 0.837 0.919

waveforms indexed by three naturally interpreted parameters:
position, scale (as in wavelet decomposition), and frequency. Con-
sequently, all of the advantages of the fast wavelet transform are
preserved since the wavelet basis is in the range of bases available
with the wavelet packet transform (Shinde 2004). The complete
structure of wavelet packet transform makes flexibility for the
(particularly in single form) which categorized most of the area as
very susceptible to landslide occurrence. In contrast, FN is high
in GLM which indicates that this model could not show susceptible
0.99
0.98
Performance criteria

signal representation to achieve more accuracy. It provides the 0.97

most suitable frequency subbands for the signal representation. 0.96
0.95 Kappa
Figure 7 shows the best landslide susceptibility maps resulted
TPR
from single and hybrid landslide susceptibility models. There are 0.94
AUC
many classification methods available such as quantiles, natural 0.93

breaks, equal intervals, and standard deviations (Ayalew and 0.92

Yamagishi 2005). The obtained susceptibility maps were classified 0.91

0.9
using the quantile method in which the data classifies into a 1 2 3 4 5
certain number of categories with an equal number of units in Decomposition level
each category (ESRI 1995). Figure 7 clearly shows that TP and TN
are high in SVM and ANN models particularly in their hybrid Fig. 6 Performance of the best hybrid model (SVM (nu-SVC) polynominal kernel
forms. On the other hand, FP is very high in the MaxEnt model function, hybrid-bior1.5) in different wavelet packet decomposition levels

Landslides 13 & (2016) 109

 Original Paper
Fig. 7 Landslide susceptibility maps
resulted from single and hybrid landslide
susceptibility models

110 Landslides 13 & (2016)

areas properly. Therefore, SVM and ANN models outperform
MAxEnt and GLM models in both single and hybrid forms.
According to the Fig. 7, the spatial distributions of the suscep-
tible area are consistent with the observed landslide distribution
pattern particularly for ones obtained from the hybrid models.
Lithology is a particularly important parameter of landslide oc-
currence, since different lithologic units have different sensitivities
to landslides. The landslide susceptibility maps suggest that the
marl and shale possess the highest landslide occurrence values,
while the massive limestone has the lowest landslide occurrence
values. In this case, the presence of weak planes in marls and
shales made the rocks weak, creating slope instability (Eberhardt
et al. 2005). The high amount of weathering and the presence of
clay-rich soils in these areas seemed to be the main reason for
making the slopes unstable. Only a few slides were found in the
limestone areas; however, the slides that were present were related
to the existence of joints and local presence of marl and shale.
These findings agree with those of El Khattabi and Carlier (2004),
Gokceoglu et al. (2005), Sidle and Ochiai (2006), Lee and Pradhan
(2007), Akgun et al. (2008) and Khanh (2009) which has found
that variations in landslide distribution are highly dependent on
lithology type. The strength of the rocks is also strongly affected by
tectonic and current weathering particularly, tectonics associated
with slope instability by fracturing, faulting, jointing, and
deforming the foliation structures (Sanchez et al. 2010). The fre-
quency ratio of landslides was higher in bare soils and dry farming
and lower in forests and agricultural areas. Several studies have
produced similar results for the relationship between landslides
and land use (Dai et al. 2001; Kanungo et al. 2006; Nagarajan et al.
2000). As it is obvious in all produced maps, the LS is significantly
affected by the pattern of roads and rivers in the study area. In
general, landslides have been observed to occur in greater spatial
densities when adjacent to roads. Actually, landslide events
decrease with the increasing the distance from road. This trend
discloses that road networks play a significant role in landslide
initiation. This is in accordance with the results of Guthrie (2002)
and Guthrie et al. (2010). Streams also can adversely affect stability
by eroding slopes or saturating the lower part of a material until
the water level increases. The convex portions in the study area
have higher LS values than the other parts because of repeated
dilation and contraction of loose debris on a disposed surface and
can provoke creeping or a mudslide due to heavy rainfall (Pradhan
2011). This finding is in accordance with those of Havenith et al.
(2006) and Conforti et al. (2011).
Conclusion
The main objective of this research was to develop hybrid wavelet
packet-statistical models for landslide susceptibility mapping. The
four single statistical methods, i.e., artificial neural network, sup-
port vector machine, generalized linear model, and maximum
entropy show moderate or approximately poor performance.
However, application of wavelet packet transform on the input
maps significantly enhanced the performance of the mentioned
models to produce landslide susceptibility maps. This enhance-
ment can be related to the ability of wavelet packet to decompose
the original maps to their components. The main advantage of
wavelet transform lies in its multi-resolution decomposition capa-
bility. In contrast to Fourier-based transform, it offers better lo-
calization properties in both spatial and frequency domains. This
result confirms that robust preprocessing approaches such as
wavelet transform can be a good idea to deal with the complexity
of modeling the natural phenomena. GLM and MaxEnt do not
perform satisfactorily in both single and hybrid forms, but ANN
and SVM performed fairly well in their hybrid forms. The results
of this study also show that the SVM outperform other methods in
landslide susceptibility mapping. The produced landslide suscep-
tibility maps using hybrid wavelet-ANN and wavelet-SVM models
show that the pattern of susceptible areas follow the pattern of
roads. It extensively confirms the effect of human-made structures
on initiation of landslides. For future works, development of other
hybrid models with wavelet transform such as wavelet-ANFIS
models can be suggested.
References
Abe S (2010) Support vector machines for pattern classification. Springer 2010, pp 435
Ablay G, Hrliman M (2000) Evolution of the north flank of Tenerife by recurrent giant
landslides. J Volcanol Geotherm Res 103:135159
Adamowski J (2007) Development of a short-term river flood forecasting method based
on wavelet analysis. Polish Academy of Sciences Publication, Warsaw, p 172
Adamowski J, Sun K (2010) Development of a coupled wavelet transform and neural
network method for flow forecasting of non-perennial rivers in semi-arid watersheds.
J Hydrol 390(12):8591
Akgun A, Turk N (2010) Landslide susceptibility mapping for Ayvalik (Western Turkey)
and its vicinity by multi criteria decision analysis. Environ Earth Sci 61(3):595611
Akgun A, Dag S, Bulut F (2008) Landslide susceptibility mapping for a landslideprone
area (Findikli, NE of Turkey) by likelihood frequency ratio and weighted linear
combination models. Environ Geol 54(6):11271143
Atkinson PM, Massari R (1998) Generalised linear modelling of susceptibility to landslid-
ing in the central apennines, Italy. Comput Geosci 24(4):373385
Ayalew L, Yamagishi H (2005) The application of GIS-based logistic regression for
landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan.
Geomorphology 65(1/2):1531
Bednarik M, Magulov B, Matys M, Marschalko M (2010) Landslide susceptibility
assessment of the KraovanyLiptovsk Mikul railway case study. Phys Chem Earth
Parts A/B/C 35(35):162171
Beven K, Kirkby MJ (1979) A physically based, variable contributing area model of basin.
Hydrol Sci Bull 24:4369
Bishop AJ (1994) Cultural conflicts in mathematics education: developing a research
agenda. Learn Math 14(2):1518
Bhner J, McCloy KR, Strobl J (2006) SAGAanalysis and modelling applications, vol.
115. Gttinger Geographische Abhandlungen, Gttinger, p 130
Borah S, Hines EL, Bhuyan M (2007) Wavelet transform based image texture analysis for
size estimation applied to the sorting of tea granules. J Food Eng 79:629639
Brenning A (2005) Climatic and geomorphological controls of rock glaciers in the Andes
of Central Chile: combining statistical modelling and field mapping, Dissertation,
Mathematisch- Naturwissenschaftliche Fakultat II, Humboldt-Universitat zu Berlin,
Berlin, urn:nbn:de:kobv:11-10049648, pp, 412
Brckl EP (2001) Cause-effect models of large landslides. Nat Hazards 23:291314
Burt T, Butcher D (1986) Stimulation from simulation? a teaching model of hillslope
hydrology for use on microcomputers. J Geogr High Educ 10(1):2339
Camps-Valls G, Bruzzone L (2005) Kernel-based methods for hyperspectral image
classification. IEEE Trans Geosci Remote Sens 43(6):13511362
Cannas B, Fanni A, Sias G, Tronei S, Zedda MK (2006) River flow forecasting using neural
networks and wavelet analysis. Proceedings of the European Geosciences Union, In,
pp 234243
Carrara A, Cardinali M, Guzzetti F, Reichenbach P (1995) GIS technology in mapping
landslide hazard. In: Carrara A, Guzzetti F (eds) Geographical information systems in
assessing natural hazards. Kluwer, The Netherlands, pp 135175
Cevik E, Topal T (2003) GIS-based landslide susceptibility mapping for a
roblematic segment of the natural gas pipeline, Hendek (Turkey). Environ Geol
44:949962
Clerici A, Perego S, Tellini C, Vescovi P (2002) A procedure for landslide susceptibility
zonation by the conditional analysis method. Geomorphology 48:349364
Landslides 13 & (2016) 111
 Original Paper
Conforti M, Robustelli G, Muto F, Critelli S (2011) Application and validation of bivariate
GIS-based landslide susceptibility assessment for the Vitravo river catchment (Cala-
bria, south Italy). Nat Hazards 61:127141
Conforti M, Pascale S, Robustelli G, Sdao F (2014) Evaluation of prediction capability of
the artificial neural networks for mapping landslide susceptibility in the Turbolo River
catchment (northern Calabria, Italy). Catena 113:236250
Cruden DM, Varnes, DJ (1996) Landslide types and processes. In: Turner AK, Schuster RL
(eds) Landslides, investigation and mitigation. Special Report 247, Transportation
Research Board, Washington, DC, pp 36-75. ISSN: 0360-859X, ISBN: 030906208X
Dai FC, Lee CF, Li J, Xu ZW (2001) Assessment of landslide susceptibility on the natural
terrain of Lantau Island, HongKong. Environ Geol 40:381391
Daubechies I (1988) Orthonormal bases of compactly supported wavelets. Commun Pure
Appl Math 41:909996
Demirhan A, Guler I (2011) Combining stationary wavelet transform and self-organizing
maps for brain MR image segmentation. Eng Appl Artif Intell 24:358367
Devkota KC, Regmi AD, Pourghasemi HR, Yoshida K, Pradhan B, Ryu IC, Dhital MR,
Althuwaynee OF (2012) Landslide susceptibility mapping using certainty factor, index
of entropy and logistic regression models in GIS and their comparison at Mugling-
Narayanghat road section in Nepal Himalaya. Nat Hazards 65:135165
Duman TY, Can T, Gokceoglu C, Nefeslioglu HA, Sonmez H (2006) Application of logistic
regression for landslide susceptibility zoning of Cekmece Area, Istanbul, Turkey.
Environ Geol 51:241256
Eberhardt E, Thuro K, Luginbuehl M (2005) Slope instability mechanisms in dipping
interbedded conglomerates and weathered marls-the 1999 Rufi landslide, Switzer-
land. Eng Geol 77:3556
Egan J (1975) Signal detection theory and ROC analysis. Academic, New York
El Khattabi J, Carlier E (2004) Tectonic and hydrodynamic control of landslides in the
northern area of the Central Rif, Morocco. Eng Geol 71(34):255264
Ercanoglu M, Gokceoglu C (2002) Assessment of landslide susceptibility for a landslide-
prone area (North of Yenice, NW Turkey) by fuzzy approach. Environ Geol 41:720730
Ercanoglu M, Gokceoglu C (2004) Use of fuzzy relations to produce landslide suscepti-
bility map of a landslide prone area (West Black Sea Region, Turkey). Eng Geol
75:229250
ESRI (1995) ArcGIS Help File. ArcGIS
Fawcett T (2006) An introduction to ROC analysis. Pattern Recogn Lett 27:861874
Foody GM, Mathur A (2004) A relative evaluation of multiclass image classification by
support vector machine. IEEE Trans Geosci Remote Sens 42(6):13351343
Fratinni P, Crosta G, Carrara A (2010) Techniques for evaluating the performance of
landslide susceptibility models. Eng Geol 111:6272
Fredlund DG (1987) Slope stability analysis incorporating the effect of soil suction. In:
Anderson MG, Richards KS (eds) Slope stability: geotechnical engineering and
geomorphology. Wiley, Chichester, pp 113144
Frizzelle BG, Moody A (2001) Mapping continuous distributions of land covera
comparison of maximum likelihood estimation and artificial neural networks.
Photogramm Eng Remote Sens 3:6769367705
Garcia C, Zikos G, Tziritas G (2000) Wavelet packet analysis for face recognition. Image
Vis Comput 18:289297
Garrett JH (1994) Where and why artificial neural networks are applicable in civil
engineering. J Comput Civ Eng 8(2):129130, special iusse
Glade T, Anderson M, Crozier MJ (2005) Landslide hazard and risk. Wiley, New York, p 824
Gokceoglu C, Aksoy H (1996) Landslide susceptibility mapping of the slopes in the
residual soils of the Mengen region (Turkey) by deterministic stability analyses and
image processing techniques. Eng Geol 44:147161
Gokceoglu C, Sezer E (2009) A statistical assessment on international landslide literature
(19452008). Landslides 6(4):345351
Gokceoglu C, Sonmez H, Nefeslioglu HA, Duman TY, Can T (2005) Kuzulu landslide (Sivas,
Turkey) and landslide-susceptibility map of its near vicinity. Eng Geol 81(1):6583
Goldberg Y, Elhadad M (2008) splitSVM: fast, space-efficient, non-heuristic, polynomial
kernel computation for NLP applications. Proc. ACL-08: HLT. Proceedings of ACL-08:
HLT, Short Papers (Companion Volume), pages 237240
Gomez H, Kavzoglu T (2005) Assessment of shallow landslide susceptibility using artificial
neural networks in Jabonosa River Basin. Venezuela Eng Geol 78:1127
Greenbaum D, Tutton M, Bowker M, Browne T, Buleka J, Greally K, Kuna G, McDonald A,
Marsh S, OConnor E, Tragheim D (1995) Rapid methods of landslide hazard mapping:
Papua New Guinea case study. British Geological Survey. Technical Report WC/95/27
Guthrie RH (2002) The effects of logging on frequency and distribution of landslides in
three watersheds on Vancouver Island, British Columbia. Geomorphology 43:273292
Guthrie RH, Hockin A, Colquhoun L, Nagy T, Evans SG, Ayles C (2010) An examination of
controls of debris flow mobility: evidence from coastal British Columbia. Geomor-
phology 114:601613
112 Landslides 13 & (2016)
Guzzetti F, Reichenbach P, Ardizzone F, Cardinali M, Galli M (2006) Estimating the quality
of landslide susceptibility models. Geomorphology 81:166184
Hall FG, Townshend JR, Engman ET (1995) Status of remote sensing algorithms for
estimation of land surface state parameters. Remote Sens Environ 51:138156
Harison JV, Falcon NL (1936) An ancient landslip at Seymareh in southwestern Iran. Geol
J Geol 46:296309
Havenith HB, Strom A, Caceres F, Pirard E (2006) Analysis of landslide suscepti-
bility in the Suusamyr region, Tien Shan: statistical and geotechnical approach.
Landslides 3:3950
Heumann BW, Walsh SJ, McDaniel PM (2011) Assessing the application of a geographic
presence-only model for land suitability mapping. Ecol Inform 6:257269
Hinton GE (1992) How neural networks learn from experience. Sci Am 267:145151
Ingls J, Darrozes J, Soula JC (2006) Effects of vertical component of ground shaking on
earthquake-induced landslide displacements using generalized Newmarks analysis.
Eng Geol 86:134147
Jordan C, OConnor E, Marchant A, Northmore A, Greenbaum D, McDonald A, Kovacik M,
Ahmed R (2000) Rapid landslide susceptibility mapping using remote sensing and GIS
modelling. Proc. 14th International Conference on Applied Geologic Remote Sensing,
Las Vegas, pp 113120
Kanevski M, Pozdnoukhov A, Timonin V (2009) Machine learning for spatial environ-
mental data: theory, applications and software. EPFL Press, Lausanne, p pp 275
Kanungo DP, Arora MK, Sarkar S, Gupta RP (2006) A comparative study of conventional,
ANN black box, fuzzy and combined neural and fuzzy weighting procedures for
landslide susceptibility zonation in Darjeeling Himalayas. Eng Geol 85:347366
Kavzoglu T, Mather PM (2003) The use of back-propagating artificial neural networks in
land cover classification. Int J Remote Sens 24(23):49074938
Khanh NQ (2009) Landslide hazard assessment in muonglay, Vietnam applying GIS and
remote sensing. Dr. rer. nat. at the Faculty of Mathematics and Natural Sciences Ernst-
Moritz-Arndt-University Greifswald
Kisi O (2009) Neural networks and wavelet conjunction model for intermittent
streamflow forecasting. J Hydrol Eng 14:773782
Kuriakose SL, van Beek LPH (2011) Plant root strength and slope stability. In Glinski J,
Horabik J, Lipiec J (eds). Encyclopaedia of Agrophysics. Springer. doi:10.1007/978-90-
481-3585-1
Kuriakose SL, Devkota S, Rossiter DG, Jetten VG (2009a) Prediction of soil depth using
environmental variables in an anthropogenic landscape, a case study in the Western
Ghats of Kerala, India. Catena 79:2738
Kuriakose SL, van Beek LPH, van Westen CJ (2009b) Parameterizing a physically based
shallow landslide model in a data poor region. Earth Surf Process Landf 34(6):867881
Lee S (2005) Application of logistic regression model and its validation for landslide
susceptibility mapping using GIS and remote sensing data. Int J Remote Sens
26:14771491
Lee S, Min K (2001) Statistical analysis of landslide susceptibility at Yongin, Korea.
Environ Geol 40:10951113
Lee S, Pradhan B (2007) Landslide hazard mapping at Selangor, Malaysia using
frequency ratio and logistic regression models. Landslides 4:3341
Lee S, Ryu JH, Won JS, Park HJ (2004) Determination and application of the weights for
landslide susceptibility mapping using an artificial neural network. Eng Geol 71:289302
Li TS (2009) Applying wavelet transform, rough set theory and support vector
machine for copper clad laminate defects classification. Expert Syst Appl
36(3):58225829
Lineback M, Andrew W, Aspinall R, Custer S (2001) Assessing landslide potential using
GIS, soil wetness modelling and topographic attributes, Payette River, Idaho. Geo-
morphology 37:149165
Liu Y, Ngan KN (2006) Embedded wavelet packet object-based image coding based on
context classification and quadtree ordering Signal Processing. Image Commun
21:143155
Mallat SG (1989) A theory for multi resolution signal decomposition: the wavelet
representation. IEEE Trans Pattern Anal Mach Intell 11:674693
Marjanovi M, Kovaevi M, Bajat B, Voenlek V (2011) Landslide susceptibility
assessment using SVM machine learning algorithm. Eng Geol 123:225234
Mihaela C, Bednarik M, Jurchescu MC, Vlaicu M (2011) Landslide susceptibility assess-
ment using the bivariate statistical analysis and the index of entropy in the Sibiciu
Basin (Romania). Environ Earth Sci 63(2):397406
Mishra AK, Desai VR (2006) Drought forecasting using feed-forward recursive neural
network. Ecol Model 198(12):127138
Moon V, Simpson CJ (2002) Large-scale mass wasting in ancient volcanic material. Eng
Geol 64:4164
Moore ID, Burch GJ (1986) Sediment transport capacity of sheet and rill flow: application
of unit stream power theory. Water Resour Res 22:13501360
Moore ID, Grayson RB, Ladson AR (1991) Digital terrain modeling: a review of hydro-
logical, geomorphological, and biological applications. Hydrol Process 5330
Moore ID, Gessler PE, Nielsen GA, Peterson GA (1993) Soil attribute prediction using
terrain analysis. Soil Sci Soc Am J 57:443452
Moosavi V, Fallah Shamsi SR, Moradi HR, Shirmohammadi B (2013a) Application of
Taguchi method to satellite image fusion for object-oriented mapping of Barchan
dunes. Geosci J. doi:10.1007/s12303-013-0044-9
Moosavi V, Vafakhah M, Shirmohammadi B, Behnia N (2013b) A wavelet-ANFIS hybrid
model for groundwater level forecasting for different prediction periods. Water
Resour Manag 27:13011321
Moosavi V, Vafakhah M, Shirmohammadi B, Ranjbar M (2013c) Optimization of wavelet-
ANFIS and wavelet-ANN hybrid models by Taguchi method for groundwater level
forecasting. Arab J Sci Eng 39:17851796
Moosavi V, Talebi A, Shirmohammadi B (2014) Producing a landslide inventory map
using pixel-based and object-oriented approaches optimized by Taguchi method.
Geomorphology 204:646656
Nagarajan R, Roy A, Vinod Kumar R, Mukherjee A, Khire MV (2000) Landslide hazard
susceptibility mapping based on terrain and climatic factors for tropical monsoon
regions. Bull Eng Geol Environ 58:275287
Nefeslioglu HA, Gokceoglu C, Sonmez H (2008) An assessment on the use of logistic
regression and artificial neural networks with different sampling strategies for the
preparation of landslide susceptibility maps. Eng Geol 97:171191
Nourani V, Alami MT, Aminfar MH (2009) A combined neural-wavelet model for
prediction of watershed precipitation, Lighvanchai, Iran. Eng Appl Artif Intell 16:112
Ogden RT (1997) Essential wavelets for statistical applications and data analysis.
Birkhauser, Boston
Oh HJ, Pradhan B (2011) Application of a neuro-fuzzy model to landslide susceptibility
mapping for shallow landslides in tropical hilly area. Comput Geosci 37(9):12641276
Park CJ, Lee HK, Song WY (2011) Thorsten Graeve Achterkirchen, Ho Kyung Kim, 2011,
Defective pixel map creation based on wavelet analysis in digital radiography
detectors. Nucl Inst Methods Phys Res A 634:101105
Peng ZK, Chu FL (2004) Application of the wavelet transform in machine condition
monitoring and fault diagnostics: A review with bibliography. Mech Syst Signal
Process 18(2):199221
Phillips SJ, Anderson RP, Schapire RE (2006) Maximum entropy modeling of species
geographic distributions. Ecol Model 190(34):231259
Pourghasemi HR, Gokceoglu C, Pradhan B, Deylami Moezzi K (2012a) Landslide suscep-
tibility mapping using a spatial multi criteria evaluation model at Haraz Watershed,
Iran. In: Pradhan B, Buchroithner M (eds) Terrigenous mass movements. Springer-
Verlag, Berlin, pp 2349. doi:10.1007/978-303-642-25495-6-2
Pourghasemi HR, Mohammady M, Pradhan B (2012b) Landslide susceptibility mapping
using index of entropy and conditional probability models in GIS: Safarood Basin, Iran.
Catena 97:7184
Pourghasemi HR, Pradhan B, Gokceoglu C (2012c) Application of fuzzy logic and
analytical hierarchy process (AHP) to landslide susceptibility mapping at Haraz
watershed, Iran. Nat Hazards 63(2):965996
Pourghasemi HR, Pradhan B, Gokceoglu C, Deylami Moezzi K (2012d) A comparative
assessment of prediction capabilities of Dempster-Shafer and Weights-of-evidence
models in landslide susceptibility mapping using GIS, Geomatics. Nat Hazards Risk
4(2):93118
Pradhan B (2010a) Application of an advanced fuzzy logic model for landslide suscep-
tibility analysis. Int J Comput Intell Syst 3(3):370381
Pradhan B (2010b) Landslide susceptibility mapping of a catchment area using frequency
ratio, fuzzy logic and multivariate logistic regression approaches. J Indian Soc Remote
Sens 38(2):301320
Pradhan B (2011) Manifestation of an advanced fuzzy logic model coupled with Geo-
information techniques to landslide susceptibility mapping and their comparison with
logistic regression modelling. Environ Ecol Stat 18:471493
Pradhan B (2013) A comparative study on the predictive ability of the decision tree,
support vector machine and neuro-fuzzy models in landslide susceptibility mapping
using GIS. Comput Geosci 51:350365
Pradhan B, Lee S (2010) Landslide susceptibility assessment and factor effect
analysis: backpropagation artificial neural networks and their comparison with
frequency ratio and bivariate logistic regression modeling. Environ Model Softw
25:747759
Saha AK, Gupta RP, Sarkar I, Arora MK, Csaplovics E (2005) An approach for GIS-based
statistical landslide susceptibility zonation with a case study in the Himalayas.
Landslides 2:6169
Sanchez G, Rolland Y, Corsini M, Braucher R, Bourls D, Arnold M, Aumatre G (2010)
Relationships between tectonics, slope instability and climate change: Cosmic ray
exposure dating of active faults, landslides and glacial surfaces in the SW Alps.
Geomorphology 117:113
Shinde AD (2004) A wavelet packet based sifting process and its application for structural
health monitoring. Master Thesis, Faculty of Worcester Polytechnic Institute
Shirmohammadi B, Moradi HR, Moosavi V, Taie Semiromi M, Zeinali A (2013) Forecasting
of meteorological drought using Wavelet- ANFIS hybrid model for different time steps
(case study: southeastern part of east Azerbaijan province, Iran). Nat Hazards
69:389402
Sidle RC, Ochiai H (2006) Landslides: processes, prediction, and landuse. American
Geophysical Union, Washington, DC. Water Resources Monograph No. 18: 312 pp
Skempton AV, Hutchinson JN (1969) Stability of natural slopes and embankment
foundations. In: State of Art Report, 7th International Conference on Soil Mechanics
and Foundation Engineering, Mexico, pp 291-335
Song S, Zhan Z, Long Z, Zhang J, Yao L (2011) Comparative study of SVM methods
combined with voxel selection for object category classification on fMRI data. PLoS
ONE 6(2), e17191
Srensen SA, Bauer B (2003) On the dynamics of the Kfels sturzstrom. Geomorphology
54:1119
Tabachnick BG, Fidell LS (1996) Using multivariate statistics, 3rd edn. Harper Collins, New
York, p 880
Tagluk ME, Akin M, Sezgin N (2010) Classification of sleep apnea by using wavelet
transform and artificial neural networks. Expert Syst Appl 37(2):16001607
Taguchi G (1990) Introduction to quality engineering. McGraw-Hill, New York, p 191
Talebi A, Uijlenhoet R, Troch PA (2008) A low-dimensional physically-based model of
hydrologic control on shallow landsliding in complex hillslopes. Earth Surf Process
Landf 33. doi:10.1002/esp.1648
Tang Z, Fishwick PA (1993) Feedforward neural nets as models for time series forecast-
ing. ORSA J Comput 5(4):374385
Tien Bui D, Pradhan B, Lofman O, Revhaug I (2012a) Landslide susceptibility assessment
in vietnam using support vector machines, decision tree, and Naive Bayes Models.
Math Probl Eng. doi:10.1155/2012/974638, 26 pp
Tien Bui D, Pradhan B, Lofman O, Revhaug I, Dick OB (2012b) Spatial prediction of
landslide hazards in Hoa Binh province (Vietnam): a comparative assessment of the
efficacy of evidential belief functions and fuzzy logic models. Catena 96:2840
Tien Bui D, Pradhan B, Lofman O, Revhaug I, Dick OB (2012c) Landslide
susceptibility assessment in the Hoa Binh province of Vietnam: a comparison
of the LevenbergMarquardt and Bayesian regularized neural networks. Geo-
morphology 171:1229
Tien Bui D, Pradhan B, Lofman O, Revhaug I, Dick OB (2012d) Landslide susceptibility
mapping at Hoa Binh province (Vietnam) using an adaptive neuro-fuzzy inference
system and GIS. Comput Geosci 45:199211
Troch PA, Paniconi C, van Loon EE (2003) Hillslope-storage Boussinesq model for
subsurface flow and variable source areas along complex hillslopes. 1. Formulation
and characteristic response. Water Resour Res 39:13161330
Tu JV (1996) Advantages and disadvantages of using artificial neural networks versus
logistic regression for predicting medical outcomes. J Clin Epidemiol
49(11):12251231
University of Utah (1984) Flooding and landslide in Utahan economic impact analysis.
bur. Of Econ. Devel. And Utah office of Plan. Budget, Salt Lake city. Bureau of
Economic Development and Utah Office of Planning and Budget, Salt Lake
City, 123 pp
Van Beek LPH, van Asch TWJ (2004) Regional assessment of the effects of land-use
change and landslide hazard by means of physically based modeling. Nat Hazards
30(3):289304
Van Beek LPH, Wint J, Cammeraat LH, Edwards JP (2005) Observation and simulation of
root reinforcement on abandoned Mediterranean slopes. Plant Soil 278:5574
Vapnik VN (2001) The nature of statistical learning theory. Statistics for Engineering and
Information Science, 2nd edn. Springer, New York
Varnes DJ (1984) Landslide hazard zonation preview of principals and practice, Paris,
UNESCO, International association of engineering geologists, commission on land-
slides and other mass movements on slopes. Nat Hazards 3:176
Viera AJ, Garrett JM (2005) Understanding interobserver agreement: the kappa statistic.
Fam Med 37(5):360363
Vijith H, Madhu G (2008) Estimating potential landslide sites of an upland sub-watershed
in Western Ghats of Kerala (India) through frequency ratio and GIS. Environ Geol
55(7):13971405
Vivas L (1992) Los Andes Venezolanos. Academia Nacional de la Historia, Caracas
Wang CM, Huang YF (2009) Evolutionary-based feature selection approaches with new
criteria for data mining: a case study of credit approval data. Expert Syst Appl
36(3):59005908
Landslides 13 & (2016) 113
 Original Paper
Wang XY, Yang HY, Fu ZK (2010) A new wavelet-based image denoising using
undecimated discrete wavelet transform and least squares support vector machine.
Expert Syst Appl 37(10):70407049
Watkins LR (2012) Review offringepatternphaserecoveryusingthe1-Dand2-Dcontinuous
wavelet transforms. Opt Lasers Eng 50:10151022
Wilson RC, Keefer DK (1985) Predicting aerial limits of earthquake-induced
landsliding. In: J.I. Ziony (ed), Evaluating earthquake hazards in the Los
Angeles region - An Earth-Science perspective. USGS Professional paper 1360:
316-345
Wischmeier W, Smith D (1978) Predicting rainfall erosion lossesa guide to conserva-
tion planning. U.S. Department of Agriculture Science and Education Administration,
Washington, DC
Wong FS (1991) Time series forecasting using backpropagation neural networks.
Neurocomputing 2:147159
Wong WK, Yuen CWM, Fan DD, Chan LK, Fung EHK (2009) Stitching defect detection and
classification using wavelet transform and BP neural network. Expert Syst Appl 36(2
(part 2)):38453856
Xu C, Dai F, Xu X, Lee YH (2012) GIS-based support vector machine modeling of
earthquake-triggered landslide susceptibility in the Jianjiang River watershed, China.
Geomorphology 145146:7080
Xu C, Dai FC, Xu XW, Lee YH (2013a) GIS-based support vector machine modeling of
earthquake-triggered landslide susceptibility in the Jianjiang River watershed, China.
Geomorphology 145146:7080
Xu C, Xu X, Yao Q, Wang Y (2013b) GIS-based bivariate statistical modelling for
earthquake-triggered landslides susceptibility mapping related to the 2008 Wenchuan
earthquake, China. Q J Eng Geol Hydrogeol 46(2):221236
114 Landslides 13 & (2016)
Xu C, Xu X, Yao X, Dai F (2014) Three (nearly) complete inventories of landslides
triggered by the May 12, 2008 Wenchuan Mw 7.9 earthquake of China and their
spatial distribution statistical analysis. Landslides 11(3):441461
Yao X, Tham LG, Dai FC (2008) Landslide susceptibility mapping based on Support
VectorMachine: a case study on natural slopes of Hong Kong, China. Geomorphology
101:572582
Yilmaz I (2009) Landslide susceptibility mapping using frequency ratio, logistic regres-
sion, artificial neural networks and their comparison: a case study from Kat landslides
(Tokat-Turkey). Comput Geosci 35:11251138
Yilmaz I (2010) Comparison of landslide susceptibility mapping methodologies for
Koyulhisar, Turkey: conditional probability, logistic regression, artificial neural net-
works, and support vector machine. Environ Earth Sci 61:821836
Zhou R, Bao W, Li N, Huang X, Yu D (2010) Mechanical equipment fault diagnosis based
on redundant second generation wavelet packet transform. Digit Signal Process
20:276288
Zhu G, Blumberg DG (2002) Classification using ASTER data and SVM algorithms; the case
study of Beer Sheva, Israel. Remote Sens Environ 80(2):233240
Zou W, Chi Z, Lo KC (2008) Improvement of image classification using wavelet
coefficients with structured-based neural network. Int J Neural Syst
18(3):195205
V. Moosavi ()) : Y. Niazi
Department of Watershed Management Engineering, College of Natural Resources,
Yazd University, Yazd, Iran
e-mail: moosavi_v66@yhoo.com
 Original Paper
Landslides (2016) 13:115127 Renato Macciotta I C. Derek Martin I Norbert R. Morgenstern I David M. Cruden
DOI 10.1007/s10346-014-0551-4
Received: 5 March 2014
Accepted: 30 December 2014 Quantitative risk assessment of slope hazards
Published online: 13 January 2015
Springer-Verlag Berlin Heidelberg 2015
along a section of railway in the Canadian
Cordilleraa methodology considering
the uncertainty in the results
Abstract Railway alignments through the Canadian Cordillera are
constantly exposed to slope instabilities. Proactive mitigation
strategies have been in place for a few decades now, and instability
record keeping has been recognized as an important aspect of
them. Such a proactive strategy has enhanced the industrys ca-
pacity to manage slope risks, and some sections have been recog-
nized as critical due to the frequency of instabilities. At these
locations, quantification of the risks becomes necessary. Risk anal-
ysis requires knowledge of some variables for which statistical data
are scarce or not available, and elicitation of subjective probabil-
ities is needed. A limitation of such approaches lies in the uncer-
tainty associated to those elicited probabilities. In this paper, a
quantitative risk analysis is presented for a section of railway
across the Canadian Cordillera. The analysis focused on the risk
to life of the freight train crews working along this section. Upper
and lower bounds were elicited to cope with the uncertainties
associated with this approach. A Monte Carlo simulation tech-
nique was then applied to obtain the probability distribution of
the estimated risks. The risk probability distribution suggests that
the risk to life of the crews is below previously published evalua-
tion criteria and within acceptable levels. The risk assessment
approach proposed focuses on providing a measure of the uncer-
tainty associated with the estimated risk and is capable of handling
distributions that cover more than two orders of magnitude.
Keywords Rock falls . Uncertainty . Quantitative risk
assessment . Monte Carlo simulation
Introduction
The valley formed by the Fraser River hosts an important transporta-
tion corridor between the City of Vancouver and the interior of British
Columbia and other provinces in Western Canada. This corridor cuts
through the Canadian Cordillera and is used by the Canadian Pacific
Railway (CP), the Canadian National Railway Company (CN), and one
of Western Canadas major highways (Highway 1).
Transportation corridors through sections of the Canadian
Cordillera required steep rock cuts to accommodate highway
and railway alignments. Rock slope instabilities such as rock
falls, slides, and topples that originate in these cuts and from
natural rock cliffs can travel downslope and potentially reach
and block highways and railway tracks. Mountainous regions
are known to be highly susceptible to these events, the Cana-
dian Cordillera being no exception (Gardner 1970; Gardner
1977; Whalley 1984; Spang and Rautenstrauch 1988; Hungr
and Evans 1989; Evans and Hungr 1993; Dorren 2003). Hence,
it is not uncommon for transportation corridors through this
type of terrain to contain sections that are highly exposed to
these slope instabilities (Peckover and Kerr 1977; Brawner
1978; Pierson 1992; Bunce et al. 1997; Budetta 2004; Lan
et al. 2010).
Recognizing the risks associated with slope cuts in the Cordil-
lera, CP engaged in the development and implementation of a rock
slope management program in the early 1970s (Brawner and
Wyllie 1975). This system has evolved into a qualitative rating
system to describe the slope hazard and its likelihood of occur-
rence (Mackay 1997). Mitigation works (such as protection walls,
ditch widening and maintenance, face stabilization and scaling)
follow site inspections in the priority indicated by the rating
system. This form of risk assessment and management is currently
being applied. However, results from this approach are not readily
amenable to comparison with other hazard assessments that the
railway may employ. Hence, the need for the development of a
quantitative risk assessment approach that minimizes the amount
of qualitative inputs.
A quantitative risk assessment (QRA) associated with slope
instabilities between milepost 2 and 15 of CPs Cascade subdivision
is developed for this study. The assessment focuses on the risk to
life of running trade employees working along this section. CPs
extensive records dating back to the 1940s serve as the main input
for the hazard analysis and are of significant value in the conse-
quence analysis stage. The risk estimation considers measures of
the uncertainty in the input parameters and how these are carried
through the analysis to reflect uncertainty in the calculated risk
values.
Study area
CPs Cascade subdivision is located in southwest British Columbia,
along the Fraser River valley, in the Canadian Cordillera. Figure 1
shows the section between mile 0 (North of Boston Bar) and mile
40 (at Hope, 150 km East of Vancouver and about 50 km North of
the U.S. Border) of CPs Cascade subdivision. This section has a
long history of slope instability (Piteau 1977; Lan et al. 2007;
Macciotta et al. 2011). In particular, the section from milepost 2
to 15 (along the west riverbank) accounts for 67 % of all recorded
slope instabilities in a length equivalent to 32.5 % of the 64 km (40
miles). Instabilities documented along this section include rock
falls, slides, and topples.
Figure 1 presents the spatial distribution of slope instabilities
recorded along the study area. Moreover, the volumes of these
slope instabilities cover several orders of magnitude (Fig. 2). Slope
instability records in this section date back to the 1940s. These
records include date, location, and volume of the events, as well as
the probable source height, weather conditions, and any site ob-
servations considered relevant by the inspector (Macciotta et al.
2011). Although these records are considered of a high quality, rock
Landslides 13 & (2016) 115
 Original Paper

Fig. 1 CPs Cascade subdivision study area and geologic context

falls of small volume that can cross the railway tracks without From Boston Bar to Hope (first 40 miles of the subdivision), the
causing damage or leaving traces of its path are not noticed and Fraser River lies along the junction between the Coast Mountains
are not included in the database. and the Cascade Mountains. Several orogenic episodes involved
folding and faulting, metamorphism, and intrusion (McTaggart
and Thompson 1967). This intense deformation resulted in com-
plexly folded rock masses cut by northsouth trending faults
(Monger 1970). These faults are part of the Fraser River fault zone
and are associated with broad zones of weak materials and differ-
ential weathering (Piteau 1977). The lithology along milepost 2
through 15 consists in mainly intrusive rocks (Diorites and Grano-
rock falls rock slides diorites) and metamorphic rocks (Schist). Shearing and alteration
is common to all rock units, being intensified at contacts between
105 units and near faults. The study area is characterized by steep
Number of Recorded

90 slopes that have been glaciated with the highest peaks staying
above the ice (Monger 1970), and significant lateral erosion by
instabilities

75 the river is evident. Figure 1 shows the geological context of the

60 area. In this figure, the Hope and Yale faults are delineated (both
part of the Fraser River fault zone). Also, the general lithology of
45
the area is presented in Fig. 1.
30
15 Quantitative risk assessment methodology
The general methodology followed is consistent with current prac-
0.1 1. 10. 100. 1000. 10 000. tice for a landslide QRA (Ho et al. 2000; Crozier and Glade 2005;
Lee and Jones 2004; Fell et al. 2005; AGS 2007). The process
Instability volume (m3) leading to a fatal accident is modelled with the aid of an event
tree analysis (ETA) as shown in Fig. 3. The ETA considers two
Fig. 2 Distribution of slope instability volumes in the study area based on 340
scenarios for a moving train: (1) falling material impacts a moving
event records along the study section that included volume information. Along the
study section, rock falls are considered to have volumes up to about 30 m3 and train and (2) the moving train encounters a blocked track. The
rock slides to have volumes of 10 m3 or higher. Slope instabilities with volumes scenario where falling material impacts a stationary train is not
between 10 and 30 m3 have shown to potentially behave as rock falls or rock slides considered representative of the section analyzed. Moreover,

116 Landslides 13 & (2016)

 A B C D E F G H I
Material Hazard Hazard Conditional
Conditional
reaches or Detection Detection Warning Derailment probability of fatality Probability
Scenario Train speed derailment
blocks the System System in place probability given derailment of fatality
probability
track present activated occurs
debris
impact a Track speed
moving train AxBxF GxH
(40 km/h)

Slow order (20

Yes AxBxCxDxExF GxH
Slope km/h)
Yes Yes
instability
Track speed
Yes No AxBxCxDxExF GxH
(40 km/h)
Train
encounters a Track speed
blocked track No AxBxCxDxExF GxH
(40 km/h)

Track speed
No AxBxCxDxExF GxH
(40 km/h)

Probability of a fatal accident given slope instability (P[fatality:SF]) column I

Fig. 3 Event tree used to estimate the probability of a fatal accident given a rock slope instability

records indicate that fatal accidents have occurred only after the a) precise input probabilities
train derails (CP personal communication). It is then decided to no estimate of output uncertainty
1
simplify the analysis by considering that a fatal accident can only
occur if the train derails.
Previous studies have applied QRA methods along transporta- 0.1
tion corridors (Wyllie et al. 1980; Bunce et al. 1997; Guzzetti et al.
2004; Pine and Roberds 2005; Shamekhi and Tannant 2010). In 0.01
these analyses, multiple variables influence the location, magni- ?
tude, and frequency of the hazards, as well as the likelihood and 0.001
severity of the consequences. Determination of the values for these

Hazard

Consequence

RISK
Vulnerability
variables is associated with different degrees of uncertainty that x x =
are carried through the analysis. Even when an extensive database
exists, only those events that damage or derail the train are
reported. Events that can be classed as near misses are not
included in the database. Consequently, there is a gap in the
statistical record used to stochastically derive the consequence b) probabilities input as range of values
estimate of output uncertainty
probabilities required for QRA. When such statistics are not avail-
able, subjective probabilities are often used to fill the missing gaps. 1
Cum. Prob.

Subjective probability can be defined as an expression of per-

sonal belief about outcomes. It is a quantified measure of the
degree of belief or confidence in the outcome, according to the
personal state of knowledge at the time of assessment (Vick 2002). 0
Such personal assessments are not unique and change with in- Min Max
Believed prob.
creasing knowledge about the situation. As a consequence, when range
subjective probabilities are used as input for QRA, the uncertainty
related to these input probabilities can be carried forward in the 1 Max
Max
n

analysis, without proper quantification (Fig. 4a). Accounting for Max

tio
vi a

the uncertainties associated to QRA inputs, and propagating them 0.1

de
ard

Min
quantitatively through the analysis, would allow for a measure of
Sta an
nd
Me

uncertainty in the calculated risk (Fig. 4b). 0.01 Min

You and Tonon (2012) presented an approach using approxi-
Min
mate methods of probabilistic analysis to incorporate measures of
0.001
uncertainty in risk estimation. They highlighted the uncertainty
Hazard

Consequence

RISK
Vulnerability

related to defining a probability density function (PDF) for vari-

x x =
ables where limited or no information is available, and they pro-
posed the use of imprecise probabilities within event tree analysis.
Imprecise probabilities allow constructing a set of PDFs. Random
sets, normalized fuzzy sets, and envelopes of cumulative probabil-
ity distributions are special cases of imprecise probabilities (You
Fig. 4 Uncertainty associated with risk analyses input variables and risk
and Tonon 2012). This method then renders a lower and upper estimations. a Input probabilities as a single value with no estimate of the
estimate of risk based on approximation techniques. Wang et al. model result uncertainty. b Subjective probabilities defined as a range of values
(2013) presented an example of uncertainty assessment in and a probability density function

Landslides 13 & (2016) 117

 Original Paper
quantification of risk related to slope instabilities. They also simulation routine solves the ETA a defined number of itera-
adopted an approximate method, a First Order Second Moment tions. The iterations of the simulation randomly pick values of
error propagation technique, which provides an estimate of the the input parameters following their probability distribution.
mean and variance of the calculated risk. One objective of this The solutions of the ETA for all iterations are recorded and
study is insight into the distribution of the uncertainty embedded presented as a probability distribution of results. Figure 5 shows
in the calculated risks. Monte Carlo random sampling techniques the iteration steps of the Monte Carlo simulation to estimate the
and simulation consist of sampling processes over input PDFs in probability of a fatality in this study.
order to populate a mathematical expression that represents the The annual probability of fatality is calculated following the
phenomena being modelled. Values of the input variables are Binomial Theorem (Bunce et al. 1997) as:
selected on the basis of random number generation and variable
mapping according to their cumulative probability distribution. P fatality  11P fatality : S FN
Each iteration of the simulation is statistically treated as an obser-
vation within a set of possible outcomes. This approach allows for where P[fatality] is the annual probability of fatality,
a PDF to be fitted to the calculated values of the mathematical P[fatality:SF] is the probability of a fatal accident given a slope
expression (Ayyub 2003). This approach has been applied to land- cut instability volume, and N is the number of slope instabilities
slide science before. Examples of this are the analyses by El-Ramly each year.
et al. (2002) to calculate the probability distribution of the stability The risk to life of a crew member is calculated as:
levels of a slope and more recently, Macciotta et al. (2014) to
calculate the probability distribution of height and kinetic energy Pfatality   DR  C
of falling rock blocks. In this study, we use Monte Carlo random R
E
sampling and simulation techniques to calculate the PDF of risk. where:
Upper and lower bounds of the subjective probabilities cap- R is the risk to life of a crew member, DR is the ratio of crew
ture their uncertainty. Given the wide range of instability vol- members killed given a fatal derailment occurs, C is the number of
umes recorded, subjective probabilities are defined as a function crew members per freight train, and E is the total number of
of this volume. A PDF is defined between the upper and lower people employed as freight train crew that travel through the
bounds. This implies that a PDF is generated for each volume, section.
depending on the expected subjective probability range. The For this study, E is estimated at 500 employees, DR is estimated
model then automatically generates the input PDFs as functions as 0.87 based on records of fatal derailments (Bunce 2008), and C
of the instability volume analyzed. Then, a Monte Carlo is set at 2 crew members.

