Professional Documents
Culture Documents
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
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
Fig. 4 A typical retrogressive thaw slump in burned area southeast of Inuvik. Note 2009 7.7 1306
dead trees in background (aerial photo of G1, Sept. 23, 2004)
(a) (b)
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
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 Fig. 10 Illustration of thaw slump process
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 As discussed earlier, the average air temperatures for 2008 and
walls. The slope orientation or the scarp facing direction at each 2009 were very close to that of 2007. It is therefore unlikely that
monitored location was measured as well. Table 3 provides the 3- the air temperature contributed to the increase of the upper limit
year monitoring data. of the data. Note that, based on the slightly lower air thawing
indices in 2008 and 2009 (Table 2), the retrogression rates should
Retrogression correlation have dropped slightly, which was actually not the case. In fact, as
An initial correlation between the thaw slump retrogression rate indicated in Table 3, amongst the 16 locations at the 13 sites
and the height of the scarp wall was identified by Wang et al. monitored in 2007, the retrogression rate dropped at five locations
(2009), which was from the first year (2007) monitoring data. and increased at seven locations in both 2008 and 2009, while the
Continued monitoring during the subsequent 2 years (2008 and other four locations experienced a drop in 1 year and an increase
2009) enhanced the data. The annual retrogression rates for all in another year. Also note from Table 3 that the six monitoring
3 years were plotted against the height of the scarps as shown in points at five slumps added in 2008 and 2009 indicated near-
Fig. 8. As noted from this figure, the distribution of the 20082009 average retrogression rates. Therefore, the increase of the upper
data followed the same pattern as established by the 2007 data. bound limit would not have been attributed to the newly added
However, the upper bound curve moved higher with the addition sites.
of the 2008 and 2009 data. The lower bound curve did not change. Nevertheless, given the 3-year duration of measurements and
The 3-year data are more scattered than the first year data alone. the near normal average air temperature conditions, the
As noted in Wang et al. (2009), the increased scattering was retrogression rates collected are believed to be representative of
expected with more data becoming available. the normal rates in the region. Based on this set of data, the
Where R denotes scarp wall retrogression rate (m/year); H is Survey stake G1-STK3/G2-STK1
height of scarp wall (m) and A, B and C are fitting constants with
values given in Table 4. G2
G1
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
Fig. 12 Two retrogressive thaw slumps (G1 and G2) retreating towards each other
retreat is associated with thawing of ice-rich permafrost. (photo taken: Sept. 8, 2007). A survey stake (labelled as G1-STK3 and G2-STK1) was
Ablation continues along the scarp wall in summer as long as installed between the two slumps. The south-facing scarp wall of G1 and the north-
there is enough ice-rich material in the wall. Depending on the facing scarp wall of G2 retreated at the same rate in 2007 and 2008. The two
water content, the ablated materials build up at the toe of the wall slumps merged together by September 2008 (see Fig. 13)
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 However, as the debris builds up, its weight increases. When the
reasons could be attributed to the debris covering the ice-rich wall buildup is high enough, the mass slides away driven by its own
from time-to-time. The materials in the active layer have a lower weight. Thus, the covered ice-rich permafrost along the face of the
moisture content compared to those of the ice-rich permafrost. wall is re-exposed, and thawing is accelerated until it is covered
The ice-rich soil behaves like slurry when thawed and flows away again. This process is illustrated in Fig. 10. The cycle repeats
relatively quickly from the scarp (Wang et al. 2005). When the throughout the summer.
slurry is mixed with the slumped materials from the active layer, In summary, the maximum scarp retrogression rate ranged
its fluidity decreases, resulting in a higher and steeper debris from about 8 m/year to about 15 m/year. The retrogression rate
buildup at the toe. The debris buildup reduces the area of ice- was found to increase with increased scarp wall height. However,
rich permafrost being exposed and therefore a reduced amount of once the scarp reaches a certain height, say about 6 to 8 m (Fig. 8),
slurry is generated. This further decreases the fluidity of the the increase of the retrogression rate starts to taper off.
slurry mix and hence more sediment accumulation at the toe.
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
Fig. 11 Distribution of scarp retrogression rate and scarp wall facing direction taken: Sept. 10, 2008)
Acknowledgments
B. Wang ()) : B. Paudel : H. Li
The authors would like to thank Wendy Sladen for her careful
review and comments on the manuscript that helped improve the Natural Resources Canada,
quality of the paper. This paper was released under an internal ESS Ottawa, Ontario K1A 0E8, Canada
Contribution Number 20140195. e-mail: bwang@NRCan.gc.ca
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 the potentialities of rockslide, rockfall, and debris flow. This re-
village in Dujianyan city as an example to investigate the forma- sulted in a misunderstanding in lower risk of geological hazards.
tion mechanism of a rapid and long run-out rockslide-debris flow This area enjoyed a high reputation stemming partly from the
of fractured/cracked slope, under the application of a rare heavy Qincheng MountainDujianyan Scenic Resort and attracted a
rainfall in July 2013. The slope site could be affected by the great many tourists every summer. Therefore, research on
Wenchuan Ms 8.0 Earthquake in 2008. The sliding involved the rainfall-induced post-earthquake geological hazards in these areas
thick fractured and layered rockmass with a gentle dip plane at has significance to the reconstruction of towns and villages and the
Sanxicun. It had the following formation process: (1) toppling due tourism safety after the earthquake. The rockslide dynamic re-
to shear failure at a high-level position, (2) shoveling the accumu- sponse characteristics triggered by the Wenchuan Ms 8.0 earth-
lative layer below, (3) forming of debris flow of the highly weath- quake and post-earthquake debris flow disaster have received
ered bottom rockmass, and (4) flooding downward along valley. much attention from researchers (Yin et al. 2009; Tang et al.
The debris flow destroyed 11 houses and killed 166 people. The 2011). Tang et al. (2009) reported a heavy-rainfall-induced debris
run-out distance was about 1200m, and the accumulative volume flow disaster on September 24 of 2008 at the earthquake-ruined
was 1.9106m3. The rockslide can be divided into sliding source, Beichuan town in the central part of the area affected by the
shear-shoveling, and flow accumulative regions. The stability of Wenchuan Ms 8.0 earthquake and presented the possibility that,
this fractured rock slope and the sliding processes are discussed at due to the ground shaking, the critical amount of accumulated
four stages of cracking, creeping, separating, and residual accu- precipitation and the hourly rainfall intensity necessary to initiate
mulating, under the applications of hydrostatic pressure and uplift debris flow were reduced compared with values before the
pressure. This research also investigates the safety factors under earthquake. Recently, Zhou and Tang (2013) summarized 11 rainfall
different situations. The double rheological model (F-V model) of events that induced debris flows between 2008 and 2012 after the
the DAN-W software is utilized to simulate the kinematic and Wenchuan earthquake and discussed the rainfall thresholds for
dynamic processes of the shear-shoveling region and debris flow. debris flow occurrence. However, the researches on rainfall-
After the shear failure occurred at a high-level position of rock, the induced rockslides after the earthquake and its transferring pro-
rockslide moved for approximately 47s downward along the valley cess into a long run-out debris flow have made little significant
with a maximum velocity of 35m/s. This is a typical rapid and long progress.
run-out rockslide. Finally, this paper concludes that the identifi- The parent rock of the Sanxicun rockslide is a slightly inclined,
cation of the potential geological hazards at the Wenchuan moun- nearly horizontal strata sandstone formation with intact rock
tain area is crucial to prevent catastrophic rockslide triggered by constitutes. This rockslide was located at southwest of Wenchuan
heavy rainfall. The identified geological hazards should be prop- Ms 8.0 earthquake fault and was only 12 km away from the
erly considered in the town planning of the reconstruction works. epicentral region. The seismic intensity of the Sanxicun area was
up to the IX degree. According to the emergency investigation
Keywords Wenchuan Ms 8.0 earthquake . 1 week after the Wenchuan Ms 8.0 earthquake, three small scale
Heavy rainfall . Rockslide . Debris flow rockfalls were induced. Two rockfalls were at the frontier of the
slope of the rockslide source area in the moving direction of
Introduction NNE10, and the other was at the southeastern corner in the
During July 712 of 2013, a heavy rainfall occurred at the moving direction of NWW288 (Fig. 2, also see Fig. 8). They
Wenchuan Ms 8.0 earthquake area. The 5 days accumulative pre- presented a disturbance, or even damage on the slope from the
cipitation in the Dujiangyan area of Sichuan province was up to earthquake. Research on this rockslide is significant for prevention
1129 mm, which was rare in previous years and exceeded the mean on post-earthquake geological hazard prevention. Therefore, this
annual precipitation of 1110 mm over the 25 years from 1987 to paper uses the Sanxicun landslide as an example to study the
2012. The heavy rainfall triggered a rapid and long run-out stability of fractured slope after the Wenchuan Ms 8.0 earthquake
rockslide in Sanxicun village. Its debris had a volume of 1.9 million and the formation mechanism of rapid and long run-out
m3. It caused a sliding distance up to 1200 m and killed 166 people rockslides triggered by heavy rainfall. The adverse conditions of
in the village. The Sanxicun landslide was located in Zhongxing the geological features prior to the rockslide are investigated,
town, Dujiangyan city of Sichuan province. The site is located at which provide a scientific consideration for classifying hazard
(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 Environmental geological condition
natural mountain scenery, lakes, and abundant vegetation. The In terms of geomorphology, the Sanxicun rockslide was located at
typical natural conditions to resorts and vocations often concealed the transition zone from the Chengdu Plain to the QinghaiTibet
1200 m, and the rockslide area was 18.1103 m2. The maximum and 54 m with a mean thickness of 40 m. The longitudinal length
thickness of the rockslide accumulation was 30 m with a mean of this area ranges between 50 and 200 m with a mean value of
thickness of 10 m and a volume of 1.915106 m3. The movement 100 m. The plane of this area exhibits an irregular trapezoidal
and accumulative characteristics of the rockslide can be classified shape whose width is between 250 and 420 m with a mean value
into rockslide source area, impact-shoveling area, and flow accu- of 335 m. The mean area is 33.5103 m2, and the estimated volume
mulative area (Figs. 6 and 7) that will be described in detail in the is 1.47106 m3.
following three paragraphs. 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
Rockslide source area applied, i.e., the Ie is suggested that the volume of the accumulation is
Rockslide source area was located approximately 100 m northwest about 1.21.4 times the volume of the original rockmass of the
from the downside slope of the Shangqing Palace Building in the rockslide (Yin et al. 2011). Therefore, by multiplying with a rockmass
Qingcheng Mountain Scenery Area. The terrain of this area is steep loose factor of 1.2, the accumulative volume of the rockslide source
with a slope of approximately 30. This area mainly consists of area can be computed as 1.21.47106 m3=1.77106 m3.
sandstone and siltstone. The surface is comprised of clay gravel According to the estimate from remote sensing images prior
soil covered by a Quaternary loose residual layer with a thickness and post-sliding, the entirety of the rockslide body moved down-
of <2 m. ward about 50 to 80 m. A sliding mass with a volume of approx-
The boundaries of the east and the west and the rear edge of the imately 300103 m3 was sheared out at a high-level position and
rockslide were controlled by joints, the frontier was a free face moved along two directions, forming a major slide area and a
(Fig. 8). The plane configuration of this area is an approximate branch slide area. The major slide moved along the N57 E direc-
parallelogram whose sections exhibit trapezoidal shapes with a tion with an estimated volume of 275103 m3 and transferred to
larger thickness of rear part and a smaller thickness of front part. debris flow with a run-out distance of >1000 m from the exit of
The elevation difference between the rockslide rear edge and the rockslide (also see Figs. 6 and 7). The major slide resulted in severe
shear outlet is 96 m with an elevation of 1140 m for the former and disaster and is the focus of this paper. The branch slide moved
1044 m for the latter. The thickness of this area ranges between 30 along the 10 NW direction with a flow distance of 380 m, a mean
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
8 m with a mean value of 6 m, a width of 60110 m with mean obviously separated by two sets of dominate steep joints with
value of 90 m, an approximate area of 60103 m2, and an approx- discontinued interface and soft layer in bedrock that had an
imate volume of 360103 m3. attitude of 11016. One was tense joint nearly vertical to the
strike of the bedrock with smooth-jointed surface with an attitude
Initiation mechanism of rockslide of 386. The other was a compressive joint nearly parallel to
The front of the rockslide source area had a scarp with an eleva- the strike of bedrock with an attitude varied between 2887080
tion difference more than 20 m (NNE direction). The west side of and 3247580. Due to the barrier of stable bedrock at the east
the rockslide source area was a deep cut valley. The upper bound- boundary, a compressional area was formed at the southeast
ary of the rockslide source area was a tension fracture created by corner where the east boundary interconnected with rear edge
preferred joints along the NWWSEE direction. It was recorded in cracks. This compressional area blocked the movement of
the literature that this crack had been formed 400 years ago and rockslide along the direction of the true dip angle and forced a
had a maximum width of 10 m and a visible depth of up to 25 m deflection of sliding direction from 110 SE to 57 NE, which
(Fig. 12a) (Li 1998). After sliding, the maximum continuous exten- formed a rockslide with an inclined dip azimuth. In other word,
sion length of this fracture was 250 m, and the width was 64 m the rockslide was forced to change the movement direction into an
(Fig. 12b). There were several tension cracks parallel to the killing apparent dip of NE from the true dip of SEE (also see Fig. 8 and
ditch in the residues of the rockslide. These cracks had lengths of 15). This kind of slide was typically studied taking a case of the
up to 100 m, splitting rocks into multiple blocks. The east bound- Jiweishan rockslide at Wulong, Chongqing city, which occurred on
ary of the rockslide extended to the NNENE direction joints June 5, 2009, illustrating a typical failure pattern of massive rock
along the valley (also see Figs. 6 and 8). The rockslide was slope that rockmass rapidly slides along apparent dip, and then
assured that the groundwater in rear edge tension fracture could pressure on the rear edge of rockmass. However, the gently in-
steadily reach the needed height before sliding and at the early clined cracks along the bottom of the rockmass were not well
stage of sliding. It could be speculated from Fig. 15 that the developed and had poor water permeability; therefore, penetrating
groundwater in the tension fracture could keep a relatively steady uplift pressure could not occur (Fig. 16a).
depth due to a closed hydrogeological condition not easily yielding Due to a lack of groundwater and runoff information in the
to discharge formed by the oblique apparent dip layered rockmass. rockslide, it is difficult to set up the seepage field in the rockmass.
Secondly, the heavy rainfall before and during the rockslide not The 2D analysis on typical longitudinal section is preferred instead
only exceeded the record of recent decades but also created a of 3D numerical simulation. This paper utilizes the Geo-Studio
duration record. This kind of continuous several-days-long ex- 2007 commercial software to simulate the stability of rockmass
treme heavy rainfall was enough to provide the necessary ground- under the application of hydrostatic pressure in the relaxed crack
water lastly recharge and seepage into the crack during the stage (Swedish method of slices). Rockslide of the Cretaceous
rockmass sliding. Thus, the slide of the rockmass could be gener- Guankou Formation has a wide distribution in the Sichuan area,
alized into four stages of cracking, creeping, separating, and re- which makes it possible to select parameters in an analogy meth-
sidual accumulation (Fig. 16). 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:
Cracking stage
This stage corresponded to the early process of rock crack defor- 1. Shear strength for sliding surface(clay): cohesion c=13.54 kPa
mation. Due to the incised valley, a terrain with two free faces was and internal friction angle =8.74
formed surrounding the east and north sides of the slope. Multiple 2. Rockslide mass(mud rock): cohesion c=70100 kPa; internal
sets of joints resulted in unloading cracks in the rock. These friction angle =2042; density =21 kN/m3.
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 A calculated model is conceptualized according to the
of siltstone and shall sandstone at the bottom of the rock. During hydrogeological condition and rockmass structure, in which the size
the rainfall, surface water infiltrated into the rockmass along steep of model has the same size as the actual rockslide with a longitudinal
inclined cracks at the rear edge, which produced hydrostatic length of about 140.6 m and a height varying from 55.6 to 34.7 m
Fig. 12 Photos showing the width of back crack of rockslide before and after sliding
(Fig. 17, also see Fig. 7). The model has three sub-blocks at the stages movement of the weak bottom layers along the inclined direction.
of cracking and creeping, and then front block sliding away accord- The penetration of rain along the steep inclined cracks produced
ing to the remote sensing images. Since the water level that filled in hydrostatic pressure, which applied to the rear edge of sliding
tension cracks of rockmass during the rainfall has not been actually mass and also formed uniform uplift pressure. This uplift pressure
tested, an approximate method using the percentage ratio of filling resulted from the evident slide at the bottom weak zone. The
level to crack thickness is used for trend analysis. pressure was applied to the free face area from the rear edge to
When cracks are not filled with groundwater, the factor of safety the front edge of sliding mass, where there was a certain thickness
(FOS) equaled 1.105 (Table 1), which indicates that rockmass is of gentle inclined crack zone and good water permeability
generally stable. However, the rear edge of rockmass is subject to (Fig. 16b, also see Fig. 12b).
hydrostatic pressure with the increase in the filling depth of ground- The sliding mass was simultaneously subject to the application
water. The corresponding safety factor decreases. When the filling of rear edge hydrostatic pressure (Ps) and bottom uplift pressure
depth (h) of groundwater reaches 60 % of the tension crack thickness (Pu) (Fig. 18 and Table 1).
at the rear edge of rock, the hydrostatic pressure (Ps) applied to the When the filling depth (h) of groundwater reached 60 % of the
rear edge of rockmass can be determined by the equation: 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
Ps h2 =2 10 32:42 =2 5248:8 kN=m also to uplift pressure at the bottom with a value determined
according to the following equation:
Under this situation, the FOS equaled 1.007, and the rockmass
is at the critical slide status. Pu hl 10 32:4 150 48; 600 kN=m
Creeping stage
This stage corresponded to the evolution process that the steep This situation corresponds to a FOS equal to 0.819 and a sliding
inclined joints separated from adjacent parent rock and cracked status. This superposition effect results in an evidently lower FOS.
rockmass transformed into sliding mass. A weak zone with good Actually, under joint application of hydrostatic pressure and uplift,
property of water penetration was formed due to the long-term if the filling depth of groundwater exceeds 20 % of the crack depth
(H) of the rear edge, i.e., 11 m, FOS would be equal to 1.0 and between the critical water head and the dip angle of the sliding
sliding would occur. 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
Separating stage same physico-mechanical parameters of rock and soil in the
This stage corresponds to the evolution process of an evident slide, rockslide. Wang et al. (2012) presented a similar result that the
gradual failure of rock, separation from parent rock, and rockslide wedge-shaped cracking expansive force of groundwater played a
formation. Multiple steep inclined joints developed, continued to key role in the course of deformation and destruction of slope for
cut the slide mass into multiple blocks, and resulted in separation the near horizontal layered rock slope.
of the slide mass. Especially at the steep location with free faces at In the early stage of rock separation, the cracks in sliding mass
the front edges, the rockmass was readily sheared to rockslide were not filled by groundwater. Therefore, the bottom uplift pres-
failure at a high-level position, the front rockslide changed to fall, sure generally exhibited a triangle distribution, i.e., P u=hl
topple, or rotate (Fig. 16c). Rainfall infiltrated into rockmass (Fig. 19a). With the increase in movement, the rockslide was
through a variety of steep inclined cracks and produced multiple gradually separated, and the wedging expansion from the ground-
sections of hydrostatic pressure. Meanwhile, the uniform uplift water in cracks correspondingly increased and expanded. This
pressure applied in the free face area from the rear edge to the resulted in the bottom uplift pressure changing overall and
front edge at the bottom gentle dip laminated crushing zone exhibiting a composition distribution of a trapezoid and triangle
produced multiple sections of uplift pressure. Many investigators (Fig. 19b, also see Fig. 16c), i.e.
studied the characteristics of groundwater pressure in the multiple
cracks. With the physical simulation, Xu et al. (2012) conducted a Pu h1 h2 l1 =2 hll1
research on the deformation and failure evolution of the transla-
tional rockslide, which was caused by the cooperation of the Under the condition of rockmass cracking, a special situation of
hydrostatic pressure and the uplift pressure. The clear relationship h2=h1 might occur in which the uplift pressure applied on the rear
block varies from trapezoid to rectangular, and the safety factor bilateral materials and colluviums. The entrainment increased
decreased accordingly (Table 2). the volume of slide to 1.0 million m3 and may also enhance the
It must be noted that during sliding, separation of rockmass mobility of the rockslide. Zhang et al. (2011) conducted a study on
structure was accelerated, and the discharge of groundwater was the transform mechanism from rockslide to debris flow with ring
accomplished more readily. Accordingly, the groundwater hydro- shear test that indicated the so-called soft base effect. Because of
static pressure at the tension cracks of the rear edge and the uplift shearing shrinkage, grain crushing, and layering, the sliding body
pressure at the bottom cracked rockmass decreased significantly. volume of soil reduces in the course of shearing, which makes
excess pore water pressure increase, and thus its shear strength
Residual accumulating stage decreases. Simultaneously, a thick liquefied layer is formed at the
This stage corresponds to the process that the decrease in ground- bottom of sliding body, of which thickness increases with shearing
water hydrostatic pressure after the shear failure of rockmass displacement increasing. The liquefied layer carries the sliding
resulted in a slow slide of material on a gentle dip surface body moving fast and reduces its stability greatly. There is a
(Fig. 16d). In the process of a large slide, rockmass further sepa- different slide-flow transferring process between the Sanxicun
rated to stone blocks. The major application of groundwater rockslide and above cases. The Sanxicun rockslide had no
during rainfall was represented by pore water pressure. However, obvious liquefied layer. After sliding, the rockslide at front edge
in relatively large rocks, hydrostatic water pressure and uplift was subject to shear failure, impacted, and then shoveled the
pressure were still jointly applied. The resultant process of a surface rock, soil, and vegetation. With the increase in sliding
laminated rockslide would be an accumulation slide. volume, the rockslide flooded into the valley and merged into a
mountain flood to form a debris flow. Hungr (2010) proposed a
Analysis of rapid and long run-out slide process variety of rheological relations and developed the DAN-W
In Southern China, it is a very typical disaster mode that a rockslide dynamic analysis software to simulate this process.
rockslide triggered by a heavy rainfall is transformed into a rapid DAN-W is an MS Windows-based program used to model the
and long run-out debris flow (Yin 2011). Xu et al. (2012) reported a post-failure motion of rapid rockslides. The basic premise of the
catastrophic avalanchedebris flow that was triggered by the heavy analysis is that, as a result of sliding or other failure, a predefined
rainfall on July 27, 2010 in Sichuan. A mass of strongly weathered volume of soil or rock (the source volume) changes into a fluid
basalts with a volume of 480,000 m3 was initiated from a valley and runs downslope, following a path of a defined direction and
side slope and then moved downstream along the valley, width. The mass can entrain additional material from the path and
entraining a large amount of unconsolidated substrate and eventually deposits, when it reaches slopes that are sufficiently flat
Table 1 Calculating results of factor of safety (FOS) with various depths of groundwater
Groundwater FOS
Depth h (m) h/Ha State 1b State2c
1 0.0 0.0 1.105 1.105
2 10.8 0.2 1.079 1.015
3 21.6 0.4 1.033 0.907
4 32.4 0.6 1.007 0.819
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
(the deposition area). The model implements a one-dimensional account factors such as terrain limits on both sides of the sliding
Lagrangian solution of the equations of motion and is capable of approach and the internal strength of the sliding mass.
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 Establishment of model
rockslides are a shallow flow approximately similar to a parallel The Double rheology model (F-V model for short) is based on a
sliding bed with smaller sliding thickness. The sliding mass is friction criterion and Voellmy criterion. It is established per the
divided into a certain number of mutually contacted blocks difference in ground rocks and divisions of sliding approach. In
under curvilinear coordinates. These blocks can freely transform the F-V model, the rheological relation is based on the friction
but have constant volume during movement. Through the criterion for the impact-shoveling area but on the Voellmy criteri-
Lagrange method, continuous equations and momentum on for the debris flow section (Hungr 2008; Hungr and Evans
equations are resolved on the basis of curvilinear coordinates 2004), of which:
and moving meshes. In the DAN method, Hungr (1995) took into
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.
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
Fig. 19 Seepage analysis with hydrostatic and uplift pressure (separating stage) turbulivity (its value is the square of the Chzy coefficient). The f
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:
volume of 296,500 m3 (Fig. 20a). When moving forward 500 m, the houses were destroyed with a resultant volume of up to
rockslide continued to shovel down the slope surface to a depth of 398,400 m3.
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 Movement velocity
and depth of 13 m was formed, and the resultant volume of the If the initial velocity at the shear failure of the rockslide is set to
rockslide increased to 338,400 m3 (Fig. 20c). zero, the sliding velocity kept increasing between 300 and 600 m
The rockslide converged with the flood in the valley at 600 m to (Fig. 22, also see Fig. 20). This sliding velocity reached a maximum
form a debris flow and continued to move forward. When moving value of 35 m/s at the location of 600 m with a movement time
forward 700 m, the mud level in the confluence was as high as 18 m t=13.6 s. After entering the valley, the rockslide was transformed
with a volume of 356,700 m3 (Fig. 20d). When it arrived at 800 m, the into a debris flow, and the corresponding velocity decreased to
mud level in confluence was as high as 22 m with a volume of 28 m/s. At the distance of 800 m, the velocity before destroying
371,100 m3 (Fig. 20e). When it arrived at 8001000 m, houses were houses was 22 m/s with a corresponding moving time of t=21.2 s
destroyed, and the volume increased to 388,800 m3 (Fig. 20f, g). and a time interval of t=21.213.6=7.6 s. Between the distances of
Subsequently, the debris flow continued to move downward and 800 and 1000 m, 14 houses were destroyed, and the movement
the mud height in the confluence decreased from 17 to 12 m with a velocity decreased to 14.2 m/s with a corresponding moving time
resultant volume of 398,400 m3 (Fig. 20h, k). The difference of the of t=31.3 s and a time interval of t=31.321.2=10.1 s. After that, the
debris flow between the estimated volume of 360,000 m3 from the rockslide continued to move forward approximately 100 m with a
field observation and the calculated volume of 398,400 m3 based on corresponding moving time of t=46.7 s and a time interval of
the computer simulation is 38,400 m3. It should be noted that, when t=46.731.3=15.4 s. Finally, the rockslide tended to stabilize at
the water in debris flow was exhausted, the thickness of the solid 1250 m with a corresponding moving time of t=46.7 s and a time
accumulation was generally <8 m (Fig. 21, also see Figs. 10 and 11). interval of t=83.446.7=36.7 s.
In general, after shear failure in a high-level position, the After shear failure occurred in a high-level position, the
rockslide moved downward about 950 m in 47 s. When the rockslide moved downward for approximately 950 m in 47 s with
rockslide moved forward 600 m, a shovel zone with a length of a maximum velocity of 35 m/s. This was a typical rapid and long
250 m and depth of 13 m was created. Its volume increased to run-out rockslide.
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 Discussions
reached 22 m. When the rockslide arrived 8001000 m away, 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
Fig. 22 Velocity of slide mass vs. distance of run-out of rockslide debris at and before earthquakes in relation to the peak ground acceleration
Sanxicun (PGA) of earthquakes and the number of rainy seasons after the
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 watershed, was taken as a case. The rainfall and materials supply-
loose materials supplies for debris flows. Many debris flows have ing conditions, the developmental processes of its velocity, dis-
occurred in watersheds in area beyond 20km2, presenting charac- charge, and density were analyzed. It is hoped that this study wll
teristics differing from those in small watersheds. The debris flows be helpful in revealing the debris flow characteristics in large
yearly frequency decreases exponentially, and the average debris watersheds in the Wenchuan Earthquake area.
flow magnitude increases linearly with watershed size. The rainfall
thresholds for debris flows in large watersheds were expressed as Study area
I=14.7 D0.79 (2h<D<56h), which is considerably higher than those The study area is located upstream on the Minjiang River, between
in small watersheds as I=4.4 D0.70 (2h<D<37h). A case study is Yingxiu Town on the south and Wenchuan County on the north
conducted in Ergou, 39.4km2 in area, to illustrate the formation (Fig. 1). The physiography is characterized by mountain ranges
and development processes of debris flows in large watersheds. A trending in an approximately north-northeasterly direction which
debris flow develops in a large watershed only when the rainfall are dissected by the Minjiang River and its tributaries. The topog-
was high enough to trigger the wide-spread failures and erosions raphy is typical of that formed by tectonic uplifting and fluvial
on slope and realize the confluence in the watershed. The debris erosion. The channel gradients range from 5 to 30, and the slope
flow was supplied by the widely distributed failures dominated by gradients range from 25 to 50. The geology conditions are shown
rill erosions (14 in 22 sources in this case). The intermittent in Fig. 2.
supplying increased the size and duration of debris flow. While Most of watersheds (D1D16) have a wet climate, with annual
the landslide dam failures provided most amounts for debris flows average precipitation of 1,200 mm, and D17 and D18 are located in
(57% of the total amount), and amplified the discharge suddenly. relative dry climate with annual average precipitation of 800 mm.
During these processes, the debris flow velocity and density in- After the earthquake, rainfall stations were established in selected
creased as well. The similar processes were observed in other large watersheds primarily for the purpose of supporting early warning
watersheds, indicating this case is representative. systems for debris flows. Compared with other regions in south-
western China, rainfall measurements in this region are conducted
Keywords Debris flows . Triggering rainfall . Formation . Large relatively intensively (Fig. 1). Four of these stations were
watershed . Wenchuan earthquake established by authors, in Ergou (Fig. 1), from source region to
the outlet (Fig. 3).
Introduction Ergou, with an area of 39.4 km2, is a tributary of Minjiang River,
The region severely affected by the Wenchuan Earthquake (May 12, and located about 6 km from the epicenter (Fig. 1). The slope
2008, in Sichuan Province, China) has been particularly vulnerable gradient averages 40. Elevations range from 930 to 4120 m. The
to debris flows (Cui et al. 2011; Tang et al. 2012). Debris flows gully length is about 12 km, and the mean channel gradient is
occurred frequently even in watersheds larger than 20 km2, while about 12, and ranges from 5 to 12 in the debris flow transporta-
in other regions most debris flows occur in watersheds smaller tion region. Landslide deposits, with depth between 1 and 30 m, are
than 10 km2 (Li et al. 2002). spatially distributed over about 24 % of the watershed, amounting
Debris flows are generally initiated in response to failure of a to 1.0108 m3. Debris flows occurred in low-order gullies in Ergou
discrete landslide mass on a hillslope or failure of channel bed before, according to the field investigation, but only the one on 10
material (Innes 1983). Various studies have investigated the initi- July 2013 flowed out and was recorded at the outlet.
ation mechanisms and established some critical thresholds (e.g.,
Takahashi 1978; Cui 1992; Iverson 1997; Iverson et al. 1997; Huang Debris flows in watersheds with different size
et al. 2009). Comparing with the case of small watershed where
debris flow originates from single or several sources, debris flow in Debris flow activities
large watershed originates in numerous sources, which requires Debris flows behaviors vary in watersheds with different size. We
more intense rainfall to trigger the slope failures and of which the collected the information of landslide area ratio in a watershed,
confluence depends on complex structures of the full watershed slope gradient, and debris flow numbers in Ergou and its neigh-
(Cui et al. 2013). Therefore, the critical rainfall or runoff thresholds boring watersheds (Table 1). Here, the landslide area ratio, the
that derived from the aforementioned models that based on indi- landslide area divided by the total watershed area, is defined to
vidual slope or in laboratorial channel and the previous proposed represent the materials supplying level, which is negatively pro-
linear debris flow discharge calculation methods (Kang and Zhang portional to the area (P=0.410.0007A). Debris flow yearly fre-
1991; Hu et al. 2010) should be limited applied on a watershed quency ranges from 0.17 to 2.67, which decreases exponentially
scale, only with full acquaintance of its formation and develop- with watershed size (Fig. 4a), while the average magnitude of
ment of debris flows. debris flow increases linearly with watershed size (Fig. 4b).