Step 1: Random selection of Step 2: Random selection of the Event Tree input probabilities
the slope instability volume
Volume independent Probabilities Volume Dependent Probabilities
C um
1 u la t
0 ive
Pro
bab
Cumulative Probability

1 Random ility
Cumulative Probability

1 Random
Number 1
ax

Number
M
Probability

P Random
Number

0
0 Min P Max Min
V
Slope Instability Volume Input Probability
0
V
Slope Instability Volume

Step 3: Population of the Event Step 4: Slope instability frequency (N) Step 5: Annual probability of
Tree Analysis a fatal accident
Cumulative Probability

1 Random
Event Tree Analysis
Number
P ... P[fatality] = 1 - (1 - P[fatality:SF])
N
P
P P ...
P
P ... 0
N
P[fatality:SF] Slope Instability Frequency

Step 6: Risk estimation

P[fatality] x DR x C
Risk to life for the average crew member =
E

Fig. 5 Monte Carlo simulation process to estimate the probability of a fatality and the risk to life of a crew member

118 Landslides 13 & (2016)

Hazard analysis
Slope instability volumes and frequencies
Rock falls, topples, and slides along the slope cuts in the study
section are ubiquitous and frequent, and for purposes of the QRA
presented in this study, they are treated statistically and differen-
tiated by volume and not by failure type. Differentiation by failure
type would have been necessary if precursory factors or triggering
mechanisms were to be included in the analysis.
The database of the volume and the temporal and spatial occur-
rence of the rock instabilities between miles 2 and 15 is used to obtain
cumulative probability distributions of the annual number of slope
instabilities and their volumes. These distributions are presented in
Fig. 6. The records of annual number of slope instabilities are
approximated with an inverse Gaussian distribution (Fig. 6a), and
their volumes are approximated with a Pearson distribution (Fig. 6c).
The relationship between the quantiles of the model results and the
fitted probability distribution can aid in assessing the goodness of
the distribution fit. A quantile N% can be expressed as the value
that divides a set or a distribution into two subsets with N% and (1-
N%) elements, respectively. The plot of the quantiles of a set of values
against the quantiles of a fitted distribution is a Q-Q plot. The Q-Q
plot for a perfect distribution fit is a line with a slope of 45. The more
the plot deviates from this line, the more likely the distribution fit is
not representative of the data set. Figures 6b and d present the Q-Q
plots for the distributions fitted to the annual number of slope
instabilities and their volumes, which suggest good fits between
records and the distributions adopted.
Probability that the instability reaches/blocks the track
CPs records include events where the instability volume was
encountered blocking the tracks and those caught within a ditch
or behind protective structures. Events of limited volume that fall
and cross the railway tracks without damage or evidence of their
paths are not noticed and, thus, not included in the database. It is
Fig. 6 Cumulative probability distribution of the annual number of slope instabilities
important to consider this limitation when evaluating the proba-
bility of a train being hit by a falling rock or debris, as the database
could be underestimating the potential for such outcome. The
fraction of the instabilities blocking the track can be estimated
from the database, but the fraction of instabilities with the poten-
tial to impact a moving train can only be roughly approximated
from the database. The ratio between the number of records where
the instability volume was encountered blocking the tracks to the
total number of slope instabilities noticed is given for three slope
instability volume ranges in order to assess any volume depen-
dency (Table 1). Increasing the number of event volume ranges
limits the number of records within each range to a level where
ratios calculated become unreliable.
The results in Table 1 suggest that the probability of an insta-
bility reaching the track increases with increasing instability vol-
ume. Hence, the upper and lower bounds of the probability
distribution are defined as continuous and volume dependent as
illustrated in Fig. 7.
There are not enough incident records of volumes less than
0.1 m3 to confidently extrapolate the linear assumptions shown in
Fig. 7. Consequently, the probability distribution is truncated at a
minimum of 0.3 for these smaller volumes. Also, the few instability
volumes larger than 1000 m3 all resulted in blocked tracks. It is
therefore assumed that for volumes over 1000 m3, the probability
of the track being blocked is 100 %. As it was previously discussed,
the track blocked ratios in Table 1 are considered approximations
of the probability that the instability volume blocks the track. This
uncertainty is reflected in the choice of upper and lower bounds
shown in Fig. 7. This information is also used as a rough approx-
imation of the potential for a falling block or debris to impact a
moving train.
Consequence analysis
Two outcomes are considered after the instability volume
reaches the track: the material impacts a train or the material
Landslides 13 & (2016) 119
 Original Paper
Table 1 Ratio of failure events blocking the track to total failuresCPs Cascade Table 2 Input parameters used to estimate the spatial probability P[S] of a slope
subdivision milepost 0 to milepost 40 instability coinciding with a moving train
Event volume Total no. No. events Track blocked Lower bound Upper bound
events track blocked ratio
L 1 km 3 km
All events* 535 156 0.29
T 20 trains/day 25 trains/day
0.11 m3 153 40 0.26
V 40 km/h (posted track speed)
110 m3 135 52 0.39
P[S] 0.021 0.078
101000 m3 40 21 0.53
*Events with known and unknown volumes

blocks the track. These then become mutually exclusive events,

and the sum of their probability, given the material reaches the no previous slope instabilities reported at the time the train travels
track, is equal to one. through the section and no slow orders are in place.

Probability that a freight train encounters a blocked track

Probability that the slope instability impacts a moving train
Given a slope instability occurs, each iteration of the simulation
approximates the probability that the instability impacts a freight
train by estimating the probability that the instability and the train
coincide in space. This spatial probability is estimated as:
L  T
PS
V  24
where P[S] is the spatial probability of the falling material
coinciding with the moving train, L is the average train length in
km, T is the number of trains per day, and V is the train speed in
km/h. Table 2 presents the input upper and lower bounds used to
calculate P[S]. This equation assumes that the length of the train is
significantly larger than the dimension of the slope instability and
that each event can only affect one train. This is consistent with the
length of the trains along the section of study, the fact that there is
no double track along this section and that operators of trains in
the area will receive warning if an incident occurs.
A uniform probability distribution is assumed for the values of
L and T within the upper and lower bounds presented in Table 2.
Note that this calculation considers constant train frequency and
an even slope instability frequency throughout the day and
throughout the year. While this is not a representative of the site,
this simplification is a valid approximation for the evaluation of
annual average risks. The posted track speed of 40 km/h assumes
Fig. 7 Conditional probability distribution for an event reaching the track given a
slope instability occurs
The probability of a freight train encountering a blocked track
depends on the instability not impacting a moving train while
travelling downslope. This is estimated as (1 - P[S]) for each
iteration of the simulation. This neglects those blocks of limited
volume crossing the railway tracks that are not noticed and, hence,
not included in the instabilities database. Events that leave no
trace of their path and cross the tracks are considered not to have
enough kinetic energy to derail a train at impact. There could be
considerable damage to a train car or locomotive, however, not
likely to pose a noticeable risk to crew members (probability of
spatial intersection between one of these events and a crew mem-
ber would be quite small if we follow the equation for P[S] with a
length equal to the size of the falling block). Approximating the
probability of a freight train encountering a blocked track as (1 -
P[S]) is therefore considered adequate for the QRA presented in
light of the available information.
There is a possibility that the first train or vehicle reaching the
blocked track is not the freight train considered in this study.
Because the freight train traffic is much more frequent than other
traffic (passenger train, high rail, maintenance equipment), it is
assumed that the freight train is the first to impact the slide
volume.
Probability that the hazard detection system is present and
operational
The Hazard Detection System (HDS) consists of a series of wire
fences along the section, between the railway track and the cut
slope. A sketch of the HDS is presented in Fig. 8 in the context of
two types of slope instability, rock falls (Fig. 8a) and wedge slides
(Fig. 8b). The spacing between wires is about 20 to 25 cm, and the
fence height varies between less than a meter and up to 2 m in
some sections. When a section of track is blocked, it is expected
that the material blocking the track would have broken one or
more of these wires in its path. This is detected by the system, and
the nearest track circuit signal shows that the track is occupied.
The probability that the HDS would detect the volume depends
on the ratio of length of HDS to the total section length and on the
total number of days per year, the HDS is active. The HDS is
installed along mileposts 2 through 15 of the study section. Con-
sidering the time required for maintenance and repairs, the prob-
ability that the HDS is present is between 0.9 and 1 for any given
year. A uniform PDF was defined in this range.

120 Landslides 13 & (2016)

Fig. 8 Hazard Detection Systems (HDS). Sketch of the HDS for rock fall type of slope instability (a), in the context of sliding type of slope instability (b), and views of the
HDS installed at mile post 3.4 (c) and 11.4 (d) of CPs Cascade subdivision

Probability that the HDS is activated
The probability that the HDS is activated depends on the instabil-
ity volume, the spacing between wires, and the height of the HDS
relative to the slope geometry. An example of the HDS layout on
the site is presented in Fig. 8. Figure 8c and d show views of the
HDS installed at mile post 3.4 and 11.4, respectively.
Insufficient records include information on HDS activation,
whether the material was encountered blocking the track and the
instability volume, for a reliable estimate of the HDS activation
probability. Based on the characteristics of the HDS and the slopes
along the study area, we judged the HDS activation probabilities.
These adopted probabilities and their justification are presented in
Table 3. Continuous upper and lower boundaries for the probabil-
ity distributions that the HDS is activated are based on Table 3 and
are presented in Fig. 9.
Probability that a warning is issued and train speeds
When a section of the HDS is activated, the nearest track circuit
signal shows a track as occupied and activates a slow order. This
implies that the first train to encounter the blocked track receives a
warning only if there is a track circuit signal between the activated
HDS and the train. The section between track circuit signals is

Table 3 Subjective probabilities for the HDS activation conditional probability

Volume (m3) Assumed probability Justification
HDS is activated
0.01 0.01 to 0.05 (residual) A 10 to 20 cm diameter block may jump the
wire fence or pass between wires.
0.1 0.1 to 0.5 A 30 to 50 cm diameter block activates the
wire fence when rolling through it, but may
jump over the fence depending on the slope
section.
10 0.6 to 0.9 A 1.5 to 2 m diameter block activates the
wire fence when rolling through it, but may
jump over the fence depending on the slope
section.
100 1 (certain) A 4 m diameter block breaks the fence and
reaches the track as a pile of debris. This
activates the HDS.

Landslides 13 & (2016) 121

 Original Paper
1.00 the instability mass, velocity, and how it disaggregates and reaches

HDS activation probability

U pp e r
0.50 the track. A comprehensive analysis of these factors is complex
and requires information rarely available. Furthermore, unless the
0.20
r impact caused a derailment or excessive damage, it may not be
we
0.10 Lo noticed until the train reaches the next inspection site. Hence, the
0.05 impact record is incomplete.
In this study, subjective upper and lower bounds probabilities
0.02
are developed based on the limited available data. According to
0.01 0.1 1 10 100
Bunce (2008), CP records indicated that rock volumes less than
1 m3 have not caused a derailment after impacting a moving train.
Instability volume (m3)
Hence, for instability volumes less than 1 m3, a residual derailment
conditional probability is judged as 0.01. Based on the historical
Fig. 9 Boundaries for the HDS activation conditional probability distributions
described in Table 3 records, this probability increases with increasing volume. For
volumes over 40 m3, the derailment probability, should an impact
occur, approaches one. The wide range of subjective probabilities
known as the signal block. We assume that if the train is outside
adopted for slope instability volumes between 1 and 40 m3 reflects
the signal block where the event occurs, and the HDS is activated,
the uncertainty at this level of the ETA. Figure 10 shows the
the warning is effective.
adopted upper and lower bounds for the derailment probabilities
The probability that a warning is issued is approximated by the
given the instability volume impacts a moving train.
complement of the probability that the train is inside the signal
block when the event occurs:
Conditional derailment probabilitytrain encounters a blocked track
The distance between the train and the farthest visible section of
BT
Pwarning 1PSignalBlock 1 rail is referred to as the sight distance. The sight distance required
V  24
for a train to stop is the stopping distance. The ratio of sight
where P[warning] is the probability a warning is issued given
distance to stopping distance can aid in estimating the probability
the HDS is activated, P[SignalBlock] is the probability the train is
that the train stops before impacting a blocked track. A ratio of 1
inside the signal block when the event occurs, B is the distance
indicates that there is just enough distance between the train and
between the activated HDS and the nearest track circuit signal, T is
the blocked section for the freight train to come to a stop. Lower
the number of trains per day, and V is the train speed in km/h. B is
ratios indicate that the train is not able to stop but can reduce its
not known but can be conservatively estimated as the entire length
speed.
of the signal block or as half this length to account for an average
The stopping distance is a complex field and depends on factors
distance. In this study, B is assumed between 0.5 and 1 km.
such as train length, weight, type, brake force and initial speed,
Table 4 presents the input parameters used to calculate P[warn-
alignment grade and curvature, interaction between the wheels
ing]. Uniform probability distributions are adopted for the values
and track, and weather conditions (Barney et al. 2001; Loumiet
of B and T within the upper and lower bounds. The input values
and Jungbauer 2005; Bunce 2008). Loumiet and Jungbauer (2005)
characteristic of the site render a high warning probability when
presented an analysis of the stopping distance for a freight train
the HDS is activated and show limited variation.
consisting of 100 loaded cars and 4 locomotives, a train of similar
The analyses assume that if warned, the train will slow down to
characteristics to the average freight train considered in this study.
a restricted speed. This restricted speed, or slow order, is at a
Following Loumiet and Jungbauer calculations, a slightly higher
maximum half the track speed (20 km/h). Records show that most
stopping distance of 400 m is adopted for trains travelling at track
slow orders in this section are to keep speeds to about 16 km/h,
speed and 250 m for slow order.
which is consistent with the above assumption. All other branches
In the study area, the average visible track length ahead of the
of the ETA with unsuccessful warning outcomes consider the train
locomotive is about 1 km. This, when compared to the stopping
to be travelling at track speed when encountering the blocked
distances, suggests that the crew have enough time to react and
track.

Conditional derailment probabilityslope instability impacts a 1.0

moving train
0.8
When the slope instability impacts a moving train, the derailment
Probability

probability is a function of kinetic energy. This energy depends on

r
Uppe

0.6
er

Table 4 Input parameters used to estimate the probability of a warning (P[warn- 0.4
Low

ing]) being issued given the HDS is activated

0.2
Lower bound Upper bound
0.0
B 0.5 km 1 km 0 20 40 60 80 100
T 20 trains/day 25 trains/day Instability volume (m 3 )

V 40 km/h (posted track speed)

Fig. 10 Elicited upper and lower bounds for the derailment conditional probability
distributions given falling slope debris impact a moving train

122 Landslides 13 & (2016)

stop the train if they observe a blocked track. However, about 10 %
of the section contains curves that decrease sight distances to
about 200 m, so the sight distance to stopping distance ratio
becomes less than one.
Judgment is required to account for the volume of material
blocking the track. In this scenario where the train impacts a
blocked track, we judge that up to 0.1 m3 of material poses a
residual probability of derailment (0.01) if the train is travelling
at track speed. As when 100 m3 or more material blocks the track,
the chance for a derailment approaches certainty, given the impact
is at enough speed. Subjective probabilities between 0.1 and 1 are
adopted for material volumes of about 10 m3. These subjective
probabilities are estimated to be between 1 and 2 orders of mag-
nitude lower for scenarios where the train is travelling at slow
order.
The sight distance to stopping distance ratio analysis and the
judgment about the influence of the material volumes blocking the
track in the derailment probability are used to elicit upper and
lower bounds for the subjective probabilities of derailment given a
train encounters a blocked track (Fig. 11). Uniform probability
distributions for each volume are adopted between the upper
and lower bounds shown in Fig. 11.
Probability of fatality should a derailment occur
Bunce (2008) presented an analysis of CPs records showing that
only 3 out of more than 230 mainline derailments resulted in fatal
accidents. This suggests that on average, 1.3 % of all derailments
result in a fatal accident. Bunce noted that all fatal derailments had
occurred when the locomotive derailed and fell into a water body
a) 1.E+00
and also suggested that the probability of a fatal accident given a
derailment in mountainous terrain would be higher than average.
However, due to the history of slope instability in the study area,
the track speed is limited to 40 km/h, lower than other sections
along the Canadian Cordillera. Table 5 shows the adopted values
for the conditional probability of fatality used to populate the ETA.
These values are based on the analyses presented by Bunce (2008)
and consider the particular conditions along the study section.
Following the approach in Bunce (2008), upper and lower bounds
corresponding to the train travelling at slow order speed are
adopted as one order of magnitude lower than those for track
speed.
Simulation and results
Defining the number of iterations for the simulation
The outcome of the Monte Carlo simulation applied to the event
tree is a normalized histogram of observations that can be approx-
imated using a PDF for the estimated risk values. Estimates of the
mean, mode, and standard deviation for the resulting PDF can be
compared against selected risk evaluation criteria. The random
nature of the approach means that point estimates derived from
the resulting PDF vary for different simulations using the same
model. Incrementing the number of iterations increases the num-
ber of results to a larger statistical sample and reduces this vari-
ability. Ten simulations for each of four different numbers of
iterations are evaluated for a total of 40 simulations. Results are
plotted in Fig. 12 in terms of the mean and variance of the
probability of fatality for each simulation. We decided to work
with the results obtained for a simulation with 100,000 iterations,
given its variability is considered negligible relative to the magni-
tude of the results (Fig. 12).

1.E-01 Visualization and interpretation of risk distributions over several

e r
pp orders of magnitude
Probability

1.E-02 U er The estimated risk values in the Monte Carlo simulation covered 5
Low
orders of magnitude. This spread reflects the input uncertainty
1.E-03 carried through the analysis. A PDF of such results would show a
long tail towards the higher magnitudes (Fig. 13a). Adopting the
1.E-04
mode of the PDF could allow assessing the central tendency of the
results; however, the shape of the distribution towards the lower
1.E-05
0.01 0.1 1 10 100 1000
values and in the proximity of the mode remains hidden by the
scale of the plot.
Volume (m3)
In order to analyze the central tendency and variability of the
b) 1.E+00 estimated risk, the calculated PDF is plotted using a base-10 semi-
logarithmic scale. Given the PDF in the semi-logarithmic scale
1.E-01 approximates a normal distribution (Fig. 13b); a mean and stan-
dard deviation could then be calculated for the base-10 logarithm
Probability

1.E-02 of the estimated risk values. This method minimizes shifting of the
p er calculated mean, while the mode suffers no change, and allows for
1.E-03 Up a better assessment of the distribution. In this study, the mean and
er
Low
1.E-04

1.E-05 Table 5 Adopted conditional probability of fatality given a derailment occurs

0.01 0.1 1 10 100 1000 Train speed Lower bound Upper bound
Volume (m3)
Track speed (40 km/h) 0.002 0.05
Fig. 11 Subjective probability that derailment occurs after a train reaches a Slow order (20 km/h) 0.0002 0.005
blocked track a at track speed of 40 km/h and b at a slow order of 20 km/h

Landslides 13 & (2016) 123

 Original Paper
It is important to note that this approach treats the orders of
magnitude of risk as risk categories, minimizing the effect of the
actual estimated values when calculating the point estimates.
However, we believe that the approach is compatible with how
probability is perceived at orders of magnitude below 101 and
compatible with how evaluation criteria are expressed. Further, it
allows for a measure of the uncertainty in the estimated risks.

Using partial results for potential model calibrationderailment

Fig. 12 Comparison of mean and variance of the calculated probability of fatality
from the Monte Carlo simulation for increasing number of model iterations
the standard deviation calculated using the base-10 logarithm of
the risk values are denoted as m and s, respectively.
a) Frequency
mean
mode
probability
The model setup presented in Figs. 3 and 5 allows for calculation of
the derailment probability as a partial result of the simulation. The
calculated annual probability of derailment is shown in Fig. 14.
This section compares these partial results to historical derailment
records to increase the confidence on the validity of the model.
Historical records of the study area indicate one freight train
derailment occurred between 1975 and 2009, a frequency of 1 in
35 years (average 0.029 occurrences annually). What is not con-
sidered in this simple frequency is the number of trains per year;
fewer trains used the corridor in the early years. We judge that
train frequencies have been similar to the ones considered for
analysis since 1980. The derailment frequency would then repre-
sent 1 derailment in 20 to 30 years or an average of 0.033 to 0.05
annually. Figure 14 shows the derailment annual probability dis-
tribution and the historical derailment frequencies considering
similar train traffic for the periods described above. The derail-
ment annual probability mean value estimated by the model is
0.04 with a mode of 0.02, consistent with the statistical data.
Analysis of the estimated derailment probability variation is
less straightforward. The probability of derailment occurring with-
in a period of time depends on the annual frequency of derailment
and can be expressed through the binomial theorem as:
Pderailment 11 F Y
where F is the derailment annual frequency and Y is the period
of time for which the derailment probability is being estimated.
The derailment annual frequency represents the average distribu-
tion of freight derailments in time and is estimated as the inverse
of the average time between derailment occurrences. Figure 15
presents the probability of derailment for different time intervals
(Y). The estimates consider derailment annual frequencies (F) of 1

Risk
0 0.2 0.4 0.6 Derailment annual frequency
1/500 1/100 1/10 1/2
b) Frequency 1.0
Recorded on site

mean (m) ~ mode 0.8 mode = 0.02

m-s m = 0.04
m = mean
Probability

0.6
s = standard
m+s
-s +s 0.4
m+2s

0.2
Risk m-2s

0.001 0.01 0.1 1 0.0

1.0 E-3 1.0 E-2 1.0 E-1 1.0 E-0
Fig. 13 Diagram of Monte Carlo simulation results covering several orders of Annual probability of derailment
magnitude in normal scale (a) and in semi-logarithmic scale assuming a normal
distribution (b) Fig. 14 Monte Carlo simulation results for the annual derailment probability

124 Landslides 13 & (2016)

 derailments in a period of time (Y)
Probability (P*) of one or more
F* = 0.5 (1 in 2 years) F* = 0.03 to 0.05 (1 in 30 to 20 years)
1.0
0.8
0.6
0.4
0.2
1 5 10 50 100
Period of time (Y) in years
* P=1-(1-F)
Y failures in Australia (ANCOLD 2003). Figure 16 also shows that the
** F = Average annual derailment frequency
Fig. 15 Probability of derailment in different time periods considering annual
derailment frequencies of 1 in 2 years, 1 in 20 to 30 years, and 1 in 500 years
in 2 years, 1 in 500 years, and the frequency estimated from records
of 1 in 20 to 30 years. Figure 15 highlights the time period between
20 and 30 years. As expected from the binomial theorem, the
probability of derailment occurring in a period of 20 to 30 years
is about 0.64 if the annual frequency is considered as 1 derailment
every 20 to 30 years.
Figure 15 shows how the probability of derailment in a 20 to 30-
year period is almost certain when considering an annual derail-
ment frequency of 1 in 2 years, and it is significantly low when
considering an annual derailment frequency of 1 in 500 years. This
suggests that given a derailment was recorded in a 20 to 30-year
period, it is very unlikely that the annual derailment frequency is
higher than 1 in 2 years or lower than 1 in 500 years. The Monte
Carlo simulation estimation of the derailment annual probability
in Fig. 14 shows that the distribution of results between the mean
and twice the standard deviation (m2 s and m+2 s) lies within
annual frequencies of 1 in 2 and 1 in 500 years.
Following this methodology, the model is assessed to have enough
accuracy in light of the available information for the estimation of
derailment probabilities. It should be noted that this approach showed
the potential for model calibration as data increase with time.
1 derailment recorded
in a valid period between
20 to 30 years
F* = 0.002 (1 in 500 years)
500
The individual risk evaluation criteria selected for comparison
are those developed for people living in landslide prone areas in
Hong Kong (ERM 1998), risks associated with dam failures in
Australia (ANCOLD 2003), and the criterion proposed for land
use planning around industries in the UK (HSE 2001). The crite-
rion proposed by HSE (2001) was included given its wide spread
application and because it proposes risk criterion for workers.
The mean (m) of the estimated risk to life for the crew (Fig. 16)
is 3.6106. This value is lower than the tolerable limit of 103 set
for workers in the UK (HSE 2001). It is also lower than the
tolerable limit of 104 set for the public exposed to landslide
phenomena in Hong Kong (ERM 1998) and associated with dam
risk value corresponding to the mean plus twice the standard
deviation (m+2 s) is 8105. Statistically, 97.7 % of the risk values
in the PDF are lower than this value, which was also below the
tolerable risk criteria selected.
According to Fell et al. (2005), the International Society of Soil
Mechanics and Geotechnical Engineering (ISSMGE) Technical
Committee on Risk Assessment and Management defines
individual risk to life as the increment of risk to the individual
in addition to the everyday risk if the hazard was not present. In
this regard, Porter et al. (2009) estimated that an increase in
individual risk of 1105 would represent an increase of less than
0.2 % over the average Canadian risk to life, which could be
considered as low. It is noted that the mean estimated for the
individual risks is 3.6106 and, hence, plots below this value.
The estimated risks, however, are above the acceptable limits
set for workers by the HSE (2001). This evaluation indicates that

Risk calculation
Risk was calculated in terms of the risk to life for a crew member
and following the model setup presented in Figs. 3 and 5. Figure 16
shows the distribution of the risk to life for a crew member. Also
shown are the selected risk evaluation criteria and common risks.
The mean and mode of the estimated individual risk PDF are
3.6106 and 3.4106, respectively. The mean annual probability
of fatality along the section (probability of at least one fatal
accident), or total risk, was estimated at 2103.

Risk evaluation
Societys risk perception and tolerance vary between different re-
gions depending on the social, cultural, and economic context
(Morgenstern 1995; Finlay and Fell 1997). The risks estimated in this
chapter are compared against widely used risk evaluation criteria.
Even though these criteria were proposed for other locations and
contexts, they are considered applicable for illustrative purposes.
Fig. 16 Monte Carlo simulation estimate of the risk to life of the average crew
member. (1) Derived from the 2007 age-standardized mortality rates for the
Canadian population (Statistics Canada 2010). (2) Data from Baecher and Christian
(2003). (3) Porter et al. (2009) suggestion that the incremental risk is low if it does
not exceed 0.2 % of the Canadian age-standardized risk of loss of life. (4) HSE
(2001). (5) ANCOLD (2003). (6) ERM (1998)

Landslides 13 & (2016) 125

 Original Paper
the risks between milepost 2 and 15 of the Cascade subdivision are
within the As Low As Reasonably Practicable (ALARP) zone and
measures are required to minimize the risks posed by the slope
hazard, should the benefits outweigh the cost of mitigation. This
conclusion is not unexpected, for this section is a highly hazardous
one, where considerable risk mitigation measures, rock fall detec-
tion fences, ditch cleaning, slope scaling, are used, thus complying
with the ALARP principle.
Conclusion
The risks from slope instabilities between mileposts 2 to 15 of CPs
Cascade subdivision are estimated using an event tree analysis. The
lack of statistical data to populate some branches of the analysis
requires elicitation of subjective probabilities. Upper and lower
bounds are estimated for these probabilities to account for the
uncertainties of these probabilities and a Monte Carlo simulation
technique used to propagate the uncertainties through the analysis.
The estimated risk probability distribution shows normality
when plotted using a semi-logarithmic scale. Calculation of the
mean and standard deviation uses the base-10 logarithm of the
estimated risk. This method minimizes the influence of the risk
values when calculating the central tendency and its variability
and treats each order of magnitude as a risk category. This ap-
proach is compatible with how risk is perceived when dealing with
several orders of magnitude. It is also compatible with how risk
evaluation criteria are expressed, while further allowing for the
uncertainty in the estimated risk to be measured.
Unmeasured uncertainties associated with the upper and lower
subjective probability limits are still present in the results. It is also
noticed that other sources of uncertainty, such as model uncer-
tainty, could still represent the major source of error. In this
regard, the validity of the model could be assessed to a certain
extent by comparing the model estimations at different levels of
the ETA against available data. In this study, the model derailment
probability is compared against derailment statistics in the section.
This also opens the possibility for model calibration and upgrade
in light of new data.
The low estimated individual risks in this analysis corre-
spond to the short period of time each individual spends in
the study section. The total risk associated to slope hazards
along this section (2103) is distributed through a large num-
ber of freight trains and their crew. This renders lower individ-
ual risks. If reduction of the total risk was to be considered,
mitigation measures should focus on lowering the probability
of a fatal accident. Site inspections, scaling works, rock fall
detection fences, ditch maintenance, and protective walls are
all in place at the site. These either reduce the hazard frequency
or the consequence probability and, hence, reduce the total risk
in the area.
The simple yet comprehensive analysis presented is readily
applicable. It demonstrates a QRA methodology applied to a
section of track that has a substantial slope instability data-
base. QRA for such conditions can be a valuable tool for
decision-making. Reviews of the analyses should be carried
out at regular intervals and probabilities updated in light of
new information. As more experience is gained with the
application of the QRA process, it may prove to be a suitable
tool for risk management of many aspects of railway
operations.
126 Landslides 13 & (2016)
Acknowledgments
The authors wish to acknowledge the Railway Ground Hazards
Research Program (RGHRP), the Canadian Railway Research Lab-
oratory (CaRRL), and the Natural Science and Engineering Re-
search Council of Canada (NSERC) for funding this study. The first
author wishes to acknowledge the assistance of Dr. Simaan
Abourizk at the University of Alberta, Department of Civil and
Environmental Engineering for his insights into simulation
through Monte Carlo techniques.
References
Australian Geomechanics Society (AGS) (2007) Practice note guidelines for landslide risk
management 2007. Aust Geomech 42(1):63114
Australian National Committee on Large Dams (ANCOLD) (2003) Guidelines on risk
assessment. Australian National Committee on Large Dams Inc, Melbourne
Ayyub BM (2003) Risk Analysis in Engineering and Economics. CRC Press
Baecher GB, Christian JT (2003) Reliability and statistics in geotechnical engineering.
Wiley, West Sussex, p 605
Barney D, Haley D, Nikandros G (2001) Calculating train braking distance. Proceedings of
the 6th Australian workshop on Safety Critical Systems and Software, Brisbane,
Australia. AustComp Soc 3:2329
Brawner CO (1978) Case examples of rock stability on rail projects. American Railway
Engineering Association-Bulletin 668, Proceedings 77th technical conference. Chicago
79:348364
Brawner CO, Wyllie DC (1975) Rock slope stability on railway projects. American Railway
Engineering Association-Bulletin 656, Proceedings 75th technical conference. Chicago
77:449484
Budetta P (2004) Assessment of rockfall risk along roads. Nat Hazards Earth Syst Sci
4:7181
Bunce CM (2008) Risk estimation for railways exposed to landslides. Dissertation,
University of Alberta
Bunce CM, Cruden DM, Morgenstern NR (1997) Assessment of the hazard from rock fall
on a highway. Can Geotech J 34:344356
Crozier MJ, Glade T (2005) Landslide hazard and risk: issues, concepts and approach,
landslide hazard and risk. Wiley, Chichester, pp 140
Dorren LKA (2003) A review of rockfall mechanics and modelling approaches. Prog Phys
Geogr 27(1):6987
EI-Ramly H, Morgenstern NR, Cruden DM (2002) Probabilistic slope stability analysis for
practice. Can Geotech J 39(3):665683
ERM-Hong Kong Ltd. (1998) Landslides and boulder falls from natural terrain: interim
risk guidelines, The Government of Hong Kong Special Administrative Region, pp 183
Evans SG, Hungr O (1993) The assessment of rockfall hazard at the base of talus slopes.
Can Geotech J 30:620636
Fell R, Ho KKS, Lacasse S, Leroi E (2005) A framework for landslide risk assessment and
management. In: Hungr F, Couture E (eds) Landslide risk management. Proceedings
of the international conference on landslide risk management, A.A. Balkema, Van-
couver, pp 325
Finlay PJ, Fell R (1997) Landslides: risk perception and acceptance. Can Geotech J
34:169188
Gardner J (1970) Rockfall: a geomorphic process in high mountain terrain. Albertan
Geographer 6:1520
Gardner J (1977) High magnitude rockfall-rockslide: frequency and geomorphic signif-
icance in the Highwood Pass area, Alberta. Great Plains Rocky Mountain Geograph J
6(2):228238
Guzzetti F, Reichenbach P, Ghigi S (2004) Rockfall hazard and risk assessment along a
transportation corridor in the Nera Valley, Central Italy. Environ Manag 34(2):191208
Health and Safety Executive (HSE) (2001) Reducing risks, protecting people. Her Majestys
Stationery Office, London
Ho KKS, Leroi E, Roberds WJ (2000) Quantitative risk assessment application, myths and
future direction. Proceedings of the International Conference on Geotechnical and
Geological Engineering (GeoEng2000), Melbourne, pp 269312
Hungr O, Evans SG (1989) Engineering aspects of rock-fall hazards in Canada. Geological
Survey of Canada, Ottawa
Lan H, Martin CD, Lim CH (2007) Rock fall analyst: a GIS extension for three-dimensional
and spatially distributed rock fall hazard modeling. Comput Geosci 33:262279
Lan H, Martin CD, Zhou C, Lim CH (2010) Rockfall hazard analysis using LiDAR and spatial
modeling. Geomorphology 118:213223
Lee EM, Jones DKC (2004) Landslide risk assessment. Thomas Telford, London, p 454
Loumiet JR, Jungbauer WG (2005) Train accident reconstruction and FELA and railroad
litigation. Lawyers and Judges Publishing Company, Tucson, p 576
Macciotta R, Cruden DM, Martin CD, Morgenstern NR (2011) Combining geology,
morphology and 3D modelling to understand the rock fall distribution along the
railways in the Fraser River Valley, between Hope and Boston Bar, B.C. International
Symposium on Rock Slope Stability in Open Pit Mining and Civil Engineering,
Vancouver, BC, Canada
Macciotta R, Martin CD, Cruden DM (2014) Probabilistic estimation of rockfall height and
kinetic energy based on a three-dimensional trajectory model and Monte Carlo
simulation. Landslides. doi:10.1007/s10346-014-0503-z
Mackay CH (1997) Management of rock slopes on the Canadian Pacific Railway. In:
Cruden F (ed) Landslide risk assessment. Proceedings of the International Workshop
on Landslide Risk Assessment, Hawaii, pp 271275
McTaggart KC, Thompson RM (1967) Geology of part of the Northern Cascades in
Southern British Columbia. Can J Earth Sci 4:11991228
Monger JWH (1970) Hope map-area, west halfBritish Columbia, Paper 6947. Geo-
logical Survey of Canada, Department of Energy, Mines and Resources
Morgenstern NR (1995) Managing risk in geotechnical engineering, The 3rd Casagrande
Lecture. Proceedings 10th Pan-American Conference on Soil Mechanics and Founda-
tion Engineering, Guadalajara, Mexico, 4:102126
Peckover FL, Kerr JWG (1977) Treatment and maintenance of rock slopes on transpor-
tation routes. Can Geotech J 14(4):487507
Pierson LA (1992) Rockfall hazard rating system. Transport Res Record No 1343:613
Pine RJ, Roberds WJ (2005) A risk-based approach for the design of rock slopes subject to
multiple failure modesillustrated by a case study in Hong Kong. Int J Rock Mech
Min Sci 42:261275
Piteau DR (1977) Regional slope stability controls and related engineering geology of the
Fraser Canyon, British Columbia. Landslides - Rev Eng Geol GSA 3:85111
Porter M, Jakob M, Holm K (2009) Proposed landslide risk tolerance criteria. 62nd
Canadian Geotechnical Conference and 10th Joint CGS/IAH-CNC Groundwater Confer-
ence, Halifax, pp 533541
Shamekhi SE, Tannant DD (2010) Risk assessment of a road cut above Highway #1 near Chase,
BC. Proceedings of the 63rd Canadian Geotechnical Conference, Calgary, pp 94101
Spang RM, Rautenstrauch RW (1988) Empirical and mathematical approaches to rockfall
protection and their practical applications. In: Bonnard (ed) Landslides: Proceedings of
the Fifth International Symposium on Landslides, Balkema, Rotterdam, pp 12371243
Statistics Canada (2010) Mortality, Summary List of Causes 2007. Catalogue no.
84F0209X, Electronic file from www.statcan.gc.ca extracted June 11, 2012
Vick SG (2002) Degrees of Belief: Subjective Probability and Engineering Judgment.
American Society of Civil Engineers, ASCE Publications, pp 455
Wang X, Frattini P, Crosta GB, Zhang L, Agliardi F, Lari S, Yang Z (2013) Uncertainty
assessment in quantitative rockfall risk assessment. Landslides. doi:10.1007/s10346-
013-0447-8
Whalley WB (1984) Rockfalls. Brunsden, Prior (ed) Slope Instability, John Wiley and Sons,
New York, pp 217256
Wyllie DC, McCammon NR, Brumund W (1980) Planning slope stabilization programs by
using decision analysis. Transportation Research Record 749, Transportation Research
Board, Washington DC, USA, pp 3439
You X, Tonon F (2012) Event-tree analysis with imprecise probabilities. Risk Anal
32(2):330344
R. Macciotta ())
Department of Civil and Environmental Engineering,
University of Alberta,
3-017 Markin/CNRL Natural Resources Engineering Facility, Edmonton, AB T6G 2W2,
Canada
e-mail: macciott@ualberta.ca
C. D. Martin
Canadian Rail Research Laboratory, Department of Civil and Environmental Engineering,
University of Alberta,
3-071 Markin/CNRL Natural Resources Engineering Facility, Edmonton, AB T6G 2W2,
Canada
N. R. Morgenstern
Department of Civil and Environmental Engineering,
University of Alberta,
3-075 Markin/CNRL Natural Resources Engineering Facility, Edmonton, AB T6G 2W2,
Canada
D. M. Cruden
Department of Civil and Environmental Engineering,
University of Alberta,
3-064 Markin/CNRL Natural Resources Engineering Facility, Edmonton, AB T6G 2W2,
Canada

Landslides 13 & (2016) 127

 Recent Landslides
Landslides (2016) 13:129140 Alena V. Kadetova I Artem A. Rybchenko I Elena A. Kozireva I Vadim A. Pellinen
DOI 10.1007/s10346-015-0661-7
Received: 16 June 2015
Accepted: 11 November 2015 Debris flows of 28 June 2014 near the Arshan village
Published online: 5 December 2015
Springer-Verlag Berlin Heidelberg 2015
(Siberia, Republic of Buryatia, Russia)
Abstract On 28 June 2014, high intensity rainfall resulted in
seven simultaneous debris flows going down the slopes of
the Tunka Ridge in the vicinity of Arshan village, which is a
balneological and alpine resort (51 54 31 N, 102 25 44 E).
The debris flows caused one life loss and several injuries, 112
buildings were damaged, and 15 were completely destroyed.
The total volume of the transported deposits amounted to
3106 m3. Debris flows formation began with the failure of
weak sediments in the hanging cirques. Similar phenomena
had not been recorded in the study area for over 40 years.
The article presents a complete picture of the event and
analysis of geological, geomorphological, and meteorological
conditions for debris flows formation, for which extreme
local rainfall was the major cause.
Keywords Central Asia . Debris flow . Tunka Ridge . Mountain
areas
Introduction
Worldwide debris flows cause catastrophic damage in moun-
tain and foothill areas (Hrlimann et al. 2006; Yueping et al.
2011; Mateos et al. 2012; Jakob et al. 2013; Xiao et al. 2013;
Seplveda et al. 2014; Youssef et al. 2014; Bregoli et al. 2015).
Investigations of debris flows show that activation of debris
flow processes is often caused by abundant rainfall (Allen
et al. 2015; Martha et al. 2015).
On 28 June 2014, at about 8 a.m. local time, as a conse-
quence of a heavy and prolonged rainfall in the vicinity of
the resort village of Arshan (Tunka District of the Republic
of Buryatia), the Kyngarga River bursted its banks, and
seven debris flows of various sizes simultaneously descended
the creek valleys of the Tunka Ridge (Laperdin et al. 2014).
Two of the seven streams came very close to the village and
caused damage to buildings and infrastructure. According to
the chief directorate of the Ministry of Emergency Situations
of Russia for the Republic of Buryatia, the debris flows
resulted in one life loss, several injuries, 112 buildings being
damaged, and 15 buildings being partially or completely
destroyed. Moreover, the debris flow demolished the
Kyngarga River bridge, severely damaged local roads, and
covered market gardens of local residents with stones and
sand. In addition, the debris flows partly affected the envi-
ronment: they changed active channels, destroyed large for-
est areas, covered large areas with sandy debris material, and
the sediments dammed the mineral springs for a few days.
Preliminary summary of the events was delivered hot on the
traces by the authors in the paper (Laperdin et al. 2014).
with analysis of the events consequences and photographs.
In this article, we have compiled a more precise chronology
of the events and have attempted to determine the main
causes of this natural phenomenon.
Study area
Geographical and geomorphological setting
Arshan village is situated at the foot of the southern slope of
the Tunka Ridge, where the Kyngarga River enters the Tunka
Valley, at the altitude of 893 m a.s.l. (Fig. 1). In the village,
healing mineral and thermal waters discharge at the faults
junction (Pinneker et al. 1968). The village is a balneological
and alpine resort which is visited annually by tens of thou-
sands of people. The resident population of the village is
2700 people, but tourists and visitors increase this figure
significantly, especially during the summer season. The vil-
lage housing is comprised of onetwo-storey houses and a
small number of threefour-storey buildings of the resorts
and hotels; the constructions are mainly wooden with some
exceptions for brick.
The Tunka Ridge is the eastern spur of the Eastern Sayan
Ridge. The maximum elevation of the Tunka Ridge is 3284 m
(Strelnikov Peak) with the altitude range of 2000 m. Within
the case area, the maximum elevation reaches 2627 m a.s.l.
The ridge is bounded from the north and south by Main
Sayan and Tunka faults, respectively (Fig. 1). The ridge is
eastwest orientated; its southern slope changes into pied-
mont plain (Fig. 2). Within the study area, the ridge is cut
by deep valley of the Kyngarga River and numerous deep
but short V-shaped shallow stream valleys. Those streams
originate from glacial hanging cirques (Shetnikov 2001).
Geological conditions
Geologically, the Tunka Ridge in the area of debris flows is
composed of Archean rocks, lower Proterozoic complex
(gneiss, gneiss-granites, schists) with thin marble inter-
l a y e r s a n d Q u a t e r n a r y s e d i m e n t s ( Su m b urg h 1 971 ) .
Bedrock differs in physical and mechanical properties, such
as abradability. Most resistant to water-rock flows are
granite-gneisses (their abradability is 0.160.18 cm 3 /cm 2 )
and gneisses (0.150.20 cm 3 /cm 2 ); marble are most prone
to abradability (6.688.72 cm 3 /cm 2) . Another important in-
dicator of the bedrock strength properties is weathering
rate; it varies from 0.11 cm/year for granites, up to
Landslides 13 & (2016) 129
 Recent Landslides
Fig. 1 Location map. The inset shows the location of the study area within East Asia. SRTM data (Jarvis et al. 2008) were used for digital elevation model
1.0 cm/year for shales, 0.7 cm/year for marbles, and 0.6 cm/
year for gneisses (Engineering geology of the Baikal region
1968). Quaternary deposits are present by fluvioglacial sed-
iments, alluvial terrace sediments, and proluvial-deluvial
fan sediments (Fig. 3). The sediment material varies from
large boulders (over 2 m in diameter) to sand fraction.
Climate
Climatically, the area is characterized by considerable am-
plitude of seasonal and daily temperatures and predomi-
nance of precipitation in the summer time. The average
annual temperature in Arshan village is 1.4 C. The
terrain features have impact on spatial distribution of pre-
cipitation. In the mountains, its amount is significantly
higher than in the lowland area (Ufimtsev and
Shchetnikov 2003). According to Arshan meteorological sta-
tion (operated until 1997), which was located in the foothill
area, the mean annual rainfall can reach 750 mm, which on
average is 200 mm over the amount recorded at the Tunka
meteorological station located 20 km to the south in the
valley (Meteorological monthly magazine 1972). In summer,
rainfall amount can reach up to 170 mm/day (Arshan me-
teorological station) (Meteorological monthly magazine
1972). The bulk of precipitation (about 70 % of mean
annual rainfall) falls in the summer months from June to
August. For most of the year, precipitation falls as snow.
130 Landslides 13 & (2016)
According to the national park staff and locals, the maxi-
mum snow depth is 30 cm in the foothills and lowlands,
and it reaches 100140 cm in the mountains (in hanging
cirques). Perennial snow cover comes off in the third and
sometimes in the second decade of April. In the mountains,
the snow cover remains for up to 180 days (data of
National Climatic Data Center). The overall picture of tem-
perature, precipitation, and snow depth reflected in Fig. 4
shows data for the Tunka weather station. Although
amount of precipitation and snow depth parameter values
within the Arshan village area may differ from the values
recorded at the Tunka weather station, meteorological data
from this station has been used in this article in order to
present an overview of the general climatic characteristics
of the study area.
Cryogenic conditions
In hanging cirques, the seasonal freeze-thaw layer is 2.0
2.5 m or more (Laperdin and Trzhtsinskii 1976; Leshchikov
1978). Stability of the overlying weak sediments on the
slopes largely depends on the presence of the frozen layer
and the regime of temperature distribution inside the layer
in the annual cycle. Soil thawing begins in April and May
and depends on the elevation, slope exposure, and vegeta-
tion. In late Juneearly July, seasonally thawed layer does
not exceed 0.7 m on southern slopes and 0.4 m on
 l

ll

l l

l l

l l ll
l
ll l l l
ll l l l
ll l l l

ll l l l l l
l l

ll

ll l

ll

Fig. 2 Geomorphological scheme of the study area (Shetnikov 2001)

northern slopes, while in the range area, the thawed layer Results
is merely about 0.2 m. Most intensive thawing of seasonal
permafrost usually occurs in July and August, with the Event chronology
arrival of warm rainfall. Penetrating into the depth of half As mentioned above, in the early morning on 28
thawed rocks, water forms a temporary aquifer. Water in- June 2014 seven debris flows of different sizes simulta-
filtrates up to the boundary of frozen layers and contrib- neously descended down the brook valleys of the Tunka
utes to soil thawing and liquefaction (Poznanin 1991). In Ridge near the village of Arshan (Fig. 5). Local residents
summer time, the waters above the frozen layer form nu- confirmed that an extreme downpour occurred during the
merous outflows with discharge zone located in the foothill night from 27 to 28 June. The Kyngarga River bursted its
area (Solonenko 1960; Leshchikov and Shats 1983). banks and flooded the streets and gardens of local

Landslides 13 & (2016) 131

 Recent Landslides
Fig. 3 Geological scheme of the study area (Sumburgh 1971)
residents in the village. In the morning of 28 June, debris
flows began to come down the brook valleys. The flows
force can be judged from the fact that huge boulders of
up to 5 m in diameter were moved down (Fig. 6); the
debris flows damaged cars and buildings and washed out
the roads; the flows left erosional forms up to 1-m deep.
The debris flow velocity varied from 4.80 to 13.07 m/s
(Table 1). The traces on the trees and buildings suggest
that the highest level of debris flow reached more than
5 m in the area of the alluvial fans (Fig. 7). By studying
scars and mud splashes on the trees, there is a way to
determine the height of the debris flow level after the
132 Landslides 13 & (2016)
event (Klime and Vilmek 2011). The second, fifth, and
sixth debris flows had the greatest force and formed the
largest accumulative forms (alluvial fans). The most sig-
nificant damage was caused by the fifth debris flow; it
damaged the resort building and destroyed some local
houses (Fig. 8). The major parameters of the debris flow
basins are presented in Table 1.
Morphology and sedimentology
Debris flows formation began with the failure of weak sedi-
ments in the hanging cirques at the height of 1640 to
1850 m a.s.l. Sediment failures in all debris flows basins,
Fig. 4 Main meteorological characteristics of the Tunka Valley for 2014 (according to the Tunka weather station): air temperature, precipitation volume, and snow depth
(data of National Climatic Data Center). The graph demonstrates climatic conditions within the study area. The date of event is highlighted by pink color

except for the third, occurred in the frontal parts of the
moraine ramparts. As a result of the extreme rainfall, hanging
cirques accumulated large volumes of water. The existing
permafrost prevented precipitation infiltration, thereby con-
tributing to the formation of concentrated surface water flow
in the direction of the moraine rampart. Extreme water vol-
ume led to failure of weak sediments in the zone of transition
from hanging cirque to the slope, where the inclination angle
changes dramatically from 2 in the hanging cirque to 50 in
the ridge slope (Fig. 9). In the debris flows origin places, in

Fig. 5 Location scheme of the debris flow basins. SRTM data were used (Jarvis et al. 2008)

Landslides 13 & (2016) 133

 Recent Landslides

Alluvial
The average angles of inclination, degrees

fans

16

16
7
9

6
7
5
Propagation
zones

16
14
16
22
17
13
13
Hanging
cirques

21
18
21
9
9
8
8
velocitya, m/s
Average
Fig. 6 Debris flow sediments of the sixth debris flow basin

11.84

12.15
13.07
4.80

5.45
9.96

addition to the frozen and thawed deposits of moraines, the

Melton
landslide material also was involved in debris flows. In the

ratio

0.88
1.00
1.04
0.80
0.81
0.58
0.58
third basin, debris flow was triggered by failure of colluvial
sediments from the nivation cirques slopes. The failure oc-
curred along the rock avalanche paths. elevation, m
Debris flow sediments have polydisperse composition; frac-
Maximum

tional size varies from fine particles to large boulders.