To reveal the formation conditions and development processes Area, landslide area, landslide proportion, and channel gradi-
in large watersheds, the debris flow event in Ergou, a 39.4-km2 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 order valleys, the magnitudes in large watersheds are gener-
by the total watershed area. Debris flow times were obtained by ally tremendous.
field investigation. The debris flow yearly frequency was deter- It is noted that 10 out of the recorded 18 gullies have an area bigger
mined as dividing debris flow times by the number of years than 10 km2, which is contrary to the usual cases that less than 30 %
(generally 6 years from 2008 to 2013, with exception of watershed debris flow occur in gullies bigger than 10 km2 according to statistics
no. 13 with debris flow prevention engineering since 2012). The in the southwest China (Li et al. 2002), showing the debris flow
average debris flow magnitude was obtained by field investiga- susceptibility in large area. Also, debris flow yearly frequency ranges
tions. Debris flow only occurred on 10 July 2013 in the watersheds 12.67 and 0.171 in areas in watersheds with area smaller and larger
with asterisks. than 20 km2, respectively. Therefore, we take 20 km2 as the boundary
The larger landslide ratio and steeper channel slope in small of large watershed to analyze the rainfall thresholds.
watersheds or slope gullies lead the more frequent debris flow
activities. Confluence processes, indicating the control effort Rainfall conditions for debris flows
of a fluvial system, significantly influence debris flow activities As shown in Table 2, the mean minimum triggering rainfall inten-
in larger watersheds. Debris flows in lower-order valleys did sity and inducing rainfall for debris flows are 14.8 and 43.5 mm/h.
not necessarily move downward to the outlet. On the other The minimum triggering rainfall intensity and inducing rainfall
hand, due to the wide-spatially distributed landsliding and for debris flows is as low as 3.7 mm/h and 5.3 mm, on 17 May, 2008,
loose materials accumulation by sediment supply from low- quite short following the earthquake. Maximum triggering rainfall
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
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.
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 characteristic size for the total aggregation of particles that can be
710 m3/s. linked to the flow density by Eq. (6) (Li et al. 2013):
2. Constituents of debris flow deposits and variations in densities
In all cases, laboratory analysis revealed that the debris flow deposits Dc 0:04718:1446 6
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 %. For these samples, the parameters C, , and Dc ranged from 75.15
The particle size distributions are indicated in Fig. 10. to 83.56, 0.047 to 0.073, and 5.93 to 22.2, respectively (Table 4).
The cumulative particle-size curve P(D) (percentage of particles The overall increase in Dc during debris flow transport is evident,
with diameters exceeding D) was plotted using the following distri- suggesting that greater proportions of coarse particles were assimi-
bution (Li et al. 2013): lated into the matrix and the kinetic energy and transportation
capacity increased. Meanwhile, debris flow density increased as well
PD CD expD=Dc 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
Where, P (D) is the percentage of particles with diameters greater viscous. On the debris flow fan, Dc and were smaller because of the
than D (in millimeter); the power exponent , can be associated with lower velocity and due to the increased amounts of coarse particles
and estimated by the criterion of shear flow; D c defines a in subsequent floods.
Deposit characteristics
Five sections can be identified in the deposit according to the of the depositional region, is about 200 m long with depths of
topography (Fig. 11). The deposit depths, varying with channel about 5 m, while sections B and C are relatively wide and with
width, slope gradient, and building distribution, were measured depths of about 510 m. These deposits are present in the key area
by marks left on constructions (e.g., Fig. 12). Section A, at the tail 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)
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 (D11, D15, D17, and D18 in Fig. 1, the debris flow magnitude are also
the head of the region as the debris flow fan with a deposit depth listed in Table 1) (Yan et al. 2014; Zeng et al. 2014), indicating that
of about 34 m. The deposit parameters are listed in the table in debris flows in a large watershed are potentially initiated and
Fig. 11. The total volume of residue in the channel and debris flow supplied by various processes than small gully or single slope,
fan was estimated to be 500103 m3. However the debris flow such as new landslides and collapses, slope rill erosion, landslide
lasted for 4 h, exhibited typical surge characteristics, and dam failures, and bed erosion of loose materials in channels. The
transported a great volume of debris flow material into the processes are common:
Minjiang River. Therefore, the total volume of this debris flow
event was significantly greater than 500103 m3. 1. Material supplies from diverse slope processes:
Figure 12 Buildings indicating deposit height. On the left, photo The failures and rill erosions on loose, coarse-grained colluviums
(a) (taken in May 2013) was Ergou power station prior in section C to were widely distributed as the most common supplying mode for
the debris flow, and photo (b) (taken in July 2013) was the same debris flows (14 of 22 in this study), and their intermittent supplying
building after the debris flow. H1, H1 and H2, H2 represent the lead the flood to behave like surges and lasting for hours, which is
vertical distances from the same white object in red circles and the also different from the common debris flows on slopes.
same line to the ground before and after the debris flow, indicating 2. Sediment assimilation:
the depth of debris flow deposit was about 5.5 m. The red line in The mobilization of granular material allows the gradual transformation
photo (c) shows debris flow deposit on against another building in of a flood into a debris flow and enlarges the discharge of debris flows,
Section D, indicating the derbsi flow deposit height was about 4 m. velocity of the flow body also possibly decrease according to the local
slope gradient, leading the debris flow discharge vary along the channel,
Discussion which is also verified by previous studies (Coe et al. 1997; Godt and Coe
2007; Cui et al. 2013).
Debris flow formation and development features in large watersheds 3. Landslide dam failures:
The debris flow in Ergou is typical for debris flow in large water- In addition, the corresponding sediments of certain landslides
sheds. Similar processes are observed in other large watersheds 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
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
Abstract The 2000 Yigong landslide was one of the most cata- initiation mechanisms of landslides under different triggering
strophic landslides worldwide, resulting in huge casualties and factors are not the same and are also influenced by the geological
property losses. The dynamic process of the Yigong landslide condition of the slope (Acharya et al. 2011). Triggering factors for
was very complicated, especially for the initiation and entrainment the landslide can be divided into two main types: long-term
mechanism during the landslide movement process. The topogra- factors, such as freeze-thaw cycles, weathering, and unloading,
phy, geological condition, traces left by the landslide, and distri- and short-term factors, such as strong earthquakes and heavy
bution characteristics of the landslide deposits were determined by rainfall (Barth 2014). Most of the catastrophic landslides are trig-
field investigations, combined with several years of monitoring the gered by the strong earthquake, such as the Tangjiashan landslide,
temperature and rainfall data in this region. The initiation mech- the Donghekou landslides, the Wenjiagou landslide, and the
anism of the Yigong landslide is presented. The main reasons for Daguangbao landslide triggered by the 2008 Wenchuan earth-
the landslide initiation are as follows: the strength reduction of quake (Yin et al. 2011) and the Tsaoling landslide triggered by the
rock masses (especially for the weak structural surface), the impact 1999 Chi-Chi earthquake (Tang et al. 2009). Because strong earth-
from years of freeze-thaw cycles, the superposition of glacier quake poses huge energy on the slopes in the earthquake hit area,
melting and heavy rainfall on the slope, and a slope that was some of the catastrophic landslides are triggered by heavy rainfall,
almost at the limit state before the landslide. Laboratory tests for example, on 9 August 2009, a catastrophic landslide (Hsiaolin
and physical modeling experiments were carried out to study the landslide) is triggered by the accompanying heavy rainfall from
entrainment process of this landslide. Combined with the topo- Typhoon Morakot in Kaohsiung County, Taiwan (Lo et al. 2011).
graphic survey data and theoretical analyses, the entrainment Previous studies have pointed out that water content within the
mechanism during the movement process of the Yigong landslide sliding mass and the sliding bed can accelerate rock mass dilation
is presented. The old landslide deposits on the lower slope collided and fragmentation and a successive increase in mobility (Crosta
with and were scraped by the high-speed debris avalanche, which et al. 2009). Landslides may be initiated by one triggering factor or
resulted in the volume amplification of the landslide. The exis- several coupling factors. Furthermore, the failure of the slope is
tence of water plays a key role during the landslide initiation and controlled by internal factors, such as terrain condition, mechan-
movement processes. ical properties of the rock mass, and distribution of joints and
structural surface; the failure mode and probability for different
Keywords Yigong landslide . Dynamic process . Initiation slopes are not the same (Mather et al. 2014).
mechanism . Entrainment . Water During the motion process of rapid giant landslides, the rapid
sliding masses will interact with the bed materials along the trav-
Introduction eling paths, especially for the saturated accumulations (Legros
Rapid catastrophic landslides, often defined as having a minimum 2002; Dufresne 2012). The high-speed sliding mass can entrain
volume of 106 m3, can cause huge property loss and human death large volumes of sediments both in a dry state and a saturated
because of their large volume and high speed (Korup et al. 2007; state (Angeli et al. 1998; Dufresne and Davies 2009). Entrained dry
Kuo et al. 2013; Pudasaini and Miller 2013; Weidinger et al. 2014). material is generally the consequence to reduce the mobility of
Previous studies have shown that in recent years, the occurrence sliding mass (Crosta et al. 2009). However, according to the
frequency of rapid and long-runout catastrophic landslides has Xiejiadianzi catastrophic landslide triggered by the 2008
significantly increased because of the changes of global climate, Wenchuan earthquake, Wang et al. (2013) have pointed out that
such as rising temperature, heavy rainfall, and retreating snowline an earthquake can simultaneously trigger liquefaction of runout
(Crosta et al. 2009). Catastrophic landslides are widely distributed path materials before the arrival of the landslide mass, thus greatly
worldwide, especially in the mountainous regions, such as the increasing the mobility of an overriding landslide. Entrainment of
southwest of China (Wang et al. 2013). On 12 May 2008, the great almost saturated material can have a more complex consequence,
Wenchuan earthquakes resulted in many catastrophic landslides in which most can increase the runout distance, because of sudden
and provided rich data to study the dynamic problem of rapid undrained loading and impact on loose deposits, and entrainment
giant landslides (Huang and Li 2009; Tang et al. 2010; Dai et al. of a saturated substrate can cause enhanced mobility (Crosta et al.
2011; Yin et al. 2011; Zhou et al. 2013a). The initiation mechanism 2009; Sassa 2000; Zhu et al. 2000). The bed materials are scoured
and motion process are very important to understand the dynam- and eroded by the sliding mass and mixed with the sliding mass
ical process of giant landslides, which can provide possible pre- and water, accreted to the main flow front or entrained, resulting
ventive measures for potential giant landslides. in the change of the landslide form (Dufresne and Davies 2009; Lo
Landslides and other slope failures can be triggered by different et al. 2011). Some studies for rainfall-induced landslides have
factors, such as earthquakes, heavy rainfall, freeze-thaw cycles, pointed out that high-speed sliding motion leads to grain crushing
glacier melting, and human activities (Evans et al. 2001). The along the sliding surface, and the liquidated bed materials are
(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- influenced by the freeze-thaw cycle, resulting in the decreasing of
ness. The powder of strongly weathered dolomite existed at dif- shear strength (Angeli et al. 1998; Borgatti and Soldati 2010).
ferent deposition locations, mainly at the middle section of
landslide depositions. Figure 5b shows the X-ray diffraction anal- The 2000 Yigong landslide
ysis results for the sample of dolomite powder, and the mineral On 9 April 2000, a catastrophic landslide took place at Yigong,
composition of dolomite is relative to a single of CaMg(CO3)2. The Southeastern Tibet, one of the largest non-seismic landslides in the
dolomite can be weathered more easily and significantly world. Interpretation of remote sensing images shows that the
(b)
(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
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.
4000
of the rock mass and thick glacier, toppling and bending failure
3000
occurred to the slate during a long geological history. The bending
2000 failure surface can provide a potential sliding surface for the
initiation of landslides. The slope stability of the initiation zone
1000
is controlled by the structural surfaces. Figure 8 shows the typical
0
36-0426> Dolomite - CaMg(CO3)2 geological condition of the initiation zone for the 2000 Yigong
10 20 30 40 50 60 70 80
landslide.
2-Theta( As shown in Fig. 8(a), there are five joint sets in the landslide
initiation zone; the occurrence parameters of those joints are as
Fig. 5 a Powder of strongly weathered dolomite and b X-ray diffraction analysis follows: (1) in joint set 1, the dip is approximately 4050 and the
for the sample of dolomite powder 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
From the field investigation results, we can determine the 7080 and the spacing is approximately 2.05.0 m; and (5) in joint
material compositions at different locations: (a) at the tail of set 5, the dip is approximately 3055 and the spacing is approxi-
deposition zone, the deposition height is approximately 410 m mately 1.54.5 m. The landslide initiation zone can be divided into
(Fig. 6b) and the landslide deposits are mainly composed of three subzones: (a) a tensile failure surface at the tail of the slope
boulders with a proportion of approximately 7090 %; (b) at the (zone 1), in which a clearly fractured surface can be seen; (b) a
middle of the deposition zone, the deposition height was approx- dislocation layer with several rock layers along the direction of
imately 3565 m (Fig. 6c) and the landslide deposits are mainly slate (zone 2); and (c) a compound weak layer (zone 3). The upper
composed of rock debris, sandy silt, and solitary stone; the pro- part is a weak layer of bending failure surface of slate (rough), and
portion of sandy silt is approximately 4060 %; and (c) at the the lower part is a rock layer (smooth). To the left of subzone 1,
lower of deposition zone, the deposition height is approximately clearly bending failure surface existed and an old landslide oc-
5080 m (Fig. 6d) and the landslide deposits are mainly composed curred in this area (Fig. 8(b)). The 2000 landslide initiation zone is
of rock debris and sandy silt; the proportion of sandy silt is a typical wedge failure mode, which is controlled by two structural
approximately 5070 %. At the middle and lower deposition surfaces and a tensile failure surface, and a free surface at the slope
zones, the new 2000 landslide deposits covered the old 1900 front provided the potential initiation of the landslide in this
landslide deposits. Figure 7 shows the distribution of the landslide region.
deposits for the 2000 Yigong landslide. Furthermore, thick glaciers covered the slope rock masses and
As shown in Fig. 7(a), the deposition of the 2000 Yigong could have resulted in three adverse impacts on the landslide
landslide can be divided into three zones: (1) the boulder deposi- initiation zone: (1) an increase in the weight of the sliding mass,
tion zone, (2) the debris and sandy silt deposition zone, and (3) the (2) the melting and freezing of the glacier effect on the rock masses
impact and scouring zone. As shown in Fig. 7(b), at deposition or structural surfaces such as the unloading-loading cycle for
zone 2, the landslide deposits are composed of two main types of rocks, and (3) the melting water could decrease the shear strength
material: rock debris and soils. Three-layered structural character- of the structural surface and increase the pore water pressure in
istics existed in the lower deposition zone (from one eroded the slope rock masses. The typical geological condition on the top
section) because of the high mobility of the debris avalanche. slope provides favorable conditions for the initiation of landslides.