2370
2544
2260
2544
2470
2363
2627
Accumulation of coarse fraction of debris flow sediments that
largely consists of boulders of up to 5 m in diameter begins
at the altitudes of 1305 to 1000 m a.s.l. and can be traced to
the 1100- to 820-m a.s.l. level (Fig. 10). Sediment thickness of
elevation, m

coarse fractions reaches 5 m.

Minimum

Debris flow material was accumulated in several stages; a

1100
1040
1060
1320
1070
1100
1120
few debris flow waves can be traced within alluvial funs. This
is evidenced by the sediment character (the presence of ero-
sion forms, lateral levees in the alluvial fans and also
The velocity was calculated by the formula Vc=(1/Nc)Hc2/3 Ic1/5 (Rickenmann 1999)
branches from the main debris flow) (Fig. 11). Accumulative
The area of the

coarse fraction,
alluvial fan of

material deposited with the first wave of debris flow was

washed out by the next waves. Alluvial fan of sixth debris
0.004
0.102
0.013
0.018
0.210
0.281

flow was divided into two parts (Fig. 10a). This happened due
km2

to the fact that the debris flow came up against a block field,
losing its energy due to infiltration of its water component
into the block field. New debris flow waves moved down in
The catchment

two separate streams. Fine fraction of debris flow sediments

Table 1 The main characteristics of debris flow basins

deposited at the 1100820-m a.s.l. marks, which covered the

area, km2

village and the surrounding area. Thickness of this fraction of

2.09
2.28
1.34
2.37
2.96
4.74
6.76

sediments was up to 1 m.

Discussions
The length of
basin include

Features of the relief and debris flows basins

fan, km

Morphometric characteristics, topographic features of each

11.61
3.96
3.84
2.97
4.50
7.13
8.21

debris flow basin and the distribution pattern of precipita-

tion had a significant influence on both the flows energy
and the volume of the solid debris flow component. Melton
ratio for all debris flow basins is high: from 0.58 to 1.04,
Debris

basin
flow

indicating proneness to the debris flow formation (Melton

1
2
3
4
5
6
7

1965; Wilford et al. 2004). The first and second debris flow
a

134 Landslides 13 & (2016)


Fig. 7 Debris flows traces on the tree. Both ripped bark and stone stuck in the
tree mark the level of the debris flow. The total debris flow level in the area of
alluvial fan was over 5 m (the thickness of alluvial fan is about 3 m, the height of
traces on the tree is 2.75 m)
basins are morphometrically similar in catchment area size,
as well as the slope gradient of the source and propagation
areas (Table 1). However, the areas of alluvial fans of those
debris flows are significantly different. This fact is attrib-
uted to the local nature of precipitation, i.e., the first valley
was touched by the rain front edge and got less
a b
Fig. 8 The consequences of the debris flows in the Arshan village: a three-storey building of Sagaan-Dali resort, the first floor is entirely covered by debris flow sediments;
b destroyed building. These photos were taken in early July 2014
precipitation. At the same time, the second debris flow
basin was affected by large proportion of precipitation.
The third debris flows basin differs from all other basins
by morphometric characteristics. The catchment basin here
is represented not by a hanging cirque, but by a nivation
cirque, which has a much smaller catchment area, steep
slope of the flow channel, and a relatively smooth longitu-
dinal profile without benches. In this debris flow basin,
failure of weak sediments started from the top of the
nivation cirque. The character of the longitudinal relief
profile of the debris flow basin enabled high flow energy,
as evidenced by a worked-out V-shaped debris flow path
with high steep sides.
The fourth and fifth debris flow basins have similar
morphometric characteristics (see Table 1), but the volume
of the carried material in the fifth debris flow basin far
exceeds that of the fourth basin. This is due to the features
of the debris flow basin microrelief. The analysis of satellite
images and field studies allowed to establish that a portion
of water stream flows from the hanging cirque, which
serves as a initiation area of the fourth debris flow basin,
into the adjacent valley of the fifth debris flow basin
through a gully (Fig. 12). Another portion of the stream is
drained through the front moraine rampart and flows down
the valley. These microrelief features led to the decrease of
the water component of the fourth debris flow and increase
of water pulse of the fifth debris flow. Thus, the fourth
debris flow did not have enough energy to wash out the
deluvial deposits, as what happened in other debris flow
basins. This is supported by shallow erosional cut in the
propagation zone of the fourth debris flow basin and by
the fact that accumulation of debris flow deposits began at
higher hypsometric marks.
The sixth and seventh debris flow basins are the largest.
The size of their catchment areas is more than twice as
large as those of other debris flow basins (Table 1).
Morphometric characteristics of the two basins are almost
identical. However, the volume of carried debris flow mate-
rial of sixth basin exceeded that of the others. Alluvial fan
of the seventh debris flow basin, on the contrary, is small
Landslides 13 & (2016) 135
 Recent Landslides
Fig. 9 Failure of weak sediments on the bottom part of hanging cirque, which
resulted in the debris flows beginning. The yellow dotted line denotes the line
of weak sediments failure. The figure shows the inclination angles of the surface:
and not clearly contoured. Different quantity of debris flow
material in two morphometrically equal debris flow basins
is due to the fact that the seventh debris flow basin, as well
as the first one, was located at the edge of the rain front
and got a smaller water volume.
Deposits
Solid part of the debris flows included proluvial-deluvial
slope deposits, as well as glacial deposits of hanging
cirques. The input of bedrock in the debris flow is so small
that it cannot be considered as a source of the solid debris
material. Small amount of bedrock enter the flow due to
the debris flow erosive action at the sites of bedrock expo-
sure in the propagation paths. Among the bedrocks, marble
outcrops are most affected by erosion. The hanging cirques
of the Tunka range are filled mostly by glacial deposits,
which are composed of coarse material with an average
diameter of 1 m . I n additio n, as a result of rock
weathering, the hanging cirques and valleys accumulate
large amounts of coluvial deposits. The quantity of glacial
deposits of the cirques in debris flows is smaller than the
volume of deluvial deposits involved in the flow. Despite
the large thickness of glacial deposits, they were not fully
involved in the debris flow for two reasons: firstly, cirque
glacial deposits are well drained, and secondly, the cirque
bottoms are flat (821). Such conditions were unfavorable
for the movement of the thick glacial deposits. The debris
flows partially involved the deposits of the front parts of
frontal moraines, located on the bend from the cirque to
the ridge slope, where the slope is steeper than in the
136 Landslides 13 & (2016)
yellow linesthe surface of the hanging cirque, red linesthe surface of the
slope, blue linesthe horizontal surface. The red arrow marks the place of
suprapermafrost water discharge. The photo was taken on 1 July 2014
cirque bottoms (Fig. 9). The bulk of the solid component
of the debris flows includes deluvial and proluvial slope
deposits (Fig. 13). Their composition ranges from large
boulders (up to 5 m in diameter) to the sand fraction.
Deluvial and proluvial deposits provide coarse and fine
fraction for the solid component of the debris flow. An
analysis of the size distribution of the deluvial deposit
samples from the margins of debris flow paths demonstrat-
ed that the presence of <1 mm fraction was 69 % (Fig. 14).
It is that particular size of fraction that combined with
water forms a debris flow suspension effect and increases
the carrying capacity of a debris flow (Perov 2012; Ni et al.
2014).
Precipitation
The extreme rainfall was responsible for the debris flows.
Unfortunately, there is no operating weather station at
present within a radius of 20 km from the Arshan village,
and we cannot accurately estimate the amount of precipi-
tation, which served as a trigger for the debris flows.
However, we can compare the scale of natural events of 18
July 1962 and 27 July 1971 in Arshan. For those events, there
is available data about the rainfall volumes. In July 1971, a
flash-flood occurred on the Kyngarga River. A water flow
swept away windmills, bridges, and houses (Makarov 2012).
According to local witnesses of both events (1971 and 2014)
the flooding on the Kyngarga River in 1971 was stronger
than in 2014. According to the data from Arshan weather
station, 90.4 mm of rainfall was recorded that day
(Meteorological monthly magazine 1972). but the rainfall
 a
Fig. 10 Alluvial fans of the debris flows: a the alluvial fun of the sixth debris flow divided into two parts, b the alluvial fun of the fifth debris flow. The arrows indicate
the flow directions
volume could be higher in the mountains in the river head.
As for the event of 18 July 1962, heavy rainfall also trig-
gered debris flows descended down the Kyngarga River and
ten brook valleys located to the east and west of it.
According to the Arshan weather station, 16 and 17
July 1962 was marked by 176 mm of rain in 2 days
(88 mm on July 16 and 88 mm on July 17) (Meteorological
monthly magazine 1972).
Conclusions
Debris flow of 28 June 2014 caused social and economic
damage to the resort village of Arshan and the loss of one
life. Two of the seven streams destroyed buildings and dam-
aged roads and facilities of the village; infrastructure of the
resort was disabled for some time. Debris coarse material
covered the area of 670103 m2. The total volume of fans
was 3106 m3.
Landslides 13 & (2016) 137
 Recent Landslides
Fig. 11 Morphological elements in the alluvial funs of debris flows: a lateral levee, b erosional incision
Geomorphological conditions in combination with the
presence of weak sediments and meteorological situation
of the area favors hazardous gravitational processes, in-
cluding debris flows. The terrain energy is favorable for
debris flow occurrence in this region, Melton ratio for the
debris flow basins ranges from 0.58 to 1.04. Such basins are
prone to formation of water flow of sufficient power for
transformation into a debris flow (Wilford et al. 2004).
Debris flows of 28 June 2014 were triggered by intense
torrential rainfall. Precipitation had a local character,
which is quite typical of the area. A 4.5-km-long rain
front passed over the catchment area from the second
138 Landslides 13 & (2016)
to the sixth debris flow basins and over the left upper
tributaries of the Kyngarga River and affected only the
edge of the first and seventh catchment basins. Although
there is lack of precise data on the rainfall amount that
day, based on the past events data (1962 and 1971), we can
conclude that the rainfall was minimum of at least
90 mm/day. Extremely large amounts of rainfall created
the conditions for watering the thickness of already
thawed weak sediments, suspension of debris sediments
of different genesis, and lithology. The shift occurred
down the frozen, inclined basement as a result of flood
and movement of avalanchine water flow; debris flow
Fig. 12 The terrain features of fourth and fifth debris flow basins. The figure shows a gully through which the water flows from the hanging cirque of the fourth debris
flow basin to the fifth one: a, b the Google space images, c, d the photo of the gully from different spots

formed. The main source of the debris flow solid phase Despite the recent debris flows that have taken out a
were deluvial-proluvial slope sediments, glacial sediments large amount of weak material, a large quantity of debris
of the lower parts of the cirques, as well as, to a minor material still remains in the valleys, which by abnormally
extent, weathered bedrock. high rainfall can lead to formation of new debris flows. The

Fig. 13 a, b Deluvial deposits exposed by debris flows in the propagation zones

Landslides 13 & (2016) 139

 Recent Landslides
Fig. 14 Particle size distribution of fine fraction of the deluvial sediment samples
from the margins of the debris flow path
data obtained in the course of research and reflected in the
article are the groundwork for future analysis of the mech-
anisms of debris flows.
References
Allen SK, Rastner P, Arora M, Huggel C, Stoffel M (2015) Lake outburst and debris flow
disaster at Kedarnath, June 2013: hydrometeorological triggering and topographic
predisposition. Landslides, 113. doi:10.1007/s10346-015-0584-3
Bregoli F, Medina V, Chevalier G, Hrlimann M, Bateman A (2015) Debris-flow suscep-
tibility assessment at regional scale: validation on an alpine environment. Landslides
12(3):437454. doi:10.1007/s10346-014-0493-x
Engineering geology of the Baikal region (1968) Publishing house BNauka^, Moscow (in Russian)
Hrlimann M, Copons R, Altimir J (2006) Detailed debris flow hazard assessment
in Andorra: a multidisciplinary approach. Geomorphology 78(3):359372
Jakob M, McDougall S, Weatherly H, Ripley N (2013) Debrisflow simulations on Cheekye
River, British Columbia. Landslides 10(6):685699. doi:10.1007/s10346-012-0365-1
Jarvis A, Reuter HI, Nelson A, Guevara E (2008) Hole-filled seamless SRTM data V4,
International Centre For Tropical Agriculture (CIAT), available from http://
srtm.csi.cgiar.org [27 October 2011]
Klime J, Vilmek V (2011) A catastrophic landslides near Rampac Grande in the Cordillera
Negra, northern Peru. Landslides 8(3):309320. doi:10.1007/s10346-010-0249-1
Laperdin VK, Trzhtsinskii YB (1976) The role of weathering in formation of the solid
mudflow phase (a case study of East Sayan). In: geological factors of landslides and
mudslide formation, and the problem of their evaluation. Moscow State University,
Moscow, pp 4955, in Russian
140 Landslides 13 & (2016)
Laperdin VK, Levi KG, Lekhatiniv AM, Kadetova AV, Pellinen VA, Rybchenko (2014) Causes
and consequences of a catastrophic mudflows on 28 June 2014 near Arshan village in the
Republic of Buryatia, Russia. Geodyn Tectonophysics 5(3):799816
Leshchikov FN (1978) Frozen rocks of Priangaria and Pribaikalia. Nauka, Novosibirsk (in Russian)
Leshchikov FN, Shats MM (1983) Permafrost in the southern part of Middle Siberia.
Nauka, Novosibirsk, p 169 (in Russian)
Makarov SA (2012) Debris flows in the Cis-Baikalia. The V.B. Sochava Institute of
Geography SB RAS, Irkutsk (in Russian)
Martha TR, Priyom R, Babu Govindharaj K, Vinod Kumar K, Diwakar PG, Dadhwal VK
(2015) Landslides triggered by the June 2013 extreme rainfall event in parts of
Uttarakhand State, India. Landslides 12:135146. doi:10.1007/s10346-014-0540-7
Mateos RM, Garca-Moreno I, Azan JM (2012) Freeze-thaw cycles and rainfall as
triggering factors of mass movements in a warm Mediterranean region: the case of
the Tramuntana Range (Majorca, Spain). Landslides 9(3):417432. doi:10.1007/
s10346-011-0290-8
Melton MA (1965) The geomorphic and paleoclimatic significance of alluvial deposits in
southern Arizona. J Geol 73:138
Meteorological monthly magazine (1972) 22(13) Irkutsk Hydrometeorological Senter,
Irkutsk (in Russian)
Ni H, Zheng W, Song Z, Xu W (2014) Catastrophic debris flows triggered by a 4 July 2013
rainfall in Shimian, SW China: formation mechanism, disaster characteristics and the
lessons learned. Landslides 11(5):909921
Perov VF (2012) Mudflow research. Study guide. MSU Faculty of Geography, Moscow (in Russian)
Pinneker EV, Pisarsky BI, Lomonosov IS, Koldysheva RY, Didenko AA, Sherman SI (1968)
Hydrogeology of Baikal region. Nauka, Moscow, in Russian
Poznanin VL (1991) Laws of hydrothermal interaction of water flows frozen soil. In:
Denudation in cryolithzone. Nauka, Moscow, pp 6273, in Russian
Rickenmann D (1999) Empirical relationships for debris flows. Nat Hazards 19:4777
Seplveda SA, Moreiras, SM Lara M, Alfaro A (2014) Debris flows in the Andean ranges of central
Chile and Argentina triggered by 2013 summer storms: characteristics and consequences
Shetnikov AA (2001) Geomorphology of Tunkin rift. Geomorphology 2:93103 (in Russian)
Solonenko VP (1960) Engineering geology of East Siberia. Irkutsk (in Russian)
Sumburgh AL (1971) Geological map of the Soviet Union. 1: 200 000. East Sayan. Sheet
M48I. Explanatory note. Moscow (in Russian)
Ufimtsev GF, Shchetnikov AA (2003) The Tunka rift. Priroda 8:4349 (in Russian)
Wilford DJ, Sakals ME, Innes JL, Sidle RC, Bergerud WA (2004) Recognition of debris-flow,
debris-flood and flood hazard through watershed morphometrics. Landslides 1:61
66. doi:10.1007/s10346-003-0002-0
Xiao H, Luo Z, Niu Q, Chang J (2013) The 2010 Zhouqu mudflow disaster: possible
causes, human contributions, and lessons learned. Nat Hazards 67(2):611625
Youssef AM, Al-kathery M, Pradhan B, El-sahly T (2014) Debris flow impact assessment
along the AlRaith Road, Kingdom of Saudi Arabia, using remote sensing data and
field investigations. Geomatics, Natural Hazards and Risk, (Published online), 119.
doi:10.1080/19475705.2014.933130
Yueping Y, Ping S, Jiliang Z, Shengyuan Y (2011) Research on catastrophic rock avalanche
at Guanling, Guizhou, China. Landslides 8:517525. doi:10.1007/s10346-011-0266-8
A. V. Kadetova ()) : A. A. Rybchenko : E. A. Kozireva : V. A. Pellinen
Institute of the Earths Crust,
Siberian Branch of Russian Academy of Sciences,
128 Lermontov Street, Irkutsk, 664033, Russia
e-mail: kadetova@crust.irk.ru
 Technical Note
Landslides (2016) 13:141151 Chia-Ming Lo I Hung-Hui Li I Chien-Chung Ke
DOI 10.1007/s10346-015-0650-x
Received: 26 June 2015
Accepted: 7 October 2015 Kinematic model of a translational slide in the Cidu
Published online: 19 October 2015
Springer-Verlag Berlin Heidelberg 2015
section of the Formosan Freeway
Abstract This study presents a kinematic model of a translational
slide in the Cidu section of the Formosan Freeway in Taiwan,
including the associated kinematic processes and the geometry
of the deposition. A numerical simulation of the slide was per-
formed using the discrete element method. This kinematic model
has been considered the distribution of bedding plane and joints,
engineering installations, and mechanical parameters in the devel-
opment of a numerical model capable of clarifying the kinematic
characteristics of this translational slide with high velocities. When
the friction coefficient of each particle was set to 0.03, the predict-
ed maximum velocity was 23.6 m/s and the debris reached the
other side of the Formosan Freeway. The simulations demonstrat-
ed that the three cars involved in the incident and the Formosan
Freeway itself were buried at 2 to 4 s during the event (the predicted
average velocity was 10.2 m/s) and that the translational slide ceased
its movement at 7.5 s, resulting in the formation of a slightly
fragmented depositional mass.
Keywords Kinematic model . Translational slide . Numerical
simulation . Discrete element method
Introduction
Numerous calamities throughout the metropolitan and mountain-
ous areas of Taiwan have been caused by translational slides. Dip
slopes, at which translational slides are more likely to occur, are
natural slopes (particularly of rock) in which the dip of the strata
and the primary planes of weakness of the strata have the same
direction with a difference in angle of less than 15. The sudden
onset and velocity of translational slides (reaching 100 km/h or
28 m/s) can lead to severe disasters (Hung 2010).
At approximately 14:29 on April 25, 2010, a dip slope slide
occurred at kilometer 3.1 in the Cidu section of the Formosan
Freeway, an area referred to as Shi-gong-ge-shan. A substantial
amount of debris (>200,000 m3) completely buried a section of
the freeway, 190 m long and 35 to 43 m wide, as well as three
vehicles passing through the area. The landslide occurred on a
sunny day with no recorded earth tremors, thus excluding the
possibility that rainfall or an earthquake triggered the event
(Wang et al. 2013). A topographical/geographical investigation
conducted by Lin et al. (2010) following the disaster revealed
that, prior to the event, the slope in the area was distinctly
wedge-shaped with the tip protruding toward the northwest.
Near Shi-gong-ge-shan, the dip slope presented a flatiron feature
in four underlying layers, as shown in Fig. 1, where A marks the
topmost rock section, the gentle terrain of which may have
formed after the dip slope slide. The Cidu landslide occurred
in the bottommost rock section, marked by D in Fig. 1. Chen
et al. (2010) identified this section as a sandstone section (SS) at
the bottom of the Miocene Shihti Formation (St), the thickness
of which is approximately 15 to 20 m. The underlying stratum
exposed by the landslide belongs to the Miocene Taliao Forma-
tion (Tl), which consists lithologically of a thick layer of shale
with occasional beds of thin sandstone formations. The contact
between the two strata was identified as the primary sliding
plane in the Cidu event (Fig. 2). It should be noted that the
strike of rock formations in this area is NE and that the dip
slopes dip toward the SE an angle of 1030. These rock slope
characteristics in conjunction with the fact that Freeway No. 3
cuts through the dip slope increased the likelihood of a dip
slope failure (Wang et al. 2013).
Following the incident, government agencies and researchers
investigated the mechanisms associated with the dip slope slide
and augmented efforts to monitor the stability of dip slopes near
national freeways, and established an early warning system.
However, there remains considerable room for improvement,
particularly with regard to determining the velocity of sliding
masses and their sphere of influence. Extremely high velocity
landslides, particularly rockslides that develop into debris ava-
lanches, are remarkable geological phenomena (Steven and Si-
mon 2006). Velocity is the most important parameter
determining the destructive potential of landslides (Hungr
2007), and terrain-related characteristics, such as mass separa-
tion, collisional interaction, and kinematic change from slope
mass sliding to particle flow, are important for estimating the
sphere of influence in order to protect properties in the down-
stream area (Lo et al. 2011, 2014). Consequently, this study
employed a discrete element program (PFC3D 3.0) to construct
a numerical model of a translational slide in the Cidu section of
the Formosan Freeway. Our goal was to compare the simulated
deposits with those observed onsite, with the ultimate goal of
clarifying the kinematic processes occurring in each stage of the
landslide. The study also conducted numerical simulations of the
collapse in order to estimate the speed of the landslide, which
enabled us to explain the processes and characteristics of this
translational slide.
Methodology
Principle of PFC3D
Particle Flow Code in 3D (PFC3D) is a program developed by
Itasca in 1999 based on the discrete element method (Itasca
2002). The PFC3D model is used mainly to simulate the move-
ment and interactions of rigid spheres (ball elements). More
complex behavior can be modeled by allowing the particles to
be bonded together at their contact points, such that when the
inter-particle forces acting at any bond exceed the bond
strength, that bond is broken. In addition, the PFC3D model
also includes wall elements. The balls and walls interact with
one another via the forces that arise at contacts. In addition,
contacts may not form between two walls; thus, contacts are
either ball-to-ball or ball-to-wall (Itasca 2002). The PFC3D mod-
el is based on the explicit finite difference method of calculating
changes in the system at each time step, when the positions,
amount of overlap, and relative movements of the particles are
Landslides 13 & (2016) 141
 Technical Note
Fig. 1 Result of the geomorphologic interpretation of the slope before the disaster (Lin et al. 2010)
calculated prior to calculation of the contact force using the
forcedisplacement law. This makes it possible to derive new
velocities and positions of the particles in accordance with
Newtons second law of motion.
PFC is used to calculate contact stiffness under the assumption
that the stiffness of the contact between two adjacent objects is
treated as normal and shear springs in series. Normal stiffness (kn)
and shear stiffness (ks) are defined as follows:
kA kB
k 1
kA kB
where [A] and [B] represent the two objects in contact, and k
denotes normal stiffness (kn) or shear stiffness (ks).
In the slip model, given settings are used for the coefficient of
friction. In the event that the slipping force at the interface be-
tween an element in contact exceeds the frictional resistance,
slippage will occur, i.e., slippage occurs when the shear contact
force exceeds the friction and when the contact force is equal to
the friction. The bonding model includes parallel bonds between
particle elements, which combine element groups into other
shapes. The bond will be broken when the external force applied
to the elements exceeds the bonding forces.
142 Landslides 13 & (2016)
Numerical modeling and a comparison of deposit formations
Material parameters of the numerical model
This study used PFC3D to construct a numerical model of the dip
slope slide in the Cidu section of the Formosan Freeway. The
advantage of this model is its ability to simulate the bonding and
separation of objects and also the substantial movements that
occur after failure along weak planes and in response to the
interaction effects of collision. This study referred to the setting
of material parameters outlined by Potyondy and Cundall (2004).
We used the results of mechanical experiments in conjunction
with those obtained from PFC to conduct a preliminary conver-
sion of macro- and micro-parameters before comparing the actual
uniaxial experiments with those of simulations (Fig. 3). Thus, we
could calibrate the conversion formula for the macro- and micro-
parameters and derive the micro-parameters of the simulated
materials.
To facilitate the simulation of energy dissipation in material
collisions, we referred to the rebound coefficient obtained by Giani
et al. (2004) in onsite tests (Table 1). This provided us with
damping parameter settings consistent with those found in the
field. The coefficient of kinetic friction is the key factor in the
movement processes of dip slope landslides; however, obtaining
such measurements can be difficult. Thus, this study compared the
deposit formations that resulted from simulations using a variety
Fig. 2 Aerial photograph and geological cross section at kilometer 3.1 in the Cidu section of the Formosan Freeway (modified from Chen et al. 2010)
of coefficient settings and selected the setting that produced the
result closest to the actual landslide deposit.
Full-scale numerical model
1. 3D terrain: The study adopted a post-event (2010) 1 m1 m
digital elevation model (DEM) for the source area and a pre-
event (2009) 5 m5 m DEM for the deposit area. A total of
25,342 wall elements were used for the construction of terrain,
and the total length and width of the numerical model were set
at approximately 842 and 376 m, respectively.
2. Sliding mass: The sliding mass included 8552 ball elements
with a radius of 1.5 m. In the source area, pre-landslide terrain
was constructed using wall elements with ball elements to fill
the sliding mass. Once the ball elements were stable, bonding
strength was applied where the ball elements came into contact
with one another to form consolidated rock masses. Joints
were added to the rock mass, the contact surfaces of which
possessed friction settings.
3. Ground anchors: According to a summary report by the Tai-
wan Geotechnical Society (2011), the total number of ground
anchors was 572, and each ground anchor had a strength of
60 t and a setting of 20 (angle with respect to a horizontal
plane) into the dip slope. Therefore, this study used ball ele-
ments with parallel bonds (setting a 60-t tensile strength and
an insertion into the dip slope of 20 for each ground anchor)
to simulate the ground anchors and retaining wall at the toe of
the dip slope (Fig. 4). Additionally, we fixed the bottom layer of
the particles so that the upper sliding mass could slide along
the top of the bottom layer of the particles. Then, the parallel
bonds will play the role of the ground anchors.
4. Contact stiffness and bonding strength: Specimens obtained
by on-site drilling (diameter of 10 cm and height of 25 cm)
were used to perform uniaxial compression tests to determine
the uniaxial compressive strength and elastic modulus. Com-
parisons between actual and simulated compression tests were
used to revise the conversion formula in order to derive the
micro-parameters of the rock in the numerical simulation
(Table 2).
5. Damping coefficient: The method proposed by Giani (1992) for
the conversion of viscous damping parameters was used in an
experiment to obtain preliminary settings for the onsite resti-
tution coefficient of rock collisions in the numerical simula-
tion. The goal was to enhance consistency with the movement
observed in the actual landslide.
Landslides 13 & (2016) 143
 Technical Note

Fig. 3 Comparison of test results from simulated uniaxial compression with actual test results

6. Friction coefficient: The results of a direct shear test by the
Taiwan Geotechnical Society (a drilling and experiment report
cited the residual shear strength of the wet shale as approxi-
mately 1417) were used to derive the initial settings of the
friction coefficients. In the numerical simulation, the deposit
formations resulting from different friction coefficient settings
were compared to enable selection of the most suitable simu-
lation parameters.
7. Joint setting: Field survey results from 22 representative mea-
surement points related to joint attitudes following the land-
slide (Fig. 5) revealed five sets of weak planes of weakness
(including the bedding plane and joints) distributed
throughout the collapse area: joint A-1, joint A-2, joint B, joint
C, and joint D. Joint A-1 exhibited a dip slope bedding plane at
an attitude of N45 E/15 E (15 measurement points). The
attitudes of the four other joint sets were N44 E/45 E (three
measurement points), N67 E/60 E (two measurement
points), N55 W/55 N (two measurement points), and N26
W/50 S (three measurement points). The distribution of the
planes of weakness has a significant influence on dip slope
sliding, crumbling of the sliding mass, and the distribution of
the deposit. We therefore included this information in the
numerical model to increase the consistency of the simulation
results.

Table 1 Conversion of on-site damping parameters (modified from Giani et al. 2004)
Normal restitution Converted normal Shear restitution Converted shear
coefficient damping ratio coefficient damping ratio
Bedrock slope 0.50 0.21 0.95 0.02
Bedrock slope covered with 0.35 0.32 0.85 0.05
broken rock
Slope covered with rock debris 0.30 0.36 0.70 0.11
and soil
Soil slope covered with lush 0.25 0.40 0.55 0.20
vegetation

144 Landslides 13 & (2016)

Fig. 4 Numerical model of a translational slide in the Cidu section of the Formosan Freeway

Results the deposit, indicating that a reasonable value would be approxi-

mately 1e10 N/m.

Parametrical analysis Comparison of full-scale numerical simulation results

Table 3 lists the parameter combinations used in this study as well
as the influence of simulation parameters on runout distance and
run-up height. As shown in Tables 3 and 4, by changing the friction
coefficient from 0.01 to 0.3, the runout distance was decreased and
the run-up height was increased. With the lower friction coefficient
of 0.01, a portion of the deposit moved south along the Formosan
Freeway and another portion extended as far as highway No.62.
These results are inconsistent with the actual case, which suggests
that a more reasonable friction coefficient would be in the range
from 0.01 to 0.3. As shown in Fig. 6 and Table 3, an increase in joint
strength reduced the spread of deposited material from north to
south and preserved the geometric appearance of the deposit. As
shown in Table 3, higher contact stiffness expanded the spread of
This study applied various friction coefficients during the simu-
lation process (Fig. 7) to identify deposit formations most con-
sistent with the actual deposits. The factors that we compared
included run-out distance, deposition length, and deposition
width. Our results revealed that a friction coefficient of 0.03
resulted in a deposited terrain that was most consistent with
the formations observed in aerial photos and photogrammetry
results (Table 4 and Fig. 8). Under these conditions, the maxi-
mum sliding velocity was 23.6 m/s, which enabled the entire
sliding mass to reach the slope on the other side of the freeway.
We therefore employed this parameter value to explain the kine-
matics of the dip slope slide in the Cidu section of the Formosan
Freeway.

Table 2 The numerical parameters of the PFC modeling

Parameters Full-scale numerical model
Rock mass Joints
Number of particles 8552
Unit weight of ball elements (kg/m ) 3
2600
Particle radius (m) 1.5
Normal stiffness (N/m) 1.44e11
Shear stiffness (N/m) 7.2e10
Friction coefficient of ball elements 0.010.3 0.010.3
Friction coefficient of wall elements 0.6
Normal stiffness of parallel bond (N/m) 2.0e9
Shear stiffness of parallel bond (N/m) 1.0e9
Normal bond strength of parallel bond (MPa) 26 0.3
Shear bond strength of parallel bond (MPa) 13 0.1
Normal damping coefficient (rock mass) 0.36
Shear damping coefficient (rock mass) 0.11
Material damping coefficient (the soil slope on the far side of the freeway) 0.40

Landslides 13 & (2016) 145

 Technical Note
Fig. 5 Orientation of bedding plane and joints at the Cidu section of the Formosan Freeway
Kinematic processes of the Cidu translational slide
In this study, the numerical simulation was divided into two main
parts. In the first part, we considered that the ground anchors were
located along the toe of the dip slope, then gradually reduced the
ground anchors strength, and simulated the failure process. In the
second part, almost all of the ground anchors were considered to
experience failure, causing translational sliding, and the dip slope
movement process was simulated. Figure 9 presents the simulation
results of the first part with a friction coefficient of 0.03. The
simulation results of this first part are summarized in the follow-
ing points:
1. Original strength of the ground anchors (60 t): The construc-
tion of the anchor system was completed at the toe of the dip
slope in 1998 (Taiwan Geotechnical Society 2011). In this kine-
matic model, due to the usage of the original strength of the
ground anchor supports (all of ground anchors were not at
failure during this stage), the dip slope did not incur displace-
ment or deformation and the overall dip slope presented a
stable state. With long-time corrosion of the anchor head, the
force of each ground anchor will be reduced gradually, even to
the point of failure. When we reduced the strength of the
ground anchors by 015 %, the dip slope still not incur any
displacement or deformation and the overall dip slope pre-
sented a stable state.
2. Reduction of ground anchor strength by 30 % (42 t): When the
ground anchor strength was reduced by 30 %, the
Table 3 Tendency of runout distance and run-up height with increasing value of the different parameters
Input parameters Value
Friction coefficient 0.010.3
Joint strength (MPa) 0.01100
Contact stiffness (N/m) 1e81e11
146 Landslides 13 & (2016)
displacement of the dip slope extended to 0.51.7 m (some
tension cracks were even found on the dip slope) and caused
about 6 % of the ground anchors to fail (Fig. 9a).
3. Reduction of ground anchor strength by 50 % (30 t): When the
ground anchor strength was reduced by 50 %, there was about
89 % failure of ground anchors, causing the dip slope to lose
support and a translational slide to begin (Fig. 9b). Field
investigations by the Taiwan Geotechnical Society performed
after the translational slide (April 2010) showed that only 58
anchors remained in place and that 48 % of the remaining
anchors showed a fracture of their steel strands (Taiwan Geo-
technical Society 2011). These field investigation results are
similar to the results of the first part of the simulation. There-
fore, the study used this model to perform the next step of
simulation (translational slide process).
Figure 10 presents the simulation results of the second part
(assume that almost all of the ground anchors have been
destroyed). The landslide kinematics were divided into the follow-
ing time periods:
1. 0.00.5 s: Almost instantaneously, the toe portion of the right
wing of the sliding mass (in the direction toward Taipei, near
the side of the overpass) reached a velocity of 8.2 m/s, such that
the mean velocity of the entire sliding mass was 2.9 m/s
(Fig. 10a).
Runout distance Run-up height
Decreasing Increasing
Decreasing Increasing
Increasing Decreasing
Table 4 Comparison result of modeling with different friction coefficients
Coefficient of friction Movement distance (m) Deposition length (m) Deposition wide (m)
0.3 0 0 0
0.2 0 0 0
0.1 0 0 0
0.05 0 0 0
0.04 185.64 170.21 144.47
0.03 253.23 182.65 173.51
0.02 263.58 189.25 176.76
0.01 326.45 201.63 253.97
Measurement results 259.78 187.78 174.17

2. 0.52.0 s: During this period, the mean velocity of the topmost
layer of the sliding mass exceeded 12.8 m/s, and the mean
velocity of the bottom layers remained steady at 9.6 m/s, which
was sufficient to bury the three lanes of the freeway in the
Taipei-bound direction. The sliding mass had not yet crumbled,
even at the time it collided with the vehicles in the first three
lanes of the highway, before swiftly pushing on toward the
Keelung-bound lanes. In this stage, the load pressure of the
topmost layer was small relative to that of the bottom layers,
and it was not constrained by the effect of the ground anchors.
A few ground anchors may still have maintained support func-
tionality at the toe of the dip slope, or the slip surface that
caused the velocity of the bottom layers was small relative to
that of the topmost layer. Additionally, almost all of the ground
anchors had lost support functionality at 0.52 s, even about
13 % of the toe of the sliding mass appeared to be in a Bfloating
state^ (this caused the contact area between the sliding mass
and sliding surface to be reduced), and the static friction had

Fig. 6 Comparison result of different joints strengths

Landslides 13 & (2016) 147

 Technical Note

Fig. 7 Numerical modeling of the Cidu translational slide of the Formosan Freeway

Fig. 8 Compaction of the numerical simulation and aerial photos of the Cidu translational slide at the Formosan Freeway