The top layer is the new 2000 landslide deposit, and the lower The variation of the external environment condition on the slope
layer is the old 1900 landslide deposit interbedded with an over- can be summarized in two aspects: decreasing shear strength and
consolidated layer (which is mainly composed of fine particles). At increasing stress.
(a)
(b)
(c)
(d)
(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 only affects the loose landslide deposits located at the creek
In the study area, the 19512011 data recorded by the Yigong but also influences the slope stability in this region. As shown
Meteorological Station shows that the average temperature was in Fig. 9b, the daily temperature difference is the difference
approximately 810 C. Since the 1970s, the temperature in this between maximum temperature and minimum temperature in
area has had a slowly rising tendency with a total ratio of 0.3 C/ 1 day. The daily temperature difference in the study area was
10 years (Shang et al. 2003). Figure 9 shows the monitoring results relatively large; the average value was approximately 12.5 C
of temperature in the study area. from 1 January 2000 to 1 January 2001, and the maximum
As shown in Fig. 9a, the average temperature gradually value attained was 26 C. The temperature difference from 1
increased during 1 March to 4 May, which means that the April to 15 April 2000 was relatively large, which had a signif-
melting of glacier and snow gradually increased during this icant effect on the initiation of landslide on 9 April 2000.
season. However, there was no significant difference between Temperature changes can provide two aspects for the initiation
the years 1998, 1999, and 2000. Along with the rises of tem- of landslides: (a) a freeze-thaw cycle on structural surfaces and
perature from April to May, glaciers and snow cover melt and (b) glacier melting resulted in the decreasing of shear strength
water flow was generated in Zhamu Creek. The water flow not and increasing pore water pressure.
(b) (d)
Over-consolidate layer
(c)
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 strength (Ma et al. 1999). Microcracks and micro-defects occur in
this area: the seasonal freeze-thaw cycle (with a period of 1 year) the rocks, and water will run into the rocks. The water in the rocks
and the daily freeze-thaw cycle (with a period of 1 day). Figure 10a begins to freeze when the temperature is lower than 0 C.
shows the temperature monitoring result from 1 January 1996 to 30 Volumetric expansion occurs during the freezing process and leads
December 2003, and Fig. 10b shows the temperature monitoring to microcrack propagations in the rock. Local damage occurs in
results from 15 March to 16 April 2000. the rocks and results in the decreasing of compressive strength
As shown in Fig. 10a, the annual temperature variation period- and shear strength (Draebing et al. 2014). However, the ice in rocks
ically changes; the minimum temperature (about 5 C) occurred melts when the temperature is greater than 0 C, while pore water
in December or January and then gradually rose to the maximum pressure is formed in the rocks, and also results in the decreasing
temperature (approximately 20 C) in June or July. This long-term of shear strength (Zhu et al. 2011; Krautblatter et al. 2013). With the
freeze-thaw cycle resulted in the gradual decreasing of shear increasing numbers of freeze-thaw cycles, the shear strength of
strength of rock masses and structural surfaces. As shown in rock masses or structural surface is gradually decreased (Lai et al.
Fig. 10b, the freeze-thaw cycle property of daily temperature is 2010). The freeze-thaw cycle not only decreases the shear strength
exhibited. However, strong cycles did not occur because the of rock masses but also results in the increasing stress distribution
daily temperature variation was disordered. The mechanical in the slope and finally causes the initiation of landslides in high
properties of rock masses were also influenced by the daily mountainous regions (Bayard et al. 2005).
freeze-thaw cycle.
Previous studies have shown that freeze-thaw cycles have a Effect of rainfall
significant influence on the mechanical properties of rocks, in- According to the 19512011 data from the Yigong Meteorological
cluding porosity, compressive strength, elastic modulus, and shear Station, the mean annual rainfall in the Yigong area was
(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 saturation of loose landslide deposits at Zhamu Creek and increas-
accounts for 90 % of the annual total rainfall, and November to ing pore water pressure in the upper slope, especially in the
February is the dry season (Shang et al. 2003). Figure 11 shows the structural surfaces. Furthermore, relatively more water flow
rainfall monitoring results in the Yigong region. formed in Zhamu Creek, which could have affected the motion
As shown in Fig. 11a, most of the daily rainfall is approximately of debris avalanche.
040 mm, but some heavy rainfall occurred on special days. The
maximum daily rainfall occurred on 20 October 1998, approxi- Comprehensive analysis
mately 111.7 mm. The daily rainfall was not distributed uniformly Through the above analyses of the geological and environmental
during the whole year; the daily rainfall during the rainy season conditions at the Yigong area, we found that the main reasons for
was greater than during the dry season. The rainfall in different the initiation of the 2000 Yigong landslide were typical slope
seasons has similar characteristics. The rainfall data are monitored geological conditions, temperature changes, and heavy rainfall
in the valley region, and the rainfall level at the top of the moun- on April 2000.
tain is greater. Thus, in the initiation zone of Yigong landslide, the Table 1 summarized the annual environment conditions at the
rainfall is relatively abundant. Figure 11b shows the comparison Yigong area from 1991 to 2001, including temperature and rainfall.
results for the monthly rainfall of 1998, 1999, and 2000. We can see Extreme maximum temperatures always occurred from June to
that rainfall occurred mainly during April to October. On April, August, with values of approximately 29 to 31 C. However, the
the monthly rainfall in 2000 was approximately 175 mm, more extreme minimum temperatures always occurred from December
than the years 1998 (38 mm) and 1999 (80 mm). Increasing rainfall to February, with values of about 10 to 13 C. The maximum
on April 2000 played some key effects on the initiation of the 2000 rainfall for the whole year of 2000 occurred on 9 April.
Yigong landslide. Figure 12 shows the daily rainfall during 1 March The reasons for the initiation of the 2000 Yigong landslide are
to 4 May 2000 at the study area. the coupling of long-term effects and short-term impact.
As shown in Fig. 12, in 2000, when the recent Yigong Landslide Regarding the long-term effects, there are two aspects, as men-
occurred, the rainfall from April 2 to April 9 was more than 60 mm tioned above: (a) the loading-unloading cycle on the rock masses
(Hu et al. 2009b) and the total rainfall in April was approximately or structural surfaces due to the melting and freezing of glaciers
175 mm, 90135 % more than the mean value from the same period and (b) the continuous seasonal and daily freeze-thaw cycles on
in southeast Tibet (Liu 2002). The daily rainfall on 9 April 2000 the rock masses or structural surfaces. These two aspects resulted
was approximately 25 mm. The abundant rainfall resulted in the in the gradual decrease of mechanical properties of rock masses or
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 structural surface finally resulted in the catastrophic landslide in
elastic modulus, especially for the shear strength of rock masses or the Yigong area in 2000.
structural surfaces. Furthermore, the tensile fracture has gradually
been increasing over a long period. Other long-term factors in- Entrainment effect during motion process
cluding weathering and unloading can also have adverse impacts For the 2000 Yigong landslide, the volume of landslide deposits
on the slope stability. was about three times the volume of initiation the zone. The
Regarding the short-term effects, the rising temperature will entrainment effect during the motion process obviously occurred
accelerate the melting of glacier and snow; by coupling with heavy because of the high motion speed and water flow.
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 Motion of high-speed landslide
increasing of pore water pressure on the structural surfaces. The entrainment of path material is an important feature of
Another key factor for the initiation of the landslide is the special many rapid landslides (Hs 1978; Evans et al. 2001; Dufresne
geological conditions of the upper slope at Zhamu Creek. The et al. 2010). During the motion process of rock avalanches,
potential sliding wedge is controlled by two structural surfaces the landslide paths are typically covered by thick surficial
and a tensile failure surface, which are sensitive to temperature deposits. These deposits may be loose and have high water
change and water content. content. The rapid loading due to the weight and momentum
Regarding the initiation mechanism of the 2000 Yigong land- of the moving masses may cause failure and mobilization of
slide, special geological conditions at the top slope provide neces- these bed materials and resulted in the volume amplification
sary conditions for the initiation of the landslide. Long-term of the landslide (Scott et al. 2005; Cepeda et al. 2010).
freeze-thaw and loading-unloading cycles resulted in the gradually Figure 13 shows the topography changes for the main geolog-
deterioration of the slope stability. The direct triggering factors are ical section of Zhamu Creek before and after 2000 Yigong
short-term heavy rainfall and rising temperatures on April 2000. landslide, which is extracted for the high-resolution digital
The increasing water content and pore water pressure in the elevation model of the study area.
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
Time (mm/dd/yy)
(b)
25
Temperature (C)
15
-5
3/15 3/23 3/31 4/8 4/16
Time (month/day)
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 As shown in Fig. 15, these test results were obtained under dry
on Zhamu Creek during the motion process of the high-speed rock conditions, without water flow. The movement distance of bed
avalanche, and these bed materials were scoured and eroded by materials increased with the weight ratio between sliding mass
the sliding masses, which finally resulted in the volume amplifica- and bed depositions, an approximately linear relationship.
tion of the landslide. The serious scouring of the bed materials at During the landslide motion process, the bed depositions are
the middle section of Zhamu Creek by sliding masses can clearly impacted and scoured by the sliding masses (Van Wyk de Vries
be seen. et al. 2001; Crosta et al. 2005; Dufresne 2012). A larger weight ratio
between the sliding mass and bed deposition means the difference
Laboratory tests of impact energy and friction resistance is increased, resulting in
Regarding the entrainment problem of high-speed landslides, the longer movement distance of bed materials. Figure 16 shows
a simple laboratory test was carried out to study the bed the experimental results for the impact and scouring of sliding
materials impact and scouring by the sliding masses. masses.
Figure 14 shows the experimental equipment for the laborato- As shown in Fig. 16a, rock blocks were used to simulate the rock
ry tests. The horizontal length of the flume is 2.0 m, the avalanche and placed at the top of the flume and the loose deposits
height is 1.5 m, and the width is 0.3 m. Loose bed materials were covered on the lower flume with a uniform thickness.
are accumulated at the lower of flume, and the sliding mass Figure 16b shows the final deposited situation of test No. 6 under
(composed by rock blocks) are placed in the topper of flume dry condition, and the WR is 1.0. The bed depositions were im-
(Fig. 14b). These experiments are solely conceptual since no pacted and scoured by the moving mass, and the total motion
scaling calculations were done. Table 2 shows the experimen- distance of the bed materials was 16.0 cm. Most of the rock blocks
tal design for the impact and scouring effect on the bed were blocked at the tail of landslide deposits, and some blocks
materials. moved to the front of the flume. When water flow was added in the
As shown in Table 2, 11 tests under different material composi- experimental process, the bed depositions were saturated, pore
tions and weight ratios, with and without water, were carried out. water pressure formed in the soils, and the friction resistance of
The weight ratio is a key factor affecting the motion distance of the bed materials decreased sharply. When the sliding mass im-
bed materials. Figure 15 shows the movement distance of bed pacted on the bed materials, the huge impact energy and excess
materials influenced by the weight ratio between sliding mass pore water pressure resulted in the failure and scouring of bed
and bed depositions. materials. Then, the bed loose deposits were mixed in the sliding
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)
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
deposits. Furthermore, during the motion process of the rock bed deposits and larger water flow. During the motion process of
avalanche, the bed material was scoured by the sliding mass. the landslide, the landslide volume gradually increased through
This former entrainment process is shown in Fig. 17b. After the the entrainment effect and finally resulted in the volume amplifi-
landslide initiation, the form and material composition of land- cation of the landslide deposits.
slide varied during the whole motion process, and the cumulative
volume of the landslide changed enormously along the path of Discussion and conclusion
movement. The path is described as a series of reaches that have a
distinct morphology, width, and slope inclination, and the struc- Discussion
tural characteristics, material compositions, and mechanical prop- The initiation mechanism and motion process of landslide are
erties are also not the same. These changes made the entrainment very complicated for the dynamical problem of high-speed land-
process of sliding mass very complicated. slides (Zhou et al. 2010). The initiation of landslides is not only
As shown in Fig. 17c, the bed deposits were eroded by the high- influenced by the geological condition but also affected by the
speed sliding mass, especially when water flow occurred. The bed short-term environmental conditions, such as earthquake, heavy
erosion at the base of the landslide is similar to bedload transport rainfall, or rising temperature (Wang et al. 2003; Tsou et al. 2011).
in fluvial hydraulics. The failure of bed material can be caused by Previous studies show that landslides may be initiated by some
the impact force and scouring effect and the generation of excess sudden events, such as strong earthquake, heavy rainfall, or rising
pore water pressures due to rapid undrained loading with possible temperature in a short time (Evans et al. 2001; Dai et al. 2011; Kuo
liquefaction. The existence of water plays a key role during the et al. 2013; Barth 2014). However, heavy rainfall is not the single
motion process of landslides, especially for the entrainment of triggering factor for the 2000 Yigong landslides, and the rising
landslides. temperature on April 2000 is another short-term triggering factor.