148 Landslides 13 & (2016)

Fig. 9 Simulation results of the translational slide at the Cidu section of the Formosan Freeway with ground anchors
changed into dynamic friction of sliding mass movement,
which caused the velocity of the sliding mass to increase sud-
denly (Fig. 10b).
3. 2.03.0 s: At this point, the toe of the right wing had reached
the slope on the other side of the freeway and had begun to
decelerating. This part of the sliding mass began to crumble
along the weak plane of joint D, and some portions began to
settle onto the freeway. However, the main portion of the
sliding mass continued to the lanes on the far side (Fig. 10c).
4. 3.05.0 s: During this time, the separated portions of the right
toe stopped moving, leaving them scattered between the two
lanes. When the main portion of the sliding mass collided with
the slope on the far side of the freeway, it continued moving up
the slope. During this period, the velocity of the main sliding
mass peaked (at 23.6 m/s), whereas the velocity of the right toe
decreased to 2.3 m/s (Fig. 10d).
5. 5.07.5 s: At approximately the 5.0-s point, the main portion of
the sliding mass was climbing up the slope on the far side of
the freeway (Fig. 10e). Because of the slope barrier and influ-
ence of increased friction, the velocity of the sliding mass
showed significant attenuation. The entire mass came to a
complete stop by 7.5 s (Fig. 10f).
Evaluation of the Cidu translational slide modeling
The success of landslide modeling depends on the ability of the
model to maintain consistency with observed deposits while si-
multaneously matching the reports of eyewitnesses and measure-
ment data. In the current case, we sought to ensure agreement
between simulation results and observations with regard to land-
slide volume, the position of the collapse area, and the terrain of
the slip surface. Agreement between other factors was made indi-
rectly, e.g., choosing the friction coefficient for each slip surface,
the bond strength, and the contact stiffness to reproduce known
patterns of deposition. Nevertheless, other consistencies that were
not determined directly or indirectly were also observed:
1. Most of the proportions of materials that slid from the north-
west onto the Formosan Freeway to the southeast were correct.
In addition, the thickness distribution in the model closely
matched the values observed throughout the deposit.
2. Approximately 20 % of the sliding mass climbed the slope on
the far side of the Formosan Freeway, approximately 35 % of
the sliding mass remained on the freeway, and the remainder
stagnated on the slide surface. These values match the actual
Landslides 13 & (2016) 149
 Technical Note
Fig. 10 Kinematic process of the translational slide at the Cidu section of the Formosan Freeway
amount of debris deposited on the freeway and opposite slope.
The rather more substantial deposits on the slip surface
formed a slightly fragmented deposit, providing evidence of
the processes involved in the Cidu translational slide.
3. Combining geomorphologic analysis and field observation is
crucial to the construction of landslide models that accurately
represent actual landslides (Lo et al. 2011). Wang et al. (2013)
suggested a mechanism for the Cidu translational slide, which
indicated that the event was caused by hollowed-out dip slope
toes and infiltration of water into the surface that resulted in
loss of strength in the clay-like material. This material, which is
assumed to be the result of shearing during landslide move-
ment and earlier deformation, forms a basically impermeable
layer at the base of the landslide material. As a result, water
pressure builds up over this layer and decreases the effective
stress. This mechanism may have been the trigger initiating the
Cidu translational slide. Following the initiation of movement,
the clay-like material probably played an important role by
forming a lubricating layer at the base of the landslide material
by maintaining high pore pressure during movement (Tsou
150 Landslides 13 & (2016)
et al. 2011). Therefore, succeeding landslide models also should
follow this mechanism to reconstruct the Cidu translational
slide event. The geomorphologic analysis revealed that the
source area of the slip surface was covered with rock masses
and joints, which we represented using ball elements with joint
settings in the numerical model. Slip surfaces and irregular
terrain in the study area were represented using wall elements
in the construction of the terrain model. The low friction
coefficient reflects a reduction in the effective friction coeffi-
cient associated with the sliding surface due to pore fluid.
Conclusion
This study used a discrete element method to simulate the kine-
matic processes involved in the translational slide in the Cidu
section of the Formosan Freeway. We included the distribution
of planes of weakness, engineering installations (the ground an-
chors and retaining wall), and mechanical parameters in the de-
velopment of a numerical model capable of clarifying the
kinematic characteristics of this translational slide with high ve-
locities. The results indicate that within 24 s from initiation, the
mean velocity of the sliding mass reached 10.2 m/s, such that the
freeway and three vehicles were buried almost instantaneously. In
7.5 s, the entire sliding mass came to a rest. The sliding mass
formed a slightly fragmented deposit on the freeway. The results
of numerical simulation are applicable of the assessment of pre-
disaster scenarios and provide a valuable reference for the desig-
nation of hazardous areas associated with dip slope slides and the
planning of early warning systems.
Acknowledgments
The research was mainly supported mainly by the Ministry of
Science and Technology of Taiwan, Grant no. MOST 103-2625-M-
270-001 and MOST 104-2625-M-270-001. The advice, comments,
and help provided by the editor and reviewer have significantly
strengthened the scientific soundness of this paper. Their kind
assists are gratefully acknowledged.
Lo CM, Lin ML, Tang CL, Hu JC (2011) A kinematic model of the Hsiaolin landslide
calibrated to the morphology of the landslide deposit. Eng Geol 123:2239
Lo CM, Lee CF, Chou HT, Lin ML (2014) Landslide at Su-Hua Highway 115.9 k triggered
by Typhoon Megi in Taiwan. Landslide 11(2):293304
Ministry of Transportation and Communications ROC (2010) Translational slide at the
Cidu section (3K+100) of Formosan Freeway preliminary examination report. Ministr
Transport Commun ROC Rep. (in Chinese)
Potyondy DO, Cundall PA (2004) A bonded-particle model for rock. Int J Rock Mech Min
Sci 41:12391364
Steven NW, Simon D (2006) Particulate kinematic simulations of debris avalanches:
interpretation of deposits and landslide seismic signals of Mount Saint Helens, 1980
May 18. Int J Geoph 167:9911004
Taiwan Geotechnical Society (2011) Summary report of investigation of reasons for
Taiwan freeway No. 3 (Formosan Freeway Cidu section) 3 k+300 m landslide. Rep.
prepared by Taiwan Geotechnical Society for Ministry of Transportation and Commu-
nication, Taiwan: 1679 [in Chinese]
Tsou CY, Feng ZY, Chigira M (2011) Catastrophic landslide induced by Typhoon Morakot,
Shiaolin, Taiwan. Geomorphology 127:166178
Wang L, Hwang JH, Luo Z, Juang CH, Xiao J (2013) Probabilistic back analysis of slope
failurea case study in Taiwan, 1980 May 18. Comput Geotech 51:1223

References

Chen MM, Wei ZY, Fei LY (2010) Analysis geology of dip slope at the Cidu section of C.-M. Lo ())
Formosan Freeway. Geology 29(2):1215 (in Chinese) Department of Civil Engineering,
Giani GP (1992) Rock slope stability analysis. Balkema, Rotterdam, p 361 Chienkuo Technology University,
Giani GP, Migliazza M, Segalini A (2004) Experimental and theoretical studies to improve Changhua, Taiwan, Republic of China
rock fall analysis and protection work design. Rock Mech Rock Eng 37:369389 e-mail: ppb428@yahoo.com.tw
Hung JJ (2010) A study on the failure and disaster prevention of dip slopes. J Civil H.-H. Li
Hydraul Eng 94:518 (in Chinese) Department of Environmental Information and Engineering,
Hungr O (2007) Dynamics of rapid landslides, progress in landslide science. Chapter 4, pp National Defense University,
4756 Taoyuan, Taiwan, Republic of China
Itasca Consulting Group Inc (2002) PFC3D Particle flow code in three dimensions theory
and background, Version 3.0. Minneapolis, pp 12 C.-C. Ke
Lin HH, Chen ZS, Chi ZC (2010) Interpretation geomorphologic of dip slope at the Cidu Geotechnical Engineering Research Center,
section of Formosan Freeway. Geology 29(2):2023 (in Chinese) Sinotech Engineering Consultants, Inc.,
Taipei, Taiwan, Republic of China

Landslides 13 & (2016) 151

 Technical Note
Landslides (2016) 13:153164 Chien-Yuan Chen
DOI 10.1007/s10346-015-0654-6
Received: 17 March 2015
Accepted: 2 November 2015 Landslide and debris flow initiated characteristics
Published online: 14 November 2015
Springer-Verlag Berlin Heidelberg 2015
after typhoon Morakot in Taiwan
Abstract Typhoon Morakot brought extreme rainfall and initiated
numerous landslides and debris flows in southern Taiwan in
August of 2009. The purpose of this study is to identify the
extreme rainfall-induced landslide frequency-area distribution in
the Laonong River Basin in southern Taiwan and debris flow-
initiated conditions under rainfall. Results of the analysis show
that debris flows were initiated under high cumulative rainfall and
long rainfall duration or high rainfall intensity. The relationship of
mean rainfall intensity and duration threshold could reflect debris
flow initiation characteristics under high rainfall intensity in short
rainfall duration conditions. The relationship of cumulative rain-
fall and duration threshold could reflect debris flow initiation
characteristics under high cumulative rainfall in long rainfall du-
ration. Defining rainfall events by estimating rainfall parameters
with different methodologies could reveal variations among inter-
mittent rainfall events for the benefit of issuing debris flow warn-
ings. The exponent of landslide frequency-area distribution
induced by Typhoon Morakot is lower than that induced by the
Chi-Chi earthquake. The lower exponent of landslide frequency-
area distribution can be attributed to the transportation and de-
position areas of debris flow that are included in the landslide
area. Climate change induced high rainfall intensity and long
duration of precipitation, for example, Typhoon Morakot brought
increased frequency of debris flow and created difficulty in issuing
warnings from rainfall monitoring.
Keywords Landslide . Power-law distribution . Climate
change . GIS
Introduction
Climate change-induced abnormal rainfall has caused an increase
in the frequency and risk of floods, landslides, and debris flows in
the world (Evans and Clague 1994; Joshi and Kumar 2006; Clarke
and Rendell 2007; Petley et al. 2007; Hilker et al. 2009; Djalante
and Thomalla 2012; Mechler et al. 2014). Climate change and the
potential of extreme rainfall conditions may be a contributing
factor for landslides (Nadim et al. 2006). The extreme rainfall
events in the USA from 1999 to 2003, with daily precipitation total
at one or more stations exceeding the 50-year recurrence amount
are generally consistent with previous studies of heavy precipita-
tion and flash floods (Schumacher and Johnson 2006). Extreme
rainfall here is defined as rainfall intensity or cumulative rainfall
exceeding the 100-year recurrent at a specified station for the
initiation characteristics of debris flow in Taiwan. Extreme
rainfall events have increased flood risks in India
(Guhathakurta et al. 2011) and the frequency of flash floods,
cloud bursts, and/or glacial lake outburst floods in the Hima-
layan region (Joshi and Kumar 2006). Taiwan is expected to
experience an increase in average annual maximum rainfall
from 322 mm in 19602008 to 371 mm in 20102099, due to
climate change (Chiang and Chang 2011).
Numerous debris flows are initiated during the peak rainfall
intensity of a rainfall event in Taiwan (Chen et al. 2005; Huang
2011). Shallow landslide-induced debris flows on vegetated slopes
usually occur only after significant antecedent rainfall and mod-
erate to long duration rainfall in the USA (Baum and Godt 2010).
High intensity and short duration rains without antecedent rain-
fall trigger mostly debris flows in Eastern Pyrenees, Spain
(Corominas and Moya 1999). Separating antecedent rainfall and
a rainfall event are required in order to estimate the critical rainfall
factors that initiate debris flows.
Debris flow warning by threshold rainfall could divide into
various categorizes. The relationships between (mean) rainfall
intensity (Im or I) and duration (D) (Caine 1980; Cannon and Ellen
1985; Wieczorek 1987; Keefer et al. 1987; Marchi et al. 2002; Chen
et al. 2005; Crosta and Frattini 2008; Guzzetti et al. 2008; Staley
et al. 2013), and intensity and (effective) cumulative rainfall (Re or
R) (Chen et al. 2005; Chang and Chao 2006) are common climate
factors used for constructing a predictive relationship for issuing
debris flow warnings. Other relationships included but not limited
to the use of effective cumulative rainfall and duration (Chen et al.
2005), rainfall intensity and antecedent rainfall (Corominas and
Moya 1999), and the daily precipitation and averaged rainfall of
raining days per year (Wilson 1997). A rainfall triggering
index (RTI) defined as the product of the rainfall intensity
multiplied by the effective accumulated rainfall was used for
real-time debris flow monitoring (Jan and Lee 2004). Radar
product QPESUMS was tested to improve the resolution of
spatial rainfall distribution for debris flow warning (Chen
et al. 2007a). The rainfall intensity per 10 min is used in
recent years for real-time monitoring in addition to the use
of the hourly intensity (Floris et al. 2010).
Landslide and debris flows are growing as a threat to human
beings due to global climate change. How could a watershed
initiate a debris flow under abnormal extreme rainfall conditions?
It is an ongoing effort to predict landslide magnitude and frequen-
cy distribution under extreme rainfall conditions (Chen 2013).
Landslide events show a power-law distribution (Stark and Hovius
2001; Guzzetti et al. 2002; Chen et al. 2007b; Van Den Eeckhaut
et al. 2007; Chen 2009; 2012). The linear trend of the landslide
frequency-area curve for larger landslides (the nearly straight line
portion) in power law (log-log plot) can be found through least
squares regression as
log N A c logA S 1
where N(A) is the number of landslides of area A, c is the slope of
the line defining the relationship, and S is the slope intercept. The
straight line with a slope of c has identical slopes in a double-
logarithmic plot with an exponent c.
The distribution curve of the power-law distribution in
frequency-area distribution exhibits an exponential gradient of
Landslides 13 & (2016) 153
 Technical Note
1.0 in the sandpile model test (Bak et al. 1987). The sandpile model 3371
3500
is the first model to exhibit self-organized criticality (SOC) behav- yrs 1951-2010
1
2870.8
3000
ior. A system organizes itself toward a critical state without any yr 2009
2
tuning (Bak et al. 1987, 1988). The SOC system displays 2500 2261

Ra inf a ll (mm)
spatial and temporal power laws and scale invariance, with-
2000
out controlling the external parameters. The evolution of the
system is inherently moved toward the critical point 1500
(Turcotte 1999; Markovi and Gros 2014). Will the landslide 1000 693.5
frequency-area statistics at Laonong during Typhoon 466
500 193 180
Morakot show a distribution similar to the sandpile model 0 4 50 129 47 26 8 7
under extreme rainfall conditions? There is limited informa- 0

Jan.

Nov.
March
Feb.

Dec.
Aug.
May
April

Total
Sep.
July

tion on landslide characteristics under extreme rainfall con-
ditions. The purpose of this study is to identify the features
of rainfall-induced landslides and debris flows by their
intensity-duration relationship and their frequency-area dis-
tribution under extreme rainfall conditions.
Study area
The Laonong River Basin is located in Kaohsiung City in southern
Taiwan. The Laonong River is 137 km in length with a basin area of
1373 km2. It is the second largest river watershed in Taiwan. The
area had been documented by numerous studies (Weng et al. 2011;
Wu et al. 2011). The basins elevations span from 27 m in the plains
to 3941 m in the mountain areas and is divided into 76 sub-
basins (Fig. 1).
The annual average rainfall in the study area was
3360 mm per year from 1982 to 2010 (WRA 2010). The
maximum rainfall intensity was 106 mm/h, and the cumula-
tive rainfall was up to 2244 mm from August 712, 2009 at
Oct.
June
Month
Fig. 2 Monthly average rainfall distribution at Jiasian rain gauge station in
southern Taiwan from year 1951 to 2010 (source WRA 2010)
the Jiasian rain gauge station (Fig. 2). The basin received
2300 mm of extreme rainfall during the Typhoon Morakot
event in Taiwan.
Debris flow hazards were initiated in the study area when
Typhoon Kalmegi landed on Taiwan in 2008. Kalmegi caused
the deaths of seven people and two missing in the study area
(NFA 2008). Debris flows initiated by Typhoon Morakot in
2009 caused the deaths of 43 people in the study area (source
SWCB 2008, 2009).

Alishan rain
gauge station
Jiasian rain
gauge station

Chiayi County Cumulative rainfall (mm): Elevation (m)

200-400
Tainan City 401-800
801-1400
1401-2000
Kaohsiung City 2001-2400
Laonong River Basin
Taiwan
Pingdong County
river

Taiwan Laonong River Basin

Fig. 1 Site location of the Laonong River Basin and Typhoon Morakot that brought rainfall distribution in Taiwan on from 710 August 2009

154 Landslides 13 & (2016)

 Methodology
The study applied Spot 5 image landslide interpretation and
augmented field investigation in the study area. The resolu-
tion of the Spot 5 land merging images is 2.5 m. Landslides
are digitalized by hand in GIS using the differences between
Spot images before and after typhoon events. The interpreta-
Gravel, sand, clay
tions of naked areas from remote sensing images were sepa-
rated into debris flow source areas and landslides and debris
Sandstone, shale Volcanic rocks flows (including source, transportation, and deposition areas)
of two categories (Fig. 4).
The landslide interpretation from images is statistics in
Phyllite, slate, power-law form for the frequency-area distribution. The
with sandstone interbeds cumulative form of landslide frequency-area distribution
could be restrictive and misleading (Pelletier et al. 1997)
and noncumulative form in logarithmic binning which
Conglomerate increases linearly with the event size (Hergarten 2003).
Argillite, slate, phyllite The noncumulative form of landslide frequency-area dis-
tribution is calculated based on the derivative of the num-
ber of landslides (N c ) with an area greater than or equal
Alluvium to the value A L and is defined as the frequency density
Rain gauge station function f(A) (Turcotte 1999; Crosta et al. 2003; Malamud
Landslide et al. 2004):

Fig. 3 Rain gauge stations, landslides, and debris flows induced by Typhoon
Morakot and geologic conditions in the Laonong River Basin (source CGS 2000)
Site characteristics
The sediment material in the source area is a crucial component to
lower the rainfall threshold of debris flow (Shieh et al. 2009). The
main geologic formation is argillite, slate, and phyllite in the basin
(Fig. 3; source CGS 2000). Landslides were located in argillite, slate,
and phyllite in the middle area. Weak geological conditions, steep
slopes, high cumulative rainfall, and earthquakes cause numerous
debris flow and landslide disasters in the basin (Jan et al. 2011).
f AL dN c =dAL 2
Landslide frequency and area statistics are used in power-law
distribution (log-log plot) and fit the least-squares regression for
larger landslides according to
dN c =dAL bA
L 3
where b is a constant and is the exponent of the fit line, and its
value is equal to c+1.0 (Malamud et al. 2004).
The definition of critical rainfall is the total amount of
rainfall from the time of a distinct increase in rainfall

(a) N
N

W E
(b) W E

S
S
(transportation and
deposition area)

(source area)
River
Identified Landslide (source area)
debris flows Debris flow
0 0.3 Kilometers

0 6 Kilometers

Fig. 4 Interpretations of naked areas from remote sensing images: a identified debris flows and b separation of debris flow source, transportation, and deposition areas

Landslides 13 & (2016) 155

 156
Table 1 Time of debris flows initiated and their corresponding rainfall characteristics
No Typhoon event X-coor. (TWD97) Y-coor. (TWD97) Ic (mm/h) Im (mm/h) Re (mm) D (h) Lag time (h)a
1 Kalmegi 217724 2556210 144.5 64.9 454 7 0
2 213187 2542470 17.0 35.1 421 12 6
3 213302 2538429 47.5 46.4 371 8 3

Landslides 13 & (2016)

4 216500 2555671 33.0 64.5 710 11 4
5 217971 2556390 144.5 64.9 454 7 0
6 217206 2554808 109.5 62.0 310 5 2
7 209613 2555185 139.0 59.9 539 9 0
8 203012 2550536 130.0 47.9 383 8 1
9 210978 2552859 130.0 42.6 383 9 0
10 210459 2551517 139.0 59.8 538 9 0
11 202139 2549040 66.0 46.0 414 9 3
Technical Note

12 218194 2571182 92.5 49.2 443 9 0

13 216522 2556264 34.5 67.7 677 10 3
14 217743 2556342 33.0 59.2 710 12 5
15 Morakot 217520 2555428 18.5 11.4 309 27 21
16 217292 2551284 63.5 28.3 1331 47 3
17 214000 2542255 56.0 27.7 1272 46 2
18 213302 2538429 15.5 13.2 290 22 22
19 215812 2547379 35.5 28.3 1586 56 11
20 214010 2533640 32.5 15.3 534.5 35 9
21 217200 2554780 71.0 24.3 1046 43 1
22 217010 2550443 63.5 28.3 1331 47 3
23 218120 2552950 56.0 27.7 1272 46 2
24 216088 2545472 27.0 28.2 1525 54 10
25 215792 2545885 95.5 22.5 921 41 3
26 221973 2555585 35.5 28.3 1586 56 12
27 215816 2547873 35.5 27.8 1586 57 12
28 214968 2544224 71.0 24.3 1046 43 1
29 213813 2546466 71.0 24.3 1046 43 1
30 218240 2555745 86.5 17.5 733 42 7
 Table 1 (continued)
No Typhoon event X-coor. (TWD97) Y-coor. (TWD97) Ic (mm/h) Im (mm/h) Re (mm) D (h) Lag time (h)a
31 231588 2572274 74.0 24.0 647 27 8
32 224914 2551728 88.5 31.8 1049 33 2
33 224124 2551583 88.5 31.8 1049 33 2
34 221506 2550485 88.5 31.8 1049 33 2
35 220139 2548369 86.5 26.2 733 28 7
36 232221 2573545 74.0 24.0 647 27 8
37 213479 2562097 43.0 27.6 1545 56 13
38 208247 2551287 42.5 26.7 1257 47 3
39 208755 2552247 54.5 28.2 1411 50 5
40 210455 2554959 42.5 26.7 1257 47 3
41 211780 2552632 22.5 27.9 1507 54 10
42 203840 2550328 42.5 27.1 1329 49 5
43 211503 2558844 42.5 19.0 626 33 11
44 209757 2556474 42.5 28.3 1328 47 3
45 210209 2552665 67.5 29.1 1396 48 4
46 211501 2558913 42.5 26.7 1257 47 3
47 206646 2545692 38.0 18.7 1031 55 11
48 202840 2548950 15.5 17.9 752 42 2
49 219053 2541064 89.0 16.6 713 43 6
50 218648 2568618 9.5 20.7 1410 68 25
51 214750 2568189 30.5 23.4 1403 60 11
52 217161 2534363 25.0 18.4 994 54 10
before, + after
a
The lag time is defined as the time difference between the time of peak rainfall intensity occurred and the outburst of debris flow

Landslides 13 & (2016)

157
 Technical Note
Table 2 Comparisons of rainfall from Typhoon Morakot and world recorded extreme rainfall (source: CWB 2009; NCDR 2010; NOAA 2012)
Duration (h) Morakot rainfall (mm) World record
Rainfall (mm) Place Date
1 136 401 Mongol 3 July 1975
6 549 840 Mongol 1 August 1977
a d
24 1624 1870 La Runion, France 78 January 1966
b
48 2361 2467 La Runion, France 89 January 1958
72 2748c 3930 La Runion, France 2425 February 2007
a
14:00 8~14:00 9 August at the Alishan station
b
18:00 7~18:00 9 August at the Alishan station
c
00:00 7~00:00 10 August at the Alishan station
d
Wikipedia 2010

intensity to the time of the triggering of the landslide (Govi
and Sorzana 1980). A rainfall event starts as intensity over
4 mm/h and ends as intensity lower than 4 mm/h lasting for
6 h. The effective cumulative rainfall, Re (mm), is defined as
the sum of current daily rainfall (d0) and antecedent rainfall
(Jan and Lee 2004; Jan et al. 2006):
Re d0 1 d 1 2 d 2 14 d 14 ; t 0:5t=T
where t is empirical attenuation coefficient, dt (mm) is daily
rainfall of t day, and T is half-life (used 1 day herein).
The critical rainfall intensity (Ic) is defined as the rainfall
intensity on the time of debris flow initiated. The mean rainfall
intensity, Im (mm/h), is defined as
I m Re =D 5
where D is rainfall duration via (average of) the nearest rain gauge
4
station during the initiation of debris flow.
The 61 debris flow cases in middle Taiwan listed by Chen
et al. (2005), 14 Kalmegi-induced cases, and 38 Morakot-
induced cases in southern Taiwan documented by SWCB

Fig. 5 Effective cumulative rainfall and duration (Ac-D) for initiation of debris flows

158 Landslides 13 & (2016)

 160
(http://www.swcb.gov.tw) were used for the analysis
(Table 1).

Rainfall characteristics during Typhoon Morakot

Typhoon Morakot hit Taiwan in August 2009, causing the most 120
serious floods southern Taiwan had ever experienced. The ty-
phoon brought a maximum rainfall of 1624 mm in 24 h and
2361 mm in 48 h, with a total amount of 2884 mm at the Alishan
rain gauge station (source CWB 2009). The rainfall was distributed
in Chiayi and Pingdong counties and Tainan and Kaohsiung cities, 80
in the mountainous area of southern Taiwan (Fig. 1).
The typhoon-induced disasters were attributed to its slow mov-
ing velocity caused by the long rainfall duration and high rainfall
intensity (Jou et al. 2010). The monthly rainfall was reduced prior
to August 2009 while it abruptly increased from 693.5 mm (aver- 40
aged between 1951 and 2010) to 2261 mm in the Jiasian station
attributing the increase to Typhoon Morakot bringing rainfall in
August in southern Taiwan (Fig. 2). The extreme rainfall event
(exceeding a 200-year recurrence amount at many rain gauge
stations) caused severe floods and triggered landslides and debris 0
0 400 800 1200 1600
flows throughout southern Taiwan (Juang 2011). This extreme
rainfall approaches the highest global rainfall in 1 and 2 days,
measuring 1870 and 2467 mm, respectively (Table 2). The Fig. 6 Comparisons of effective cumulative rainfall and critical rainfall intensity to
typhoon, bringing rainfall intensity over a flood frequency of initiate debris flows during typhoons Kalmegi (2008) and Morakot (2009)
2000 years, was recorded in many rain gauge stations (EPA
2009). Typhoon Morakot rainfall occupied nine of the top ten
cumulative rainfalls in the history of Taiwan (WRA 2010). The Selection of rainfall parameters for debris flow threshold
typhoon also brought long durations of rainfall for 100 h Debris flows could be initiated by high cumulative rainfall and
from the 6 to 11 August 2009. The maximum rainfall intensity long duration or solely by high rainfall intensity. Typhoon
was up to 130 mm/h. Morakot brought extreme rainfall showing high cumulative rain-
fall and long duration. The critical rainfall intensity (Ic) and
Results and discussion duration relationship (duration time from the start of rain to the
time of debris flow initiated) show that Morakot induced debris
Debris flow initiated characteristics
It was found that 94 % (30 out of 32 events) of the investigated 160
debris flows occurred within 0.5 h before and 1.0 h after the peak of
rainfall intensity by the study of Huang (2011). This suggests that
debris flows occurred around the peak of rainfall intensity, espe-
cially for rolling rainfall (Chen et al. 2005). Total, there were 6800
landslides and 166 debris flows identified from the Spot 5 images 120
after Typhoon Morakot (Chen 2013). The largest historical land-
slides occurred in the study area after Typhoon Morakot. Debris
flows initiated by Typhoon Morakot were attributed to a longer
duration than the Kalmegi-induced (Fig. 4). The minimum rainfall
duration to initiate debris flow was 5 h among historical cases. The 80
effective cumulative rainfall to initiation of debris flows during
Morakot was also higher than Kalmegi-induced (Fig. 5).
Debris flows initiated by Kalmegi were attributed to higher
rainfall intensity than Morakot (Fig. 6). In general, the debris flows
initiated by Typhoon Morakot were attributed to long rainfall 40

duration and high cumulative rainfall, while Typhoon Kalmegi

had high rainfall intensity in short duration.
Rainfall characteristics affect the rainfall duration time to initi-
ate debris flow (Chen et al. 2005). Debris flow caused by high
0
rainfall intensity has the triggering time access to the time of peak -30 -20 -10 0 10 20 30
rainfall intensity as occurred during Typhoon Kalmegi (Fig. 7).
Debris flow initiated by high cumulative rainfall and long duration
is less concerned with the time of peak rainfall intensity as oc- Fig. 7 Comparisons of the debris flow triggering time to peak rainfall intensity
curred during Typhoon Morakot (Fig. 8). during typhoons Kalmegi (2008) and Morakot (2009)

Landslides 13 & (2016) 159

 Technical Note
14 14
12 (a) Typhoon Kalmegi 12 (b) Typhoon Morakot
10

No . o f Ev ent
No . o f Ev ent
10 10 9
8 6 8
6
6 6
3 4 3
4 4
1 1 1 2 1 2 1 2
2 0 0 0 0 0 0 2 0 0 0 0
0 0
>-5 -5 -4 -3 -2 -1 0 1 2 3 4 >5 >-5 -5 -4 -3 -2 -1 0 1 2 3 4 5 >5
Time Lag (hr) Time Lag (hr)

Fig. 8 Statistics of debris flow initiated time during typhoons a Kalmegi and b Morakot (+ and indicate after and before peak rainfall intensity)

flows with lower critical intensity and had a longer of duration
than Kalmegi (Fig. 9).
The debris flows sourced from Chen et al. (2005) were located
in middle Taiwan with some basins affected by the Chi-Chi earth-
quake in 1999 that were initiated under the lowest critical rainfall
intensity. It also means that earthquake shaking causes the debris
flow to be initiated under lower critical rainfall intensity for the
earthquake-induced landslides upstream before the initiation of
debris flow. The Ic-D threshold line (Ic=38.86D0.343) is identical to
the results of Chen et al. (2005) for the new added data (Kalmegi
and Morakot) located in southern Taiwan. The result shows a
higher threshold than the available results (Caine 1980; Ceriani
et al. 1992; Calcaterra et al. 2000) and comparable to the result of
Guadagno (1991). Typhoon Kalmegi showed the high mean rainfall
intensity under short rainfall duration (Im-D) to initiate debris
flows (Fig. 10). It shows that the Im-D threshold could improve
the uncertainty of the threshold for debris flow, for example, when
affected by earthquake, under low critical rainfall intensity.
Due to climate change, Taiwan is expected to experience increased
rainfall in the flooding seasons and decreased rainfall during dry seasons
in 20202039 (IPCC 2012). Rainfall intensity will increase in summer,
and extreme rainfall events will frequently occur (TCCIP 2011). It shows
that high rainfall intensity or high cumulative rainfall is expected and
prone to initiate landslide and debris flows during the climate change.
Different rainfall parameters reflect different rainfall characteristics. A
prediction model to issue a warning of debris flow should reflect the
threshold characteristics of rainfall, for example, under high cumulative
rainfall and long duration in an intermittent rainfall (e.g., Morakot) or in
high rainfall intensity and short duration in a rolling rainfall (e.g.,
Kalmegi). The relationship of Ac-D (Fig. 5) reflects debris flow initiation
characteristics under high cumulative rainfall and long duration. The
relationship of Ic-D reveals the effects of earthquake-induced low critical
rainfall intensity to initiate debris flows (Fig. 9). The relationship of Im-D
reflects debris flow initiation characteristics under high rainfall intensity
and low duration conditions and eliminates the effects of an earthquake
(Fig. 10). Defining rainfall events by estimating rainfall parameters with

Fig. 9 Critical rainfall intensity and duration (Ic-D) to initiate debris flows

160 Landslides 13 & (2016)

Fig. 10 The mean rainfall intensity and duration (Im-D) to initiate debris flows
different methodologies could reveal variations among intermittent
rainfall events for the benefit of issuing debris flow warnings.
Characteristics of debris flow frequency-area distribution
The frequency-area distribution of landslides and debris flows can
be expressed as (Fig. 11)
ln f 1:53lnA 4:42r 2 94% 6 and that shows a larger exponent (exponent=1.53 in Eq. (6)) in the
Debris flows and land slides in Laonong basin
The interpretations of naked areas from remote sensing images
were separated into two categories for debris flow source areas and
landslides and debris flows including source, transportation, and
deposition areas. The debris flow source areas are overlapped, but
the two categories are drawing their distribution separately. Its
purpose is to show the different area interpretation from images
(including debris flow transport and deposition or excluded)
changing the power-law exponent distribution. Although, it is hard
to differentiate the source and transport area of debris flows. We
identified some debris flow cases, and the result exhibit an in-
creased exponent distribution for instance. If we can well separate
all the dataset, then the exponent of the distribution will become
higher. The least square best fit equation for debris flow source
areas and landslides can be expressed as
ln f 1:39lnA 5:07r 2 89%
Debris flow source areas and land slides in Laonong basin
The frequency-area distribution for the Chi-Chi earthquake-
induced larger landslides can be expressed as
ln f 2:06lnA 5:79r 2 93%ChiChi EQ 8
The distribution of debris flow (ADF) and landslides (AL) areas
(AT=ADF+AL) has a larger naked area under the same frequency (fT=fL)
frequency-area distribution (Fig. 11) compared with Eq. (7) (expo-
nent=1.39). Debris flow transportation and deposition areas increase
the frequency of larger landslide area and cause a lower
exponent in the landslide frequency-area distribution curve
(Fig. 12).
The concept of critical states (barely stable) could be extended to
landslides in a basin and its potential to debris flow to present the
current state of a basin during rainfall (Chen 2012) and to sediment
budgets. The sediment budget is an estimation of volumes (or rates of
change) for sediments entering (source from landslide and soil erosion)
and leaving a basin (Bowen and Inman 1966). A basin is in sediment
balance (equilibrium state) as the residual is equal to zero in the
sediment budget equation (Kraus and Rosati 1999).
If we divide a basin state under landslide into three categories for the
exponent of landslide frequency-area distribution equal to (critical
state), over and smaller than , then landslide in a basin having an
exponent of frequency-area distribution equal to is defined as the
critical state herein. This is similar to the critical exponent for cumulative
7 frequency-area distribution with exponent c=1.0 in the conceptual
sandpile model (Bak et al. 1987, 1988) and =2.0 in noncumulative form
Landslides 13 & (2016) 161
 Technical Note

Fig. 11 Debris flow and landslide frequency-area distribution

(c=1.0) (Guzzetti et al. 2002; Malamud et al. 2004). The exponent of
coseismic landslide frequency-area distribution of the Chi-Chi earth-
quake is 2.06 in noncumulative form and 1.06 for the cumulative form,
which is explained by the critical exponent. The coseismic landslides
during the Chi-Chi earthquake are primarily shallow landslides (Lee
et al. 2008). It shows that more small areas of landslides and a lower
exponent occurred during Typhoon Morakot in the study area than the
Chi-Chi earthquake-induced (ML=7.3) in the middle of Taiwan. Al-
though, the higher landslide ratio (the sum of landslide areas in a basin
divided by the basin area) could have been affected by the earthquakes
Log(Frequency)
in the study area on 4 March 2008 (ML=5.2) and others (Weng et al.
2011).
The exponent value varied with the triggering mechanism and types
of natural disasters (Hergarten 2003; Van Den Eeckhaut et al. 2007).
Landslide frequency-area distribution in an unstable state basin shows
an exponent close to 1.0 in Chushui creek (Chen et al. 2007b), Dachia
(Chen 2009), and in the Lushan basins (Chen 2012) in Taiwan. The
exponent in the study basin is affected by the occurrence of the numer-
ous debris flows. It is sourced from the fact that the naked area
interpretation from aerial image for debris flow includes the source,
transportation, and deposition areas, in which a naked area increased
and its exponent in landslide frequency-area distribution increased. In
the meantime, the sandpile model is conducted in dry conditions with

surficial movement that is inherent more access to the case of the Chi-
Chi earthquake-induced coseismic landslides in triggering mechanism.
AT = AL + ADF
critical state However, the exponent is limited by the basin area for the available
fT = fL
> number and area of landslides that could occur. The laboratory test
mechanism of the sandpile model with critical exponent 1.0 is not
validated for debris flow. The area of debris flows causing the exponent
fT to increase and the value of critical exponent () need further efforts.

fDF Conclusions
<
Under extreme rainfall conditions, debris flows were initiated by high
ADF AL AT Log(Area) cumulative rainfall and long rainfall duration or solely by high rainfall
intensity. The relationship of mean rainfall intensity and duration
Fig. 12 Debris flow transportation and deposition areas increase the frequency of threshold could reflect debris flow initiation characteristics under high
a larger landslide area (AL increasing to AT) and caused a larger exponent in the rainfall intensity and rolling rainfall in short duration conditions. The
landslide frequency-area distribution curve relationship of cumulative rainfall and duration threshold could reflect

162 Landslides 13 & (2016)

debris flow initiation characteristics under high cumulative rainfall and
intermittent rainfall in long duration conditions. Defining rainfall events
by estimating rainfall parameters with different methodologies could
reveal variations among intermittent rainfall events for the benefit of
issuing debris flow warnings. In basins under extreme rainfall condi-
tions where debris flows are included, the area-frequency distribution
could not be estimated by the sandpile model with a critical exponent of
1.0 for a larger landslide in power-law distribution. The varied exponent
could be attributed to the compound area of landslides and debris
flow. The types of events, such as landslide, and debris flow,
will affect the trend of frequency-area distribution of naked
areas.
Acknowledgments
The authors would like to thank the anonymous reviewers for
their useful comments, and SWCB in Taiwan provided valuable
data for the analysis.
References
Bak P, Tang C, Wiesenfeld K (1987) Self-organized criticality: an explanation of 1/f
noise. Phys Rev Lett 59:381384
Bak P, Tang C, Wiesenfeld K (1988) Self organized criticality. Phys Rev A 38:364
374
Baum RL, Godt JW (2010) Early warning of rainfall-induced shallow landslides and debris
flows in the USA. Landslides 7:259272
Bowen AJ, Inman DL (1966) Budget of littoral sand in the vicinity of point Arguello,
California. U.S. Army Coastal Engineering Research Center, Technical Memorandum
No. 19, 56p
Caine N (1980) The rainfall intensity duration control of shallow landslides and debris
flows. Geogr Ann 62:2327
Calcaterra D, Parise M, Palma B, Pelella L (2000) The influence of meteoric events
in triggering shallow landslides in pyroclastic deposits of Campania, Italy. In:
Proceedings 8th International Symposium on Landslides, Cardiff, pp. 209214
Cannon SH, Ellen SD (1985) Rainfall conditions for abundant debris avalanches, San
Francisco Bay region. Calif Geol 38:267272
Ceriani M, Lauzi S, Padovan N (1992) Rainfall and landslides in the Alpine area of
Lombardia region, central Alps, Italy. In: Proceedings of Interpraevent, Bran, Switzer-
land, pp. 920
CGS (2000) Central Geologic Survey, Ministry of Economic Affairs in Taiwan. Geological
map of Taiwan, scale 1:50,000
Chang TC, Chao RJ (2006) Application of back-propagation networks in debris flow
prediction. Eng Geol 85:270280
Chen CY, Chen TC, Yu FC, Yu WH, Tseng CC (2005) Rainfall duration and debris-
flow initiated studies for real-time monitoring. Environ Geol 47:715724
Chen CY, Lin LY, Yu FC, Lee CS, Tseng CC, Wang AH, Cheung KW (2007a)
Improving debris flow monitoring in Taiwan by using high-resolution rainfall
products from QPESUMS. Nat Hazards 40:447461
Chen CY, Yu FC, Lin SC, Cheung KW (2007b) Discussion of landslide self-organized
criticality and the initiation of debris flow. Earth Surf Proc Land 32:197209
Chen CY (2009) Sedimentary impacts from landslides in the Tachia River Basin,
Taiwan. Geomorphology 105:355365
Chen CY (2012) Landslide and basin self-organized criticality in the Lushan hot spring
area. J Mt Sci 9:463471
Chen CY (2013) Landslide and debris flow characteristics during extreme rainfall
in Taiwan. World Acad Sci Eng Technol 78:775778
Chiang SH, Chang KT (2011) The potential impact of climate change on typhoon-
triggered landslides in Taiwan, 20102099. Geomorphology 133:143151
Clarke M, Rendell H (2007) Climate, extreme events and land degradation. Climate and
Land Degradation, Environmental Science and Engineering, pp. 137152
Corominas J, Moya J (1999) Reconstructing recent landslide activity in relation to
rainfall in the Llobregat River basin, Eastern Pyrenees, Spain. Geomorphology
30:7993
Crosta GB, Negro PD, Frattini P (2003) Soil slips and debris flows on terraced
slope. Nat Hazards Earth Syst Sci 3:3142
Crosta GB, Frattini P (2008) Rainfall-induced landslides and debris flows. Hydrol Process
22:473477
CWB (2009) Central Weather Bureau in Taiwan, http://www.cwb.gov.tw
Djalante R, Thomalla F (2012) Disaster risk reduction and climate change adaptation in
Indonesia: institutional challenges and opportunities for integration. Int J Disaster
Resilience Built Environ 3:166180
EPA (2009) Environmental Protection Administration, Executive Yuan (R.O.C., Taiwan).
Impacts due from climate change, extreme events and disasters are the biggest threat
to Taiwan, http://unfccc.epa.gov.tw/unfccc/english/_uploads
Evans SG, Clague JJ (1994) Recent climatic change and catastrophic geomorphic
processes in mountain environments. Geomorphology 10:107128
Floris M, DAlpaos A, Squarzoni C, Genevois R, Marani M (2010) Recent changes in rainfall
characteristics and their influence on thresholds for debris flow triggering in the
Dolomitic area of Cortina dAmpezzo, north-eastern Italian Alps. Nat Hazards Earth
Syst Sci 10:571580
Govi M, Sorzana PF (1980) Landslide susceptibility as function of critical rainfall amount
in Piedmont basin (North-Western Italy). Stud Geomorphol Carpatho-Balcanica
14:4360
Guadagno FM (1991) Debris flows in the Campanian volcaniclastic soils (Southern Italy).
In: Chandler RJ (ed) Proceedings of International Conference on Slope Stability
Engineering. Isle of Wight, UK, pp 109114
Guhathakurta P, Sreejith OP, Menon PA (2011) Impact of climate change on extreme
rainfall events and flood risk in India. J Earth Syst Sci 120:359373
Guzzetti F, Malamud BD, Turcotte DL, Reichenbach P (2002) Power-law correlations of
landslide areas in central Italy. Earth Planet Sci Lett 195:16918
Guzzetti F, Peruccacci S, Rossi M, Stark CP (2008) The rainfall intensityduration control
of shallow landslides and debris flows: an update. Landslides 5:317
Hergarten S (2003) Landslides, sandpiles, and self-organized criticality. Nat Hazards Earth
Syst Sci 3:505514
Hilker N, Badoux A, Hegg C (2009) The Swiss flood and landslide damage database
19722007. Nat Hazards Earth Syst Sci 9:913925
Huang CC (2011) Gaussian-distribution-based hyetographs and their relationships with
debris flow initiation. J Hydrol 411:251265
IPCC (2012) Intergovernmental Panel on Climate Change (IPCC). Managing the Risks of
Extreme Events and Disasters to Advance Climate Change Adaptation, IPCC, AR5, WG1
Jan CD, Lee MH (2004) A debris-flow rainfall-based warning mode. J Chin Soil Water
Conserv 35:275285 (in Chinese with English abstract)
Jan CD, Lee MH, Kuo FH (2006) A rainfall-based debris-flow spatial and temporal
warning model. Sino-Geotechnics 110:5564 (in Chinese with English abstract)
Jan CD, Hsu YC, Wang JS, Huang WS (2011) Debris flows and landslides caused by
typhoon Morakot in Taiwan. 5th International Conference on Debris-Flow Hazards
Mitigation: Mechanics, Prediction and Assessment, Padua, Italy, pp. 675683
Joshi V, Kumar K (2006) Extreme rainfall events and associated natural hazards in
Alaknanda valley, Indian Himalayan region. J Mt Sci 3:228236
Jou BJD, Yu YC, Feng L, Chen YM, Lee CS, Cheng MD (2010) Synoptic environment and
rainfall characteristics of Typhoon Morakot. Atmos Sci 38:2138 (in Chinese)
Juang CH (2011) Reconnaissance of extreme natural disasters of Morakot Typhoon,
Taiwan. Eng Geol 123:12
Keefer DK, Wilson RC, Mark RK, Brabb EE, Brown WM III, Ellen SD, Harp EL, Wieczorek GF,
Alger CS, Zatkin RS (1987) Real-time landslides warming during heavy rainfall. Science
238:921925
Kraus NC, Rosati JD (1999) Estimating uncertainty in coastal inlet sediment budgets.
Proceedings 12th Annual National Conference on Beach Preservation Technology
(Tallahassee, FL, Florida Shore & Beach Preservation Association), pp. 287302
Lee CT, Huang CC, Lee JF, Pan KL, Lin ML, Dong JJ (2008) Statistical approach to
earthquake induced landslide susceptibility. Eng Geol 100:4358
Malamud BD, Turcotte DL, Guzzetti F, Reichenbach P (2004) Landslide inventories and
their statistical properties. Earth Surf Proc Land 29:687711
Marchi L, Arattano M, Deganutti AM (2002) Ten years of debris flow monitoring in the
Moscardo Torrent (Italian Alps). Geomorphology 46:117
Markovi D, Gros C (2014) Power laws and self-organized criticality in theory and nature.
Phys Rep 536:4174
Mechler R, Bouwer LM, Linnerooth-Bayer J, Hochrainer-Stigler S, Aerts JCJH, Surminski S,
Williges K (2014) Managing unnatural disaster risk from climate extremes. Nat Clim
Chang 4:235237
Nadim F, Kjekstad O, Peduzzi P, Herold C, Jaedicke C (2006) Global landslide and
avalanche hotspots. Landslides 3:159173
NCDR (2010) National Science and Technology Center for Disaster Reduction, Disaster
Survey and Analysis of Morakot Typhoon, http://satis.ncdr.nat.gov.tw/morakot/ (in
Chinese)
Landslides 13 & (2016) 163
 Technical Note
NFA (2008) National Fire Agency, Ministry of the Interior. Typhoon Kalmegi emergency
response report, http://www.nfa.gov.tw/
NOAA (2012) National Oceanic and Atmospheric Administration. Available
http://www.noaa.gov/, accessed 2 March 2012
Pelletier JD, Malamud BD, Blodgett TA, Turcotte DL (1997) Scale invariance of soil
moisture variability and its implications for the frequency-size distribution of land-
slides. Eng Geol 48:254268
Petley DN, Hearn GJ, Hart A, Rosser NJ, Dunning SA, Oven K, Mitchell WA (2007) Trends
in landslide occurrence in Nepal. Nat Hazards 43:2344
Schumacher RS, Johnson RH (2006) Characteristics of U.S. extreme rain events during
19992003. Weather Forecast 21:6985
Shieh CL, Chen YS, Tsai YJ, Wu JH (2009) Variability in rainfall threshold for debris flow
after the Chi-Chi earthquake in central Taiwan, China. Int J Sediment Res 24:177188
Staley DM, Kean JW, Cannon SH, Schmidt KM, Laber JL (2013) Objective definition of
rainfall intensity-duration thresholds for the initiation of post-fire debris flows in
southern California. Landslides 10:547562
Stark CP, Hovius N (2001) The characterization of landslide size distributions. Geophys
Res Lett 28:10911094
SWCB (2008; 2009) Soil and Water Conservation Bureau, Debris flow annual report.
http://www.swcb.gov.tw/english/. Accessed 6 May 2013 (in Chinese)
Turcotte DL (1999) Self-organized criticality. Rep Prog Phys 62:13771429
TCCIP (2011) Taiwan Climate Change Projection and Information Platform.
http://satis.ncdr.nat.gov.tw/ccsr/doc/00_Summary.pdf (in Chinese)
164 Landslides 13 & (2016)
Van Den Eeckhaut M, Poesena J, Goversa G, Verstraetena G, Demoulin A (2007)
Characteristics of the size distribution of recent and historical landslides in a
populated hilly region. Earth Surf Proc Land 256:588603
Weng MC, Wu MH, Ning SK, Jou YW (2011) Evaluating triggering and causative factors of
landslides in Lawnon River Basin, Taiwan. Eng Geol 123:7282
Wieczorek GF (1987) Effect of rainfall intensity and duration on debris flows in central
Santa Cruz Mountains. Rev Eng Geol 7:93104
Wilson RC (1997) Normalizing rainfall/debris-flow thresholds along U. S. pacific
coast for long-term variation in precipitation climate. Proceedings of the First
International Conference on Debris-Flow Hazards Mitigation, San Francisco,
pp. 3243
Wikipedia (2010) http://www.wikipedia.org/. Accessed 12 December 2012
WRA (2010) Water Resources Agency, Hydrological Year Book of Taiwan Republic of
China 2010, Ministry of Economic Affairs. http://gweb.wra.gov.tw/wrhygis/
Wu CH, Chen SC, Chou HT (2011) Geomorphologic characteristics of catastrophic
landslides during Typhoon Morakot in the Kaoping Watershed, Taiwan. Eng Geol
123:1321
C.-Y. Chen ())
Department of Civil and Water Resources Engineering, National Chiayi University,
Room A05B-401, No. 300 Syuefu Rd., Chiayi City, 60004, Taiwan, Republic of China
e-mail: chienyuc@mail.ncyu.edu.tw
 Technical Note

Landslides (2016) 13:165172
DOI 10.1007/s10346-015-0630-1
Received: 5 May 2015
Accepted: 10 August 2015
Published online: 26 August 2015
Springer-Verlag Berlin Heidelberg 2015
Massimo Melillo I Maria Teresa Brunetti I Silvia Peruccacci I Stefano Luigi Gariano I
Fausto Guzzetti
Rainfall thresholds for the possible landslide occurrence
in Sicily (Southern Italy) based on the automatic
reconstruction of rainfall events

Abstract Review of the literature on the reconstruction of the
rainfall responsible for slope failures reveals that criteria for the
identification of rainfall events are lacking or somewhat subjective.
To overcome this problem, we developed an algorithm for the
objective and reproducible reconstruction of rainfall events and
of rainfall conditions responsible for landslides. The algorithm
consists of three distinct modules for (i) the reconstruction of
distinct rainfall events, in terms of duration (D, in h) and cumu-
lated event rainfall (E, in mm), (ii) the identification of multiple
ED rainfall conditions responsible for the documented landslides,
and (iii) the definition of critical rainfall thresholds for possible
landslide occurrences. The algorithm uses pre-defined parameters
to account for different seasonal and climatic settings. We ap-
plied the algorithm in Sicily, southern Italy, using rainfall mea-
surements obtained from a network of 169 rain gauges, and
information on 229 rainfall-induced landslides occurred between
July 2002 and December 2012. The algorithm identified 29,270
rainfall events and reconstructed 472 ED rainfall conditions as
possible triggers of the observed landslides. The algorithm
exploited the multiple rainfall conditions to define objective
and reproducible empirical rainfall thresholds for the possible
initiation of landslide in Sicily. The calculated thresholds may be
implemented in an operational early warning system for shallow
landslide forecasting.
Keywords Algorithm . Landslide . Rainfall . Rainfall
event . Thresholds
Introduction
Rainfall is the primary trigger of landslides in Italy. In large areas,
the prediction of rainfall-induced landslides relies on the defini-
tion of empirical rainfall thresholds. Generally, rainfall thresholds
are the lower bounds to empirical distributions of rainfall condi-
tions that have resulted in landslides, including rainfall intensity, I,
vs. rainfall duration, D, (e.g., Caine 1980; Aleotti 2004; Guzzetti
et al. 2007, 2008; Cannon et al. 2008; Martelloni et al. 2012; Staley
et al. 2012) and cumulated event rainfall, E, vs. rainfall duration, D,
(e.g., Innes 1983; Cannon and Ellen 1985; Wieczorek 1987; Crosta
1998; Kanji et al. 2003; Vennari et al. 2014). In other cases, rainfall
thresholds are defined in order to separate the rainfall conditions
that have and have not resulted in landslides (e.g., Jibson 1989;
Crozier 1997; Corominas and Moya 1999; Marchi et al. 2002; Berti
et al. 2012). Most of the empirical rainfall thresholds available in
the literature are defined using nonobjective and poorly reproduc-
ible methods (Guzzetti et al. 2007, 2008). Moreover, rainfall
thresholds are affected by uncertainties that limit their use in
landslide warning systems (Aleotti 2004; Guzzetti et al. 2008;
Bach-Kirschbaum et al. 2012; Gariano et al. 2015). Sources of
uncertainty include (i) the availability and quality of rainfall and
landslide information, (ii) the methods used to reconstruct the
rainfall conditions responsible for the failures, and (iii) the
methods used to determine the thresholds. A specific source of
uncertainty lays in the characterization of the rainfall event re-
sponsible for the landslides (Guzzetti et al. 2007). Despite their
importance, criteria for the objective and reproducible
reconstruction of rainfall conditions that triggered landslides are
still lacking, poorly formalized, or ambiguous. In order to
overcome this problem, Melillo et al. (2015) proposed an algorithm
to define automatically rainfall events from standard rainfall re-
cords, with an objective and reproducible procedure. The algo-
rithm was implemented using the R open-source software for
advanced statistical computing and graphics (http://www.r-
project.org, release 2.15.2).
In this work, we use a new release of the algorithm that allows
calculating reproducible rainfall thresholds from multiple ED rain-
fall conditions that have resulted in landslides. We tested the
algorithm in Sicily, southern Italy.
Rainfall and landslide data
Rainfall data consist of hourly rainfall measurements collected
in the period between July 2002 and December 2012 by a
network of 169 rain gauges managed by the Sistema
Informativo Agrometeorologico Siciliano (Agro-meteorological
Information Service of Sicily) (Fig. 1). For each rain gauge,
data are stored in a worksheet, where the header contains
information on rain gauge name, code, geographical coordi-
nates, and elevation. Each row in the worksheet reports the
date (day, month, and year), the time (h), and the hourly
rainfall (in mm).
We used the catalogue of rainfall events with landslides
described in Gariano et al. (2015). The catalogue lists 229 rainfall
events that have triggered 265 shallow landslides in Sicily (red
dots in Fig. 2) in the period from July 2002 to December 2012.
In particular, 204 rainfall events induced a single landslide, 18
rainfall events induced two landslides, 5 rainfall events induced
three landslides, and 2 rainfall events induced five landslides, for
a total of 265 landslides. In the subsequent analysis, we use only
the first triggered landslide of each rainfall event (a total of 229
landslides).
Each record in the catalogue includes (i) event identification
(code number, reference source), (ii) landslide localization (place,
geographic accuracy), (iii) landslide type (if available from the
source of information), and (iv) temporal information of the
failure (day, month, year, time, date, temporal accuracy). We
defined three classes of geographic accuracy and three classes of
temporal accuracy, depending on the type and quality of the
information (Gariano et al. 2015). As an example, the first class
of geographic accuracy was attributed to landslides located with an

Landslides 13 & (2016) 165

 Technical Note

Fig. 1 Map showing the geographical location of the 169 rain gauges (black triangles) in Sicily

accuracy of 1 km2 or less, while the first class of temporal accuracy
includes landslides for which the exact time of occurrence is
known. Figure 3 shows the number and percentage of each type
of landslides in the catalogue according to Cruden and Varnes
(1996). Classification is based primarily on the landslide descrip-
tion given in the information sources. Since many reports failed to
specify the landslide type, the majority of the failures was classified
as generic shallow landslide (SL).
Method
We applied the automatic procedure proposed by Melillo
et al. (2015) to reconstruct rainfall events, starting from hourly

Fig. 2 Map showing the geographical location of 265 rainfall-induced landslides (red dots) occurred in the period July 2002December 2012, in Sicily

166 Landslides 13 & (2016)

Fig. 3 Number and percentage of landslides per each type: RF rock fall, EF earth flow, DF debris flow, MF mud flow, SL generic shallow landslide
rainfall measurements. The algorithm exploits a continuous record
of rainfall measurements to determine the duration D (in h) and the
cumulated rainfall E (in mm) of distinct rainfall events. To single out
rainfall events from rainfall series, the algorithm requires a set of
input parameters varying with the season and the climate. In partic-
ular, a minimum dry period (i.e., a period without rainfall or with a
negligible amount of rainfall) between two consecutive rainfall pe-
riods is defined. The length of the dry period may vary, depending on
seasonal and climatic conditions. In particular, in Sicily, where cli-
mate is Mediterranean (mild and wet winters, hot and dry summers),
we separated the rainfall events selecting a dry period of 96 h for the
wet season from November to March, and of 48 h for the dry season
from April to October.
For each of the 229 landslides, the algorithm reconstructed a
variable number of equally (multiplicity) probable rainfall condi-
tions. Overall, 472 multiple rainfall conditions were associated
with the 229 landslides. In the following, we describe how we
improved the algorithm.
Fig. 4 Map of the spatial distribution of the median values of the annual cumulated rainfall. Yellow triangles show the 85 rain gauges with rainfall series lasting at least
10 years. In some areas (in white), rain gauges have shorter rainfall series
New release of the algorithm
In the new release, the algorithm allows to select the representative
rain gauge in the neighborhood of each landslide. As stated above,
using information on the date/time of landslide occurrence, the
algorithm reconstructs the duration D (in h) and the cumulated
rainfall E (in mm) for multiple rainfall conditions responsible for
the initiation of the failures. All the possible rainfall conditions
associated with landslides are then used to determine empirical
rainfall thresholds. In the previous release, only one rainfall con-
dition arbitrarily chosen was used to define the thresholds.
A bootstrapping statistical technique (Peruccacci et al. 2012)
and a frequentist method (Brunetti et al. 2010) were included in
the new release of the algorithm. This improvement allows the
definition of thresholds and their associated uncertainties for
different exceedance probabilities. The general form of the thresh-
old curves is a power law:
E   D 1
Landslides 13 & (2016) 167
 Technical Note
Table 1 Summary statistics for the 29,270 reconstructed rainfall events in Sicily; RE number of rainfall events per year (#), D rainfall duration (h), E cumulated event
rainfall (mm), I mean rainfall intensity (mm/h), IP peak hourly rainfall intensity (mm/h)
Variable Min 1% 5% Median Mean 95 % 99 % Max
RE (#) 6 6 20 24 24 29 30 30
D (h) 1 1 2 28 71 295 591 1571
E (mm) 1.1 1.2 1.6 14.4 29.3 107.4 196.8 711.0
I (mm/h) 0.03 0.07 0.12 0.53 1.05 3.60 9.10 150.00
IP (mm/h) 0.20 0.60 0.90 4.40 6.40 19.80 34.20 150.00

where is a scaling constant (the intercept), is the shape
parameter (that defines the slope of the power law curve), and
and represent the uncertainties of and , respectively.
More in detail, the bootstrapping technique generates k series of m
randomly selected events from a data set of n events (which
correspond to all the possible rainfall conditions associated with
each landslide calculated by the algorithm). The m events were
selected with a probability of selection inversely proportional to
their multiplicity. As an example, events with multiplicity equal to
2 have probability of selection equal to 0.5. Analysis of the
synthetic k series allows calculating the mean value of and
parameters and their uncertainty and .
The use of the algorithm to determine the rainfall conditions
responsible for landslides allows to avoid the previous manual
reconstruction and to decrease the uncertainty associated with
the definition of the rainfall events. Moreover, the subjectivity
inherent to the selection of the (D,E) pair is overcome by the use
of all the multiple rainfall conditions able to trigger the individual
landslide.
Results and discussion
Using information of the rain gauges, reconstructed rainfall
events, and a catalogue of the landslides in Sicily region (Southern
Italy), we performed three different types of analysis. The applica-
tion and testing of estimation methods developed in this study are
reported in this section.

Rainfall data analysis

For each rain gauges with rainfall series lasting at least 10 years (85
rain gauges), we calculated the median values of the annual rain-
fall. Using the kriging method (Bivand et al. 2008), we obtained a
map of the median annual rainfall values (Fig. 4). Annual rainfall
values greater than 1000 mm are located in the northeastern area
of the region. In the Southern area of the island, values are lower
than 800 mm. This outcome is in agreement with results obtained
by Alecci and Rossi (2007) and Di Piazza et al. (2011).

Rainfall event analysis

In Sicily, the algorithm reconstructed 29,270 rainfall events. Table 1
reports descriptive statistics for the reconstructed rainfall events,
corresponding to minimum, mean, and maximum values, and to
different quantiles (1, 5, 50, 95, and 99 %).

Table 2 Number (#RE) and percentage of rainfall events in the six categories of
cumulated rainfall proposed by Alpert et al. (2002)
Category E (mm) #RE %
Light (L) 4 6690 22.9
Light-moderate (LM) 416 11326 38.7
Moderate-heavy (MH) 1632 6608 22.6
Heavy (H) 3264 3550 12.1
Heavy-torrential (HT) 64128 958 3.3
Fig. 5 Two-dimensional frequency density distribution of the reconstructed events
Torrential (T) >128 138 0.5
a in the (D,E) plane and b in the (longitude, latitude) plane

168 Landslides 13 & (2016)

 a - Light b - Light-Moderate

Probability of rainfall events Probability of rainfall events

0.0 0.005 0.01 0.05 0.1 0.5 1.0 0.0 0.005 0.01 0.05 0.1 0.5 1.0

c - Moderate-Heavy d - Heavy

Probability of rainfall events Probability of rainfall events

0.0 0.005 0.01 0.05 0.1 0.5 1.0 0.0 0.005 0.01 0.05 0.1 0.5 1.0

e - Heavy-Torrential f - Torrential

Probability of rainfall events Probability of rainfall events

0.0 0.005 0.01 0.05 0.1 0.5 1.0 0.0 0.005 0.01 0.05 0.1 0.5 1.0

Fig. 6 Maps showing the spatial distribution of the reconstructed rainfall events classified into the six categories reported in Table 2

The number of rainfall events per rain gauge per year ranges is lower than 20 %. Maximum values of D, E, I, and IP are much
from 6 to 30, with an average value equal to 24. Considering the 5 greater than the corresponding quantiles at 99 %. IP values are at
95 % inter-quantile interval, the variation around the median value least a factor of three greater than I values.

Table 3 Summary statistics for the 472 reconstructed ED rainfall conditions in Sicily; D rainfall duration (h), E cumulated event rainfall (mm), I mean rainfall intensity
(mm/h), IP peak hourly rainfall intensity (mm/h)
Variable Min 1% 5% Median Mean 95 % 99 % Max
D (h) 2 2 4 50 89 289 484 605
E (mm) 5.6 11.2 18.0 66.3 84.1 212.4 361.5 659.6
I (mm/h) 0.23 0.27 0.35 1.26 2.31 7.89 14.25 26.73
IP (mm/h) 1.40 3.10 4.20 10.70 15.20 37.60 65.50 86.60

Landslides 13 & (2016) 169

 Technical Note
Fig. 7 Number and percentage of landslides for each multiplicity
Figure 5 shows the two-dimensional frequency density distri-
bution (Kernel Density Estimation, KDE2d function, Bivand and
Gebhardt 2000) of the reconstructed events (a) in the (D,E) loga-
rithmic plane and (b) in geographic coordinates (longitude, lati-
tude). In Fig. 5a, the highest density values are in the area
delimited by 40 h<D<150 h and 15 mm<E<50 mm. In Fig. 5b, the
highest frequency of rainfall events is located in the northeastern
area of the island.
According to the classification proposed by Alpert et al. (2002),
we subdivided the reconstructed rainfall events in six categories of
daily rainfall, from Blight^ to Btorrential^ (Table 2). In order to
assign each event to a specific category, we adopted the following
procedure. For events lasting up to 24 h, we assigned a category on
the basis of the cumulated event rainfall. For events longer than
24 h, we calculated the maximum value of cumulated rainfall in
24 h using a moving window and assigned the corresponding
category. Almost 40 % of the events are classified as Blight-
moderate,^ and only 3.8 % as Bheavy-torrential^ or Btorrential.^
Fig. 8 Multiple ED rainfall conditions responsible for landslides in Sicily,
distinguished in five categories of mean rainfall intensity (dots in grayscale)
and related rainfall thresholds at 1 % (T1,S, green curve) and 5 % (T5,S, red
curve) exceedance probability levels. Shaded areas portray uncertainty
associated with the threshold curves. Data are in log-log coordinates
170 Landslides 13 & (2016)
Using the kriging method (Bivand et al. 2008), we interpo-
lated the cumulated rainfall obtained from the rain gauge net-
work, and we calculated the probability density for each
category (Fig. 6). BLow^ events are mostly localized in the
southeastern part of the region, while Bmoderate-heavy^ events
are concentrated in the northwestern area of the island. Heavy-
torrential and torrential events are mainly localized in the
Eastern side of the region.
Rainfall thresholds
In order to define rainfall thresholds, we used data collected in the
period between July 2002 and December 2012, described in Rain-
fall and landslide data. Table 3 shows statistics of the 472 rainfall
conditions. The rainfall duration D varies from 2 to 605 h, with an
average value of 89 h, and cumulated event rainfall E from 5.6 to
659.6 mm, with an average value of 84.1 mm. Half of the ED
rainfall conditions have a mean rainfall intensity greater than
2.31 mm/h.
Fig. 9 Multiple ED rainfall conditions reconstructed by the algorithm (light blue
dots) and of single rainfall conditions identified by Gariano et al. (2015) (blue
dots). The red and the blue curves are the 5 % threshold defined in this work
(T5,S) and by Gariano et al. (2015) (T *5;S ), respectively. Shaded areas portray
uncertainty associated with the threshold curves. Data are in log-log
coordinates
 Figure 7 shows the number and percentage of landslides for
each multiplicity. As an example, for 103 landslides (45 %), the
algorithm reconstructed only one rainfall condition responsible
for the failure (multiplicity=1), while for 57 landslides (25 %), it
calculated two different rainfall conditions (multiplicity=2).
We applied the bootstrapping technique generating 10,000 se-
ries of 229 randomly selected events from the 472 multiple rainfall
conditions, and we calculated the parameters of the thresholds.
ED frequentist thresholds at 1 % (T1,S) and 5 % (T1,S) exceedance
probabilities, and their associated uncertainties (shaded areas) are
shown in Fig. 8. Threshold equations are reported in the same
figure. The relative uncertainties in the threshold parameters are
/=14.3 % and /=7.5 % for T5,S, and /=15.8 % and / Department for Civil Protection (DPC). MM was supported by
=7.5 % for T1,S. Rainfall conditions that have resulted in slope
instabilities in Sicily cover the range of duration 2D605 h, which
is the range of validity for the thresholds. In Fig. 8, the multiple ED
rainfall conditions are depicted in five out of the six categories of
mean rainfall intensity described in Rainfall data analysis.
BHeavy^ (H) rainfall conditions are the most numerous (205,
43.4 % of the total), followed by heavy-torrential (HT, 113,
23.9 %) and moderate-heavy (MH, 106, 22.5 %). Torrential (T)
rainfall conditions are 24 (5.1 % of the total) and have durations
lasting between 14 and 231 h. On the contrary, light-moderate (LM)
rainfall conditions (24, 5.1 %) are mainly characterized by shorter
duration (2D115 h). Light-moderate (LM) rainfall conditions are
also likely to trigger landslides due to the amount of the cumulated
rainfall. Figure 9 shows the T5,S thresholds defined in this work
compared with the thresholds at 5 % exceedance probability de-
fined in Gariano et al. (2015) for the same study area (T *5;S ).
The two thresholds are only slightly different. In particular, T *5;S is
steeper than T5,S. This is due to the presence of few rainfall condi-
tions characterized by low (D,E) values. The relative uncertainty for
the parameter is lower for T *5;S (/=11.7 %) than for T5,S (14.3 %).
Conversely, for the parameter, the uncertainty is higher for T *5;S
(/=10.0 %) than for T5,S (7.5 %). For a duration D=10 h, consid-
ering the uncertainties associated to T *5;S , the cumulated rainfall E
ranges from 15.5 to 22.4 mm (E=6.9 mm). Considering the uncer-
tainties associated to T5,S, the cumulated rainfall E ranges from 11.3 to
17.5 mm (E=5.9 mm). More generally, for duration D<65 h, E for
T *5;S higher than for T5,S, vice versa for D10 h.
Conclusions
We used an algorithm for the objective reconstruction of rainfall
events and of multiple ED rainfall conditions responsible for slope
failures in Sicily. We reconstructed 29,270 rainfall events in Sicily
in the period 20022012, and 472 rainfall conditions responsible
for 229 shallow landslides. Rainfall events are more frequent in the
northeastern area of the island. Following the classification pro-
posed by Alpert et al. (2002), heavy-torrential and torrential events
are more frequent in the eastern side of the island.
The algorithm was used to exploit the 472 multiple rainfall
conditions in order to define objective and reproducible empirical
rainfall thresholds for the possible initiation of landslide. The
calculated frequentist threshold at 5 % is slightly steeper than that
defined by Gariano et al. (2015) for the same study area.
This work is a first attempt to define rainfall thresholds starting
from multiple rainfall conditions. The main advantage is to use a
larger number of possible (D,E) pairs (from one to seven per each
landslide) to be included in the bootstrapping technique.
We expect that the proposed algorithm will be used (i) for the
objective and reproducible definition of large sets of rainfall events
and rainfall conditions responsible for failures and (ii) for the
objective and reproducible definition of rainfall thresholds from
the ensemble of the ED rainfall conditions. The obtained thresh-
olds may be implemented in an operational early warning system
for shallow landslide forecasting.
Acknowledgments
The work was supported by a grant of the Italian national
DPC grant. We thank Provincial Fire Brigades in Sicily for
providing the landslide information. We are grateful to the
editor and the anonymous reviewers for their comments and
suggestions that have helped us to improve the paper
considerably.
References
Alecci S, Rossi G (2007) Controllo di qualit dei dati pluviometrici ed analisi delle serie
temporali. In: Rossi G (ed) Siccit. Analisi, monitoraggio e mitigazione. Applicazioni in
Sicilia. Nuova Editoriale Bios, Cosenza, in Italian
Aleotti P (2004) A warning system for rainfall-induced shallow failures. Eng Geol 73:247265
Alpert P, Ben-Gai T, Baharan A, Benjamini Y, Yekutieli D, Colacino M, Diodato L, Ramis C,
Homar V, Romero R, Michaelides S, Manes A (2002) The paradoxical increase of
Mediterranean extreme daily rainfall in spite of decrease in total values. Geophy Res
Lett 29(11):31-131-4. doi:10.1029/2001GL013554
Bach-Kirschbaum D, Adler RF, Hong Y, Kumar S, Peters-Lidard C, Lerner-Lam A (2012)
Advances in landslide nowcasting: evaluation of a global and regional modeling
approach. Environ Earth Sci 66:16831696. doi:10.1007/s12665-011-0990-3
Berti M, Martina MLV, Franceschini S, Pignone S, Simoni A, Pizziolo M (2012) Probabilistic
rainfall thresholds for landslide occurrence using a Bayesian approach. J Geophys Res
117, F04006. doi:10.1029/2012JF002367
Bivand RS, Gebhardt A (2000) Implementing functions for spatial statistical analysis using
the R language. J Geogr Syst 2:307317. doi:10.1007/PL00011460
Bivand RS, Pebesma E, Gomez-Rubio V (2008) Applied spatial data analysis with R. Use R!
series. Springer, New York, 410 pp
Brunetti MT, Peruccacci S, Rossi M, Luciani S, Valigi D, Guzzetti F (2010) Rainfall
thresholds for the possible occurrence of landslides in Italy. Nat Hazards Earth Syst
Sci 10:447458. doi:10.5194/nhess-10-447-2010
Caine N (1980) The rainfall intensity-duration control of shallow landslides and debris
flows. Geogr Ann 62A:2327
Cannon SH, Ellen SD (1985) Rainfall conditions for abundant debris avalanches in the San
Francisco Bay region, California. Calif Geol 38(12):267272
Cannon SH, Gartner JE, Wilson RC, Bowers JC, Laber JL (2008) Storm rainfall conditions
for floods and debris flows from recently burned areas in southwestern Colorado and
southern California. Geomorphol 96(34):250269
Corominas J, Moya J (1999) Reconstructing recent landslide activity in relation to rainfall
in the Llobregat River basin, Eastern Pyrenees, Spain. Geomorphol 30:7993
Crosta GB (1998) Regionalization of rainfall thresholds: an aid to landslide hazard
evaluation. Environ Geol 35:131145
Crozier MJ (1997) The climate-landslide couple: a southern hemisphere perspective. In:
Matthews JA, Brunsden D, Frenzel B, Glser B, Wei MM (eds) Rapid mass movement as
a source of climatic evidence for the Holocene 19. Gustav Fischer, Stuttgart, pp 333354
Cruden DM, Varnes DJ (1996) Landslides types and processes. In: Turner AK, Schuster RL
(eds) Landslides: investigation and mitigation. Transportation research board special
report. National Academy Press, WA, pp 3675
Di Piazza A, Conti FL, Noto LV, Viola F, La Loggia G (2011) Comparative analysis of
different techniques for spatial interpolation of rainfall data to create a serially
complete monthly time series of precipitation for Sicily, Italy. Int J Appl Earth Obs
Geoinf 13:396408. doi:10.1016/j.jag.2011.01.005
Gariano SL, Brunetti MT, Iovine G, Melillo M, Peruccacci S, Terranova O, Vennari C,
Guzzetti F (2015) Calibration and validation of rainfall thresholds for shallow landslide
Landslides 13 & (2016) 171
 Technical Note
forecasting in Sicily, southern Italy. Geomorphol 228:653665. doi:10.1016/
j.geomorph.2014.10.019
Guzzetti F, Peruccacci S, Rossi M, Stark CP (2007) Rainfall thresholds for the initiation of
landslides in central and southern Europe. Meteorog Atmos Phys 98:239267.
doi:10.1007/s00703-007-0262-7
Guzzetti F, Peruccacci S, Rossi M, Stark CP (2008) The rainfall intensity-duration control of
shallow landslides and debris flows: an update. Landslides 5(1):317. doi:10.1007/
s10346-007-0112-1
Innes J (1983) Debris flows. Prog Phys Geogr 7:469501
Jibson RW (1989) Debris flow in southern Porto Rico. Geol Soc Am 236:2955
Kanji MA, Massad F, Cruz PT (2003) Debris flows in areas of residual soils: occurrence and
characteristics. Int. Workshop on Occurrence and Mechanisms of Flows in Natural Slopes and
Earthfills, Iw-Flows 2003, 1416 May 2003, Sorrento, Patron Ed., Bologna. 2:111
Marchi L, Arattano M, Deganutti AM (2002) Ten years of debris-flowmonitoring in the
Moscardo Torrent (Italian Alps). Geomorphol 46:117
Martelloni G, Segoni S, Fanti R, Catani F (2012) Rainfall thresholds for the forecasting of
landslide occurrence at regional scale. Landslides 9:485495
Melillo M, Brunetti MT, Peruccacci S, Gariano SL, Guzzetti F (2015) An algorithm for the
objective reconstruction of rainfall events responsible for landslides. Landslides
12(2):311320. doi:10.1007/s10346-014-0471-3
Peruccacci S, Brunetti MT, Luciani S, Vennari C, Guzzetti F (2012) Lithological and
seasonal control on rainfall thresholds for the possible initiation of landslides in
central Italy. Geomorphol 139140:7990. doi:10.1016/j.geomorph.2011.10.005
172 Landslides 13 & (2016)
Staley DM, Kean JW, Cannon SH, Schmidt KM, Laber JL (2012) Objective
definition of rainfall intensityduration thresholds for the initiation of post-
fire debris flows in southern California. Landslides 10:547562. doi:10.1007/
s10346-012-0341-9
Vennari C, Gariano SL, Antronico L, Brunetti MT, Iovine G, Peruccacci S, Terranova O,
Guzzetti F (2014) Rainfall thresholds for shallow landslide occurrence in Calabria,
southern Italy. Nat Hazards Earth Syst Sci 14:317330
Wieczorek GF (1987) Effect of rainfall intensity and duration on debris flows in central
Santa Cruz Mountains. In: Costa JE, Wieczorek GF (eds) Debris flow/avalanches:
process, recognition, and mitigation. Geological Society of America, Reviews in Eng
Geology, vol 7., pp 93104
M. Melillo ()) : M. T. Brunetti : S. Peruccacci : S. L. Gariano : F. Guzzetti
Istituto di Ricerca per la Protezione Idrogeologica,
Consiglio Nazionale delle Ricerche,
via Madonna Alta 126, 06128, Perugia, Italia
e-mail: massimo.melillo@irpi.cnr.it
S. L. Gariano
Dipartimento Fisica e Geologia,
Universit degli Studi di Perugia,
via A. Pascoli, 06123, Perugia, Italia
 Technical Note
Landslides (2016) 13:173181 Xiao-ping Chen I Hong-hu Zhu I Jing-wu Huang I Dong Liu
DOI 10.1007/s10346-015-0629-7
Received: 4 May 2015
Accepted: 10 August 2015 Stability analysis of an ancient landslide considering
Published online: 1 September 2015
Springer-Verlag Berlin Heidelberg 2015
shear strength reduction behavior of slip zone soil
Abstract In this paper, the stability of an ancient landslide during
the first impounding of a nearby reservoir is investigated through
the analyses of the shear strength reduction behavior of slip zone
soil. In view of the experimental observations, an empirical strain-
dependent soil model is established and is then incorporated in
finite element analyses. The numerical analysis results show that
the failing sections progressively develop due to the soil strength
declines from peak toward residual, and the shear zone propagates
within the front slope. It is demonstrated in the numerical results
that the toe weighting measure has a significant effect on
restraining the shear displacements of the soils and preventing
the progressive failure of the landslide. The field observations
further confirmed the stability condition of the reinforced
landslide.
Keywords Slip zone . Residual shear
strength . Landslide . Strain-softening . Slope stability . Finite
element analysis
Introduction
The stability analysis of natural and engineering slopes is a com-
plex problem, especially when the residual shear strength behavior
of slip zone soils has to be taken into account. Owing to the
strength reduction phenomenon of the soil with increasing shear
strain, the progressive slope failure process is characterized by a
nonuniform mobilization of the shear strength along the potential
slip surface. Many factors contribute to this process, one of which
is the toe slipping triggered by submerging or trimming. It is well
known that an ancient landslide, generally characterized by ample
dimensions and existing slip zones, can remain stable or meta-
stable for centuries on the river bank. However, the landslide will
be probably reactivated by the first impounding of newly built
reservoirs in the vicinity (Wang et al. 2008). Extensive studies have
been performed to investigate the main instability mechanism of
bank slopes, most of which considered the hydromechanical in-
teractions as a direct triggering factor (e.g., Li et al. 2008, 2010; Jia
et al. 2009; Paolo et al. 2013; Zhu et al. 2014; Pamuk et al. 2015).
However, the literature gives little information on how shear zones
propagate when the slope toe is gradually submerged in water. On
the other hand, during the first filling of a reservoir, the shear
strength of the soil mass involved significantly reduces from peak
toward residual due to the strain-softening effect. This may result
in nonuniform interruptions of the equilibrium state of bank
slopes, which has not been well addressed. Therefore, an in-
depth understanding of the progressive failure process induced
by the submergence of the slide toe is essential for enhancing the
capability of landslide mitigation and prevention.
In the past few decades, numerous quantitative studies have
been carried out for estimating the progressive failure and post-
failure deformation of slopes in soils with strain-softening behav-
ior. Potts et al. (1990, 1997) proposed an approach to simulate the
strain-softening properties of brittle soils by reducing the strength
parameters with the accumulated deviatoric plastic strains, which
is a major contribution to progressive failure analysis of slopes.
Troncone (2005) presented the results of a numerical study on
a landslide that was triggered by deep excavations at the slope
toe and further extended this analysis into a three-
dimensional problem (Troncone et al. 2014). The elasto-
viscoplastic model was adopted to account for the pro-
nounced strain-softening behavior of the soil in the landslide.
Other analyses of progressive failure based on elastic-plastic
strain-softening constitutive models had been reported in lit-
erature. For instance, Chai et al. (Chai and Caeter 2009)
proposed an improved model to simulate the failure of an
embankment using the Modified Cam Clay model. Through a
similar approach, Conte et al. (2010) investigated slope failure
cases trigged by weathering. More recently, Samaneh (2013)
proposed a numerical method that can be used to evaluate
the post-failure large deformation of slopes. In addition to
these numerical studies, the residual strengths of different
types of slip zone soils have been investigated over the past
few decades (Skempton 1985; Wen et al. 2007; Chen and Liu
2014). The comparison between the residual strengths obtain-
ed in laboratory tests and those measured in the field has
been made to gain a better understanding of the reactivation
of ancient landslides (Mesri and Nejan 2012).
In this study, the stability analysis of an ancient landslide
during the first filling of a reservoir is studied considering the
strength reduction characteristics of the slip zone soil. In view of
the experimental observations in laboratory, an empirical relation-
ship between shear strength parameters and shear displacement is
established and is then incorporated into a plastic criterion to
account for the strain-dependent behavior of the slip zone soil.
Based on the proposed model, finite element analyses are per-
formed to study the progressive failure of this landslide. The
numerical modeling results indicate that toe weighting is an effec-
tive measure to prevent the reactivation of the landslide.
Landslide features
The typical geological condition of the ancient landslide is shown
in Fig. 1a. This landslide is situated on the Wujiang River, a
tributary of the Beijiang River, 300 km northern of Guangzhou
City, China (Fig. 1b). It has a steep front slope, and the total
volume of the landslide is about 2.4 million m3. The aerial view
of the site before construction of the reservoir is shown in Fig. 1c.
The subsoil is essentially composed of sericite quartz sandstones
and slates. Drilling and other geological evidence revealed a
preexisting slip zone within the landslide, which is composed by
clay with sand and silty clay. The thickness of the slip zone is
between 30 and 150 cm (see Fig. 1a).
According to engineering experience, during the first
impounding of the reservoir, the steep front slope will be fully
submerged and become locally unstable. Will the toe instability
cause the overall failure of the landslide? Because the landslide is
Landslides 13 & (2016) 173
 Technical Note
a)
b)
c)
Fig. 1 The existing landslide prior to the reservoir construction. a Schematic geological section. b Location of the landslide. c Aerial view of the landslide
near the reservoir dam, the risk of landslide reactivation induced
by toe slipping must be carefully evaluated in detail before a safe
design of the reservoir can be made.
Strain-dependent soil model
A series of laboratory tests were conducted to study the physical
and mechanical behavior of the slip zone soil. The undisturbed soil
specimens were collected from trial pits where the slip zone forms
an outcropping (see point A in Fig. 1a). The grain size distribution
of the soil is 30.7 % clay (<5 m), 40.6 % silt (510 m), 15.0 % fine
sand (75250 m), and 13.7 % mid-sized sand with gravel
(>250 m). The soil has an initial water content w=16.0 % and a
174 Landslides 13 & (2016)
density =2.05 g/cm3. According to the Chinese National Standard
(GB 50021-2001 2001), the soil is classified as clay containing sand
with low liquid limit (CSLL). Based on the results of laboratory
shear tests (Chen and Liu 2014), the relationship between the
Mohr-Coulomb (MC) shear strength parameters and the shear
displacement was obtained (Fig. 2). It is seen that both the
cohesion c and friction angle were dependent on shear
displacement. In particular, the post-peak cohesion of the soil
sharply decreased to zero when the shear displacement
accumulated.
The strength reduction characteristic of the slip zone soil is of
great importance for slope stability analysis. Assuming that the
 35
30
25
c (kPa)
20
15
10 c 10
5 5
0 0
      
d (mm)
Fig. 2 Relationship between shear strength parameters and shear displacement
shear displacement occurs at the slope toe due to environmental
factors, such as water level fluctuation or toe excavation, the shear
strength of the soil involved will gradually reduce with accumu-
lated shear displacement until it reaches a residual state. This
behavior leads to a progressive failure process, which starts with
a section of the soil of large shear displacements, and then prop-
agates within the whole shear zone due to stress redistribution
effect. In finite element analyses, if the strength reduction behavior
of soil can be truly modeled, the progressive failure of landslide
can be numerically simulated with high reliability.
Following Potts et al. (1990), the strength reduction behav-
ior of soil can be described by the strain-softening function in
terms of both effective and total stresses, as shown in Fig. 3.
In this figure, p and cp are MC peak strength parameters,
and r and cr are residual strength parameters, respectively.
When the strain level esesp, the cohesion c and friction
angle maintain the peak value. The strength reduction
phenomenon begins when es>esp, following a linear decreas-
ing function in terms of c and . When the strain level
reaches esr, the strength parameters become constant again.
The application of the piecewise tri-linear function to nu-
merical analysis faces two challenges. One problem is how to
define the shear strain, which is not homogeneously distrib-
uted in the soil sample during laboratory testing. The other
problem lies in determining the strain thresholds esp and esr,
which should be done based on abundant experimental data.
Fig. 3 An empirical relationship between shear strength parameters and shear
strain
35
Potts et al. (1987) assumed a mean shear strain xy given by
30 u/h, where u is shear displacement and h is the initial sample
25 thickness. More generally, in finite element analyses, the
strain is normally represented by the deviatoric plastic strain
20
(o)
(Conte et al. 2010). In plane strain condition, it is es p16
15 q
 
x y 2 2x 2y 2xy , where x and y are normal
strains in the x and y directions, respectively, and xy is shear
strain. In this study, the strain thresholds esp and esr are
experimentally determined using laboratory test results.

Numerical analysis method of the landslide
Analysis model
The numerical analyses of the landslide were carried out
using the finite element software ANSYS. The typical cross-
section of the landslide shown in Fig. 1a was selected for the
finite element analyses. Figure 4 shows the finite element
mesh and boundary conditions used in the analyses. The
whole landslide consists of the slope mass, the slip zone,
and the bed rock, which are denoted as upper soil, slip zone
soil, and lower soil, respectively. Because the behavior of the
slip zone soil depends on both the strain-softening effect and
the size of the elements in which the collapse occurs, a
technique of adaptive mesh refinement was adopted for the
red zone in Fig. 4, which means that the mesh will be locally
refined to improve the calculation precision. The main pa-
rameters used in the numerical analyses are listed in Table 1.
Using the elasto-plastic failure criterion coupled with the em-
pirical strength-strain relationship given in Fig. 3, a strain-
dependent elasto-plastic soil model is derived. This model can be
regarded as an extension of the elastic-perfectly plastic model. In
this model, the soil is assumed to obey the elasto-plastic failure
criterion with an added strain-softening feature. The approach in
combination with the MC criterion has been successfully used to
characterize the behavior of various strain-softening clays in the
past few years (Potts et al. 1990; Troncone 2005; Kovacevic et al.
2013). In the ANSYS software, the MC failure surface is normally
substituted by the Drucker-Prager (DP) failure cone, which is
circumscribed to the MC hexagon pyramid. From the point of
view of numerical efficiency, this will bring much convenience to
numerical calculation.
The Drucker-Prager yield criterion F can be written as follows:
p
F I 1 J2 k 1
where I1 is the invariant of the spherical stress tensor, J2 is the
invariant of the deviatoric stress tensor, and and k are model
constants that are related to the shear strength parameters c and ,
respectively.
Equation (1) shows that the yield surface is controlled by
and k. A comparison of the results of a simple slope stability
analysis shows that these two model constants can be calcu-
lated using the MC strength parameters, i.e., 31 sin and
k=ccos (Zhao et al. 2006). In this way, a smooth version of
the MC yield surface for plane strain condition can be
established based on the DP model. It should be noted that
Landslides 13 & (2016) 175
 Technical Note

(%)
esr

6
7

0
Strain threshold
esp
(%)
4
3

0
25.3
25

35
()
r
Residual strength
Fig. 4 Finite element mesh and boundary condition used in the numerical
analyses (with an existing slip zone)

(kPa)

300
cr

8
0
these relationships are based on a nonassociated flow rule,

26.6
i.e., the dilatancy angle =0. In the sections to follow, this is

30

35
()
called the Mohr-Coulomb matching Drucker-Prager (MC-DP)

Peak strength
yield criterion. Furthermore, if the strength-strain relationship
shown in Fig. 3 is incorporated into the MC-DP yield criteri-

(kPa)
on, the elasto-plastic constitutive model of the strain-softening

24.6

300
23
cp
type (S-MC-DP) can be derived.
In this study, the MC-DP and S-MC-DP models are used to
characterize the behavior of the soils involved in the land-
slide. The various parameters used in the numerical analyses

Poissons ratio
are summarized in Table 1. These parameters are derived from
laboratory test data and field measurements (Chen and Liu
2014). In particular, the strain thresholds are not only mainly

0.25
0.30

0.25
deduced from test data but also estimated upon general cases
in available literature.

Analysis procedure
Toe-slipping is a hydraulic effect that occurs during the first
Youngs modulus

impounding of a reservoir, which will result in a nonuniform

mobilization of the shear strength along the slip zone. For sim-
E (MPa)

plicity, the dynamic seepage and the strength reduction of soil due
to water immersion are not taken into account in the following 1000
40
30
Table 1 Soil parameters and strain thresholds used in the numerical analyses

procedures.
Numerical analysis was first performed on the landslide before
the construction of the reservoir dam. The gravitational stress
weight ' (kN/m3)

state within the landslide is established, while the corresponding

Submerged unit

displacement and strain are set to zero. Then, the numerical

analysis is performed using the MC-DP plastic criterion and the
peak strength parameters. The stress and strain obtained in this
11.3
11.6

step are used as the initial state of the following progressive failure
-

analysis.
The process of impounding the reservoir is simulated in two
stages, until the water level elevation, Zel, reaches the normal
level, Zel=154.5 m (meters above sea level, as below). Two
Unit weight
(kN/m3)

simulations, referred to as nonsoftening and softening, are

conducted in this stage, and the results are compared to each
20.3
20.1

23.0

other. In the former, all soils involved are assumed to obey the
MC-DP criterion, and the calculating is performed in one step.
In the latter, the S-MC-DP model is employed to characterize
the behavior of the slip zone soil and the upper soil, and the
Lower soil
Upper soil
Slip zone
Soil type

strength parameters are reduced in those elements where the

soil

equivalent plastic strain reaches the thresholds (see Table 1).