In the Yigong area, the special terrain condition allowed the Furthermore, the rock masses in a slope generally experienced a
water flow to be easily collected in the Zhamu Creek. Heavy long-term geological evolution process, such as the tectonic effect,
rainfall and glacier on April 2000 resulted in the saturation of weathering and unloading, freeze-thaw cycles, and loading-
unloading cycle. These long-term geological effects can result in
60 the change of the slope stability state, including the decreasing of
y = 37.058x - 14.533 shear strength of rock masses and increasing of pore water pres-
Movement distance (cm)
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
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 mobility of sliding masses but also poses a great impact on the
easily be happened during the motion process of high-speed land- entrainment process during mass movement process.
slide, part of the debris rushed over a high-level terrace and spread For the entrainment problem during the movement process of
along the hillslope (Hungr et al. 2005). Special terrain condition high-speed landslide, well-documented landslide events have ver-
plays a key role during the mass movement process. Regarding to ified this phenomenon (Staron 2008). The landslide entrainment
the 2000 Yigong landslide, the sliding mass moved down the steep includes three main types: impact failure, scouring, and erosion
terrain with a high speed and then encountered the bed deposits. (Pudasaini and Miller 2013). Using a small-scale physical models
Huge impact energy is posed on the bed deposits, and a large (linear scale of 1:104), Dufresne (2012) found that substrates are
volume of bed deposit was impacted and scoured by the sliding failing by the effect of upper sliding masses due to low internal and
mass, resulting in the rock avalanche transformed into debris ava- basal frictional resistance and then turn into the active basal
lanche. Meantime, a part of the sliding masses are stopped and sliding layer, increasing basal boundary roughness of erodible,
accumulated at the foot of the steep slope, because of the energy loss deformable substrates resulting in shorter runout and longer de-
and the narrow terrain. The flat and straight topographical condi- posits (decreased mobility of sliding mass). Using laboratory tests,
tions along the runout path provide a favorable condition for the Shea and van Wyk de Vries (2008) found that, in the presence of a
mass movement. The presence of water not only increases the 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
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
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 study area is higher than 4500 m above sea level (ASL). The QTH,
along the Qinghai-Tibet Highway from Wuddaoliang to the Qinghai-Tibet Railway (QTR), and other linear engineering
Fenghuoshan correlates with ice content, slope gradient, and facilities are located along this corridor (Fig. 2). The objectives
ground temperature. Slope failures are of two types. (1) Retrogres- are to (1) summarize the distribution and morphological charac-
sive thaw slumps result from icy permafrost being exposed by teristics of thaw-induced slope failures, (2) investigate the thermal
either man-induced excavation or fluvial-thermal erosion and regime and failure development of a typical failure, (3) determine
undercutting of basal slopes. (2) Active-layer-detachment failures shear strength parameters in the basal zone of the active layer
are caused by thaw of icy permafrost at the active layer-permafrost using in situ direct shear tests, and (4) assess slope stability in both
interface. After initial failure, active-layer-detachment failures can its natural state and under the influence of an earthquake.
lead to retrogressive thaw-slumping and localized surficial land-
slide. Common trigger mechanisms for failure include high sum- Study area
mer air temperatures and heavy summer precipitation. A third Data from the meteorological station in Wudaoliang town (Fig. 2)
possible trigger mechanism for slope failure is earthquake occur- shows that both mean annual air temperature (MAAT) and pre-
rence. A geotechnical slope stability analysis was undertaken for cipitation have increased during the past 3050 years (Fig. 3). The
an active-layer-detachment failure that had progressed into a MAAT in the study area is lower than 4 C, while the extreme low
retrogressive thaw slump. A safety factor (Fs) of 1.24 for the natural is about 30 C at the end of January and highest is approximately
slope was determined using in situ tested strength parameters. 25 C in July. The annual precipitation, about 300 mm, mostly falls
However, the slope would lose stability when either the ground- between May and August.
water level over the permafrost table exceeded 1.42m or seismic The active-layer thickness ranges between 1.5 and 3.0 m and
acceleration reached, or exceeded, 0.03g. near-surface permafrost may contain ice that exceeds 20 % by
volume; this characterizes about 80 % of the permafrost in the
Keywords Permafrost . Active-layer-detachment study area (Liu and Wu 2000; Wu et al. 2002, 2004). The MAGT
failure . Retrogressive thaw slumping . Stability monitored at a depth of 15 m varies between 3.5 and 0.5 C, with
assessment . Qinghai-Tibet Plateau lowest values in Fenghuoshan and highest in the Beiluhe Basin.
Soil in the study area is mainly loose alpine steppe and alpine
Introduction meadow; these support a sparse vegetation cover that is dominat-
Most permafrost on the Qinghai-Tibet Plateau (QTP) is continu- ed by monotone community physiognomy, characterized by low
ous and characterized by warm ground temperatures (1 to 0 C) stature (10 to 15 cm), short growth period, and low biomass (Niu
(Zhou et al. 2000). The region has experienced noticeable warming et al. 2011).
during the past 4050 years (Cheng and Wu 2007); for example,
both mean annual ground temperature (MAGT) and active-layer Thaw-induced slope failures
thickness have increased in the last two decades (Jin et al. 2006,
2008). These changes can lead to slope instability and failure. Such Types and morphological characteristics
failures generally occur in the active layer in summer. They are A total of 42 slope failures were recognized using SPOT-5 imagery
triggered by the formation of a shear zone near the base of the and field investigation within the 10-km lateral zone along the
active layer that results from thaw of the ice-rich zone at the active QTH from Wudaoliang to Fenghuo Mountain pass. The failures
layer-permafrost interface. Thawing is usually attributed to either are classified as being either active-layer-detachment slides
high summer air temperatures or extreme precipitation events. (failures) or retrogressive thaw slumps (see French 2007). They
The QTP is also an earthquake-prone region and seismic activity are well known in the North American literature and have been
can accelerate the possibility of slope instability. For example, the described from forested, tundra, and high Arctic environments
Kunlun Mountain earthquake (magnitude 8.1) on November 14, (Burn and Lewkowicz 1990; Lewkowicz and Harris 2005a, b).
2001 attracted significant attention from engineers and scientists. Retrogressive thaw slumps are triggered by the thaw of perma-
It follows therefore that an assessment of slope stability is essential frost and the melt of exposed ground ice. This then leads to the
for engineering planning and implementation. retreat of a steep frozen back scarp, or headwall. Thaw slumps
Here, we describe thaw-induced slope failures that have oc- along the Qinghai-Tibet Engineering Corridor (QTEC) are usually
curred within a 10-km-wide zone along the Qinghai-Tibet Highway initiated following excavation of material for use in construction
(QTH), from Wudaoliang (35 13 N, 93 4 E) in the north to of local infrastructures (Ma et al. 2006; Niu et al. 2012). They
Fenghuo Mountain (34 43 N, 92 53 E) in the south (Fig. 1). The generally have a width of 20 to 90 m and a length of 40 to 150 m
zone includes the Hoh Xil Hill region, the Beiluhe Basin region, and occur at concave slope segments. Most are arc-shaped and
and the Fenghuo Mountain region. Approximately 85 % of the usually consist of a series of slumped arcuate blocks. Compression
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 oriented in the NW direction (Fig. 5b), indicating that they
sliding distances are often only a few meters. For example, developed on SE facing slopes. Since ground ice content does
Fig. 4a, b shows a typical thaw slump on a NW facing 7 slope not vary significantly with aspect, the main reason for the
developed at K3035 mile of the QTH. The actively retreating observed distribution is probably the greater amount of solar
headwall is shown in Fig. 4c. Behind the headwall, a series of radiation received by SE facing slopes. However, other studies
fissures can be observed. Once slumping has occurred, newly (Lewkowicz and Harris 2005b; Wang et al. 2009) from the
collapsed material pushes earlier collapsed blocks and forms a Mackenzie Valley, Canada, have shown that the majority of
draped and superposed structure in the failure zone (Fig. 4d). slope failures are concentrated on NE facing slopes where
The thaw depth in the failure area is about 2.2 m, slightly higher higher ice contents are found, indicating that in this environ-
than the 2.0 m measured in undisturbed natural ground. ment, solar insolation was not a primary factor influencing
In the case of active-layer-detachment failures, the active layer the occurrence of slope failures.
may slide over the permafrost table for distances as much as
several tens of meters; they constitute localized surficial landslides. Spatial distribution of thaw-induced slope failures
As such, these can be larger than thaw slumps, with widths of 80 to The spatial distribution of slope failures appears to be closely
160 m and lengths of 120 to 280 m. The mean slope angle at failure related to ice content, slope gradient, and ground temperature
sites was 7.6, much lower than that for landslides in non- (Fig. 6). As regards to ice content, statistical analysis shows
permafrost regions of China (Huang 2007). These types of slope that the vast majority (94 %) of slope failures occur in areas of
failure are usually arc-shaped or bell-shaped, and often with ver- icy permafrost or where massive ground ice occurs. As regards
tical headwalls and compression ridges across the toe zone. After to slope gradient, failures occur predominantly on slopes of
failure, and due to thaw of the icy layers at the active layer- moderate angle (Fig. 6); for example, about 88 % of the
permafrost interface, the headwall may develop into a retrogres- failures occur on slopes with gradients of between 6 and 10.
sive thaw slump. However, the occurrence of slope failures in permafrost-free
regions is positively correlated with slope gradient (Akgun
Geometric measurement 2012; Gnther et al. 2013). As regards to ground temperature,
Morphometric variables including horizontal width and length, about 40 % of failures occur in areas where the MAGT is above
slope angle, and length-to-width ratio were calculated for each 0.5 C, 32 % occur in areas with MAGT of between 1.0 and
slope failure using satellite imagery validated with field observa- 0.5 C, and 28 % occur in areas where the MAGT is lower
tions. Both the widths and lengths of slope failures showed con- than 1.0 C.
siderable variability. The mean width and length of the thaw-
induced landslides were 1.52.0 times larger than those of the thaw The slope failure at K3035W
slumps. The length-to-width ratios of the features were commonly A typical active-layer-detachment failure that developed into a
between 1 and 3, which demonstrates that the slope failures in the localized surficial landslide was selected for detailed study. It
study area were dominantly compact in morphology rather than was first examined on October 12, 2010. It is located on the
elongated (Fig. 5a). west side of K3035 mile of QTH (34 59 N, 92 58 E). To
The long axis orientation was defined as the direction toward distinguish it from the K3035 thaw slump, we refer to it as the
the head scarp of the feature. The slope failures are commonly K3035W landslide. An aerial view, taken on March 27, 2010
(Fig. 4f), shows tensional fissures (red dash line). However, Thermal regime
the landslide occurred sometime before October 12, 2010, In August 2012, thermistors were installed within three 15-m-deep
resulting in wheel ruts visible in the landslide mass that had boreholes (No. 1, 3, and 4) and one 10-m-deep borehole (No. 2)
moved downslope 28 m (Fig. 4g). Therefore, the landslide along the major axis of the K3035W landslide (Fig. 7). Borehole No.
probably occurred at the end of September 2010 when the 1 is located in undisturbed terrain, borehole No. 2 in recently
thaw depth reached its maximum. The size of this failure is disturbed terrain close to the foot of back scarp, borehole No. 3
about 230 m in length from the back scarp to the edge of the in the area between the scarp and the sliding mass at a distance of
compressed area, and 90 m wide in its middle part. A 1.8-m- 30 m from the back scarp, and borehole No. 4 was drilled in the
high headwall formed after failure occurred; this initiated main sliding mass. Each thermistor string consisted of 20 therm-
retrogressive thaw slumping; retreat of about 9 m between istors at 0.5-m interval from the surface down to 10 m and 5
2011 and 2013 was monitored. The total volume of material thermistors at 1.0-m interval below 10-m depth. The temperature
moved was 22,500 m3, several shear fissures formed on the measurement accuracy is 0.02 C. The data acquisition system is
surface (Fig. 4e), and compression ridge of about 3050 cm made up of a CR3000 data logger and a solar panel recharging a
high formed at the toe zone (Fig. 4h). According to borehole 12-V battery (Fig. 7).
data obtained in September 2012, the thaw depth was 1.8 m in Figure 8 plots MAGT against depth from September 2012 to
the undisturbed natural conditions, 1.0 m in the failure plane, September 2013 for the four boreholes. Due to the thermal distur-
and 2.1 m in the main displaced mass. bance caused by the landslide, the MAGT in the sliding mass area
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 retreat of the headwall from October 2011 to October 2013. Table 1
difference mainly existed between the 26-m depth. The maximum shows that the maximal headwall retreat from October 2011 to
differences in MAGT were about 1.3 C (between the scar area and October 2012 was 7.4 m at transect No. 4, while maximal retreat
undisturbed terrain), and 0.5 C (between recently disturbed ter- from October 2012 to October 2013 was only 1.6 m, also at the same
rain and undisturbed terrain). Such differences were caused by the transect. In general, both the laser imagery and the transect mon-
thinner active layer in the failure area than that in the undisturbed itoring indicate that the rate of headwall retreat was slower in the
area during the initial period of the landslide occurrence. second year. However, more data over a longer time period are
In order to further investigate the thermal regime of the land- required before any average rate of headwall retreat can be stated.
slide, ground temperature isotherms as a function of depth and
time, from September 2012 to October 2013, are plotted for the Geotechnical properties of the active layer
undisturbed terrain (Fig. 9a), the recently disturbed terrain Grain size analyses conducted on material from within the active
(Fig. 9b), the scar area (Fig. 9c), and the sliding mass (Fig. 9d). layer indicated that the clay content varied significantly with
As might be expected, the depth to the permafrost table in the depth. The highest amount, 48.7 %, was present in the layer closest
recently disturbed area and the scar area was significantly less than to the permafrost table (Fig. 11). This is consistent with studies of
that in undisturbed terrain; in 2012, the thaw penetration depth active-layer-detachment failures in northern Canada (Lewkowicz
reached only about 100 cm in recently disturbed terrain and the and Harris 2005a). Therefore, high clay content near the perma-
scar area. However, the thickness of the active layer increased in frost table may be another important condition facilitating land-
2013 and reached 150 cm in depth. With time, it is probable that the slide initiation in permafrost regions.
thickness of the active layer will eventually reach, or even exceed, Atterberg limits and plasticity indices of soil samples were also
the active-layer thickness in the adjacent undisturbed terrain. measured (Table 2). Results indicate that liquid limits are consis-
tent with plasticity indices. The maximal liquid limit and plasticity
Failure process index was also found in soil closest to the permafrost table;
In order to investigate the retrogressive nature of the initial active- specific values are 32.12 % and 12.25, respectively. In addition, the
layer-detachment failure (K3035W landslide), 3D laser images of peak shear strengths of undisturbed soil samples collected from
the area were acquired for October 2011, June 2012, October 2012, depths of 0.5, 1.0, and 1.7 m and the strength parameters tested by
June 2013, and October 2013. Based on these images, morpholog- the direct shear test are given in Table 2.
ical changes between 2011 and 2013 are shown in Fig. 10. The total
collapsed area was 670 m2 from October 2011 to October 2012, Shear strength of the basal zone
while it was only 65 m2 from October 2012 to October 2013. This Active-layer-detachment failures are generally caused by low shear
indicates that the rate of headwall retreat decreased in 2013. This strength in the basal zone of active layer. Knowledge of the
was confirmed by a number of manual transects that monitored strength parameters of the material that comprises the active layer
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)
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
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
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
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 of the water table above the slip surface to the depth of the slip
Most thaw instability studies in permafrost regions adopt an surface (z).
infinite slope model because the thaw of icy permafrost will lead In undisturbed terrain at the site of the K3035W failure, the
to high pore water pressure at the base of active layer (McRoberts angle of slope is 7.5, the unit weight of saturated soil and soil
and Morgenstern 1974). This is generally considered to be the main above the groundwater level are 21.66 and 18.84 kN/m3, respective-
trigger for slope failure in permafrost regions (Savigny et al. 1995). ly, the depth to the permafrost table is 1.8 m, and the level of the
Therefore, slope stability should be assessed using effective stress water table above the permafrost table was 0.8 m at the end of
analysis. Slope stability can be estimated from the following equa- September 2013. Using the strength parameters of the basal zone
tion, assuming that seepage direction is parallel to the ground ice obtained from the in situ shear tests (Table 3), the calculated safety
surface (Niu et al. 2005): factor (Fs) of the slope was 1.24 when thaw depth reached its
maximum, indicating that the slope is stable in its natural status.
c0 f1m m sat w gzcos2 tan0 Three factors trigger the instability of slopes in permafrost
Fs 1
1m m sat zsincos regions. First, extreme high air temperature in summer, or forest
fires that reduce the thickness of insulating layer, can lead to an
where Fs is the safety factor, c is effective cohesion, is effective increase in pore water pressure near the permafrost table (Savigny
frictional angle in the basal zone of active layer, is the angle of et al. 1995; Lewkowicz and Harris 2005a). Second, persistent rain-
slope, sat is unit weight of saturated soil, is unit weight of the fall can elevate the water table in the warm season. The calculated
soil above groundwater level, w is unit weight of water, z is result indicates that the safety factor would reach its critical value
vertical depth of the slip surface, and m is the ratio of the height 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
occurrence of an earthquake will result in an instantaneous in- undertaken using in situ tested strength parameters. The slope at
crease in the sliding force and pore water pressure. This last trigger the K3035W site appears to be stable in its undisturbed state but
factor is discussed below. could reach failure threshold when either the groundwater level in
the active layer is higher than 1.42 m or earthquake acceleration
Slope instability under the influence of earthquake reaches, or exceeds, 0.03g.