The calculation for the softening case is progressive with the

176 Landslides 13 & (2016)

increasing equivalent plastic strain. The increment step corre-
sponding to each stage is outlined in Table 2.
During the computing procedure, the factor of safety of slope
stability was computed by the shear strength reduction finite
element method (SSRFEM). This method was proposed by
Zienkiewicz et al. (1975) and has been widely used in slope stability
analysis. The SSRFEM is an analytic technique that integrates the
shear strength reduction with finite element analysis. Using a
series of trial factors of safety FFr, the soil strength parameters c
and are reduced as follows:
Fig. 5 Potential failure surface before the construction of the reservoir dam
c Fr c= F Fr  
(F=1.24)
2
Fr arctan tan= F Fr

approximately 1.5 times the lateral displacement of the

During the numerical computing process, cFr and Fr are used
as the input parameters. The trial factor of safety FFr incrementally
increases until the convergent criterion is not satisfied. In this
case, the slope is in the limit equilibrium state, and the corre-
sponding FFr is considered as the real factor of safety F. The failure
surface of the slope can be located simultaneously.
Numerical analysis results of the landslide
Predictions and analysis
In accordance with the landslide field condition and the comput-
ing schemes mentioned above, the results of the finite element
analysis are shown in the following figures, which are related to the
increment steps 24 (see Table 2).The initial stability of the land-
slide was first evaluated by the SSRFEM. Figure 5 shows the
distribution of plastic zone (potential failure surface) before the
construction of the reservoir dam (increment step 2). The overall
factor of safety is F=1.24, indicating that the landslide is stable at
this stage. Such a result is consistent with the actual case.
Although the dynamic seepage processes induced by the reser-
voir impounding were not simulated in the finite element analysis,
the increase of the water levels in the reservoir and the landslide
body is considered by using the submerged unit weight (see
Table 1). The numerical results in Fig. 6 show that, as the slope
toe is gradually submerged in water during reservoir impounding,
the effective geostatic stress and sliding resistance are reduced.
The toe first slides when the soil strain drastically increases and
the strength simultaneously declines. It is the initial inducement
that causes a nonuniform mobilization of shear strength because
of stress redistribution. Figure 6ad presents the comparison of
horizontal displacement contours within the landslide between
nonsoftening and softening cases. The displacement magnitudes
between these two cases are utterly distinct. The maximum differ-
ence occurs at the front slope, where the softening case produces
nonsoftening case at Zel=135.1 m (for step 3). At Zel=154.6 m (for
step 4), the discrepancies are 1.9 times and 0.93 m, respectively.
The results reveal that some elements undergo strain that exceeds
the threshold esp. Due to strain softening, part of the loads carried
by these elements are transferred to the neighboring elements,
resulting in the propagation of deformation.
The comparison of the simulated equivalent plastic strain dur-
ing reservoir impounding is shown in Fig. 7. For the nonsoftening
case (Fig. 7a), the simulated maximum plastic strain (at
Zel=154.6 m) is approximately 6.8 % at the slope toe (only in a
very small zone) and 2.2 % at the crest. However, as a result of
strain-softening of the slip zone soil, a failure zone clearly propa-
gates within the slope, and the evolution of the plastic strain is
evident (Fig. 7bd). At Zel=154.6 m, the plastic strain reaches a
maximum value of 11 % at the slope toe and 4.4 % at the crest of
the front slope. There are large shear zones where the shear strains
are over 6 % (the strain threshold of residual strength esr). The
circular failure surface exists at the front slope. The numerical
results indicate that the softening behavior of soils has an impor-
tant influence on the progressive failure of the landslide during
reservoir impounding.
Table 3 shows the variation of factor of safety F with the
reservoir water level. The factors of safety are also obtained using
SSRFEM. For the nonsoftening and softening cases, the value of F
is 1.031.01 at a water level of 135.1 m, which is defined as the
critical water level. The factor of safety indicates that the slope is
almost in the state of limit equilibrium. When the water level rises
from Zel=135.1 m to Zel=154.6 m, the factor of safety further de-
creases for the softening case, which causes an overall failure of the
landslide. The predictions were consistent with the results of
centrifuge model tests (Chen et al 2011).
In summary, the toe deformation triggered by reservoir
impounding results in the strength reduction within the soils

Table 2 Modeled analysis sequence

Increment step Description
1 Initial gravitational stress state within the landslide
2 Stability analysis before the construction of the reservoir dam
3 Initial reservoir filling from the initial level Zel=95.6 m toZel=135.1 m, the slope stability state in nonsoftening and softening
4 Continuation of filling from Zel=135.1 m to Zel=154.6 m, the slope stability state in nonsoftening and softening

Landslides 13 & (2016) 177

 Technical Note
Fig. 6 Comparison of horizontal displacement contours. a Nonsoftening case at Zel=135.1 m. b Nonsoftening case at Zel=154.6 m. c Softening case at Zel=135.1 m. d
Softening case at Zel=154.6 m
involved and it in turn induces more slope deformation. A failure
surface develops progressively, wherein the average soil strength is
less than the peak strength. It is evident that a potential slip can
take place if no reinforcement measure is adopted. That is, the
reservoir impounding causes the slope toe to slide and the slope
body to deform, which may lead to progressive failure because of
the strain dependent behavior of the slip zone soil.
Reinforcement measure
In order to improve the stability of the bank slope, reinforc-
ing the slope toe should be considered during the design
phase of the reservoir. Figure 8a illustrates the slope
a)
c)
Fig. 7 Predicted equivalent plastic strain contours during the reservoir impounding process. a Nonsoftening case (at Zel=154.6 m). b Softening case (at Zel=95.6 m). c
Softening case (at Zel=135.1 m). d Softening case (at Zel=154.6 m)
178 Landslides 13 & (2016)
stabilization scheme. The landslide is designed to be rein-
forced by backfilling rock masses at the slope toe, which is
called Btoe weighting.^ Figure 8b shows the simulated lateral
displacements of Zel=154.6 m, in which the Bcumulative^ in-
dicates the accumulated displacements at the end of the
analysis, and the Bincremental^ means the incremental dis-
placements as the reservoir water level rises from 135.1 to
154.6 m. The plastic strains at the end of the analysis are
shown in Fig. 8c.
The numerical results indicate that, after constructing a
berm at the toe, the slope deformation is significantly de-
creased. The deformation pattern is different from that of
b)
d)
Table 3 Variation of factor of safety with the water level elevations for nonsoftening and softening cases
Zel (m) F
Nonsoftening Softening
95.6 1.24 /
135.1 1.03 1.01
154.6 1.01 Failure

the non-toe reinforcement case. The upper landslide body
tends to move downward due to the softening of the slip
zone soil. However, the displacements of the front slope are
effectively restrained. As a consequence, the landslide has a
relatively uniform distribution of deformation, and the maxi-
mum accumulated displacement is only 64 mm, which occurs
at the middle height of the slope (Fig. 8b). The equivalent
plastic strain occurs in the upstream along the slip zone
(Fig. 8c), and the maximum strain is only 2.9 %. As the
plastic strain is well below the threshold esp, there is no
difference in the simulation results between the nonsoftening
and softening cases. According to the above analyses, the
stability of the landslide can be guaranteed once the sub-
merged part of the slope is stabilized after reservoir
impoundment.
A few additional explanations can further verify the nu-
merical modeling results. Figure 9 shows the slope toe after
toe weighting. After the reservoir construction work, some
field measurements were taken in the landslide.
Unfortunately, the detailed deformation data during the first
reservoir impoundment have not been recorded. When the
reservoir water level rose from 135.0 to 150.0 m (the current

a)

b)
Cumulative Incremental

Toe weighting berm

c)

Toe weighting berm

Fig. 8 Displacement and plastic strain after the reinforcement measure (Zel=154.6 m). a Slope stabilization scheme. b Displacement contour. c Plastic strain contour

Landslides 13 & (2016) 179

 Technical Note
Fig. 9 Photograph of the slide toe after toe weighting
water level), the lateral displacement increments captured by
two borehole inclinometers D1 and D2 are shown in Fig. 10.
The locations of the inclinometers are shown in Fig. 8a.The
displacement monitoring data (Fig. 10) are smaller than the
numerical results (Fig. 8b) due to the uncertainties involved
in the finite element analyses. But, the stability condition of
the landslide is confirmed as there was no apparent crack on
the landslide according to the field observation. It can be
concluded that the landslide after the reinforcement measure
was not reactivated by the first reservoir impoundment.
strength reduction behavior of the slip zone soil. This model
was implemented in the finite element software ANSYS, which
was applied to simulate how the failure zone propagates
within the landslide during reservoir impoundment.
The simulation results show that the shear zone is initiated
and appears to extend within the front slope after the toe is
submerged in water. As the failing sections progressively develop,
the strength of soils involved declines from peak toward residual,
and the slip zone propagates within the landslide. Therefore, the
toe slipping trigged by the reservoir impounding will result in the
reactivation of the landslide. The reinforcement measure of toe
weighting should be carried out to prevent this problem.
The predictions further demonstrate that, after the toe
weighting measure, the shear displacements are significantly
restrained at the slide toe, and the plastic strains are not
large enough to induce softening of the soil involved. The
field measurements confirmed that the toe weighting mea-
sure has effectively prevented the progressive failure of the
landslide.
Acknowledgments
This research was sponsored by the National Natural Science
Foundation of China (Grant Nos. 11072088 and 41472279), Guang-
dong Provincial Water Resources Science & Technology Project
(Grant No. ysk2009-01).

Summary and conclusions References

In this paper, the stability of an ancient landslide during the Chai JC, Caeter PC (2009) Simulation of the progressive failure of an embankment on soft
first reservoir impoundment has been investigated through soil. Comput Geotech 36:10241038
finite element analyses. Laboratory tests were conducted to Chen XP, Liu D (2014) Residual strength of slip zone soils. Landslides 11:306
obtain the peak and residual strengths of the slip zone soil. 314
Chen XP, Huang JW, Ng CWW, Ma SK (2011) Centrifugal model test on ancient bank
An empirical strain-dependent model was used to describe the
landslide. Chin J Geotech Eng 33(10):14961503 (in Chinese)
Chinese National Standard GB 50021-2001 (2001) Code for investigation of geotechnical
engineering. China Architecture and Building Press, Beijing, in Chinese
Displacement increments (mm) Conte E, Silvestri F, Troncone A (2010) Stability analysis of slopes in soils with strain-
0 2 4 6 8 10 12 14 softening behavior. Comput Geotech 37(5):710722
0 Jia GW, Zhan TLT, Chen YM, Fredlund DG (2009) Performance of a large-scale slope
model subjected to rising and lowering water levels. Eng Geol 106:92103
Kovacevic N, Hight DW, Potts DM, Carter IC (2013) Finite-element analysis of the failure
and reconstruction of the main dam embankment at Abberton Reservoir, Essex, UK.
10
Geotechnique 63(9):753767
Li SJ, Feng XT, Knappett JA (2008) Numerical analysis of slope stability influenced by
varying water conditions in the reservoir area of the Three Gorges, China. In: Chen Z,
20
Zhang J, Ho K, Wu F, Li Z (eds) Landslides and engineered slopes. From the past to
the future. Taylor & Francis Group, London, pp 803807
Depth (m)

Li DY, Yin KL, Leo C (2010) Analysis of Baishuihe landslide influenced by the effects of
30 reservoir water and rainfall. Environ Earth Sci 60(4):677687
Mesri G, Nejan HS (2012) Residual shear strength measured by laboratory tests and
mobilized in landslides. J Geotechnical Geoenviron Eng 138(5):585593
40 Pamuk A, Zimmie T, Adalier K, Mahmud M (2015) Ocean rise instability in coastal clay
slopes and possible countermeasures: a centrifuge modeling study. Mar Georesour
Geotechnol 33(1):5866
50 D1 Paolo P, Elia R, Alberto B (2013) Influence of fillingdrawdown cycles of the Vajont
reservoir on Mt. Toc slope stability. Geomorphology 191:7593
D2 Potts DM, Dounias GT, Vaughan PR (1987) Finite element analysis of the direct shear box
test. Geotechnique 37(1):1123
60 Potts DM, Dounias GT, Vaughan PR (1990) Finite element analysis of progressive failure
of Carsington embankment. Geotechnique 40(1):79101
Potts DM, Kovacevic N, Vaughan PR (1997) Delayed collapse of cut slopes in stiff clay.
Fig. 10 Measurement of horizontal displacement increments after toe weighting Geotechnique 47(5):953982

180 Landslides 13 & (2016)

Samaneh M, Taiebat HA (2013) A large deformation analysis for the assessment of failure X.-p. Chen ())
induced deformations of slopes in strain softening materials. Comput Geotech Key Laboratory of Disaster Forecast and Control in Engineering, College of Science and
49:279288 Engineering,
Skempton AW (1985) Residual strength of clays in landslides folded strata and the Jinan University,
laboratory. Geotechnique 35(1):318 Guangzhou, China
Troncone A (2005) Numerical analysis of a landslide in soils with strain-softening e-mail: chenxp@jnu.edu.cn
behavior. Geotechnique 55(8):585596
Troncone A, Conte E, Donato A (2014) Two and three-dimensional numerical analysis of H.-h. Zhu ())
the progressive failure that occurred in an excavation-induced landslide. Eng Geol School of Earth Sciences and Engineering,
183:265275 Nanjing University,
Wang F, Zhang Y, Huo Z, Peng X, Wang S, Yamasaki S (2008) Mechanism for the Nanjing, China
rapid motion of the Qianjiangping landslide during reactivation by the first e-mail: zhh@nju.edu.cn
impoundment of the Three Gorges Dam reservoir, China. Landslides 5(4):379386 J.-w. Huang
Wen BP, Aydin A, Duzgoren-aydin NS, Li YR, Chen HY, Xia SD (2007) Residual Guangdong Hydropower Planning and Design Institute,
strength of slip zones of large landslides in the three Geoges area, China. Eng Guangzhou, China
Geol 93:8298
Zhao SY, Zheng YR, Liu MW, Qian KD (2006) Definition and transformation of slope J. Huang
safety factor based on Drucker-Prager criterion. Chin J Rock Mech Eng 25(S):2730 e-mail: 17786726@qq.com
2734 (in Chinese) D. Liu
Zhu HH, Shi B, Yan JF, Zhang J, Zhang CC, Wang BJ (2014) Fiber Bragg grating-based Shenzhen Comprehensive Geotechnical Engineering Investigation & Design Co. LTD,
performance monitoring of a slope model subjected to seepage. Smart Mater Struct Shenzhen, China
23:095027
Zienkiewicz OC, Humpheson C, Lewis RW (1975) Associated and non-associated visco- D. Liu
plasticity and plasticity in soil mechanics. Geotechnique 25(4):671689 e-mail: liudong04@126.com

Landslides 13 & (2016) 181

 Technical Note
Landslides (2016) 13:183199 Martin Klose I Philipp Maurischat I Bodo Damm
DOI 10.1007/s10346-015-0643-9
Received: 11 November 2014
Accepted: 7 October 2015 Landslide impacts in Germany: A historical
Published online: 16 October 2015
Springer-Verlag Berlin Heidelberg 2015
and socioeconomic perspective
Abstract Landslide impacts on infrastructure and society in the
Federal Republic of Germany are associated with damage costs of
about US$300 million on annual average. Despite the large overall
losses due to widespread landslide activity, there is a lack of
historical impact assessments, not just for Germanys low moun-
tain areas but those of entire Central Europe as well. This paper is
a collection of three case studies from Germany that seek a better
understanding of landslide impacts and their economic relevance
at local and regional level. The first case study investigates damage
types and mitigation measures at a representative landslide site in
ways that support to gain insight into historical hazard interac-
tions with land use practices. This case history is followed by a case
study dealing with fiscal cost impacts of landslide damages for an
example city and the highway system of the Lower Saxon Uplands,
NW Germany. In addition to a cost-burden analysis for affected
public budgets, an overview of the principles of disaster financing
in landslide practice is given. The third case study is focused on
the conflicts of urban development in hazard areas, with an eco-
nomic approach to balancing safety and public welfare interests.
Each case study is based on historical data sets extracted from
Germanys national landslide database. This paper presents three
different case studies that in combination are a first step towards
assessing landslide impacts in integrated perspective.
Keywords Landslide impacts . Germany . Case
studies . Mitigation . Damage costs
Introduction
The nature of landslide impact in western and central European
countries such as the Federal Republic of Germany (abbr.
Germany) is still poorly understood and the severity heavily
underestimated (cf. Spizzichino et al. 2010, 2013; Gibson et al.
2013; Klose et al. 2014a). In contrast to Bglobal landslide hotspots^
(Himalayan Arc, Central America, etc.), where regional clusters of
triggered landslides often result in catastrophic disasters (e.g.,
Dilley et al. 2005; Nadim et al. 2006; Petley 2012), landslides in
Germany and adjacent low mountain areas are local phenomena
with limited areal extension (cf. Bednarczyk 2009; Damm and
Klose 2015). The associated damages vary spatially in frequency,
timing, and intensity and thus are difficult to summarize and
quantify, either in Germany or elsewhere (e.g., Blchl and Braun
2005; Vranken et al. 2013; Bowman 2015; Klose et al. 2015a). Even
though costly, high-magnitude landslides are reported on a regular
basis (Figs. 1 and 2), the nations landslide problem is rather driven
by a large number of small, less destructive damage events and
widespread insidious losses, which rarely receive broader attention
(e.g., Terhorst and Kreja 2009; Damm and Klose 2014, 2015). From
a national perspective, these separate, widely dispersed landslide
impacts, however, add up to annual direct costs of roughly US$300
million (Klose 2015), exceeding the losses of 100-year events of
other hazards, especially winter storms and river floods, within a
few years or decades (e.g., Klawa and Ulbrich 2003; Kron et al.
2012). Like elsewhere in the world (e.g., Walkinshaw 1992; Hearn
et al. 2008; Winter et al. 2014), annual landslide losses at country
level peak in the transportation sector (roads, railways, etc.), as
indicated by damage costs of US$7080 million for Germanys
highway system (~39,400 km) alone (Klose 2015; see also Krauter
et al. 2012). More than 6 % of the countrys territory shows
significant levels of landslide hazard (based on Dikau and Glade
2003), wherefore costs per capita reach a total of US$3.7 per year
(Table 1).
Landslide impacts are mainly addressed by two types of studies,
which are (i) impact assessments, either as cost surveys (ex post)
or risk analyses (ex ante) (cf. Corominas et al. 2014; Meyer et al.
2013), and (ii) vulnerability assessments, i.e., the analysis of the
exposure and sensitivity to landslide damage, most often from a
technical and retrospective (ex post) point of view (cf. Alexander
2005; Papathoma-Khle et al. 2011). The two approaches of impact
assessment are always applied separately, with landslide databases
and Geographic Information Systems (GIS), however, playing a
major role in both of them (cf. Chacn et al. 2006; Van Den
Eeckhaut and Hervs 2012; Van Westen 2013). As systematically
compiled data archives of past and current landslide events
(Hervs and Bobrowsky 2009; Hervs 2013), todays landslide
databases provide a foundation for studying hazard impacts, es-
pecially by means of GIS and geospatial analysis (cf. Dikau et al.
1996; Harp et al. 2011; Guzzetti et al. 2012). The key to understand-
ing damages in future is to track, record, and analyze their occur-
rences in the past (Klose et al. 2015b). In doing so, cost surveys and
historical studies occupy center stage in impact assessment,
gaining strongly in importance through progress in web-based
data mining and archiving (e.g., Battistini et al. 2013; Baum et al.
2014; Taylor et al. 2015). Nonetheless, there has been a shift from
cost survey to risk analysis in recent years, with a focus on
correlating costs with the physical vulnerability of elements at risk
(see above). Empirical economics yet suggests significant uncer-
tainty of such values at risk (cf. Van Westen et al. 2006; Galli and
Guzzetti 2007), wherefore some few studies turn the attention on
historical losses again, trying to integrate cost survey into risk
analysis (e.g., Crovelli and Coe 2009; Wills et al. 2014; Klose et al.
2015a).
Today, global landslide impact is estimated at US$20 billion in
overall losses annually (Table 1), given the available national cost
estimates and the assumption that losses are widely comparable
between industrialized countries (Schuster 1996). These reference
costs amount to 17 % of the 19802013 annual average global
natural disaster losses (US$121 billion, Munich Re 2014) and with
that provide a first idea of the economic relevance of landslide
impact worldwide. The true damage costs in terms of the financial
burdens on society and public budgets, however, are poorly
reflected in these benchmarks, either globally or nationally (e.g.,
Fleming and Taylor 1980; Olshansky and Rogers 1987; Olshansky
1996; Kjekstad and Highland 2009). For a better understanding of
the social vulnerability and resilience to landslide hazards, strict
Landslides 13 & (2016) 183
 Technical Note

Fig. 1 Overview map of landslide damage clusters in Germany with the location of example damage events between 2005 and 2015. The photographs refer to the
following example damage events: a 2009 Nachterstedt landslide in the overburden of a coalmine near the city of Nachterstedt, Saxony Anhalt. The landslide caused the
death of three people and direct damage costs of roughly US$60 million (photo: courtesy of Lausitzer und Mitteldeutsche Bergbau-Verwaltungsgesellschaft). b 2013
schingen, Baden-Wuerttemberg, landslide at the Swabian Jurassic escarpment. The losses related to landslide mitigation and the evacuation of 15 private homes are
estimated at US$510 million (photo: M. Klose). c 2006 Immenstadt landslide at the city limits of Immenstadt, Bavaria. This damage event is classified to the cost category
of US$25 million (photo: M. Klose). d 2005 Lohme (Rgen), Mecklenburg-Western Pomerania, landslide at the cliff coast of the Baltic Sea. The total costs due to the
demolition of a nursing home, slope stabilization, and litigation amount to about US$510 million (photo: B. Damm). Note that the loss information for (a) to (d) are
retrieved from different sources of the Landslide Database of Germany (cf. Damm and Klose 2015)

empiricalbut not necessarily newapproaches and improved
regional data sets are required, including ones that (i) distinguish
public and private losses (e.g., Taylor and Brabb 1972; Shearer et al.
1983; Creasey 1988; Godt 1999), (ii) inform on disaster financing,
insurance, or budgetary impacts (e.g., Risley 1993; Olshansky 1996;
Paus 2005), and (iii) provide insights into cost factors and rela-
tionships to landslide types, magnitudes, and mitigation (e.g.,
Smith 1958; Quinn et al. 2011; Bowman 2015). In terms of archive
studies, case histories, or cost surveys, these approaches support
analyzing the complexity of historical damage costs and thus
create added value for impact assessment in integrated perspective
(e.g., Wasowski 1998; Cornforth 2005; Van Den Eeckhaut et al.
2010).
The types and impacts of urban landslide damages are complex
(cf. Alexander 1989; Glade and Crozier 2005; Schuster and
Highland 2007; Petley 2009), with hazard and risk being closely
interlinked with the technical and financial coping capacity
(resilience) at community level, not just in developing countries
(cf. Anderson and Holcombe 2013) but large parts of the developed
world as well (cf. Fuchs 2009; Klose et al. 2014a). Although
influencing the severity of landslide impact, some of the most
important societal risk factors are still too narrowly considered
in recent studies, including (i) human activity is often a root cause
of landslides, not only by predisposing or triggering them (e.g.,
Huang and Chan 2004; Nadim et al. 2011) but also as a result of
inadequate (low-cost) mitigation, (ii) the level of tolerable or
acceptable risk, a measure driving a large part of landslide costs
in industrialized countries, is highly variable, differing between
individuals, public or private organizations, and societies, with its
nature being to change over time (cf. Harmsworth and Raynor
2005; Winter and Bromhead 2012; Scolobig et al. 2014), and (iii)
decision makers are faced with the difficult task of finding the
right balance in landslide mitigation, since being required to
evaluate complex information (e.g., Howell et al. 1999; Michaels
2005), while considering best practices and economic rationales
(cf. Crozier 2005). The complex societal dimensions of landslide

184 Landslides 13 & (2016)

Fig. 2 Examples of historical landslide events in Germany. a 1876 Kaub am Rhein, Rhineland-Palatinate, landslide in the Upper Middle Rhine Valley, a UNESCO World
Heritage site exposed to multiple natural hazards. A total of 28 fatalities were attributed to this landslide whose direct damage costs amounted to at least US$510 million
(source: unknown, photo taken in 1876; cf. Grunert and Hess 2010; Landslide Database of Germany). b The Dohlenstein landslide site near the city of Kahla, Thuringia.
Several phases of landslide activity (1740, 1780, 1828, 1881, 1970) were reported by Schmidt and Beyer (2002) for this site, whereby the year-1780 event had caused
major flood damages due to damming the Saale river below the Dohlenstein (source: unknown, photo taken in 1912)

risk along with basic geophysical factors make it necessary to turn
towards integrated impact assessments (e.g., Casale and Margottini
1999). First attempts have recently been undertaken to combine cost
survey with risk analysis (e.g., Crovelli and Coe 2009; Wills et al. 2014;
Klose et al. 2015a); a systematic integration of approaches, however,
is still at the very beginning, wherefore individual case studies
continue to play a major role (cf. Klose 2015).
Our knowledge about landslide impacts is still based on very
limited loss data. The aim of the study was therefore to review and
analyze landslide impacts in Germany from a historical and socio-
economic perspective. This paper is organized as follows. The
second section gives a brief overview of the case study areas of
this research work. In the third section, the methods used and the
data material are presented in detail. The focus of the fourth
section is on three different case studies of landslide impact. A
discussion in Section 5 puts the key findings of these historical,
economic, and societal analyses into a broader perspective. The
final section concludes with outlining the importance of integrated
approaches to impact assessment in the future. Note that all
landslide losses in this study are given in US dollar and 2013
2014 values, if not specified otherwise. Losses published in previ-
ous studies are adjusted for inflation by using the US consumer
price index (CPI-U 08-2013 or CPI-U 05-2014).
Study area
The case studies presented in this investigation refer to (i) the
highway system (i.e., BBundesstraen^ equal to US routes) in the
Lower Saxon Uplands, NW Germany, operated by the Lower
Saxony Department of Transportation (NLStBV), and (ii) the city
of Hann. Mnden (~20,000 inhabitants), an example city located

Landslides 13 & (2016) 185

 Technical Note
Table 1 Total annual losses caused by landslides in different countries worldwide
Country Total annual loss (in USD billion)
USA 2.14.3
Japan >3.0
Italy 3.9
India 2.0
China >1.0
Germany 0.3
World c
~20 d
The cost estimates given in 2014 values include direct and indirect losses and are also shown as percentage of national gross domestic product (GDP) or costs per capita (source:
modified after Klose (2015), based on data from Li (1989), Schuster (1996), and Schuster and Highland (2001))
a
According to International Monetary Fund (https://www.imf.org/external/data.htm)
b
According to the UN Population and Vital Statistics Report (cf. United Nations 2014)
c
Based on the assumption that losses are comparable between industrialized countries (cf. Schuster 1996)
d
17 % of the 19802013 annual average global natural disaster losses (US$121 billion; Munich Re 2014)
in the south of the Lower Saxon Uplands (Figs. 3 and 4). This
region in the northwestern part of the Central European Uplands
constitutes a low mountain area with moderate topographic relief
and elevations ranging from 50 to 950 m a.s.l. The entire Lower
Saxon Uplands cover an area of 7,400 km2 and are moderately
populated (~170 persons per km2), except for some few regional
population centers (Gttingen, Hildesheim, etc.) with ~100,000
inhabitants (Destatis 2015). A large part of this region is charac-
terized by hilly to mountainous relief with deeply incised river or
mountain valleys and steep cuesta scarps (e.g., Reicherter et al.
2008; Figs. 3 and 4).
In the study area, landslides are spatially clustered according to
relief intensity, land use patterns, and the areas geologic structure,
especially the dominant soil or bedrock types and properties (cf.
Varga et al. 2006; Damm et al. 2009, 2010; Klose et al. 2014b).
Throughout the Lower Saxon Uplands, there are 77 km of high-
ways (6 % of total network length) that show significant hazard
exposure, with clusters of areas at risk along highway sections
located below steep and deeply weathered valley slopes and on
hillsides covered by moisture-sensitive periglacial sediments
(Klose et al. 2014b, 2015a). The city of Hann. Mnden is built to
a large extent on mountainous terrain and large slopes above a 300
m deep valley basin. This urban area is prone to landslides due to a
combination of predisposing factors such as local topography and
relief, cohesive soils, and disintegrated bedrock, as well as histor-
ical population pressures and slope disturbance (cf. Damm 2000,
2005; Varga et al. 2006). Thus, roughly 10 % of the citys total area
are located in potential hazard zones (Fig. 3b), with major impli-
cations to all types of urban infrastructures (roads, sewer systems,
etc.). As often having reached their expected lifespans, these age-
ing infrastructures are particularly vulnerable to landslides, where-
by tight maintenance budgets hamper to mitigate deterioration
and improve asset life (cf. Damm and Pflum 2004; Maurischat and
Klose 2014).
A wide range of landslide types and processes occurs on an
annual basis in the Lower Saxon Uplands (e.g., Ackermann 1953;
Mortensen 1960; Schunke 1971; Meyer 2005; Bense et al. 2011; Klose
et al. 2014b), most often triggered by heavy rainfall and/or con-
struction during prolonged wet periods (e.g., Tilch 1999; Damm
186 Landslides 13 & (2016)
Loss as percentage of GDPa Loss per capitab (in USD)
0.010.03 714
>0.06 23
0.19 68
0.11 1.7
0.01 0.7
0.01 3.7
2005; Klose et al. 2012a). The most frequent types of landslides
damaging highways are shallow or deep-seated rotational slides in
soil material and embankment fills (e.g., Gidde 2012; Klose et al.
2015a). Furthermore, small to medium large rockfalls and planar
rockslides are widespread in this region, especially at large cut
slopes or below steep rock outcrops (e.g., Gnther et al. 2004;
Schnborn and Heucke 2012). Landslide activity within the city
limits of Hann. Mnden is also characterized by slow-moving
landslides as well as dormant and periodically reactivated slide
masses of late modern age (Fig. 4). These urban landslides extend
in part over more than 10,000 m2 and often affect highly developed
hillsides (cf. Damm 2000; Damm and Pflum 2004). Landslide
volumes, sizes, and velocities vary greatly in the Lower Saxon
Uplands (Klose 2015), but since large slope failures rarely show
high velocities (e.g., THW 1957; Meyer 2005), landslide magnitudes
range from low to moderate intensity in most cases (e.g., Schunke
1971; Tilch 1999; Klose et al. 2014b).
Methods and data material
Landslide database
The present study makes use of historical data sets extracted from
Germanys national landslide database (Damm and Klose 2014,
2015). This data pool of ~5,000 landslide events covers the entire
country and a time period of about the past 150 to 200 years, with
the oldest landslide, however, being recorded as early as 1137. A
top-down approach to data collection enables to condense data-
base coverage in case study areas such as the Lower Saxon
Uplands. The collection of data at regional level is focused on
adding detail to the data pool by targeted archive studies, whereby
most sources are now available in digital format, often compiled in
historical catalogs, online records, and project management data-
bases (cf. Damm and Klose 2014, 2015; Klose 2015). To access this
information, database development takes advantage of a continu-
ously growing data network, especially with partners like the
NLStBV or the municipal office of Hann. Mnden. As regards
knowledge exchange with the NLStBV, a systematic transfer of
data from the Project Information and Management System
(PRIMAS) has been established, providing access to data sets on
Fig. 3 a Hillshade and overview map of the Lower Saxon Uplands, NW Germany, and b the city of Hann. Mnden in the south of this low mountain area. Both maps also
show the distribution of landslides at local or regional level as well as the location of hazard areas identified on the basis of a regional landslide susceptibility model. The
spatial landslide inventory (n=597) illustrated in (a) was used to apply the information value method from Yin and Yan (1988), which is a bivariate statistical approach to
landslide susceptibility modeling at regional scales (source: modified after Klose et al. 2014b, 2015a)

current damage events and all landslide-related business processes
(cf. Klose et al. 2015a; Jaeger and Klose 2015). Alternatively, the
science-practice cooperation with Hann. Mndens municipal of-
fice is based on many years of archive works and expert interviews,
which enabled compiling a widely complete landslide record since
the mid-nineteenth century (cf. Damm 2000; Damm and Pflum
2004; Maurischat and Klose 2014).
Expert interviews
The involvement of practitioners by conducting expert interviews
to collect first-hand loss information is a critical step for both
impact assessment and the development of landslide databases
(e.g., Guzzetti et al. 2012; Van Den Eeckhaut and Hervs 2012).
However, as compared with database development, expert inter-
views for the purpose of historical or socioeconomic case studies,
especially at local and regional level, are organized in a slightly
different manner. Instead of widely standardized damage or cost
surveys (e.g., Walkinshaw 1992; Blchl and Braun 2005; Highland
2006), semi-structured interviews that leave room for an open and
lively discussion occupy center stage, including options for multi-
ple interview waves after rethinking of the survey results (see also
Vranken et al. 2013). This requires strategic partnerships with
experts from landslide practice (e.g., National Research Council
2004), particularly with practitioners who provide practical in-
sights into how landslide damage is managed and financed, either
from geotechnical or economic and legal perspective (e.g.,

Landslides 13 & (2016) 187

 Technical Note

Fig. 4 a The city of Hann. Mnden with the main landslide sites and the location of selected example landslides (photo: modified after Michael Paetzold, license: http://
creativecommons.org/licenses/by-sa/3.0/de/legalcode). The landslide photographs refer to b a shallow year-1981 landslide with a size of 200 m2 at a railway cut in the
Andreesberg area, c and d the main scarp of the 19391940 active part of the Andreesberg landslide illustrated in Fig. 9, e a December 2013 slope failure in a road
embankment located above a small summerhouse at the banks of the Werra river (photo: courtesy of Photo Burkhardt), and f insidious landslide damage at a private
home in Rattberg-Sittigsruh causing the demolition of the building in 1976 (further photos: retrieved from Landslide Database of Germany)

Olshansky 1996; Damm 2006; Popescu and Sasahara 2009). The
interviews that underpin the case studies presented here were
conducted with experts in planning, controlling, and finance from
the municipal office of Hann. Mnden as well as the NLStBV
headquarters based in Hannover and the relevant regional high-
way offices. This investigation mainly refers to the expertise of
interviewees affiliated with the NLStBV Regional Office
Gandersheim, which is the highway office that is responsible for
maintaining large parts of the highway system in the Lower Saxon
Uplands.
Case histories of landslide impact
Historical information on past and current landslide damages is
critical for achieving a better understanding of landslide impacts
in the future. As originally proposed by Varnes and IAEG (1984)
for hazard mapping, this principle also holds true for impact and
cost assessments. Case histories of landslide impact illustrate the
scientific importance of historical data sets and their application in
hazard studies (e.g., Glade 2001; Calcaterra et al. 2003; Raka et al.
2014, 2015). The main purpose of such historical studies is to track
and analyze the interactions between land use practices, landslide
activity, and hazard mitigation at specific locations of infrastruc-
ture over time (cf. Klose 2015). More specifically, case histories are
focused on analyzing the temporal exposure and vulnerability of
urban infrastructures, the changing risk perception of decision
makers, and the resilience to damages at local level, wherefore
historical landslide databases play a key role in their development.
Best practices in compiling landslide databases useful for histori-
cal impact assessments include, among others, traditional archive
works, the search of web and press archives, and data mining of

188 Landslides 13 & (2016)

digital agency records. These types of sources usually provide
information with the required data quality necessary for develop-
ing case histories and further types of impact assessments (e.g.,
Hervs and Bobrowsky 2009; Van Den Eeckhaut and Hervs 2012).
The most effective methods to structure, condense, and visualize
complex historical landslide data sets are path diagrams or illus-
trated flowcharts. Time series also serve as valuable data modeling
techniques to investigate historical landslide impacts on the basis
of information extracted from landslide databases (e.g., Devoli
et al. 2007; Rossi et al. 2010; Simon et al. 2015). The scientific
potential of case histories depends on the strategy and methods
of database development. Historical case studies dealing with land
use effects and landslide damages at local and regional level have
been presented by, among others, Wasowski (1998), Cornforth
(2005), and Raka et al. (2014, 2015).
Case studies
Historical analysis of landslide impact
A promising approach to investigate landslide impacts is to recon-
struct and analyze local infrastructure development over previous
decades and to correlate it with past landslide activity. Such a
historical analysis was performed in the present case study for
the landslide site BAltmndener Wand^ at the highway B3 within
the city limits of Hann. Mnden. The Altmndener Wand is a 300
m long and 5070 m high cut bank of the Fulda river with a natural
slope gradient between 30 and 50. This site is prone to landslides
Fig. 5 Schematic profile of the landslide site BAltmndener Wand^ at the highway B3 within the city limits of Hann. Mnden. The profile provides an overview of the
local site characteristics and the main geological predisposing factors for landslide activity since the mid-nineteenth century. This information is complemented by
technical data on some of the key slope stabilization measures of the 20012006 mitigation project (source: modified after Damm 2005)
for two main reasons (cf. Damm 2000, 2005): (i) a geologic struc-
ture characterized by disintegrated and deeply weathered Triassic
bedrock complexes composed of sand-, clay-, and siltstones dip-
ping downslope and exposed by a road cut and (ii) the presence of
weak and cohesive periglacial sediments with high moisture sen-
sitivity on top of the bedrock, especially in the upper part of the
slope (Fig. 5).
The landslide problem at this site started with the beginning of
modern road construction in the period 18801882, requiring the
preparation of a vertical, up to 15 m high cut slope and an earth
embankment by using the remaining waste material (Fig. 6).
Several soil slides and small rockfalls (50150 m3) occurred during
and shortly after the construction phase and caused its delay since
having made first mitigation measures (masonry gravity wall)
necessary. The costs of landslide mitigation and the increase in
construction costs at that time are unknown today, but legal
opinions about the responsibility for road maintenance and traffic
protection (federal, state, or local government agencies) suggest
their high financial significance. The use of moisture-sensitive
sediments from nearby as embankment fills resulted in the settle-
ment and failure of the roadway in 1894, which was repaired until
1898 with costly gravity retaining walls and by pavement renewal.
Most of the subsequent landslides (n30) primarily affected the
highway from upslope, with clusters of landslide activity in the
1920s, 1960s, and 1970s, especially the years 19361937 (n=4), 1961
(n=2), 19701974 (n=5), and 19902001 (n=7). The ambiguity
about maintenance responsibilities yet hampered early mitigation
Landslides 13 & (2016) 189
 Technical Note

Fig. 6 Case history of landslide impact for the landslide site BAltmndener Wand^ at the highway B3 in the city of Hann. Mnden (photos: retrieved from Landslide
Database of Germany)

190 Landslides 13 & (2016)

efforts and thus caused a complete shutdown of road traffic from
time by time (Fig. 6; see also Damm 2000, 2005).
Landslide mitigation throughout the twentieth century was aimed
at protecting the highway against soil slides and rockfalls with
volumes less than 50 to 100 m3. The most frequently applied methods
were very simple and less expensive, including (i) removal of loose
rock and vegetation (1924, 1936, 19611962, 1994), (ii) rock blasting
(1936), (iii) catch barriers (1937, 1994), and (iv) temporary or perpet-
ual closure of traffic lanes (1982, 1994). As landslide volumes in
excess of 100150 m3, which would have probably raised public
awareness for effective mitigation, were relatively rare, the willing-
ness to pay for expensive slope stabilization was very low in the past.
Most of the root causes of landslide events affecting this site (i.e.,
presence of slope and fissure water) were already recognized several
years after highway construction. This explains why the drainage of
the entire slope was already planned in 19361937 after several years
of increased landslide activity. The related construction works were
officially tendered in November 1937 but had never been realized in
the following years. It is important to note that rock blasting and
construction had preceded the series of landslides in the early twen-
tieth century. These improper land use practices destabilized the
entire slope and intensified landslide susceptibility by a further
disintegration of the weak and deeply weathered bedrock complexes
(cf. Damm 2000, 2005; Fig. 6).
The first engineering solution to mitigate landslide risk at this
road section was implemented in 1962 as a consequence of the
rapid increase in traffic density and thus the increased probability
for vehicles to be hit by loose debris or small landslides. At the
beginning of the 1960s, passive systems such as simple rockfall
drapery were the preferred type of landslide mitigation, but these
low-cost methods to prevent road burial were barely effective and
already destroyed by a very small landslide (<20 m3) in 1966. It
was due to repeated landslide damage that the entire slope was
clear-cut by the year 1970, since hazardous trees were also per-
ceived as significant risk factor for road traffic and vehicles.
Further landslides in winter 1974 caused the burial of the traffic
lane directly beneath the cut slope three times in one month,
which raised public hazard awareness and again illustrated the
high exposure of this road section to landslides. Despite high levels
of temporal hazard exposure throughout the mid-1970s, the con-
struction of a gravity retaining wall and rockfall barrier, which
should had been planned and realized in late fall 1978, was never
implemented during that period of the twentieth century (Fig. 6;
see also Damm 2000, 2005).
Fig. 7 Simplified model of landslide disaster financing for road infrastructures operated by the Lower Saxony Department of Transportation (NLStBV). The flowchart illustrates
the main steps of disaster financing as well as the affected budgets and the different types of expenditures (source: modified after Maurischat and Klose 2014; Klose 2015)
A year-1982 fatal landslide at the highway B80 (US$4.6 million)
few kilometers north this site suddenly raised hazard awareness in
the entire region. In response to this damage event nearby, a geo-
technical site inspection (~US$0.01 million) was conducted in 1982,
with the objective of developing an effective concept for landslide
mitigation. The first idea was to modify the slope by constructing
berms combined with a system of several retaining walls. In the year
1988, this project was estimated at costs of US$24.6 million, but its
realization in the following years, however, was canceled as a result
of high financial resource requirements for road construction after
the German reunification. The solution chosen instead was of pro-
visional character and included (i) the installation of a 200 m long
and 2.5 m high wooden catch barrier (US$0.06 million) at the slope
toe, (ii) stepwise renewal of rockfall drapery (US$0.23 million), and
(iii) rockfall protection with low-energy catch fences (~US$0.04
million). Most of these protection structures, whose total costs
amounted to roughly US$0.3 million, were heavily damaged by a
series of landslides (400500 m3) between 1999 and 2001. These
damage events forced decision makers to launch an expensive slope
stabilization project in the year 2001. Extending over several years
and three main construction phases (2001, 2003, 2006), this project
included the reinforcement of the entire slope (8,000 m2) by soil and
rock nailing with grouted bars or rock anchors (621 m) and the
stabilization of the slope toe by using a tied-back anchor wall in
combination with a steel-reinforced shotcrete facing (Figs. 5 and 6).
The total costs of landslide mitigation between 2001 and 2006
reached the total of US$7.1 million. Little is known so far about the
costbenefit and long-term effectiveness of this investment, although
no further landslide problems have been reported since 2006 (cf.
Damm 2000, 2005; Klose et al. 2015a).
Economic analysis of landslide impact
Budgeting as part of fiscal planning in Lower Saxony involves,
among others, cost analysis (ex post) and risk identification (e.g.,
Rose 2008). As showing the potential to affect public budgets, direct
losses caused or expected to be caused by landslides should be taken
into account in both steps; but this, however, is not the case in Lower
Saxony, neither at municipal nor at state level. Financing of landslide
damage is rather the opposite today, with hazard management being
characterized by reactive instead of proactive thinking (see also
Maurischat and Klose 2014; Klose 2015). The common practice can
be best described by using a simple three-tier model which indicates
that disaster financing is organized as ex post financing based on
budget reallocation (Fig. 7).
Landslides 13 & (2016) 191
 Technical Note
A reserve of funds for repair or mitigation does not exist in this
model in which actions are taken and money is spent only if
landslide damage results in a disruption of infrastructure opera-
bility. Disaster response as first step of the disaster cycle is often
financed with cash from operating budgets for which maintenance
depots are responsible. The upper limit of cash-based disaster
financing in transportation planning is set to a threshold value
of US$0.07 million. In case of landslide damage exceeding this cost
limit, payments correspond to capital expenditures, with the
NLStBVs regional offices having the responsibility for any kind
of investment. To avoid long-term disruption of road traffic,
provisional repair or mitigation is the starting point of subsequent
disaster recovery. The only way to finance such immediate mea-
sures is to shift funds from regular projects that are already
considered in budgeting. Temporary solutions are replaced by
final ones when enough time has passed to treat repair or mitiga-
tion as regular projects whose costs are included in the annual
budgets. This last step of disaster financing implicates that
permanent measures are paid in the same way as regular construc-
tion works. A key principle of hazard management defining the
prevailing culture of reaction is thus to wait until damage costs
become manageable with the standard types and methods of
infrastructure financing (cf. Maurischat and Klose 2014; Klose
2015).
On the basis of the results of a regional cost estimation for
highways (Klose et al. 2015a) and additional loss data for urban
infrastructures published in Klose et al. (2012b), a fiscal cost
impact study was conducted using the example of the NLStBV
Regional Office Gandersheim and the city of Hann. Mnden. The
methods used in this study included cost survey and analysis of
budget data from both administrative bodies. Figure 8 shows the
results of this impact study indicating that landslide losses affect
public budgets in temporally variable intensity. A large part of the
reference costs relates to insidious landslide losses or maintenance
expenditures that have been classified to certain accounting years.
As not distributing these losses over time, there is the tendency to

Fig. 8 Annual comparison of direct landslide costs and regular construction budgets for different types of public infrastructure in the Lower Saxon Uplands. The data sets
serve as an example for the temporarily high financial burdens of landslide losses and relate to a the highway network of the Regional Office Gandersheim, Lower Saxony
Department of Transportation (NLStBV) and b the sewer systems or c urban roads in the city of Hann. Mnden, Upper Weser area. Note that the costs calculated for
highwaysin contrast to the relevant budget datarefer exclusively to the Upper Weser area that covers only about 20 % of the entire highway network operated by the
NLStBV Regional Office Gandersheim (source: modified after Maurischat and Klose 2014; Klose 2015)