In order to assess slope instability on the QTP under an earth-
quake influence, the Fs in the conditions of static equilibrium is Acknowledgments
given as follows: This work was supported by the Western Project Program of the
Chinese Academy of Sciences (KZCX2-XB3-19), the State Key
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
2 Groups of the National Natural Science Foundation of China
(Grant No. 41121061). The authors are indebted to Professor
where the Fs, c, , , sat, , w, z, and m have been mentioned Fuchu Dai, Institute of Geology and Geophysics, Chinese Acad-
above, a is the seismic acceleration, and sat and are the densities emy of Sciences, and Dr. Sergey Marchenko, University of Alas-
of, respectively, saturated soil and soil above groundwater level ka Fairbanks, for constructive comments. Professor Hugh
(2.21 and 1.92 g/cm3 in the K3035W landslide site). French, Emeritus Professor, University of Ottawa, undertook
Based on Eq. 2, the critical value of seismic acceleration for Fs=1 final editing.
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 References
with the same conditions as the K3035W failure site. Furthermore,
the above calculation does not consider the influence of pore water Akgun A (2012) A comparison of landslide susceptibility maps produced by logistic
pressure; this will increase when an earthquake occurs and will regression, multi-criteria decision, and likelihood ratio methods: a case study at zmir,
Turkey. Landslides 9(1):93106
result in more unstable conditions. Therefore, the slope may reach Burn CR, Lewkowicz AG (1990) Retrogressive thaw slumps. Can Geogr 34(3):273276
its failure threshold even when seismic acceleration is less than Cheng GD, Wu TH (2007) Responses of permafrost to climate change and their
0.03g. environmental significance, Qinghai-Tibet Plateau. J Geophys Res 112:F02S03.
doi:10.1029/2006JF00063
Conclusions 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
Thaw-induced slope failures are widespread on the QTP, espe-
(2013) Tier-based approaches for landslide susceptibility assessment in Europe.
cially along the QTEC. The distribution of failures is closely Landslides 10(5):529546
related to the occurrence of icy permafrost, slope gradients of Huang RQ (2007) Large-scale landslides and their sliding mechanisms in China since the
between 6 and 10, and mean annual ground temperatures 20th century. Chin J Rock Mech Eng 26(3):433454 (in Chinese)
(MAGT) that are higher than 0.5 C. Slope failures in the study 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
area are classified as being either thaw slumps or active-layer-
Chinese
detachment failures that incorporate retrogressive thaw Jin HJ, Yu QH, Wang SL (2008) Changes in permafrost environments along the Qinghai-
slumping of the headwall. Tibet engineering corridor induced by anthropogenic activities and climate warming.
Monitoring of a typical failure indicates that the MAGT and Cold Reg Sci Technol 53(3):317333
depth to the permafrost table in the failure area are lower than Lewkowicz AG, Harris C (2005a) Morphology and geotechnique of active-layer detach-
ment failures in discontinuous and continuous permafrost, northern Canada. Geo-
similar values in undisturbed terrain. Headwall retreat is highest in
morphology 69(1):275297
the initial years after failure. The shear strength parameters in the Lewkowicz AG, Harris C (2005b) Frequency and magnitude of active-layer detachment
basal zone of the active layer were determined by in situ shear failures in discontinuous and continuous permafrost, northern Canada. Permafrost
tests. Slope stability analysis for the K3035W failure site was Periglac 16(1):115130
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 decades (Huggel et al. 2012), some studies from the Alps point to
diverse slope movements are compared to environmental factors an increasing number of rockfalls (Ravanel and Deline 2010),
such as ground surface temperature (GST) and snow cover, in order debris flows (e.g. (Hirschmugl 2003)) and large rock slope failures
to reveal and compare velocity fluctuations caused by changing (>105 m3, (Fischer 2010)) that initiated in permafrost areas over the
environmental conditions. The data cover 2years (20112013) of past two decades. Further, for many rock glaciers, an acceleration
Global Positioning System (GPS) and GST measurements at 18 (Roer et al. 2008; Delaloye et al. 2013) and for one even a total
locations on various slope movement types within an alpine study collapse (Krysiecki et al. 2008) has been observed and hypothe-
site in permafrost (Mattertal, Switzerland). Velocities have been sized to be caused by increasing air temperature (e.g. (Roer 2005;
estimated based on accurate daily GPS solutions. The mean annual Delaloye et al. 2010)). Increased thawing at rock glacier tongues,
velocities (MAV) observed at all GPS stations varied between 0.006 debris loading of gullies and the development of process chains
and 6.3ma1. MAV were higher in the period 2013 compared to could further increase the potential for debris flows and floods
2012 at all stations. The acceleration in 2013 was accompanied by a (Haeberli et al. 1990; Zimmermann and Haeberli 1992; Harris et al.
longer duration of the snow cover and zero curtain and slightly lower 2009; Lugon and Stoffel 2010; Graf et al. 2013). Geertsema et al.
GST. The amplitude (0600%) and the timing of the intra-annual (2006) also noted decadal increases in large landslides in British
variability were generally similar in both periods. At most stations, Columbia between 1973 and 2003 with most rock slides in their
an annual cycle in the movement signal was observed, with a phase study initiating in permafrost areas.
lag of 14months to GST. Maximum velocity typically occurred in Detecting spatial and temporal variation in the movement of
late summer and autumn, and minimum velocity in late winter and individual slopes is therefore central to our understanding of
beginning of spring. The onset of acceleration always started in underlying slope processes, and ultimately for application in risk
spring during the snowmelt period. At two stations located on steep reduction. The timing of acceleration or deceleration of slope
rock glacier tongues, overprinted on the annual cycle, short-term movements has the potential to allow us to identify governing
peaks of velocity increase, occurred during the snowmelt period, controls, for example, snowmelt infiltration and ground tempera-
indicating a strong influence of meltwater. ture. One simple and effective way to study the dynamics of
potentially unstable areas is the observation of surface displace-
Keywords GPS . Slope movements . Permafrost . ment. To date, most displacement measurements focus on either
Rock glacier . Landslide spatial resolution and coverage (e.g. (Kb et al. 1997; Strozzi et al.
2012)) with limited time steps or higher temporal resolution,
Introduction typically for a small number of punctual measurements (e.g.
Many mountain ranges, such as the Alps and the Himalayan (Malet et al. 2002)). To analyse short-term velocity fluctuations,
Hindu-Kush, have relatively dense populations at risk from natu- measurements with a high temporal resolution (e.g. daily) are
ral hazards related to slope instabilities (e.g. (Heim 1932; Govi required. In addition, effective methods for their interpretation
1989)). Since thawing processes reduce the strength of ice-rich are needed because many slopes have low rates of displacement (a
sediments (Morgenstern and Nixon 1971; Harris and Davies 1998) few cm per year). Continuous Global Positioning System (GPS)
and frozen jointed bedrock (Gruber and Haeberli 2007; Hasler measurements have proven to be particularly suitable for moni-
et al. 2011), permafrost degradation can develop or accelerate slope toring slope movements with high accuracy over long time periods
instabilities, such as deep-seated bedrock failures, creep-related and have been utilized to monitor landslides (e.g. (Malet et al.
processes, rockfalls, active layer detachment failures or increased 2002; Baldi et al. 2008)) and rock glaciers (e.g. (Limpach and
debris flow activity (Harris et al. 2001). Further, if average and Grimm 2009)).
extreme precipitation increase, as predicted for some areas due to Alpine slope movement types range from slow continuous
climate change (IPCC 2013), the frequency of rainfall-triggered displacement of entire slopes, where the main process is creep,
slope instabilities may rise (Beniston et al. 1996). Thus, while some to rapid, catastrophic landslides, with varied movement types
factors controlling slope stability (e.g. topography or lithology) often coexisting on the same slope or in close proximity. We use
remain rather constant overtime, others are sensitive to climate the term cryosphere-related mountain slope movements (CM-
forcing. Melting of surface or subsurface ice may respond rapidly movements, adapted from (Wirz et al. 2013)) to refer to slope
to climate forcing and may cause unexpected types of slope move- movements that (a) occur on moderate to steep mountain slopes
ments (Gruber 2013). Therefore, anticipated global climate change (angle: 1550) and (b) are related to cryosphere, i.e. strongly
is likely to increase instability of rock slopes and movement of ice- influenced by occurrence or changes to permafrost, seasonal frost,
rich debris (Haeberli et al. 1997). glacier debuttressing and/or snow. Studies on short-term variabil-
Although, there is still no unambiguous evidence that the fre- ity of CM-movements are few and mainly focused on one type,
quency or the magnitude of landslides has changed over recent such as rock glaciers (e.g. (Haeberli 1985; Krainer and He 2006;
Water flow is expected to occur in the subsurface, where perma- (250 m3, at 2973 m a.s.l.) in a less steep zone. L1c is located at
frost and bedrock may act as flow barriers. The slope of Breithorn 2870 m a.s.l. in a zone with comparably finer surface material and
supplies four main debris flow initiation zones with debris (Bhler more vegetation.
and Graf 2013). Within the unit Breithorn, three permanent GPS The unit RG Breithorn (R2) includes several distinct lobes with
stations were installed: L1a is located on a double-ridge (height: characteristic flow patterns of a rock glacier. The lower part of the rock
1050 m, length: 100 m) close to the top of Breithorn at an glacier has a convex profile curvature: approximately 100 m above the
elevation of 3140 m a.s.l. L1b is mounted on a huge block front, the slope suddenly increases from about 20 to 35. In this area,
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
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
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)
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 (MAV1213 4.3 ma1) were observed at the steep Patterns of temporal variability
fronts of RG Breithorn and RG Dirru. Both rock glaciers have In this section, the main movement patterns of the inter-annual
previously been identified as exceptional rapidly moving and intra-annual variabilities found in this study (Results and
(MAV 5 ma1, (Delaloye et al. 2013)). Cracks (Fig. 2) and fast interpretation) are discussed and compared to previous studies.
changes at their steep fronts further indicate a destabilization of We also discuss similarities and differences between landslides and
those landforms (cf., (Delaloye et al. 2010)). Currently, 15 such rock glaciers.
rapidly moving (destabilized) rock glaciers have been identified At all sites with data for both periods, MAV13 were higher than
in the European Alps (Delaloye et al. 2010). As expected, measured MAV12 (3969 %). This observed change accords well with the
velocities on slow-moving landslides (0.0070.401 ma1) are gen- measured acceleration (448 %) in the hydrological year 2013
erally lower than on rock glaciers, but are within the wide range of compared to 2012 at seven rock glaciers in the Valais and
measured surface velocities in previous studies (e.g. (Strozzi et al. Bernese Alps made by the Swiss permafrost monitoring net-
2005; Coe et al. 2003; Baldi et al. 2008; Brueckl et al. 2006; Strozzi work (PERMOS 2014).
et al. 2010)). However, velocities measured within the units We detected three main intra-annual patterns of movement
Grabengufer (3140 cma1) and Breithorn (1822 cma1 are high in rock glaciers and landslides: (a) an annual cycle, (b) a
when compared to previously observed velocities of a few to decrease in winter that is smoother than the rapid acceleration
several cma1 on a deep-seated landslides in permafrost, located in spring and (c) additional short and strong peak(s) in spring
in a neighbouring valley (Saas valley, CH; (Strozzi et al. 2005)). (during snowmelt period, Fig. 3). By contrast to previous
Generally, no correlation between observed slope movements studies (e.g. (Krainer and He 2006)), no continuous displace-
and slope angle was found, although at the fastest two sta- ment (without intra-annual variability) was observed at sta-
tions (R2b and R6c), slope angles are highest (R2b 33 and tions with sufficient temporal resolution to derive intra-annual
R6c 34, Table 1). variations.
1. Annual cycle. An annual cycle was observed at various stations, smoother compared to the acceleration in spring (Figs. 3, 8 and
including both rock glaciers and landslides (Fig. 3). The timing of 9). Such an asymmetric annual cycle seems typical for rock
the minima (Apr.Mar.) was similar for stations on landslides glaciers (Krummenacher et al. 2008; Perruchoud and Delaloye
and rock glaciers and agrees well with previous studies on rock 2007; Delaloye et al. 2010; Buchli et al. 2013), but was not
glaciers (e.g. (Lambiel et al. 2005; Perruchoud and Delaloye 2007; always detected in previous studies (e.g. on Muragl rock gla-
Buchli et al. 2013)), as well as on landslides in permafrost (Shan cier, (Arenson et al. 2002; Kb 2005)) most likely due to
et al. 2013) and seasonal frost (Coe et al. 2003). However, the coarser temporal resolutions (monthly).
observed timing maxima in velocity at stations on landslides 3. Short-term peaks. At two stations located on steep fast-flowing
(Nov.Dec.) differ from previous observations on landslides in rock glacier tongues (R7c and R2b with MAV4.3 ma1), short
seasonal frost (Shan et al. 2013) or permafrost (Coe et al. 2003), velocity peaks occurred in spring, in addition to the annual
where the maximal velocity was found in spring or early summer. cycle. The highest velocity during those peaks was more than
By contrast, the timing of maxima in velocity on rock glaciers three times higher than the maximum velocity in late summer
(Sept.Oct.) that occurs 1 to 3 months earlier than on landslides (cf. maxima of annual cycle). These peaks occurred during
agrees well with previous observations on rock glaciers (Haeberli snowmelt and related zero curtain period, but mostly termi-
1985; Roer 2005; Kb et al. 2005; Perruchoud and Delaloye 2007; nated (slowdown) before the end of the zero curtain period.
Buchli et al. 2013), although differences between individual rock On a rock glacier in a nearby valley, similar short peaks in
glaciers and from year to year have been observed (maxima in velocity during the snowmelt period, with the highest velocities
velocity between late summer and beginning of winter have been exceeding 140 % of the summer maximum, were detected
observed in the Alps, (Delaloye et al. 2010)). based on borehole-inclinometer data with a temporal resolu-
2. Smooth winter decrease/rapid spring acceleration. For stations tion of 6 h (Turtmann valley, CH; (Buchli et al. 2013)). This
on rock glaciers (except for R6a), the decrease in winter was rock glacier has high velocities (> 3ma1 ) and visible signs of
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 submerged in water and its effect on the free surface decreases with
element method for the simulation of impulse waves generated by time (Abadie et al. 2010). There are at least three phases, liquid (water),
landslides. The fluid-like landslide is modeled as a generalized gaseous (air), and solid (the landslide material), in landslide wave
non-Newtonian visco-plastic fluid. A three-phase flow model generation and propagation phenomena. Therefore, to simulate this
based on the incompressible viscous NavierStokes equations is problem correctly, a model capable of representing the interface
solved using the finite element method to describe the motion of between all these phases must be employed.
the three types of fluid in landslide. The conservative level set Over the last decades, large numbers of experimental and
method is expanded to n-phase flow cases and employed to cap- numerical studies have been conducted concerning landslide-
ture the interface of the three phases: air, water, and the landslide. generated waves, and both 2D and 3D experiments (Fritz 2002;
The overall performance of the approach is checked by a number Grilli and Watts 2005; Liu et al. 2005; Enet and Grilli 2005; Enet
of validation cases: a RayleighTaylor instability problem to illus- and Grilli 2007; Walder et al. 2003; Fritz et al. 2004; Panizzo et al.
trate the capability of the proposed method to deal with interface 2005; Heller and Hager 2010; Ataie-Ashtiani and Nik-Khah 2008;
capturing, a benchmark test of a subaerial landslide generated by de Carvalho and do Carmo 2007; Di Risio et al. 2009; Shigihara
an impulse wave is carried out and compared with the published et al. 2006; Zweifel et al. 2006; Najafi-Jilani and Ataie-Ashtiani
experimental data and numerical results, and finally, the 1958 2008) have been devoted to studying the behavior of landslide-
Lituya Bay landslide generated impulse wave, and its results are generated impulse waves, such as landslide motion, effective pa-
compared against a scaled-down experiment and other published rameters, wave characteristics, and associated run-up. A
numerical results. It can be noted that the current model has an comprehensive review of these studies can be found in
excellent ability to capture the complex phenomena that occurs Mohammed and Fritz (2012).
during the whole process of the landslide-generated impulse wave, However, in order to help understand the nature of the pro-
and considering the simplified treatment of the landslide and the cesses involved and predict the detailed outcome in specific loca-
numerical model, fairly good agreement between computed and tions, the development of computational tools for the modeling of
experimental results has been observed for all simulation cases. water waves induced by landslide seems to be inevitable
(Cremonesi et al. 2011). If there is no impulse wave problem
Keywords Landslide . Impulse wave . Three-phase involved, both continuum and discontinuous method can be
flow . Conservative level set . Finite element method employed to descript the propagation of the landslide such as
the finite element method (FEM), the discrete element method
Introduction (DEM), smoothed particle hydrodynamics (SPH), and discontinu-
Landslides represent one of the most destructive hazards in nature, ous deformation analysis (DDA). One of the main difficulties in
and the enormous damage associated with this phenomena is not developing a numerical method for simulating waves generated by
only due to the large amount of mud and debris material involved landslide movement is to model the strong interaction and energy
but also because of the potentially large generated impulse waves transfer between the moving deformable slide and the surround-
(Serrano-Pacheco et al. 2009), also referred to as landslide ing water and air. This has received enormous effort over the last
tsunamis(Ward 2001). The impulse waves are characterized by a several decades; however, due to the complexity of the phenome-
short wave length and high amplitude and can be more devastating na, different levels of simplification have had to be made in the
than meteorological flood waves (Cremonesi et al. 2011) as they can various computational models used. Prior to the development of
overtop dams or run-up along the shoreline destroying villages computational tools, only analytical or empirical formula was
located along the shore with catastrophic consequences. Numerous proposed to give rough estimates of the wave amplitude. With
instances have been observed all over the world, and two of the most the development of the powerful numerical techniques and in-
devastating examples include the 1958 Lituya Bay case, where a creasing computational capacity, a large number of numerical
landslide tsunami destroyed a forest up to a run-up height of methods have been proposed based on a variety of models and
524 m above mean sea level, and the 1963 Vaiont case in Northern hypotheses (Ataie-Ashtiani and Yavari-Ramshe 2011). According to
Italy, where an impulse wave overtopped a dam by more than 70 m the degree of simplification of the fluid, the developed models can
and caused about 2000 casualties (Heller and Spinneken 2013). It is, be categorized as the Boussinesq type (Ataie Ashtiani and Najafi
therefore, of great importance to understand and estimate the be- Jilani 2007; Hanes and Inman 1985; Lynett and Liu 2002;Watts et al.
havior and characteristics of landslide-induced impulse waves. 2003; Fuhrman and Madsen 2009; Zhou and Teng 2009), models
The motion of a subaerial landslide is typically divided into three based on the shallow water equations (Fernndez-Nieto et al. 2008;
processes: (1) after triggering, the slide propagates over some distance, Medina et al. 2008; Saut 2003; Chaudhry et al. 1983; Townson and
only interacting with air; (2) the slide penetrates the water, while part Kaya 1988), and the potential flow equations (Grilli et al. 2002;
of the slide still interacts with air; and (3) the whole slide becomes Grilli and Watts 1999, 2005). It should be noted that these analyses
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
u u* u** 21 Following the basic idea of the TSTG procedure, Eq. (25) can be
simulated as following two steps:
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)
Fig. 3 RayleighTaylor problem, non-dimensional linear growth rate against non-dimensional wave number
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
Fig. 7 Snapshots of simulated interface profile at t=0.4 s (left panel) and t=0.8 s (right panel)
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- channel, so that the experiments are equivalent to a two-
dimensional times between the presented results (the right- dimensional problem in a vertical plane. The dimensions of the
hand-side panels) and the results from Pochet et al. (2013) sand mass in cross-section are 0.65 m0.65 m and the mean
(the left-hand-side panels) are shown in Fig. 2. It can be seen apparent density is 1950 kg/m3. Because the fluid yield stress is
that the results using the proposed methods match perfectly unknown, the current simulation assumes a zero yield stress. A
with the published data throughout the whole simulation. frictional fluid model is adopted with the internal friction angle
Both the large mushroom structure and the small vortex =15. The water depth is 1.60 m and the top of the triangular mass
structures are clearly visible and well resolved. The simulation is initially 0.1 m below the water surface. A free-slip condition is
is numerically stable at all times and shows good ability in applied along the slope. The computation domain is mesh using
handling complex flow problems. 73,628 four nodes quadrilaterals with 74,029 nodes depicted in
The numerical results are now compared to the analytical Fig. 5. More finer mesh density is employed surrounding the
solution obtained in Tryggvason (1988). In the early stages submarine landslide, and the mesh size varies from 0.008 to
when the amplitude of the perturbation is much smaller than 0.04 m. The automatic adjustment of the time step is according
the wave length, the perturbation of the fluid interface has an to the following stable conditions C2+2C/Pe1, where C is the
exponential growth 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
h h0 et 34
(lower panel). The results are acceptable but show differences
with the experiments. It can be seen that the model captures
where h is the amplitude at time t, h0 is the initial amplitude,
the trend of the interface of the wave generated by the
and is the growth rate of the perturbation. With a kinetic
underwater landslide correctly but slightly over-predicts the
viscosity ratio of one and negligible surface tension, the
amplitude. Similar numerical results were obtained by other
growth rate is a function of both the Atwood number and
researchers like Rzadkiewicz et al. (1996) and Gangfeng Ma
wave number k 2=. Chandrasekhar (1961) derived the et al. (2013) with a much larger discrepancy with the experi-
following non-dimensional relationship between the growth mental data. The discrepancies are introduced by the simpli-
rate and the wave number of the initial disturbance: fied treatment of the landslide in which granular stresses
which could retard the initial failure of the sliding mass have
=g 2 = 1=3 ; k k=g= 2 1=3 35 been neglected (Ma et al. 2013). The second reason for the
deviation is that the sediments are modeled by an ideal fluid.
where and k is the non-dimensional growth rate and wave In Fig. 7, the snapshots of the simulated interface profiles are
number, respectively. In Fig. 3, the normalized growth rate plotted for each time instance investigated. From that, it can be
and the wave number are plotted and compared with the seen that the landslide has penetrated deeper into the water
analytical solutions of Gretar Tryggvason, and very good
agreement can be observed.
channel from t=0.4 s to t=0.8 s and causes a change in the water Application to the Lituya Bay impulse wave
air interface. Also, at t=0.8 s, the landslide mass has lost its original In all the documented cases of landslide-generated impulse waves,
triangular shape due to the shearing effect of the water and several there are only a few experiments with sufficient field details for
vortex-like bubbles can be seen to form at the landslide/water conclusive comparison with a simulation. Among those, the 1958
interface which slowly detach from the main mass. This phenom- gigantic wave generated in Lituya Bay has received extensive
enon is similar to the RayleighTaylor instability problem men- attention from a number of researchers (Quecedo et al. 2004;
tioned above and provides an insight into deformable landslide Pastor et al. 2005, 2009; Biscarini 2010; Mader 1999; Fritz 2001;
and water interaction. Weiss et al. 2009).
Based on the above discussion, it can be noted that the pro- Lituya Bay is a large inlet on the northeast shore of the Gulf of
posed method can be used to capture the impulse wave generated Alaska, a seismically active area particularly susceptible to
by a subaerial landslide, and an improved match with the exper- landslide-induced waves. On July 8, 1958, an 8.3 magnitude earth-
imental data has been obtained over existing numerical models. To quake along the Fair Weather fault triggered a major subaerial
further address the problem, an application to the Lituya Bay rockslide into the Gilbert Inlet. The initial slide geometry is as-
impulse wave generation is presented in the next section. 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.)
Fig. 14 Effect of the internal friction angle on the final stage of run out: a =0; b =5; c =10; d =15
Abstract Landslides are recognized as one of the most important main objective of modeling the landslide susceptibility is the
natural hazards in many areas throughout the world. Producing prediction of landslide-prone areas in space and/or time.
landslide susceptibility maps have received particular attention Several landslide susceptibility mapping methods such as direct
from a wide range of scientists. The main objective of this study geomorphological mapping, analysis of landslide inventories, heu-
was to produce landslide susceptibility maps using hybrid wavelet ristic methods, neural networks, fuzzy logic, expert systems, and
packet-statistical models (WP-SM). In the first step, landslide process-based models have been used to assess the vulnerability of
susceptibility maps were produced using single artificial neural different parts of an area to landslide (Gokceoglu and Sezer 2009;
network (ANN), support vector machine (SVM), maximum entro- Bednarik et al. 2010; Mihaela et al. 2011; Pourghasemi et al. 2012a,
py (MaxEnt), and generalized linear model (GLM). In the next b, c, d; Xu et al. 2012, 2013a, b; Tien Bui et al. 2012a, b, c, d). So, two
step, the input maps were preprocessed using different mother approaches can be mentioned establishing models for predicting
wavelets in different levels. Then, the hybrid models were devel- the geographical distribution of a phenomenon. The first one
oped using the wavelet-based preprocessed maps. Results showed attempts to disclose the fundamental mechanisms that the envi-
that the wavelet packet transform can be effectively used to pro- ronmental factors exert on a certain phenomenon. The second
duce precise landslide susceptibility maps. It was shown that tries a mathematical function, based on various supervised classi-
wavelet packet transform significantly enhanced the ability of the fication algorithms, to correlate the spatial distribution of the
single statistical models. The kappa coefficients were increased phenomenon with the environmental factors (Brenning 2005). A
from 0.829 to 0.941, 0.846 to 0.978, 0.744 to 0.829, and 0.735 to number of statistical methods (also called niche modeling ap-
0.817 in hybrid ANN, SVM, MaxEnt, and GLM, respectively. The proaches), i.e., artificial neural network (ANN), support vector
best wavelet transform was performed using bior1.5 with a three- machine (SVM), maximum entropy (MaxEnt), and generalized
level decomposition. It was also recognized that MaxEnt and GLM linear model (GLM) which lie in the second category have been
produced approximately poor results. However, SVM performed increasingly used in LS mapping (Yao et al. 2008; Yilmaz 2010;
better than the other three models both in single and hybrid forms. Marjanovi et al. 2011; Xu et al. 2012, 2013a, b, 2014; Pradhan 2013;
ANN also outperformed MaxEnt and GLM models. Spatial distri- Conforti et al. 2014). Neural network is a method that is first
bution of the susceptible area is consistent with the observed trained from known data and then used to categorize unknown
landslide distribution pattern particularly in maps obtained from pixels. SVMs demonstrate a set of theoretically superior machine-
the hybrid models. The produced maps showed that the general learning algorithms. Development of SVM was initially triggered
pattern of susceptible area intensively followed the pattern of by the exploration and formalization of learning machine capacity
roads and sensitive geological formations. control and over-fitting issues (Vapnik 2001). GLM forms a mul-
tivariate regression namely link function, between a generalized
Keywords Landslide susceptibility . Wavelet dependent variable and several independent variables (Atkinson
packet . SVM . ANN . Generalized linear model . Maximum and Massari 1998). The MaxEnt method is a generative approach
entropy which makes predictions using statistical concepts, analyzes the
weights of environmental factors, and finally calculates a contin-
Introduction uous probability value for the mentioned parameter distribution
Landslide is a natural phenomenon that plays a critical role in the (Phillips et al. 2006). Wavelet transform (WT), as a signal process-
landscape evolution and represents a serious vulnerability in many ing approach, can be used for multi-scale analysis of a signal
areas of the world (Glade et al. 2005). Different phenomena can through dilation and translation that can extract signal features
cause landslides, including intense or extended rainfall, swift snow from both time (or position) domain and frequency domain,
melting, earthquakes (Wilson and Keefer 1985; Ingls et al. 2006), effectively (Peng and Chu 2004). Wavelet transforms offer useful
volcanic activity (Ablay and Hrliman 2000; Moon and Simpson decompositions of the original signals, so that wavelet-based
2002), and human activities (Skempton and Hutchinson 1969). transformed data enhance the ability of prediction models by
Landslides involve flowing, sliding, toppling, falling, or a combi- capturing useful information on various resolution levels (Kisi
nation of two or more types of movements at the same time or 2009; Nourani et al. 2009; Adamowski and Sun 2010; Moosavi
during the lifetime of a landslide (Cruden and Varnes 1996; Brckl et al. 2013b, c). Initial researches indicated that wavelet analysis
2001; Srensen and Bauer 2003). They are the most important, appears to be a more effective tool than the Fourier transform
costly, and damaging natural hazards in mountainous regions (Adamowski 2007). The wavelet packet transform (WPT), which is
(University of Utah 1984). Consequently, it is valuable to assess a generalization of wavelet transform, is alternative bases formed
and evaluate the landslide susceptibility and vulnerability in a by taking linear combinations of usual wavelet functions (Ogden
region. Landslide susceptibility (LS) can be defined as the proba- 1997). The wavelet transform technique has been explored in
bility of the spatial occurrence of known slope failures given a set numerous studies (Liu and Ngan 2006; Borah et al. 2007; Li
of geological and environmental conditions or the spatial proba- 2009; Zhou et al. 2010; Park et al. 2011; Demirhan and Guler 2011;
bility of landslide occurrence (Guzzetti et al. 2006). Therefore, the 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 the approach proposed by Moosavi et al. (2014). They produced
wavelet packet analysis with statistical GIS-based models for land- inventory maps using advanced pixel-based and object-oriented
slide susceptibility modeling. The main objective of this research is image classification approaches optimized by Taguchi method
to evaluate the effect of wavelet packet transform as a preprocess- (Taguchi 1990). Their results showed that object-oriented ap-
ing approach on the GIS-based models in landslide susceptibility proaches outperform pixel-based ones. Therefore, we used
mapping. Taguchi-based optimized object-oriented approaches in this study.
Then, an intensive field survey has been conducted to validate and
Study area precisely detect and record the landslides geographical coordi-
The study area covers a region of nearly 272 km2 around the Ilam nates using a GPS. A total number of 102 landslides were recorded.
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 roads, highways in the mountainous regions of the study area
level, respectively. Climatic condition in the area is semi-arid (Fig. 3). The maximum and minimum size of landslides in the
(typical Mediterranean climate) with 595 mm rainfall, and average study area is 0.53 and 0.01 km2, respectively.
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, Landslide conditioning factors
forest, grass land, garden, and water body. The Ilam dam is one Determination of the landslide conditioning factors is a key step in
of the most important dams in the western Iran that supplies the landslide susceptibility assessment (Ercanoglu and Gokceoglu
potable water for the Ilam City. The geology of the study area is 2002). Fourteen landslide conditioning factors including elevation,
very complex, and the lithology units comprise several formations. slope gradient, slope aspect, slope length, topographic wetness
In this region, geological units exist from Cretaceous period to index (TWI), stream power index (SPI), plan curvature, profile
Quaternary sediments. Among various lithological formations in curvature, geology, normalized difference vegetation index
the area, the marl, shale, and silt deposits are known to be the (NDVI), land use, and distance from drainage, fault, and road
most susceptible formations to landslides (Table 1). Tectonic fea- have been considered in this research according to the literature
tures of the study area include folds, thrusts, and strike-slip faults review. Thematic layers were acquired from different resources
(Harison and Falcon 1936). using ArcGIS, SAGA GIS, and ENVI packages (Bhner et al.