192 Landslides 13 & (2016)

overestimate cost volatility, which is important to keep in mind
when interpreting Fig. 8.
The cost survey for the city of Hann. Mnden shows that
landslide losses at city level add up to US$2.4 million (urban
roads) and US$1.0 million (sewer systems) in costly years
(Fig. 8b, c). These annual costs correspond to 141 % (urban roads)
or 47 % (sewer systems) of the relevant annual budget for regular
projects. The average financial burden of landslides in relation to
the annual reference budget amounts to 44 % (urban roads) or
20 % (sewer systems) in the fiscal years between 2010 and 2015.
Expert interviews revealed that landslide activity reduces the av-
erage life cycle of sewer lines at about 50 %, while the average costs
of earthworks in this example city are increased by a factor of 1.4.
Urban planners estimate the annual maintenance costs for roads
in Hann. Mnden to be at US$1.5 million, which is probably two to
three times the average costs of cities with a comparable size. The
annual average direct costs for highways in the greater Hann.
Mnden area (Fig. 3b), which covers about 20 % of the road
network of the NLStBV Regional Office Gandersheim, amount to
US$0.76 million between 1980 and 2010. These costs represent
2.2 % of the annual average construction budget (US$34.8 million)
for the period 20042008 (Fig. 8a). When updating and extrapo-
lating this percentage value for the entire road network, a best
estimate of the budgetary burden accounts for roughly 5 % on
annual average. In costly years such as 2006 (US$7.1 million,
including third construction phase BAltmndener Wand^) or
2007 (US$5.1 million), however, the NLStBV Regional Office
Gandersheim was required to spend money for landslide mitiga-
tion equivalent to up to 16 % of the regular construction budget
(see also Maurischat and Klose 2014; Klose 2015).
The financial data collected in this case study clearly illustrate
that landslides show the potential to be a relevant cost factor in
short- to mid-term fiscal planning. An economically justified ar-
gument for either reactive or proactive hazard management is yet
hard to find on the basis of these data sets. Despite the lack of such
an economic argument, it is evident that the prevailing culture of
reaction contradicts the basic principles of fiscal planning, while
being accompanied by significant opportunity costs.
Societal analysis of landslide impact
The following case study refers to a societal analysis of landslide impacts
at local level and the implications on urban planning by using the
example of the city of Hann. Mnden. This analysis is based on different
perspectives that were discussed during several years of collaboration
with urban planners in numerous in-depth interviews. The results
presented below in part have the character of expert opinions and
standpoints wherefore having some degree of subjectivity.
The BAndreesberg^ is an urban residential area (~40 ha) above
downtown Hann. Mnden whose development started in the mid-
nineteenth century (Figs. 4 and 9). Despite its high attractiveness
for private investors and developers, new construction was
prohibited in recent years, explaining why the last building had
been raised in the mid-1980s. This is mainly because periodic
landslide activity had previously caused damages to private prop-
erty, urban infrastructures (roads, sewer systems, etc.), and a
railway line, especially in the periods 19391941, 1960, and 1980
1982 (Fig. 4). As a result of the landslide event in 19801982, a
former hotel building (~20 rooms) lost structural integrity, which
finally required its demolition in December 1985. The costs of this
particular damage event were about US$4.7 million at that time (cf.
Damm 2000; Damm and Pflum 2004). Today, large parts of the
Andreesberg area are moderately developed, including numerous
single-family homes and a few small apartment buildings (Figs. 4
and 9). This urban landslide site still shows significant levels of
hazard exposure (Fig. 3b), especially during prolonged wet periods
such as in summer 2013. The question therefore is whether strat-
egies to protect existing infrastructures are still reasonable today.
Given the experiences of the 19801982 damage event, the city
of Hann. Mnden prohibited further development in the
Andreesberg area. This was despite the building permit application
of private investors who purchased the piece of land (22,000 m2)
belonging to the former hotel building to raise 1012 single-family
houses. The decision to declare parts of this landslide area as open
space and to prohibit new housing activity caused substantial
indirect losses to the disadvantage of the new landowners (prop-
erty devaluation) and the regulating authority (loss of fee and tax
income). With regard to the surrounding private homes, which are
still occupied and show similar hazard exposure, the citys urban
planners continue to ask the following key questions: (i) is there a
serious risk of loss of life justifying direct action, (ii) is it tolerable
to ask property owners for proactive landslide mitigation, and (iii)
is effective risk reduction feasible and are the accompanying costs
proportionate (costbenefit). Knowing that risks to human life
have not been documented so far, the planners argue that as long
as the probability and magnitude of future landslides are not
clearly specified, there is no urgent necessity to demand landslide
mitigation from potentially affected property owners at their own
expense. This explains why no further prevention measures have
been implemented so far, except for the restriction of urban
redevelopment at the location of the former hotel building (see
also Damm 2000; Damm and Pflum 2004).
The behavior of private investors and developers at this or
comparable landslide sites in Germany is understandable yet not
logical; they make a buying decision under imperfect information
(potential hazard) by balancing purchase price and locational
advantage, whereby the risk of potential damages (direct or indi-
rect loss) is widely excluded in this trade-off. When considering
the role of urban planners, who seek for the citys economic well-
being, the situation is quite different, mainly because information
duties are poorly defined, both at local and national level. As a
result, planners generally have no financial incentive to gather and
provide information on potential landslide hazards, if not forced
by law and policy. From a pure economic and fiscal standpoint, the
development of landslide-prone hillsides is reasonable under cer-
tain conditions, meaning that land use restriction is not necessar-
ily the best practice. Thus, it holds true that hillside development
always involves a trade-off between conflicting interests, with
public welfare and urban growth on the one side and private safety
and well-being on the other. The decision to prohibit development
is far-reaching and affects private landownership and capital assets
as well as the communitys economic viability as a whole. A strict
prohibition of building activity in hillside areas within the city
limits of Hann. Mnden would have paralyzed urban development
even centuries ago (Fig. 4). Profitability of course plays an impor-
tant role in urban planning, requiring the willingness of planners
to take and minimize certain risks. While anticipating the costs
and benefits of hillside development, urban planning is confronted
with ensuring the economic well-being of single households and
Landslides 13 & (2016) 193
 Technical Note

Fig. 9 Geomorphological map and building register plan of the Andreesberg area showing the location and extent of this urban landslide site in the city of Hann.
Mnden. The information on landslide size and features was gathered by geomorphological mapping as well as core drillings and near-surface sounding (source: modified
after Damm 2000)

the entire local community, not just for today and the near future
but also over periods of decades (cf. Damm 2000; Damm and
Pflum 2004).
The consequence is a (non-official!) rule of thumb indicating
that isolated private property losses occurring in periods of 30
50 years are publicly tolerable from an overall economic
perspective. According to this rule of thumb applied in landslide
practice at local level, urban planners do not necessarily feel
obliged to fully protect all types of capital assets at any given costs.
However, if there is a serious risk of loss of life, landslide hazards
are usually perceived as publicly unacceptable, and planners are
forced to take direct actions. This inevitably implies a strict

194 Landslides 13 & (2016)

prohibition of any new development and the protection or evac-
uation of existing buildings that are severely exposed to landslide
hazards (see also Damm and Pflum 2004). A proper weighting of
potential private landslide losses in public profit maximization is
considered to be extremely difficult. The most effective way to
reduce financial risks for property owners while taking advantage
of the profits of hillside development is full landslide insurance
coverage. Given the fact that insurance programs are often lacking
orlike in Germanynot mandatory by law, urban risk manage-
ment is potentially faced with significant allocative inefficiencies,
either in Germany or worldwide.
Discussion
The case studies presented in this paper have shown that landslide
impacts in Germany are strongly influenced by the sociotechnical
setting and its evolution over time. Landslides are associated with
specific types of damage and a temporally changing hazard intensity,
which requires taking a historical and socioeconomic perspective in
impact assessment. The impacts of landslides at local and regional
level vary as a function of hazard awareness and acceptance and thus
the risk culture that summarizes both terms in a broader concept
(see also Ammann 2006; Angignard et al. 2014). This specific risk
culture undergoes a rapid temporal change as being subject to a
variety of societal influences and the psychological effects related to a
periodic variability in hazard exposure. The approaches to risk
management described in the three case studies reflect the common
practice for a large part of Germany (cf. Damm 2000; Merz and
Emmermann 2006; Damm and Klose 2014, 2015), except for some
very few key hazard areas (e.g., Grunert and Hess 2010). There are
also examples from other European countries or elsewhere
supporting these results in the most general sense (cf. Harmsworth
and Raynor 2005; Scolobig et al. 2014).
The analyses revealed that urban planners are faced with the
conflict of balancing public and private interests in ways that seek
economic efficiency and community resilience. It was found that
planning principles, economic rationales, and technical know-how
along with a wide range of psychological factors have a strong
influence on the decision-making processes in risk management.
Landslide hazards are thus often ignored or underestimated and
perceived as negligible not before people and property are at risk
or have already been impacted. The same holds true for the
implementation of risk reduction measures. Instead of being used
proactively for the purpose of landslide prevention, a large part of
these measures is implemented in response to damage events, with
financial considerations and cost pressure implying low effective-
ness and durability in many cases (cf. Damm 2000). The case
history illustrates that the frequency, timing, and intensity of
landslide damage are positively correlated with the past and cur-
rent risk culture of local or regional government agencies.
Consequently, public hazard awareness and the willingness to
invest in landslide mitigation are rapidly decreasing over time,
especially during periods with very few damage events. There is
historical evidence that low-risk aversion at a certain point in time
is equivalent to reduced mitigation efforts, with result that pre-
vention costs are kept to a minimum in subsequent time steps,
while damage and disruption costs show the potential to reach a
maximum in the long run (see also Klose 2015). Widely similar
results were also reported by Finlay and Fell (1997), Solana and
Kilburn (2003), and Winter and Bromhead (2012), whereas the
findings of Lin et al. (2008) do not support some of the results
presented here.
The single most striking observation to emerge from the analyses
was that landslide damages result in significant financial burdens on
public budgets at community level or higher. A culture of reaction is
characterizing risk management in the examples given herein, with
operational efficiency and cost minimization (short term!) being at
the heart of todays management strategy. The principle of organiz-
ing disaster financing on the basis of budget reallocation implies that
larger investments are widely reserved for landslide repair and post-
disaster mitigation. These findings seem to be generally valid for
Germany (see also Mayer 2012) andat least to some extentother
industrialized countries, including adjacent European member states
(Pelling et al. 2011) as well as parts of China (Choi and Cheung 2013).
The results provide no clear indications of an economic argument
for a culture of reaction or, more specifically, a cost advantage of
post-disaster mitigation over proactive landslide prevention. By con-
trast, the analyses highlighted municipal budgetary burdens of 20
40 % on annual average (20102015), which might suggest the failure
of the prevailing policies, especially for this specific urban area in
Germany (see also Maurischat and Klose 2014; Klose 2015). However,
when taking a closer look on the cost structure of the 19802010
losses for highways in the case study area, landslide repair or miti-
gation accounts for 42 or 55 % of the total direct costs, given the
reference data for the NLStBV Regional Office Gandersheim. Bearing
in mind that the majority (8090 %) of the mitigation costs are
associated with post-disaster mitigation, the proportion of costs
reflecting true direct costs increases on roughly 75 % (Klose et al.
2015a). From a historical and costbenefit perspective, the question is
thus whether a shift to a culture of prevention is economically
reasonably, and if yes, how and to which extent a cultural change
would had influenced or reduced costs in the past. Future studies will
need to focus on this key issue of hazard and risk assessment. Even
though a historical validation of the results is often lacking, todays
costbenefit analyses (ex ante) are an important step in this direc-
tion, especially the studies by Agliardi et al. (2009), Brndl et al.
(2009), and Holcombe et al. (2012).
The results of the second case study clearly illustrate that
landslide damages are required to be considered in fiscal planning,
both at local and regional level or in operational or strategic
perspective. It is mainly due to significant cost volatility, and
problems in evaluating insidious losses that associated fiscal cost
impacts are strongly varying over time. Nevertheless, large parts of
the damage costs are identified as major cost factor in short- and
mid-term fiscal planning, as indicated by budgetary burdens rang-
ing between 5 % (highways) and 2044 % (urban infrastructure)
on annual average. Fiscal planning in the sense of cost analysis (ex
post) and risk identification implies to budget a reserve of funds
for periodic and unexpected expenditures (see above). A culture of
reaction in disaster financing is in contrast with this guiding
principle, even though no economic rationale was found to sup-
port a rethinking of this common practice. The analyses yet indi-
cate that landslide damages result in significant indirect losses and
opportunity costs, especially due to financing based on budget
reallocation. This is in good agreement with basic economic prin-
ciples (e.g., Rose 2004a; Hallegatte and Przyluski 2010; Meyer et al.
2013) and landslide studies such as Blchl and Braun (2005),
Crozier (2005), and Zzere et al. (2007). Hardly any reference data
were available to cross-check and validate the fiscal cost impacts
Landslides 13 & (2016) 195
 Technical Note
calculated in this investigation. However, if taking the losses (local
to national level) compiled in cost surveys for the USA or other
industrialized countries as a benchmark (e.g., Fleming and Taylor
1980; Brabb and Harrod 1989; Walkinshaw 1992; Osuchowski and
Roberts 2011; Vranken et al. 2013), there is some first evidence that
these fiscal cost impacts could represent the economic relevance of
landslides in comparable spatial settings, especially as regards
highways at regional level.
The lack of scientific data sets on fiscal cost impacts is seen as a
major obstacle for assessing landslide hazard and risk in integrated
perspective. This is mainly because the ability of the public sector to
withstand hazard impacts (resilience) strongly depends on the avail-
able financial resources. The term Bsocial or economic resilience^ is
of key importance in this regard as describing to which extent
communities and government bodies are able to afford, finance,
and implement mitigation measures (see also Rose 2004b; Cutter
et al. 2008). It is crucial to note that the effectiveness of risk reduction
relates to the types of disaster financing (cost sharing, credits, etc.)
and the instruments for risk transfer (insurance) or budgeting. Given
the significance of todays raising budget deficits at community level,
there is the need to take greater account of fiscal cost impact studies
in future research, especially with a focus on landslide hazards in old
industrialized countries.
Conclusions
Landslide impacts are a pressing societal issue and affect the
economic well-being of single households and local communities
in hazard areas worldwide. This paper has highlighted the impacts
of landslides for selected case study areas in Germany from a
historical and socioeconomic perspective. The investigation re-
vealed that landslide damages are a major cost factor at local
and regional level, with the potential of resulting in multi-million
dollar losses and critical financial burdens, either privately or
municipally. Most of the damage costs in Germany directly relate
to the public risk culture or, more specifically, the level of hazard
awareness and acceptance of affected individuals or government
agencies. This risk culture was poorly developed in the past,
despite widespread insidious losses and frequent damage events.
Even from a global perspective, these findings contrast with to-
days continental-scale hazard and risk assessments, classifying a
large part of Germany and Central Europe as a Blow-risk^ area.
There is thus further research needed to start closing this knowl-
edge gap by estimating the true economic significance of land-
slides in regions outside areas often referred to as Bglobal landslide
hotspots.^ A major challenge of impact assessment originates
from the difficulty of quantifying hazard and risk considering
the wide range of landslide types and processes and the various
cost factors independent of size or magnitude. Recent studies
stress the importance of integrated approaches that use damage
statistics and data on societal risk acceptance to explore potential
landslide impacts in all their facets. The case studies presented in
this paper provide a stimulus for future method development to
obtain novel concepts that support an integrated assessment of
landslide hazard and risk.
Acknowledgments
This research project was funded by the German Research Foun-
dation (DFG DA 452/6-1). The funding of the project is gratefully
acknowledged. Financial support was also given by the Ministry of
196 Landslides 13 & (2016)
Science and Culture of Lower Saxony (MWK Niedersachsen 11.2-
76202-10-1/07) and the German Academic Exchange Service (DAAD
D/12/45096). The authors very much appreciate the funding received
from these institutions. Thanks are also due to the Lower Saxony
Department of Transportation (NLStBV) and the municipal office of
Hann. Mnden for the fruitful collaboration over the past years.
References
Ackermann E (1953) Der aktive Bergrutsch sdlich der Mackenrder Spitze in
geologischer Sicht. Nachrichten der Akademie der Wissenschaften in Gttingen,
Mathematisch-Physikalische Klasse, IIa. Mathematisch-Physikalisch-Chemische
Abteilung 5:6783
Agliardi F, Crosta GB, Frattini P (2009) Integrating rockfall risk assessment and counter-
measure design by 3D modelling techniques. Nat Hazards Earth Syst Sci 9:10591073
Alexander D (1989) Urban landslides. Prog Phys Geogr 13(2):157189
Alexander D (2005) Vulnerability to landslides. In: Glade T, Anderson MG, Crozier MJ
(eds) Landslide hazard and risk. Wiley, Chichester, pp 175198
Ammann WJ (2006) Risk concept, integral risk management and risk governance. In:
Ammann WJ, Dannenmann S, Vulliet L (eds) RISK21 Coping with risks due to
natural hazards in the 21st century. Taylor & Francis Group, London, pp 323
Anderson MG, Holcombe E (2013) Community-based landslide risk reduction: Managing
disasters in small steps. The World Bank, Washington
Angignard M, Garcia C, Peters-Guarin G, Greiving S (2014) The relevance of legal aspects,
risk cultures and insurance possibilities for risk management. In: Van Asch T,
Corominas J, Greiving S, Malet J-P, Sterlacchini S (eds) Mountain risks: From
prediction to management and governance. Springer, Dordrecht, pp 327340
Battistini A, Segoni S, Manzo G, Catani F, Casagli N (2013) Web data mining for
automatic inventory of geohazards at national scale. Appl Geogr 43:147158
Baum RL, Highland LM, Lyttle PT, Fee JM, Martinez EM, Wald LA (2014) BReport a
Landslide^ a website to engage the public in identifying geologic hazards. In: Sassa K,
Canuti P, Yin Y (eds) Landslide Science for a Safer Geoenvironment, The International
Programme on Landslides (IPL), vol 1. Springer, Berlin, pp 95100
Bednarczyk Z (2009) Landslide risk and mitigation measures in Poland. In: Tagungsband
der 17. Tagung fr Ingenieurgeologie und Forum Junge IngenieurgeologenB.
Hochschule Zittau-Grlitz, 6.-9. Mai 2009, Zittau, pp 15
Bense FA, Ertl G, Vollbrecht A, Battaglia M (2011) Gravitative Massenverlagerungen an der Rt/
Muschelkalk-Schichtstufe des Gttinger Waldes GPS-gesttzte Strukturkartierungen. In:
Leiss B, Tanner D, Vollbrecht A, Arp G (eds) Neue Untersuchungen zur Geologie der
Leinetalgrabenstruktur: Bausteine zur Erkundung des geothermischen Potentials der Region
Gttingen. Universittsverlag Gttingen, Gttingen, pp 125131
Blchl A, Braun B (2005) Economic assessment of landslide risks in the Swabian Alb,
Germany Research framework and first results of homeowners and experts survey.
Nat Hazards Earth Syst Sci 5:389396
Bowman ET (2015) Small landslides Frequent, costly, and manageable. In: Davies T
(ed) Landslide hazards, risks, and disasters. Elsevier, Amsterdam, pp 405439
Brabb EE, Harrod BL (eds) (1989) Landslides: Extent and economic significance. A.A.
Balkema, Rotterdam
Brndl M, Romang HE, Bischof N, Rheinberger CM (2009) The risk concept and its
application in natural hazard risk management in Switzerland. Nat Hazards Earth Syst
Sci 9:801813
Calcaterra D, Parise M, Palma B (2003) Combining historical and geological data for the
assessment of the landslide hazard: A case study from Campania, Italy. Nat Hazards
Earth Syst Sci 3:316
Casale R, Margottini C (eds) (1999) Floods and landslides: Integrated risk assessment.
Springer, Berlin
Chacn J, Irigaray C, Fernndez T, El Hamdouni R (2006) Engineering geology maps:
Landslides and geographical information systems. Bull Eng Geol Environ 65:341411
Choi KY, Cheung RWM (2013) Landslide disaster prevention and mitigation through
works in Hong Kong. J Rock Mech Geotech Eng 5:354365
Cornforth DH (2005) Landslides in practice: Investigations, analysis, and remedial/
preventive options in soils. Wiley, Hoboken
Corominas J, van Westen C, Frattini P, Cascini L, Malet JP, Fotopoulou S, Catani F, Van
Den Eeckhaut M, Mavrouli O, Agliardi F, Pitilakis K, Winter MG, Pastor M, Ferlisi S,
Tofani V, Hervs J, Smith JT (2014) Recommendations for the quantitative analysis of
landslide risk. Bull Eng Geol Environ 73:209263
Creasey CL (1988) Landslide damage: A costly outcome of the storm. In: Ellen SD, Wieczorek
GF (eds) Landslides, floods, and marine effects of the storm of January 35, 1982, in the
San Francisco Bay Region, California. U.S. Geological Survey Professional Paper 1434.
United States Government Printing Office, Washington D.C., pp 195203
Crovelli RA, Coe JA (2009) Probabilistic estimation of numbers and costs of future
landslides in the San Francisco Bay region. Georisk 3:206223
Crozier MJ (2005) Management frameworks for landslide hazard and risk: Issues and
options. In: Glade T, Anderson MG, Crozier MJ (eds) Landslide hazard and risk. Wiley,
Chichester, pp 331350
Cutter SL, Barnes L, Berry M, Burton C, Evans E, Tate E, Webb J (2008) A place-based
model for understanding community resilience to natural disasters. Glob Environ
Chang 18:598606
Damm B (2000) Hangrutschungen im Mittelgebirgsraum Verdrngte NaturgefahrB?
Standort, Zeitschrift fr Angewandte Geographie 24:2734
Damm B (2005) Gravitative Massenbewegungen in Sdniedersachsen. Die Altmndener
Wand Analyse und Bewertung eines Rutschungsstandortes. Zeitschrift fr
Geomorphologie NF, Suppl.-Bd. 138:189209
Damm B (2006) Simplified planning regulations in Germany and the effects of landslides.
J Environ Plan Manag 49:777790
Damm B, Klose M (2014) Landslide database for the Federal Republic of Germany: A tool
for analysis of mass movement processes and impacts. In: Sassa K, Canuti P, Yin Y
(eds) Landslide Science for a Safer Geoenvironment, Methods of Landslide Studies, vol
2. Springer, Berlin, pp 787792
Damm B, Klose M (2015) The landslide database for Germany: Closing the gap at
national level. Geomorphology 249:8293
Damm B, Pflum S (2004) Geomorphologische Naturgefahren und Raumplanung
Bewertungsprobleme am Beispiel von Rutschgefahren in Sdniedersachsen.
Zeitschrift fr Geomorphologie NF, Suppl.-Bd. 135:127146
Damm B, Varga K, Heckmann T, Becht M (2009) The impact of bedrock stratification on
landslide susceptibility An example of GIS-based landslide modelling in the Bunter
Sandstone areas of northern Hesse and southern Lower Saxony (Germany). Die Erde
140:175193
Damm B, Becht M, Varga K, Heckmann T (2010) Relevance of tectonic and structural
parameters in Triassic bedrock formations to landslide susceptibility in Quaternary
hillslope sediments. Quat Int 222:143153
Destatis (2015) Kreisfreie Stdte und Landkreise nach Flche, Bevlkerung und
Bevlkerungsdichte am 31.12.2013. https://www.destatis.de/DE/ZahlenFakten/
LaenderRegionen/Regionales/Gemeindeverzeichnis/Administrativ/Aktuell/
04Kreise.html. Accessed July 6, 2015
Devoli G, Morales A, Heg K (2007) Historical landslides in Nicaragua Collection and
analysis of data. Landslides 4:518
Dikau R, Glade T (2003) Nationale Gefahrenhinweiskarte gravitativer Massenbewegungen. In:
Liedtke H, Musbacher R, Schmidt K-H (eds) Nationalatlas Bundesrepublik Deutschland
Relief, Boden und Wasser. Spektrum, Heidelberg, pp 9899
Dikau R, Cavallin A, Jger S (1996) Databases and GIS for landslide research in Europe.
Geomorphology 15:227239
Dilley M, Chen RS, Deichmann U, Lerner-Lam AL, Arnold M, Agwe J, Buys P, Kjekstad O,
Lyon B, Yetman G (2005) Natural disaster hotspots: A global risk analysis. The World
Bank, Washington
Finlay PJ, Fell R (1997) Landslides: Risk perception and acceptance. Can Geotech J
34:169188
Fleming RW, Taylor FA (1980) Estimating the costs of landslide damage in the United
States. Geological Survey Circular 832. U.S. Geological Survey, Arlington
Fuchs S (2009) Susceptibility versus resilience to mountain hazards in Austria
Paradigms of vulnerability revisited. Nat Hazards Earth Syst Sci 9:337352
Galli M, Guzzetti F (2007) Landslide vulnerability criteria: A case study from Umbria,
central Italy. Environ Manag 40:649664
Gibson AD, Culshaw MG, Dashwood C, Pennington CVL (2013) Landslide management in
the UK The problem of managing hazards in a low-risk environment. Landslides
10:599610
Gidde A (2012) Niederschlagsabhngige Bschungsschden in Niedersachsen. In:
Bundesanstalt fr Straenwesen (Hrsg.) 42. Erfahrungsaustausch ber Erdarbeiten
im Straenbau. Berichte der Bundesanstalt fr Straenwesen, Heft S 76.
Wirtschaftsverlag NW, Bremerhaven, S. 9092
Glade T (2001) Landslide hazard assessment and historical landslide data An insepa-
rable couple? In: Glade T, Frances F, Albini P (eds) The use of historical data in natural
hazard assessments. Springer, Dordrecht, pp 153168
Glade T, Crozier MJ (2005) The nature of landslide hazard impact. In: Glade T, Anderson
MG, Crozier MJ (eds) Landslide hazard and risk. Wiley, Chichester, pp 4374
Godt JW (1999) Maps showing locations of damaging landslides caused by El Nio
rainstorms, winter season 199798, San Francisco Bay region, California. Pamphlet
Field Studies Maps MF-2325-A-J. U.S. Geological Survey, Reston
Grunert J, Hess S (2010) The upper middle Rhine valley as a risk area. Nat Hazards
55:577597
Gnther A, Carstensen A, Pohl W (2004) Automated sliding susceptibility mapping of
rock slopes. Nat Hazards Earth Syst Sci 4:95102
Guzzetti F, Mondini AC, Cardinali M, Fiorucci F, Santangelo M, Chang K-T (2012)
Landslide inventory maps: New tools for an old problem. Earth Sci Rev 112:4266
Hallegatte S, Przyluski V (2010) The economics of natural disasters: Concepts and
methods. Policy Research Working Paper 5507. The World Bank, Washington D.C.
Harmsworth G, Raynor B (2005) Cultural consideration in landslide risk perception. In:
Glade T, Anderson MG, Crozier MJ (eds) Landslide hazard and risk. Wiley, Chichester,
pp 219249
Harp EL, Keefer DK, Sato HP, Yagi H (2011) Landslide inventories: The essential part of
seismic landslide hazard analyses. Eng Geol 122(12):921
Hearn GJ, Hunt T, Aubert J, Howell JH (2008) Landslide impacts on the road network of
Lao PDR and the feasibility of implementing a slope management programme.
International Conference on Management of Landslide Hazard in the Asia-Pacific
Region, Sendai, Japan
Hervs J (2013) Landslide inventory. In: Bobrowsky PT (ed) Encyclopedia of Natural
Hazards. Springer, Berlin, pp 610611
Hervs J, Bobrowsky P (2009) Mapping: Inventories, susceptibility, hazard and risk. In:
Sassa K, Canuti P (eds) Landslides Disaster Risk Reduction. Springer, Berlin, pp 321
349
Highland LM (2006) Estimating landslide losses Preliminary results of a seven-state
pilot project. Open-File Report 20061032. U.S. Geological Survey, Reston
Holcombe E, Smith S, Wright E, Anderson MG (2012) An integrated approach for
evaluating the effectiveness of landslide risk reduction in unplanned communities
in the Caribbean. Nat Hazards 61:351385
Howell DG, Brabb EE, Ramsey DW (1999) How useful is landslide hazard information?
Lessons learned in the San Francisco Bay region. Int Geol Rev 41:368381
Huang RQ, Chan LS (2004) Human-induced landslides in China: Mechanism study and its
implication on slope management. Chin J Rock Mech Eng 23(16):27662777
Jaeger A-K, Klose M, Damm B (2015) Landslides along highways: GIS-based inventory
and planning issues. Geophys Res Abstr 17:EGU20151541
Kjekstad O, Highland L (2009) Economic and social impacts of landslides. In: Sassa K,
Canuti P (eds) Landslides Disaster Risk Reduction. Springer, Berlin, pp 573587
Klawa M, Ulbrich U (2003) A model for the estimation of storm losses and the
identification of severe winter storms in Germany. Nat Hazards Earth Syst Sci
3:725732
Klose M (2015) Landslide databases as tools for integrated assessment of landslide risk.
Springer Theses Recognizing Outstanding Ph.D. Research. Springer, Berlin
Klose M, Damm B, Gerold G (2012a) Analysis of landslide activity and soil moisture in
hillslope sediments using a landslide database and a soil water balance model. GEO-
KO 33(34):204231
Klose M, Damm B, Terhorst B, Schulz N, Gerold G (2012b) Wirtschaftliche Schden durch
g r a v i t a t i v e M a s s e n b e w e g u n g e n E n t wi c k lu n g e i n e s em p i r i s c h e n
Berechnungsmodells mit regionaler Anwendung. Interpraevent 12:979990
Klose M, Highland L, Damm B, Terhorst B (2014a) Estimation of direct landslide costs in
industrialized countries: Challenges, concepts, and case study. In: Sassa K, Canuti P,
Yin Y (eds) Landslide Science for a Safer Geoenvironment, Methods of Landslide
Studies, vol 2. Springer, Berlin, pp 661667
Klose M, Gruber D, Damm B, Gerold G (2014b) Spatial databases and GIS as tools for regional
landslide susceptibility modeling. Zeitschrift fr Geomorphologie NF 58(1):136
Klose M, Damm B, Terhorst B (2015a) Landslide cost modeling for transportation
infrastructures: A methodological approach. Landslides 12:321334
Klose M, Damm B, Highland LM (2015b) Databases in geohazard science: An introduc-
tion. Geomorphology 249:13
Krauter E, Kumerics C, Feuerbach J, Lauterbach M (2012) Abschtzung der Risiken von
Hang- und Bschungsrutschungen durch die Zunahme von Extremwetterereignissen.
Berichte der Bundesanstalt fr Straenwesen, Heft S75. Wirtschaftsverlag NW,
Bremerhaven
Kron W, Steuer M, Lw P, Wirtz A (2012) How to deal properly with a natural catastrophe
database Analysis of flood losses. Nat Hazards Earth Syst Sci 12:535550
Li T (1989) Landslides: Extent and economic significance in China. In: Brabb EE, Harrod BL
(eds) Landslides: Extent and economic significance. A.A. Balkema, Rotterdam, pp
271287
Lin S, Shaw D, Ho MC (2008) Why are flood and landslide victims less willing to take
mitigation measures than the public? Nat Hazards 44:305314
Landslides 13 & (2016) 197
 Technical Note
Maurischat P, Klose M (2014) Financing and budgetary impact of landslide losses for
highways and urban infrastructures in NW Germany An economic analysis using
landslide database information and cost survey data. Geophys Res Abstr 16:EGU2014
1995
Mayer J (2012) Naturrisiken und Vorsorge Eine systemtheoretische Beobachtung von
Hangrutschungsrisiken und staatlichen Vorsorgemanahmen. Dissertation,
Rheinische Friedrich-Wilhelms-Universitt Bonn. http://hss.ulb.uni-bonn.de/2012/
3049/3049.pdf. Accessed July 6, 2015
Merz B, Emmermann R (2006) Zum Umgang mit Naturgefahren in Deutschland: Vom
Reagieren zum Risikomanagement. Gaia 15(4):265274
Meyer S (2005) Der rezente Bergsturz am Messingsberg im niederschsischen
Wesergebirge. Mitteilungen des Verbandes der Deutschen Hhlen- und Karstforscher
51(3):96100
Meyer V, Becker N, Markantonis V, Schwarze R, van den Bergh JCJM, Bouwer LM, Bubeck
P, Ciavola P, Genovese E, Green C, Hallegatte S, Kreibich H, Lequeux Q, Logar I,
Papyrakis E, Pfurtscheller C, Poussin J, Przyluski V, Thieken AH, Viavattene C (2013)
Review article: Assessing the costs of natural hazards State of the art and
knowledge gaps. Nat Hazards Earth Syst Sci 13:13511373
Michaels S (2005) Addressing landslide hazards: Towards a knowledge management
perspective. In: Glade T, Anderson M, Crozier M (eds) Landslide hazard and risk. Wiley,
Chichester, pp 311328
Mortensen H (1960) Neues ber den Bergrutsch sdlich der Mackenrder Spitze und
ber die holozne Hangformung an Schichtstufen im mitteleuropischen
Klimabereich. Zeitschrift fr Geomorphologie NF, Suppl.-Bd. 1:114123
Munich Re (2014) NatCatSERVICE. http://www.munichre.com/en/reinsurance/business/
nonlife/natcat service/index.html. Accessed July 15, 2014
Nadim F, Kjekstad O, Peduzzi P, Herold C, Jaedicke C (2006) Global landslide and
avalanche hotspots. Landslides 3:159173
Nadim F, Hydal , Haugland H, McLean A (2011) Analysis of landslides triggered by
anthropogenic factors in Europe. SafeLand FP7, Deliverable D1.6. http://
www.safeland-fp7.eu/results/Documents/D1.6.pdf. Accessed August 13, 2014
National Research Council Committee on the Review of the National Landslide Hazards
Mitigation Strategy (2004) Partnerships for reducing landslide risk: Assessment of the
National Landslide Hazards Mitigation Strategy. The National Academies Press,
Washington D.C.
Olshansky RB (1996) Financing landslide hazard mitigation in the United States. J Environ
Plan Manag 39(3):371385
Olshansky RB, Rogers JD (1987) Unstable ground: Landslide policy in the United States.
Ecology Law Quart 13(4):9391006
Osuchowski M, Roberts J (2011) Landslide costs in the Wollongong Region. Record
2011/32. Geoscience Australia, Canberra
Papathoma-Khle M, Kappes M, Keiler M, Glade T (2011) Physical vulnerability assess-
ment for alpine hazards: State of the art and future needs. Nat Hazards 58:645680
Paus HL (2005) Reply of insurance industry to landslide risk. In: Glade T, Anderson MG,
Crozier MJ (eds) Landslide hazard and risk. Wiley, Chichester, pp 251283
Pelling M, Bye L, Zehra Zaidi R, Scolobig A, Sharma U, Mafttei R, Tudor E, Mihai V,
Porumbescu C, Angignard M (2011) The cultures of landslide risk management in
Europe and India. SafeLand FP7, Deliverable D5.5. http://www.safeland-fp7.eu/re-
sults/documents/d5.5.pdf. Accessed August 13, 2014
Petley DN (2009) On the impact of urban landslides. In: Culshaw MG, Reeves HJ,
Jefferson I, Spink TW (eds) Engineering geology for tomorrows cities. Engineering
Geology Special Publication No. 22. The Geological Society, London, pp 8399
Petley DN (2012) Global patterns of loss of life from landslides. Geology 40:927930
Popescu ME, Sasahara K (2009) Engineering measures for landslide disaster mitigation.
In: Sassa K, Canuti P (eds) Landslides Disaster Risk Reduction. Springer, Berlin, pp
609631
Quinn PE, Hutchinson DJ, Diederichs MS, Rowe RK (2011) Characteristics of large
landslides in sensitive clay in relation to susceptibility, hazard, and risk. Can Geotech
J 48(8):12121232
Raka P, Zbransk V, Dubiar J, Kadlec A, Hrbov A, Strnad T (2014) Documentary proxies
and interdisciplinary research on historic geomorphologic hazards: A discussion of the
current state from a central European perspective. Nat Hazards 70:705732
Raka P, Klimes J, Dubisar J (2015) Using local archive sources to reconstruct historical
landslide occurrence in selected urban regions of the Czech Republic: Examples from
regions with different historical development. Land Degrad Dev 26:142157
Reicherter K, Froitzheim N, Jarsiski M, Badura J, Franzke H-J, Hansen M, Hbscher C,
Mller R, Poprawa P, Reinecker J, Stackebrandt W, Voigt T, von Eynatten H,
Zuchiewicz W (2008) Alpine tectonics north of the Alps. In: McCann T (ed) The
geology of Central Europe, Mesozoic and Cenozoic, vol 2. The Geological Society,
London, pp 12331285
198 Landslides 13 & (2016)
Risley R (1993) Landslide peril and homeowners insurance in California. UCLA Law
Review 40:1145
Rose A (2004a) Economic principles, issues, and research priorities in hazard loss
estimation. In: Okuyama Y, Chang SE (eds) Modeling spatial and economic impacts
of disasters. Springer, Berlin, pp 1336
Rose A (2004b) Defining and measuring economic resilience to disasters. Disaster Prev
Manag An Int J 10:270277
Rose J (2008) Kommunale Finanzwirtschaft Niedersachsen: Grundriss fr die Aus- und
Fortbildung. Kohlhammer, Kiel
Rossi M, Witt A, Guzzetti F, Malamud BD, Peruccacci S (2010) Analysis of historical
landslide time series in the Emilia-Romagna region, northern Italy. Earth Surf Process
Landf 35:11231137
Schmidt K-H, Beyer I (2002) High-magnitude landslide events on a limestone-scarp in
central Germany: Morphometric characteristics and climatic controls. Geomorphology
49:323342
Schnborn C, Heucke K (2012) Bedeutung von Felsen aus der Sicht des Naturschutzes
und ihre Bercksichtigung bei Manahmen zur Verkehrssicherheit von Straen.
Naturschutz im Land Sachsen-Anhalt 49:4049
Schunke E (1971) Die Massenverlagerungen an den Schichtstufen und Schichtkmmen
des Weser-Leine-Berglandes. Nachrichten der Akademie der Wissenschaften in
Gttingen, II. Mathematisch-Physikalische Klasse 3:135
Schuster RL (1996) Socioeconomic significance of landslides. In: Turner AK, Schuster RL
(eds) Landslides: Investigation and mitigation. Transportation Research Board, Special
Report 247. National Academy Press, Washington D.C., pp 1235
Schuster RL, Highland LM (2001) Socioeconomic and environmental impacts of land-
slides in the western hemisphere. Open-File Report 01-0276. U.S. Geological Survey,
http://pubs.usgs.gov/of/2001/ofr-01-0276/. Accessed July 15, 2014
Schuster RL, Highland LM (2007) The third Hans Cloos lecture. Urban landslides:
Socioeconomic impact and overview of mitigative strategies. Bull Eng Geol Environ
66:127
Scolobig A, Linneroth-Bayer J, Pelling M (2014) Drivers of transformative change in the
Italian risk policy. Int J Disaster Risk Reduct 9:124136
Shearer CF, Taylor FA, Fleming RW (1983) Distribution and costs of landslides in San
Diego county, California, during the rainfall years of 197879 and 197980. Open-File
Report 83-582. U.S. Geological Survey, Reston
Simon N, Crozier M, de Roiste M, Rafek AG, Roslee R (2015) Time series assessment on
landslide occurrences in an area undergoing development. Singap J Trop Geogr
36:98111
Smith R (1958) Economic and legal aspects. In: Eckel EB (ed) Landslides and engineering
practice. Highway Research Board, Special Report 28. National Academy of Sciences,
Washington D.C., pp 619
Solana MC, Kilburn CRJ (2003) Public awareness of landslide hazards: The Barranco de
Tirajana, Gran Canaria, Spain. Geomorphology 54:3948
Spizzichino D, Margottini C, Trigila A, Iadanza C, Linser S (2010) Chapter 9:
Landslides. In: European Environment Agency (ed) Mapping the impacts of
natural hazards and technological accidents in Europe: An overview of the last
decade. EEA Technical report 13/2010. European Environmental Agency, Co-
penhagen, pp 8193
Spizzichino D, Margottini C, Trigila A, Iadanza C (2013) Landslide impacts in Europe:
Weaknesses and strengths of databases available at European and national scale. In:
Margottini C, Canuti P, Sassa K (eds) Landslide Science and Practice, Landslide
Inventory and Susceptibility and Hazard Zoning, vol 1. Springer, Berlin, pp 7380
Taylor FA, Brabb EE (1972) Maps showing distribution and cost by counties of structurally
damaging landslides in the San Francisco Bay region, California, winter of 196869.
USGS Miscellaneous Field Studies Map: 327
Taylor FE, Malamud BD, Freeborough K, Demeritt D (2015) Enriching Great Britain's
National Landslide Database by searching newspaper archives. Geomorphology
249:5268
Terhorst B, Kreja R (2009) Slope stability modelling with SINMAP in a settlement area of
the Swabian Alb. Landslides 6:309319
THW (1957, Autor unbekannt) Dammbruch mit schweren Folgen. Das Technische
Hilfswerk 4(12):78
Tilch N (1999) Rutschungs-Suszeptibilitt im sdlichen Niedersachsen Von der
Anatomie der Rutschung bei Brunkensen/Alfeld zur Prognose instabiler Hanglagen.
Braunschweiger Geowissenschaftliche Arbeiten 22, 184 S.
United Nations (2014) Population and vital statistics report. Statistical Papers, Series A
Vol. LXVI. UN Department of Economic and Social Affairs, New York
Van Den Eeckhaut M, Hervs J (2012) State of the art of national landslide databases in
Europe and their potential for assessing landslide susceptibility, hazard and risk.
Geomorphology 139140:545558
Van Den Eeckhaut M, Poesen J, Vandekerckhove L, Van Gils M, Van Rompaey A (2010)
Human-environment interactions in residential areas susceptible to landsliding: The
Flemish Ardennes case study. Area 42(3):339358
Van Westen CJ (2013) Remote sensing and GIS for natural hazards assessment
and disaster risk management. In: Bishop MP (ed) Treatise on Geomorphology,
Remote Sensing and GIScience in Geomorphology, vol 3. Academic Press, San
Diego, pp 259298
Van Westen CJ, Van Asch TWJ, Soeters R (2006) Landslide hazard and risk zonation
Why is it still so difficult? Bull Eng Geol Environ 65:167184
Varga K, Becht M, Damm B (2006) Anstze der GIS-gesttzten rumlichen Modellierung
von Rutschgefahren in Buntsandsteingebieten Nordhessens und Sdniedersachsens
(BRD). Angewandte Geoinformatik 2006:679684
Winter MG, Palmer D, Sharpe J, Shearer B, Harmer C, Peeling D, Bradbury T (2014)
Economic impact assessment of landslide events. In: Sassa K, Canuti P, Yin Y (eds)
Landslide Science for a Safer Geoenvironment, The International Programme on
Landslides (IPL), vol 1. Springer, Berlin, pp 217222
Yin KL, Yan TZ (1988) Statistical prediction models for slope instability of metamor-
phosed rocks. In: Bonnard C (ed) Proceedings of the 5th International Symposium on
Landslides, Vol. 2, Balkema, Rotterdam, pp 12691272
Zzere JL, Oliveira SC, Garcia RAC, Reis E (2007) Landslide risk analysis in the area north
of Lisbon (Portugal): Evaluation of direct and indirect costs resulting from a motorway
disruption by slope movements. Landslides 4:123136

M. Klose ()) : B. Damm

Varnes DJ, IAEG (1984) Landslide hazard zonation: A review of principles and practice.
UNESCO, Paris
Vranken L, Van Turnhout P, Van Den Eeckhaut M, Vandekerckhove L, Poesen J (2013) Department II - Applied Physical Geography,
Economic valuation of landslide damage in hilly regions: A case study from Flanders, University of Vechta,
Belgium. Sci Total Environ 447:323336 Driverstrae 22, 49377, Vechta, Germany
Walkinshaw J (1992) Landslide correction costs on US state highway systems. Transp Res e-mail: martin.klose@uni-vechta.de
Rec 1343:3641 B. Damm
Wasowski J (1998) Understanding rainfall-landslide relationships in man-modified envi- e-mail: bodo.damm@uni-vechta.de
ronments: A case-history from Caramanico Terme, Italy. Environ Geol 35:197209
Wills C, Perez F, Branum D (2014) New method for estimating landslide losses from P. Maurischat
major winter storms in California and application to the ARkStorm scenario. Nat Institute of Geography,
Hazard Rev. doi:10.1061/(ASCE)NH.1527-6996.0000142 University of Osnabrck,
Winter MG, Bromhead EN (2012) Landslide risk: Some issues that determine societal Seminarstrae 19ab, 49074, Osnabrck, Germany
acceptance. Nat Hazards 62:169187 e-mail: p.maurischat@gmx.de

Landslides 13 & (2016) 199

 Technical Note

Landslides (2016) 13:201210
DOI 10.1007/s10346-015-0626-x
Received: 20 October 2014
Accepted: 4 August 2015
Published online: 1 September 2015
Springer-Verlag Berlin Heidelberg 2015
Maricar L. Rabonza I Raquel P. Felix I Alfredo Mahar Francisco A. Lagmay I Rodrigo Narod C. Eco I
Iris Jill G. Ortiz I Dakila T. Aquino
Shallow landslide susceptibility mapping using high-
resolution topography for areas devastated by super
typhoon Haiyan

Abstract Super typhoon Haiyan, considered as one of the most
powerful storms recorded in 2013, devastated the central Philip-
pines region on 8 November 2013 with damage amounting to more
than USD 2 billion. Hardest hit is the province of Leyte which is
located in central Philippines. Rehabilitation of the areas that were
devastated requires detailed hazard maps as a basis for well-
planned reconstruction. Along with severe wind, storm surge,
and flood hazard maps, detailed landslide susceptibility maps for
the cities and municipalities of Leyte (7246.7 km2) province are
necessary. In order to rapidly assess and delineate areas suscepti-
ble to rainfall-induced shallow landslides, Stability INdex MAP-
ping (SINMAP) software was used over a 5-m Interferometric
Synthetic Aperture Radar (InSAR)-derived digital terrain model
(DTM) grid. Topographic, soil strength, and hydrologic parame-
ters were used for each pixel of a given DTM grid to compute for
the corresponding factor of safety. The landslide maps generated
using SINMAP are highly consistent with the landslide inventory
derived from high-resolution satellite imagery from 2002 to 2014
with a detection percentage of 97.5 % and missing factor of 0.025.
These demonstrate that SINMAP performs well despite the lack of
an extensive geotechnical and hydrological database in the study
area. The detailed landslide susceptibility classification is useful to
identify safe and unsafe areas for reconstruction and rehabilitation
efforts. These maps complement the debris flow and structurally
controlled landslide hazard maps that are also being prepared for
rebuilding Haiyans devastated areas.
Keywords Landslide . Natural hazard . SINMAP . Susceptibility
map . Spatial analyses . Philippines
Introduction
On 8 November 2013, typhoon Haiyan (Typhoon BYolanda^ in the
Philippines) reached its peak intensity and swept through the
central Philippines with 1-min sustained winds of 315 km/h at
1800 UTC (JTWC 2013). Powerful winds, heavy rains, and storm
surges triggered widespread flooding and landslides, particularly
affecting the provinces of Leyte (Fig. 1). Typhoon Haiyan left 6300
fatalities, 28,689 injured, and 1061 missing, as well as damage
amounting to over USD 2 billion (NDRRMC 2014).
Although strong winds and storm surges of up to 56 m
(Lagmay et al. 2015) were the primary causes of damage in Leyte,
landslides were also a significant factor. Shallow landslides are
rapid downhill flows of rocks and mudtypically induced by
heavy rainfallwith speeds exceeding 55 km/h (Hong et al.
2007). They constitute one of the most significant hazards in
complex terrain exposed to frequent heavy rainfall (Liao et al.
2010). Relevant shallow landslide-controlling factors include local
conditions such as slope materials and slope surface morphology,
soil properties, topography, groundwater, lithology, bedrock
permeability, and land cover (DAmato Avanzi et al. 2004; Hong
et al. 2007; Liao et al 2010; Wieczorek and Sarmiento 1988). Usu-
ally, the triggering mechanism for shallow landslides on
destabilized slopes usually begins with excessive rainfall. It is
characterized by infiltration into the soil leading to the buildup
of pore pressure, reduced soil suction value, reduced shear
strength, and increased shear stress, weakening slope stability
(Antronico and Gull 2000; Campus et al. 2001; DAmato Avanzi
et al. 2003; Giannecchini 2006; Van Asch et al. 1999). This could
cause surface sliding due to the textural-granulometric disconti-
nuity within the soil-bedrock interface, causing drastic changes in
the infiltration rate (Wieczorek 1987). Slope instability brought
about by the complex interaction of interrelated factors becomes
a major concern when they lie in close proximity to people and
property (Soeters and Van Westen 1996).
Conventionally, landslide susceptibility assessments are con-
ducted through field surveys to account for multiple relevant
controlling factors (Hong et al. 2007). However, the zonation of
shallow landslide potential for provinces with large areas may be
difficult in the Philippines, especially given the difficulty of
accessing mountainous areas and the cost of ground monitoring
networks. Thus, remote sensing may be the only practical means
to obtain data for landslide susceptibility mapping (Catani et al.
2005; Hong et al. 2007; Nadim et al. 2006). High-resolution satel-
lite remote sensing products can provide detailed provincial-wide
shallow landslide susceptibility maps. Studies conducted by Fabbri
et al. (2003) show that the layers extracted solely from a digital
elevation model (DEM) generated more accurate results than
layers derived from geological data, surficial materials, and land
use, indicating that topography is the primary determinant of
landslide location (Coe et al. 2004; Hong et al. 2007).
It is also important to develop accurate hazard maps
pinpointing the locations of areas vulnerable to landslides for
rehabilitation work. The identification of potential landslides and
the delineation of unstable and suitable areas for development can
help ensure public safety through improved land use and disaster
management planning, management of residential and recreation-
al areas, and infrastructural development (AGS 2007; Sarkar and
Kanugo 2004). In the case of areas heavily devastated by Haiyan,
reliable, detailed, and readily accessible landslide susceptibility
maps are critical for well-planned rehabilitation and
reconstruction.
Current approach to landslide susceptibility mapping in the Philippines
Prior to this work, government agencies in the Philippines imple-
ment a qualitative method for landslide susceptibility mapping.
This involves the use of landslide inventories, geologic maps, and
geomorphologic information to identify possible sites of failure.

Landslides 13 & (2016) 201

 Technical Note

Fig. 1 Track of super typhoon Haiyan across the central Philippines. The location of the study area (Leyte province) is enclosed in red. Typhoon track from IBTrACS (Knapp et al. 2010)

However, weights of contributing factors depend on expert
opinion of specialists on the subject. The experts use a checklist
to give subjective weights for the factors (e.g., slope angle, slope
weight, material type, nearness to fault, degree of gully erosion,
vegetative cover, presence of road cuts on hillsides, etc.) and then
derive a cumulative influence factor representing the degree of
hazard (Table 1). Other than the slope gradient, there are no other
quantitative bases for the hazard level assigned to an area (MGB
2013). Thus, the designating weights are highly subjective.
To date, quantitative and deterministic methods for landslide
susceptibility assessment has not been used on a nationwide scale
in the Philippines. The recent availability of high-resolution topo-
graphic models for the country has made this possible. Quantita-
tive procedures are advantageous in rehabilitation purposes
wherein rapid assessment is highly critical to conduct well-
planned reconstruction and rehabilitation.
A sample of maps made by Mines and Geosciences Bureau -
Department of Environment and Natural Resources (MGB-DENR)
Philippines is shown in Fig. 2. These belong to the current national
maps for landslide susceptibility. In the context of disaster manage-
ment, hazard maps of 1:100,000 to 1:50,000 are not suitable for localized
emergency response. Current enhancement efforts by the MGB include
producing hazard maps at a finer scale of 1:10,000 (Fig. 2b).
The same scenario occurs in all the available susceptibility
maps in the country. The major goal is to improve these existing
government maps of the Philippines to enhance post-disaster
rehabilitation using newly available high-resolution terrain data
and landslide information. It is also imperative to expand options

Table 1 Landslide susceptibility parameters currently used MGB-DENRPhilippines to determine the landslide hazard susceptibility of a given area (MGB 2013)
Landslide susceptibility Low Moderate High Very high
parameters
Slope gradient Low to moderate Moderate to steep Steep to very steep Steep to very steep
(<18) (1835) (>35) (>35)
Weathering/soil Slight to moderate Moderate Intense; soil usually Intense; soil usually
characteristics non-cohesive non-cohesive
Rock mass strength Very good to good Fair Poor to very poor Poor to very poor
Ground stability Stable with no Soil creep and other Inactive landslides Active landslides evident;
identified landslide indications for evident; tension tension cracks, bulges,
scars, either old, possible landslide cracks present terracettes, seepage
recent, or active occurrence are present
present.
Human-initiated May be an aggravating
effects factor

202 Landslides 13 & (2016)

Fig. 2 a MGBs landslide susceptibility map for Tacloban City, Leyte, Philippines, 1:50000 scale (modified from MGB 2006). b MGBs detailed landslide and flood hazard
map of Tacloban City, Leyte, Philippines, 1:10000 scale (modified from MGB 2014)
of evacuation and relocation to prevent repeat of disasters within
the area in the future.
Here, we implement a widely used method to characterize shallow
landslides that can be employed in the nationwide mapping effort as
a rapid shallow landslide assessment method. Using the typhoon-
and landslide-prone Leyte province as a case study, available high-
resolution digital elevation data is utilized to generate shallow land-
slide susceptibility maps. Unlike delineating hazard zones using
historical data, maps generated using a deterministic slope stability
geographic information system (GIS) model incorporates pixel-
specific soil strength parameters and hydraulic characteristics. Also,
it takes full advantage of the fact that landslide source areas are, in
general, strongly controlled by shallow groundwater flow conver-
gence accompanied by the reduction of shear strength and increased
pore water pressure (Pack et al. 2001, 2005).
The most recent approaches to deterministic slope stability
research involves diverse hydrology models using geotechnics at
different levels of complexity (Connell et al. 2001; Dietrich et al.
1992, 1993, 1995, 2001; Grayson et al. 1992a, b; Lan et al. 2004;
Montgomery and Dietrich 1994; Pack et al. 1998, 2001, 2005; Wu
and Sidle 1995). Successful physically based stability model exam-
ples are SHALSTAB (shallow slope stability model; Montgomery
and Dietrich 1994), SINMAP (Stability Index Mapping; Pack et al.
2005; Tarboton 1997), TRIGRS (Transient Rainfall Infiltration and
Grid-Based Regional Slope-Stability Model; Baum et al. 2008), and
SLIP (Shallow Landslide Instability Prediction; Montrasio 2000;
Montrasio and Valentino 2008). Other examples of advanced
hydrology models that involve modelling of groundwater and
vadose zone flows are GEOtop-FS (Rigon et al. 2006), TOPMODEL
(Beven and Kirkby 1979), dSLAM (Wu and Sidle 1995), and
SHETRAN (Burton and Bathurst 1998). SINMAP has been widely
used and tested in several landslide-prone areas because of its
applicability, minimum requirement for input while allowing
multi-calibration of the studied area, and model modification in
smaller areas (Calcaterra et al. 2004; Deb and El-Kadi 2009; Lan
et al. 2004; Meisina and Scarabelli 2007; Morrissey et al. 2001;
Zaitchik and van Es 2003; Zaitchik et al. 2003). It also adds the
capability of representing results into a probabilistic framework.
The reliability and degree of SINMAP model has been tested by
Zizioli et al. (2013) by using receiver operating characteristic (ROC)
analysis as compared to other widely used physically based
models: SHALSTAB, TRIGRS, and SLIP. The values in the ROC
curve (sensitivity against one specificity) represent the ability of a
model to correctly classify between negative and positive observa-
tions. The global accuracy obtained for the SINMAP model
expressed by the area under the curve (AUC) was equal to 0.79.
The global accuracy for SHALSTAB, TRIGRS, and SLIP was 0.78,
0.79, and 0.78 respectively. The values demonstrate that the four
models have similar degrees of success for prediction of shallow
landslide source areas. Considering this, SINMAP model was se-
lected for the case study due to limitation on geotechnical data or
soil strength parameters of Leyte province. With SINMAP, the
spatial variability of soil strength parameters can be expressed by
a range of values.
Landslides 13 & (2016) 203
 Technical Note
After the SINMAP simulation, the maps were then compared to
a landslide inventory derived by interpreting high-resolution sat-
ellite imagery dated 20022014. We then tested for accuracy and
performance to determine the detection percentage (DP) and miss
factor (MF) values. The shallow landslide susceptibility maps can
be complemented with debris flow and structurally controlled
hazard maps to identify safe and unsafe areas for disaster
rehabilitation.
Methods
The Stability INdex MAPping approach to stability index modelling
The methodology (Fig. 3) involves the use of SINMAP, an ArcView
GIS extension and an objective terrain stability mapping tool that
combines steady-state hydrologic concepts with the infinite slope
stability model and incorporates grid-based digital terrain model
(DTM).
The infinite slope stability model balances the resisting com-
ponents of cohesion and the destabilizing components of gravity
on a failure plane parallel to the ground surface. It is applicable to
shallow translational landslides controlled by shallow groundwa-
ter flow convergence. The slope stability theory does not apply to
deep-seated instability (Montgomery and Dietrich 1994).
The input data required to implement the methodology include
topographic slope, specific catchment area, and parameters quan-
tifying material properties (such as soil strength) and climate
(hydrologic wetness parameter). The topographic variables are
computed from the DTM data. SINMAP does not require numer-
ically precise input and accepts a range of soil strength and
hydrologic values to account for uncertainty. In an approximate
sense, this capability predicts landslide zones without additional
Fig. 3 Process flow of shallow landslide susceptibility mapping using SINMAP
204 Landslides 13 & (2016)
input of weather data and analysis by specifying a range of values
for the hydrologic wetness parameter.
Each of the input parameters is delineated on a numerical grid
over the study area. The accuracy of the output is therefore heavily
reliant on the accuracy of the input DTM (Pack et al. 2005).
The primary output of this modelling approach is a stability
index (SI), the numerical representation of the stability of terrain
and hence the probability that a pixel is stable assuming uniform
distributions of parameters over the uncertainty ranges. It does
not predict if shallow translational slope movements will occur,
but it forecasts where they are more likely to initiate given the
assumptions and input parameters used in the analysis. The indi-
ces are not to be interpreted as numerically precise and are most
appropriately interpreted as indications of relative shallow land-
slide susceptibility. SI values range between 0 (most unstable) and 1
(least unstable) (Pack et al. 2001).
The selection of breakpoints of stability index values is subjec-
tive. For mapping purposes in the Philippines, three major sus-
ceptibility ratings are used for shallow landslide susceptibility
maps: high (1.0SI>0), moderate (1.25SI>1.0), and low
(1.5SI>1.25). Stability index values greater than 1.5 are classified
as stable zones. Moreover, SI less than 0 classify regions where the
slope is held in place by forces not represented in the model (e.g.,
bedrock outcrops, man-made slope protection).
SINMAP uses the following formula to calculate the stability
index based on the infinite slope equation proposed by Hammond
et al. (1992).
   
R a
C cos 1min ; 1 r tan
T sin
FS 1
sin
where portions of Leyte. The soil type at most mountainous areas are
undifferentiated.
FS Factor of safety
a Topographic catchment area Digital terrain model grid data
C Dimensionless cohesion=(Cr+Cs)=hsg The study used an Interferometric Synthetic Aperture Radar
Cr Root cohesion (InSAR)-derived DTM of Leyte provided by the National Mapping
Cs Soil cohesion and Resource Information Authority (NAMRIA).
h Soil thickness InSAR-derived products can be obtained from airborne surveys
Soil density using single-pass systems or from space-borne satellite platforms
g Gravity constant using single- and repeat-pass systems (Dowman 2004). This tech-
hw Height of water nique is used to generate a high-resolution DEM suited for broad
R Recharge survey coverage. There are two types of DEM products: the digital
r Water density (w) to soil density (s) ratio surface model (DSM) and the DTM. The former is derived directly
T Soil transmissivity=soil hydraulic conductivityh from the survey and represents the actual earth surface which
Soil internal angle of friction includes vegetation and infrastructure. The DTM is produced by
Slope removing the vegetation and buildings from the DSM using filter-
ing algorithms leaving only a representation of the actual terrain
The variables and a are obtained from the DTM topography. (Dowman 2004).
Other parameters, C (cohesion), (soil angle of friction), T/R The raw data is acquired using Intermaps STAR-3 InSAR sys-
(transmissivity divided by recharge), and r (the ratio of water tem that includes an X-band InSAR mounted on Learjet 36A
and soil density) are manually entered into the model. These are aircraft at a flight height of 412 km above mean sea level and a
the more uncertain parameters and are set in terms of lower and baseline tilt angle of 1.3. One of the obtained raw data from the
upper bound values. The smallest C and tan phi together with the survey is the DSMs. These are processed to generate DTMs with a
largest R/T define the most conservative (worst case) scenario 5-m spatial resolutions and 2- and 1-m horizontal and vertical
within the assumed variability in the input parameters (Pack accuracy, respectively (NAMRIA 2013) The projection datum used
et al. 2001). is the Philippine Reference System of 1992 (PRS92) which is based
on the World Geodetic System of 1984 (WGS84) (Floris et al. 2010;
Soil data of the study areas Geospatial 2011).
Leyte province has an extensive cover of clay soil (Fig. 4). Clay
loam and sand are also present but cover only relatively smaller Geotechnical data
Parameters not determined by the DTM such as soil strength,
cohesion, angle of friction, transmissivity, and hydraulic conduc-
tivity are considered more uncertain and are specified in terms of
upper and lower boundary values (Pack et al. 2001). The calibra-
tion parameters used for the simulation encompasses the typical
soil parameter values for clay and clay loam, which are the dom-
inant soil types in Leyte province (Table 2).
The equation used to determine the dimensionless cohesion
combines root and soil cohesion. Theoretically, this is the ratio of
the cohesive strength of the roots and soil relative to the weight of
a saturated thickness of soil (Pack et al. 2005).

C r C s
C 2
hs g

Cr Root cohesion
Cs Soil cohesion
h Soil thickness
s Soil density
g Acceleration due to gravity

Internal angle of friction is a measure of the shear strength of

Fig. 4 Leyte soil map (digitized from BSWM 2014)
soil due to friction determined in the laboratory using direct shear
strength or triaxial stress test.
The effective recharge rates used in this study have been de-
rived from lower and upper precipitation limits (i.e., 50 and
200 mm/day). A 50-mm/day rain rate was chosen as a minimum
rate and 200 mm/day was used as the maximum (i.e., an extreme

Landslides 13 & (2016) 205

 Technical Note
Table 2 Soil calibration parameters for SINMAP simulation
Soil density Cohesion
(kg/m3) Min Max
Clay to clay loam 1900 0 0.8
example of the precipitation that can produce shallow translation-
al movement). This range matches the maximum accumulated
rainfall of 440 mm for a 72-h period or 147 mm/day on average
(MapAction 2013).
The transmissivity rate (m2/h) was calculated using the equa-
tion, T=Kb, where K is the hydraulic conductivity of the soil and b
is the soil depth in meters. In the context of shallow landslide
movement, soil thickness of 1.5 m is assumed to be uniform in the
simulation. Data on hydraulic conductivity is obtained from a
database of soils in the Philippines (Hamazaki et al. 1990). Field
and laboratory descriptions and test results, along with literature
values for soil properties given in Hammond et al. (1992) were also
used to constrain reasonable ranges of soil input parameters for
the stability index modelling.
Validation
The validation process of the generated SINMAP model is sum-
marized on Fig. 5. Landslide inventory map consisting of 363
landslides was used to validate the SINMAP model for Leyte
province. The inventory utilized remote sensing techniques using
various satellite images that have imagery dates from 2002 to 2014.
Satellite imagery includes SPOT of 5 to 10 m spatial resolution,
QuickBird of 0.6 to 2.4 m, and WorldView of 0.5 m. These images
were inspected using manual detection method to identify land-
slide events in Google Earth, freeware program, depending on its
availability.
Satellite images were loaded and inspected in Google Earth.
Detected landslide events were delineated with polygons by outlining
the visible disturbed body. Each identified landslide was classified
based on Cruden and Varnes classification (Varnes 1958, 1978). The
types of landslides indentified were mainly slides and flows. The
Fig. 5 Process flow of validation of SINMAP model
206 Landslides 13 & (2016)
Internal angle of friction Approximate T/R
Min Max Min Max
25 35 20 200
assigned polygons were saved as vector files and analyzed in ArcGIS.
The landslide inventory statistics including number of landslide,
total surface area damage, and density per locality were computed
using the satellite images of Leyte province.
Visual comparison was done with the SINMAP model and the
landslide inventory. Accuracy assessment is performed by calcu-
lation of detection percentage and miss factor.
Results
Visual comparison of model and landslide inventory
The area distribution for low, medium, and high susceptibility to
shallow landslides generated by SINMAP is shown in Fig. 6a.
Unsafe areas from shallow landslides were found to be 31 % of
the total land area of Leyte. Safe areas from shallow landslides
comprise 69 %. The previous 1:10,000 landslide susceptibility map
by MGB (Fig. 2b) presents noticeably less stable areas compared to
the SINMAP model (54 %) as shown in Fig. 6b.
Initially, visual comparison was done by manual overlaying of
the landslide inventory over the SINMAP model and MGB map.
The number of landslide polygons that fall within each suscepti-
bility class is tabulated (Tables 3 and 4). Among the 363 landslide
polygons in Leyte, 9 were within the stable zones of the SINMAP
model (Table 3). Majority of the polygons fell within the medium
(66.1 %) susceptibility zones. For the MGB map, most polygons fell
within the high and very high susceptibility zones (Table 4).
Accuracy assessment
The number and extent of mapped landslides are critical to disas-
ter relief and deterministic landslide assessment (Yange et al.
2013). Therefore, in this procedure, the susceptibility zones defined
Fig. 6 Distribution of landslide susceptibility by area in Leyte province generated by SINMAP (a) and MGB (b)
by the SINMAP model were compared with the manually delin-
eated landslide inventory. This accuracy assessment procedure is
also done to the MGB map to compare its accuracy with the
SINMAP model.
By visual comparison, the landslide polygons were either
marked as true positive or false negative using the following
category definitions:
& True positive (TP): a landslide event occurred at the area of
instability defined by SINMAP
& False negative (FN): a landslide event occurred outside the area
of instability defined by SINMAP
Once the number of landslide polygons belonging to each
susceptibility category is determined, the performance of the
SINMAP model for Leyte was evaluated using the following sta-
tistical measures:
M F FN=TP
TP
DP 100 
TP FN
The miss factor (MF) is a measure of omission error where the
SINMAP model defined actual landslide areas as stable. The de-
tection percentage (DP) is the percentage of landslide areas cor-
rectly classified by the SINMAP model as unstable (Lee et al. 2003;
Shufelt 1999).
The 354 polygons captured by the SINMAP model were consid-
ered as TP, whereas the 9 landslide polygons outside the model
were considered as FN. For the MGB map, 351 polygons are
captured (TP) and 12 are outside the susceptibility zones (FN).
Table 3 Summary of landslide polygons captured by SINMAP model for Leyte province
Stable Low
No. of landslides 9 43
% of slides 2.5 11.8
The SINMAP results showed that the DP for the study area was
97.5 % with MF of 0.025. These values demonstrate that the
SINMAP approach performs well despite the lack of an extensive
geotechnical and hydrological database in the study area. This also
suggests a consistency in the accuracy of the InSAR-derived DTM
and the high-resolution imagery used for the landslide polygon
delineation. The MGB map showed a DP of 96.7 % and MF of
0.034. The accuracy of the MGB map is high which is typical for
qualitative methods involving historical landslide data and rele-
vant geomorphological information. Both susceptibility mapping
methods establish high accuracy when compared to the landslide
inventory; however, they differ much in the amount of stable areas
they define.
Discussion
The SINMAP simulations for Leyte (Fig. 7) capture very well the
landslides that have been mapped by high-resolution imagery over
the span of 12 years (20022014). Almost all of the 363 landslides in
the inventory of Leyte fall within the high (red) and moderately
3 (orange) susceptible landslide areas. Only 2.5 % of the landslide
events were not captured by the landslide susceptibility maps
generated by SINMAP (Tables 3 and 4).
4
The stable areas identified by SINMAP within Leyte is 69 % of
its total land area, 15 % more than the safe areas identified by
previous landslide susceptibility map generated by the MGB
(Fig. 2b) which is 54 % of the total Leyte area (Fig. 6). Nonetheless,
the accuracy of the SINMAP results remain high, with the 97.5 %
DP value and a MF of 0.025.
The idea of using a model that identifies landslide susceptible
areas with high accuracy, and at the same time, maximizing safe
and habitable areas is beneficial to the development plans of the
local government. In post-disaster rehabilitation, it also expands
options for siting evacuation and relocation areas. This is critical
Medium High Total
240 71 363
66.1 19.6 100
Landslides 13 & (2016) 207
 Technical Note
Table 4 Summary of landslide polygons captured by MGB map for Leyte province
Stable Low Medium Highvery high Total
No. of landslides 12 0 61 290 363
% of slides 3.3 0 16.8 79.9 100

especially for the rehabilitation work of Haiyan-affected commu-
nities because it is imperative that rebuilding be carefully planned
to prevent future disasters.
The landslides that were not captured by the SINMAP simula-
tions for Leyte could be due to a different failure mechanism not
considered by the infinite slope stability model. For instance,
SINMAP does not consider structural controls for slope instability.
These landslides occur much deeper and are controlled primarily by
fracture planes that pervade in mountainous terrain. Structurally
controlled landslide maps are also being prepared under the new
Nationwide Hazards Mapping Program called Project NOAH where
high-resolution topographic maps derived from Light Detection and
Ranging (LiDAR) and InSAR are used. Along with the identification
of alluvial fans and simulation of debris flows, the SINMAP and the
structurally controlled landslide hazard maps will constitute the
detailed hazard maps at 1:10,000 scale for landslides in Haiyan-
devastated areas. These maps are available online for free in the
Project NOAH portal (http://noah.dost.gov.ph).
It is critical to provide rapid assessment and delineation of
areas susceptible to rainfall-induced shallow landslides for well-
planned reconstruction and rehabilitation. The method of shallow
landslide susceptibility mapping we implement in the Philippines
addresses the need for rapidly generated landslide hazard maps.
These shallow landslide hazard maps have been made freely avail-
able to different relief and rehabilitation agencies in typhoon
Haiyan-ravaged areas. The shallow landslide maps that were pro-
vided are suited to the requirements of the rehabilitation and
reconstruction process. It can still be and should be improved
for better accuracy. More detailed geological and geotechnical
assessments are necessary in the future. One area of improvement
can be the replacement of InSAR-derived DTM data with more
accurate LiDAR topographic data.
Conclusion
The landslide map generated using SINMAP are found to be highly
accurate with DP value of 97.5 % and MF of 0.025 when compared

Fig. 7 Shallow landslide susceptibility map of Leyte province with a close up view in an area with identified landslides

208 Landslides 13 & (2016)

with the landslide inventory derived from high-resolution satellite
imagery for Leyte. In addition, the maps can be further improved
with more detailed geological and geotechnical assessment of the
study areas. Additional satellite and aerial imagery can provide
more landslide inventory data to better determine the spatial
extent of landslide occurrences vis--vis the results of the SINMAP
model. The landslide susceptibility classification found in the
landslide hazard maps is useful to identify unsafe and safe areas
from shallow landslides. These maps complement the debris flow
and structurally controlled landslide hazard maps that are also
being prepared for rebuilding Haiyan-devastated areas.
Acknowledgments
The work described in this paper is supported by the Landslide
Hazard Mapping Component of Project NOAH (Nationwide Opera-
tional Assessment of Hazards) under the initiatives of the Depart-
ment of Science and Technology (DOST) for an improved disaster
prevention and mitigation system in the Philippines. This support
from the Project NOAH team is gratefully acknowledged. The au-
thors are also grateful to the National Mapping and Resource Infor-
mation Authority (NAMRIA) for providing the digital terrain model,
Bureau of Soil and Water Management (BWSM) for the digitized soil
maps, and the reviewers for the valuable comments.
References
AGS (2007) Guideline for landslide susceptibility, hazard, and risk zoning for land use
planning. Aust Geomech 42(1):1336
Antronico L, Gull G (2000) Slopes affected by soil slips: validation of an evolutive model.
8th International Symposium on Landslides, Cardiff, Wales. Thomas Telford, London,
pp 7784
Baum RL, Savage WZ, Godt JW (2008) TRIGRSa FORTRAN program for transient rainfall
infiltration and grid-based regional slope stability analysis, version 2.0 US Geological
Survey Open File Report 2008-1159. Reston, Virginia
Beven KJ, Kirkby MJ (1979) A physically-based variable contributing area model of basin
hydrology. Hydrol Sci Bull 24(1):4369. doi:10.1080/02626667909491834
BSWM (2014) Soil and physiography map of Leyte. Bureau of Soils and Water Manage-
ment (BSWM) Map Library Platform. World Bank and Department of Agriculture.
http://www.bswm.maps.da.gov.ph/maps-library. Accessed 15 August 2014
Burton A, Bathurst JC (1998) Physically-based modelling of shallow landslide sediment
yield at a catchment scale. Environ Geol 35(2-3):8999. doi:10.1007/s002540050296
Calcaterra D, de Riso R, Di Martire D (2004) Assessing shallow debris slide hazard in the
Agnano Plain (Naples, Italy) using SINMAP, a physically based slope-stability model. In:
Lacerda WA, Ehrlich M, Fontoura SAB, Sayao ASF (eds) 9th International Symposium on
Landslides, Rio de Janeiro Brazil. Taylor and Francis Group, London, pp 177183
Campus S, Forlati F, Sarri H, Scavia C (2001) Shallow landslides hazard assessment based
on multidisciplinary studies. In: Ho KKS, Li KS (eds) 14th Southeast Asian Geotechnical
Conference Hong Kong. AA Balkema, Rotterdam, pp 703708
Catani F, Casagli N, Ermini L, Righini G, Menduni G (2005) Landslide hazard and risk
mapping at catchment scale in the Arno River basin. Landslides 2(4):329342.
doi:10.1007/s10346-005-0021-0
Coe JA, Godt JW, Baum RL, Buckman RC, Michael JA (2004) Landslide susceptibility from
topography in Guatemala. In: Lacerda WA, Ehrlich M, Fontoura SAB, Sayao ASF (eds)
Landslides evaluation and stabilization, 10th Symposium on Landslides, Balkema
Leiden. Taylor and Francis Group, London, pp 6978
Connell LD, Jayatilaka CJ, Nathan R (2001) Modelling flow and transport in irrigation
catchments: 2. Spatial application of subcatchment model. Water Resour Res
37(4):965977. doi:10.1029/2000WR900269
DAmato Avanzi G, Giannecchini R, Puccinelli A (2003) A contribution to an evaluation of
landslide susceptibility in the Apuan Alps (Italy): geologic and geomorphic factors of
1996 soil slip-debris flows. In: Picarelli L (ed) International Conference, Fast Slope
Movements: Prediction and Prevention for Risk Mitigation, Naples Italy. Patron
Editore, Bologna, pp 125130
DAmato Avanzi G, Giannecchini R, Puccinelli A (2004) The influence of the geological
and geomorphological settings on shallow landslides. An example in a temperate
climate environment: the June 19, 1996 event in northwestern Tuscany (Italy). Eng
Geol 73(3-4):215228. doi:10.1016/j.enggeo.2004.01.005
Deb SK, El-Kadi AI (2009) Susceptibility assessment of shallow landslides on Oahu,
Hawaii, under extreme-rainfall events. Geomorphology 108(3-4):219233.
doi:10.1016/j.geomorph.2009.01.009
Dietrich WE, Wilson CJ, Montgomery DR, McKean J, Bauer R (1992) Erosion thresholds
and land surface morphology. J Geol 20(8):675679. doi:10.1130/0091-
7613(1992)020<0675:ETALSM>2.3.CO;2
Dietrich WE, Wilson CJ, Montgomery DR, McKean J (1993) Analysis of erosion thresholds,
channel networks, and landscape morphology using a digital terrain model. J Geol
101:259278
Dietrich WE, Reiss R, Hsu M, Montgomery DR (1995) A process-based model for colluvial soil
depth and shallow landsliding using digital elevation data. Hydrol Process 9:383400
Dietrich WE, Bellugi D, Real De Asua R (2001) Validation of shallow landslide model,
SHALTAB, for forest management. In: Wigmosta MS, Burges SJ (eds) Land use and
watersheds: human influence on hydrology and geomorphology in urban and forest
areas. American Geophysical Union, pp 195-227
Dowman I (2004) Integration of Lidar and IfSAR for mapping. Int Arch Photogramm
Remote Sens, Istanbul, Turk 35(B2):90100
Fabbri AG, Chung CF, Cendrero A, Remondo J (2003) Is prediction of future landslides possible
with a GIS? Nat Hazards 30(3):487503. doi:10.1023/B:NHAZ.0000007282.62071.75
Floris M, Squarzoni C, Hundseder C, Mason M, Genevois R (2010) The use of IfSAR data in
GIS-based landslide susceptibility evaluation. Geophys Res Abstr 12
Geospatial World (2011) PRS92 surveying standard compulsory in the Philippines.
Geospatial World. http://geospatialworld.net/News/View.aspx?id=21573_Article.
Accessed 29 May 2015
Giannecchini R (2006) Relationship between rainfall and shallow landslides in the
southern Apuan Alps (Italy). Nat Hazards Earth Syst Sci 6:357364. doi:10.5194/
nhess-6-357-2006
Grayson RB, Moore ID, McMahon TA (1992a) Physically based hydrologic modelling: 1. A
terrain-based model for investigative purposes. Water Resour Res 28(10):26392658.
doi:10.1029/92WR01258
Grayson RB, Moore ID, McMahon TA (1992b) Physically-based hydrologic modelling: 2. Is
the concept realistic? Water Resour Res 28(10):26592666. doi:10.1029/92WR01259
Hamazaki T, Paningbatan EP Jr, Pampolino MF (1990) Data base on red-yellow and
related soils in the Philippines: Part 2 Visayas and Mindanao soils. Technical Paper at
College of Agriculture. University of the Philippines Los Baos and Tropical Agriculture
Research Center, Tokyo
Hammond C, Hall D, Miller S, Swetik P (1992) Level I stability analysis (LISA) documen-
tation for Version 2.0: General Technical Report INT-285. USDA Forest Service,
Intermountain Research Station
Hong Y, Adler R, Huffman G (2007) Use of satellite remote sensing data in the mapping of
global landslide susceptibility. Nat Hazards 43(2):245256. doi:10.1007/s11069-006-9104-z
JTWC (2013) Joint Typhoon Warning Center Best Track ArchiveTyphoon Haiyan. Naval
Meteorology and Oceanography Command, Stennis Space Center, Mississippi. http://
www.usno.navy.mil/JTWC/. Accessed 30 November 2013
Knapp KR, Kruk MC, Levinson DH, Diamond HJ, Neumann CJ (2010) The International
Best Track Archive for Climate Stewardship (IBTrACS): unifying tropical cyclone best
track data. Bull Am Meteorol Soc 91:363376. doi:10.1175/2009BAMS2755.1
Lagmay AM, Agaton RP, Bahala MAC, Briones JBLT, Cabacaba KMC, Caro CVC, Dasallas LL,
Gonzalo LAL, Ladiero CH, Lapidez JP, Mungcal MTF, Puno JVR, Ramos MMAC,
Santiago J, Suarez JK, Tablazon JP (2015) Devastating storm surges of typhoon
Yolanda. Int J Disaster Risk Reduct 11:112. doi:10.1016/j.ijdrr.2014.10.006
Lan HX, Zhou CH, Wang LJ, Zhang HY, Li RH (2004) Landslide hazard spatial analysis and
prediction using GIS in the Xiaojiang watershed, Yunnan, China. Eng Geol 76(1-
2):109128. doi:10.1016/j.enggeo.2004.06.009
Lee DS, Shan J, Bethel JS (2003) Class-guided building extraction from IKONOS imagery.
Photogram Eng Remote Sens 69(2):143150
Liao Z, Hong Y, Wang J, Fukuoka H, Sassa K, Karnawati D, Fathani F (2010) Prototyping
an experimental early warning system for rainfall-induced landslides in Indonesia
using satellite remote sensing and geospatial datasets. Landslides 7(3):317324.
doi:10.1007/s10346-010-0219-7
MapAction (2013) Philippines typhoon Haiyan (Yolanda) accumulated rainfall 06-Nov-
2013 to 09-Nov-2013. Situational data source from: PAGASA, Joint Typhoon Warning
Centre Philippines. http://mapaction.org. Accessed 24 March 2014
Meisina C, Scarabelli S (2007) A comparative analysis of terrain stability models for
predicting shallow landslides in colluvial soils. Geomorphology 87(3):207223.
doi:10.1016/j.geomorph.2006.03.039
Landslides 13 & (2016) 209
 Technical Note
MGB (2006) Landslide susceptibility map of Tacloban City quadrangle Leyte Philippines.
Department of Environment and Natural Resources Mines and Geosciences Bureau
(MGB) Lands Geological Survey Division. Sheet Number 3953 I. http://
gdis.denr.gov.ph/mgbpublic. Accessed 1 October 2014
MGB (2013) Guidebook for the conduct of landslide and flood susceptibility assessment
and mapping (1:10000). Department of Environment and Natural Resources Mines
and Geosciences Bureau (MGB) Lands Geological Survey Division. North Avenue,
Quezon City, Philippines
MGB (2014) Detailed landslide and flood hazard map of Tacloban City, Leyte, Philip-
pines. Department of Environment and Natural Resources Mines and Geosciences
Bureau (MGB) Lands Geological Survey Division. North Avenue, Quezon City,
Philippines
Montgomery D, Dietrich W (1994) A physically based model for the topographic control
on shallow landsliding. Water Resour Res 30(4):11531171. doi:10.1029/93WR02979
Montrasio L (2000) Stability analysis of soil slip. Proc of Int Conf Risk 2000, In: Brebbia CA
(ed) Wit Press, Southampton, pp 357-366
Montrasio L, Valentino R (2008) A model for triggering mechanism of shallow landslides.
Nat Hazards Earth Syst Sci 8:11491159. doi:10.5194/nhess-8-1149-2008
Morrissey MM, Wieczorek GF, Morgan BA (2001) A comparative analysis of hazard
models for predicting debris flows in Madison Country, Virginia. Open file report
01-0067. USGS (ed) US Department of the Interior, US Geological Survey, Washington
Nadim F, Kjekstad O, Peduzzi P, Herold C, Jaedicke C (2006) Global landslide and
avalanche hotspots. Landslides 3(2):159173. doi:10.1007/s10346-006-0036-1
NAMRIA (2013) The Philippine IfSAR project. Internal report jointly by NAMRIA, Intermap
Technologies Inc. (Denver CO) and Certeza Infosys Corp. NAMRIA Main Office, Taguig
City, Philippines
NDRRMC (2014) National Disaster Risk Reduction and Management Council (NDRRMC)
Updates regarding the effects of Typhoon BYolanda^ (HAIYAN) 17 April 2014.
National Disaster Risk Reduction and Management Center, Camp Aguinaldo, Quezon
City, Philippines. http://www.ndrrmc.gov.ph/index.php/21-disaster-events/1329-situ-
ational-report-re-effects-of-typhoon-yolanda-haiyan. Accessed 2 June 2014
Pack RT, Tarboton DG, Goodwin CN (1998) The SINMAP approach to terrain stability
mapping. In: 8th Congress of the International Association of Engineering Geology,
Vancouver, British Columbia, Canada 21-25 September 1998. http://
www.engineering.usu.edu/cee/faculty/dtarb/iaeg.pdf. Accessed 4 May 2013
Pack RT, Tarboton DG, Goodwin CN (2001) Assessing terrain stability in a GIS using
SINMAP. In: GIS 2001. 15th annual GIS conference, Vancouver, British Columbia.
http://hydrology.usu.edu/sinmap/gis2001.pdf. Accessed 4 May 2013
Pack RT, Tarboton DG, Goodwin CN, Prasad A (2005) SINMAP 2: a stability index
approach to terrain stability hazard mapping technical description and users guide
for version 2.0. Utah State University. http://hydrology.usu.edu/sinmap2/
sinmap2.PDF. Accessed 4 May 2013
Rigon R, Bertoldi G, Over TM (2006) GEOtop: a distributed hydrological model with coupled
water and energy budgets. J Hydrometeorol 7(3):371388. doi:10.1175/JHM497.1
Sarkar S, Kanugo DP (2004) An integrated approach for landslide susceptibility mapping
using remote sensing and GIS. Photogramm Eng Remote Sens 70(5):617625
Shufelt JA (1999) Performance evaluation and analysis of monocular building extraction
from aerial imagery. IEEE Trans Pattern Anal Mach Intell 21(4):311326
Soeters R, Van Westen CJ (1996) Slope instability, recognition, analysis and zonation. In:
Turner AK, Schuster RL (eds) Landslides: investigation and mitigation Special Report
247. Transportation Research Board. National Research Council. National Academy
Press, Washington, pp 129177
Tarboton DG (1997) A new method for the determination of flow direction and upslope
areas in grid digital elevation models. Water Resour Res 33(2):309319. doi:10.1029/
96WR03137
210 Landslides 13 & (2016)
Van Asch TWJ, Buma J, Van Beek LPH (1999) A view on some hydrological triggering
systems in landslides. Geomorphology 30(1):2532
Varnes DJ (1958) Landslide types and processes. In: Eckel EB (ed) Landslides and
engineering practice, Special Report 29. Highway Research Board National Research
Council, Washington DC, pp 2047
Varnes DJ (1978) Slope movement types and processes. In: Schuster RL, Krizek RJ (eds)
Landslides: analysis and control, Special Report 176. Transportation Research Board
National Academy of Sciences, Washington DC, pp 1133
Wieczorek GF (1987) Effect of rainfall intensity and duration on debris flows in central
Santa Cruz Mountains, California. In: Costa JE, Wieczorek GF (eds) Debris flows/
avalanches: processes, recognition and mitigation. Reviews in Engineering Geology,
vol 7. Geological Society of America, pp 23-104
Wieczorek GF, Sarmiento J (1988) Rainfall, piezometric levels and debris flows near La
Honda, California, in storms between 1975 and 1983. In: Ellen SD, Wieczorek GF (eds)
Landslides, floods and marine effects of the storm of January 3-5, 1982 in the San
Francisco Bay vol 1434. USGS Professional Paper, pp 43-62
Wu W, Sidle R (1995) A distributed slope stability model for steep forested basins. Water
Resour Res 31:20792110
Yange L, Guangqi C, Bo W, Lu Z, Yingbin Z, Chuan T (2013) A new approach of
combining aerial photography with satellite imagery for landslide detection. Nat
Hazards 66:649669. doi:10.1007/s11069-012-0505-x
Zaitchik BF, van Es HM (2003) Applying a GIS slope-stability model to site-specific
landslide prevention in Honduras. J Soil Water Conserv 58(1):4553
Zaitchik BF, van Es HM, Sullivan PJ (2003) Modeling slope stability in Honduras:
parameter sensitivity and scale of aggregation. Soil Sci Soc Am J 67(1):268278
Zizioli D, Meisina C, Valentino R, Montrasio L (2013) Comparison between different
approaches to modelling shallow landslide susceptibility: a case history in Oltrepo Pavese,
Northern Italy. Nat Hazards Earth Syst Sci 13:559573. doi:10.5194/nhess-13-559-2013
R. P. Felix : A. M. F. Lagmay : R. N. C. Eco : I. J. G. Ortiz
National Institute of Geological Sciences,
University of the Philippines,
Diliman 1101, Quezon City, Philippines
M. L. Rabonza ()) : R. P. Felix : A. M. F. Lagmay : R. N. C. Eco : I. J. G.
Ortiz : D. T. Aquino
Nationwide Operational Assessment of Hazards, Department of Science and Technology,
Quezon City, Metro Manila, NCR, Philippines
e-mail: mrabonza@noah.dost.gov.ph
R.P. Felix
email: raquel.felix@noah.dost.gov.ph
A.M.F.A. Lagmay
email: mlagmay@noah.dost.gov.ph
R.N.C. Eco
email: narod.eco@noah.dost.gov.ph
I.J.G. Ortiz
email: iris.ortiz@noah.dost.gov.ph
D.T. Aquino
email: daquino@noah.dost.gov.ph