2006). The descriptions of the used input data, their spatial reso-
Materials and methods lution, and the sources are given in Table 2 and briefly explained
Figure 2 shows the flowchart of this study. As this flowchart shows, below. As mentioned in this table the resolution of all conditioning
four general stages have been defined in order to develop hybrid factors is 15 m because we used ASTER satellite imagery (15-m
wavelet packet-statistical models. In the first stage, the most im- resolution) to produce several base maps. Total number of pixels is
portant conditioning factors were selected according to the litera- 1,211,275 in the study area from which 14,125 pixels are landslides.
ture review. Then, the base maps were produced using Elevation has been used as a controlling parameter for land-
topographic, geological, and remotely sensed data. In the next slides in several research studies (Greenbaum et al. 1995; Jordan
stage, the mentioned base maps were classified and the frequency et al. 2000; Dai et al. 2001; Yilmaz 2010). It is well known that
ratio (FR) for each attribute class was calculated crossing each elevation influences a large number of biophysical parameters,
map with the landslide inventory map. Afterward, the FR values anthropogenic activities, and soil properties (Vivas 1992; van
were normalized between 0.1 and 0.9. In the third stage, two Beek and van Asch 2004; Kuriakose et al. 2009b). The other
modeling approaches (i.e., single and hybrid) were conducted important parameter in landslide susceptibility is slope gradient.
separately. In the single modeling part, landslide susceptibility It determines the convergence and divergence of water on a
maps were produced using single ANN, SVM, MaxEnt, and GLM hillslope and thereby influences soil water content, soil forma-
models. In the hybrid modeling process, first, the base maps were tion, erosion potential, and slope stability (Lee and Min 2001;
decomposed to approximation and detail components using dif- Clerici et al. 2002; Lee et al. 2004; Lee 2005; Vijith and Madhu
ferent mother wavelets in different levels. Second, their approxi- 2008). Slope aspect which is defined as the compass direction of
mation and detail components were used as inputs (instead of the the maximum rate of change is also considered by several
original maps) for the abovementioned statistical single models. In researchers as a landslide conditioning factor (Gokceoglu and
the last stage, TPR, AUC, and kappa were used to evaluate the Aksoy 1996; Saha et al. 2005; Ercanoglu and Gokceoglu 2004;
performance of both single and hybrid models. Details of these Lee et al. 2004). It is reported by Lineback et al. (2001) that
stages are explained in the following sections. 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
Landslide inventory map initiation to some extent.
Producing a landslide inventory map is necessary to describe the Slope length can be considered as another important factor in
relationship between the landslide distribution and the condition- landslide activity (Carrara et al. 1995). It is the distance along a
ing factors. Detecting the accurate location of landslides is of great slope subject to uninterrupted overland flow, from of the point at
importance in landslide susceptibility analysis. As the detection of which overland flow begins to where deposition starts, or where
landslide location is often time-consuming and labor-intensive, flow enters a well-defined channel (Wischmeier and Smith 1978),
remote sensing methods have been employed in addition to field since longer slope lengths increase the potential of erosive agents
surveys. The initial landslide inventory map was produced using to dislodge and transport materials downslope. Moreover,
(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 for shallow landslides. It is defined as follows (Beven and Kirkby
Kavzoglu 2005). The slope length is of paramount importance for 1979; Moore et al. 1991):
the travel distance of materials. Slope length (LS) factor can be
calculated by the equation proposed by Moore and Burch (1986) as As
TWI ln 2
follows: tan
As 0:6 sin 1:3 where As is the specific catchment area (SCA) and is the local
LS 1:4 1 slope angle (in degrees).
22:13 0:0896
SPI is a measure of the erosive power of water flow based on the
where AS is the specific watershed area (m2/m) and is slope angle assumption that discharge (q) is proportional to specific catch-
(in degrees). ment area (As) (Eq. 3) (Moore et al. 1991).
TWI represents a theoretical measure of the accumulation of
flow at any point within a drainage basin and the tendency of SPI AS tan 3
water to move downslope by gravitational forces (Fredlund 1987;
Kuriakose et al. 2009a). The index can be thought of as an abstract where AS is the specific catchments area (m2/m) and is the slope
parameter to be used as a basis for estimating the local soil gradient (in degrees). Moore et al. (1993) stated that the SPI
moisture status (Burt and Butcher 1986; Moore et al. 1991). Soil controls the potential erosive power of overland flow. Therefore,
moisture plays an important role in slope instability, particularly these processes can be considered as one of the components of
landslide occurrence (Lee and Min 2001; Gokceoglu et al. 2005; and on the side of the slopes affected by roads. This is mainly due
Nefeslioglu et al. 2008; Yilmaz 2009; Akgun and Turk 2010). to the fact that the natural condition of the slope is damaged
Distance from drainage, distance from fault, and distance from during the process of road construction when engineering princi-
road are the other important landslide conditioning factors ples, particularly principles related to drainage network are not
(Nagarajan et al. 2000; Dai et al. 2001; Cevik and Topal 2003; taken into account (Ayalew and Yamagishi 2005). Highly faulted
Ercanoglu and Gokceoglu 2004; Pradhan 2010a, b). Drainage net- zones are areas of particularly high incidence of unstable slopes
works have negative impacts on landslide susceptibility since they and the degree of fracturing, and shearing plays an important role
abrade the slope base and saturate the underwater section of the in determining slope stability (Varnes 1984). The plan and profile
material forming the slope (Vijith and Madhu 2008). Similar to the curvature which control the concentration, acceleration, and de-
effect of the distance to streams, landslides may occur on the road celeration of subsurface flow can affect the occurrence of
landslides (Ercanoglu and Gokceoglu 2002; Troch et al. 2003; one where crops are cultivated can increase the moisture in the
Talebi et al. 2008; Nefeslioglu et al. 2008; Oh and Pradhan soil enough to cause landslide problems. In this study, land cover/
2011). Landslides are also significantly controlled by the lithol- land use map was extracted from ASTER satellite imagery (August
ogy properties of the land surface (Dai et al. 2001; Duman 2013) using a pixel-based classification approach with the SVM
et al. 2006). Geological factor is considered as an independent algorithm. NDVI is a measure of surface reflectance and gives a
variable in landslide formation in which lithology and its quantitative estimate of the vegetation growth and biomass (Hall
varied structures tend to lead to a variation in stone stability et al. 1995). The NDVI was calculated using the ASTER satellite
and strength and also to a varied soil texture (Ayalew and imagery. Figure 4 shows the input parameters (conditioning fac-
Yamagishi 2005). tors) in the study area.
The next important factor which affects the occurrence of
landslide is land cover and/or land use (Gokceoglu and Aksoy Normalization of the input dataset
1996; van Beek et al. 2005; Kuriakose and van Beek 2011; Devkota In this step, the frequency ratio (FR) for each individual attribute
et al. 2012). For example, converting a forested area to grassland or 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
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 developed using wavelet packet transform. These approaches are
FR value. Then, the max-min normalization method was used explained below.
to rescale in the range 0.1 to 0.9 using Eq. (4) (Wang and
Huang 2009): Single statistical models
An ANN can be defined as a computational mechanism able to
Min acquire, represent, and compute a mapping from one multivariate
0 UL L 4
Max Min space of information to another, given a set of data representing
that mapping (Garrett 1994). The advantage of neural networks is
where v' is the normalized data matrix, v is the original data related to the high computation rate accomplished by their
matrix, and U and L are the upper and lower normalization inherent parallelism that is a result of a potent arrangement
boundaries. Table 3 shows the values of FR for each class of of interconnections (weights) and simple processors (neu-
conditioning factors. rons), which allows processing of very large datasets (Frizzelle
and Moody 2001).
Landslide susceptibility mapping In this part, 102 landslide points were randomly split into two
Here, two approaches have been used to produce landslide sus- datasets (75 % of the data for training phase and 25 % for valida-
ceptibility maps. Initially, landslide susceptibility maps were pro- tion phase). The feed-forward back-propagation network which is
duced using four single statistical models, i.e., ANN, SVM, GLM, the most frequently used neural network method was used in this
and MaxEnt. Then, hybrid wavelet-statistical models were study. This algorithm is a multi-layered neural network, which
consists of an input layer, hidden layers, and an output layer. The gradient, slope aspect, slope length, TWI, SPI, plan curvature,
hidden and output layer neurons process their inputs by multi- profile curvature, distance from drainage, geology, distance to
plying each input by a corresponding weight, summing the prod- fault, land use, distance to road, and NDVI). There are several
uct, and then processing the sum using a nonlinear transfer methods to determine the optimum number of nodes in the
function to produce a result. It learns by regulating the weights hidden layer including 2n+1 (Kavzoglu and Mather 2003), 2n
between the neurons in response to the errors between the actual (Wong 1991), n (Tang and Fishwick 1993) hidden neurons, where
output values and the target output values. In this study, a three- n is the number of nodes in the input layer. Mishra and Desai
layered, interconnected neural network consisting of an input (2006), Moosavi et al. (2013b) and Shirmohammadi et al. (2013)
layer, a hidden layer and an output layer was used. In this specific declared that the trial and error is an appropriate way to deter-
case, there are 14 input nodes (one each for the elevation, slope mine the best number of neurons in the hidden layer. As the
like the RBF for certain parameters (Song et al. 2011). In the
present study, the four mentioned kernel functions were tested where j and k are integers and s0, which is greater than zero,
to determine the one which gives the best result. is a fixed dilation step (Cannas et al. 2006). The DWT per-
The principle of maximum entropy states that, subject to pre- forms two sets of functions namely high-pass and low-pass
cisely stated prior data (such as a proposition that expresses filters. A two-dimensional wavelet decomposition can be de-
testable information), the probability distribution which best rep- scribed as a one-dimensional discrete wavelet transform ap-
resents the current state of knowledge is the one with the largest plied consecutively along the horizontal (rows of the map or
entropy. The MaxEnt model is a niche-based model that suppose image) and vertical (columns of the map or image) axes
the distribution of the presence data, represents the realized niche (Mallat 1989).
(Heumann et al. 2011). MaxEnt method is a generative approach The wavelet packet decomposition which is a generalization
which is used to make predictions using statistical concepts of the conventional wavelet decomposition that offers a richer
(Phillips et al. 2006). It analyzes individually the weights of signal analysis (Garcia et al. 2000) was used in this study. In
environmental factors and calculates a continuous probability the first step, the original input map was decomposed into
value for the distribution of a certain phenomenon. As with coarser approximation coefficients (CA) which contains low
other machine learning models, the maximum entropy model frequency horizontal and vertical components of the map,
requires some optimization procedures. Thus, before the gen- horizontal detail coefficients (Dhorizontal), vertical detail coeffi-
eration of the landslide susceptibility map, optimal settings cients (Dvertical), and diagonal detail coefficients (Ddiagonal).
are first searched by predictive performance measures based This procedure was iteratively performed in different levels.
on cross-validation. Using optimal settings, the landslide sus- Consequently, the original signal was decomposed to its ap-
ceptibility map over the study area is generated, and the proximation and detail components. It finally resulted in a
relationships between the input causal factors and landslide wavelet decomposition tree. In this study, the input maps
susceptibility are interpreted. were decomposed using different mother wavelets such as
GLM which allows one to form a multivariate regression Daubechies (db), Symlets (Sym), Coiflets (coif), and
relation between a dependent variable and several indepen- BiorSplines (bior) in different levels (1, 2, 3, 4, and 5) using
dent variables is also widely used for landslide susceptibility Matlab software.
mapping (Ayalew and Yamagishi 2005; Brenning 2005). Logis- Figure 5 shows the decomposition tree obtained from de-
tic regression as a generalized linear model with a logit link composition of original NDVI map using bior1.5 mother wave-
function, which is one of the most frequently used method in let in level 3 as an example. In this figure, the first node (0,
landslide susceptibility modeling, was used in this study 0) is the original map (i.e., NDVI) that is decomposed to four
Level 1
Level 2
Level 3
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 decomposition tree. This function uses an entropy type crite-
(which is shown as (1, 0), (1, 1), (1, 2), and (1, 3) nodes, rion. However, as in several cases, the optimal subtree that is
respectively) in the first level. In the next level, each of these obtained by this function is still large, the five components
components is decomposed to its subcomponents (which are that had the least entropy values were selected. Therefore, the
shown as (2, 0), (2, 1), (2, 2), etc.). The same procedure is best approximation and detail components (the five compo-
carried out in level 3. As it is shown from this figure, a large nents with the least entropy values) of each conditioning
number of components are made especially when the decom- factor were used as inputs for all the aforementioned models.
position level increases. It is not possible to use all of these
components in the modeling process. Classical entropy-based Model performance criteria
criteria are suitable for this purpose. Entropy is a common Using several statistical evaluation criteria such as true positive
concept in many fields, mainly in signal processing (TP), false positive (FP), true negative (TN), and false negative
(Daubechies 1988). Matlab has a function called besttree in (FN), the overall accuracy of the landside susceptibility models
order to compute the optimal subtree from the initial were calculated. The reliability of the landslide susceptibility
waveforms indexed by three naturally interpreted parameters: (particularly in single form) which categorized most of the area as
position, scale (as in wavelet decomposition), and frequency. Con- very susceptible to landslide occurrence. In contrast, FN is high
sequently, all of the advantages of the fast wavelet transform are in GLM which indicates that this model could not show susceptible
preserved since the wavelet basis is in the range of bases available
with the wavelet packet transform (Shinde 2004). The complete 0.99
structure of wavelet packet transform makes flexibility for the 0.98
Performance criteria
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 1978; Pierson 1992; Bunce et al. 1997; Budetta 2004; Lan
constantly exposed to slope instabilities. Proactive mitigation et al. 2010).
strategies have been in place for a few decades now, and instability Recognizing the risks associated with slope cuts in the Cordil-
record keeping has been recognized as an important aspect of lera, CP engaged in the development and implementation of a rock
them. Such a proactive strategy has enhanced the industrys ca- slope management program in the early 1970s (Brawner and
pacity to manage slope risks, and some sections have been recog- Wyllie 1975). This system has evolved into a qualitative rating
nized as critical due to the frequency of instabilities. At these system to describe the slope hazard and its likelihood of occur-
locations, quantification of the risks becomes necessary. Risk anal- rence (Mackay 1997). Mitigation works (such as protection walls,
ysis requires knowledge of some variables for which statistical data ditch widening and maintenance, face stabilization and scaling)
are scarce or not available, and elicitation of subjective probabil- follow site inspections in the priority indicated by the rating
ities is needed. A limitation of such approaches lies in the uncer- system. This form of risk assessment and management is currently
tainty associated to those elicited probabilities. In this paper, a being applied. However, results from this approach are not readily
quantitative risk analysis is presented for a section of railway amenable to comparison with other hazard assessments that the
across the Canadian Cordillera. The analysis focused on the risk railway may employ. Hence, the need for the development of a
to life of the freight train crews working along this section. Upper quantitative risk assessment approach that minimizes the amount
and lower bounds were elicited to cope with the uncertainties of qualitative inputs.
associated with this approach. A Monte Carlo simulation tech- A quantitative risk assessment (QRA) associated with slope
nique was then applied to obtain the probability distribution of instabilities between milepost 2 and 15 of CPs Cascade subdivision
the estimated risks. The risk probability distribution suggests that is developed for this study. The assessment focuses on the risk to
the risk to life of the crews is below previously published evalua- life of running trade employees working along this section. CPs
tion criteria and within acceptable levels. The risk assessment extensive records dating back to the 1940s serve as the main input
approach proposed focuses on providing a measure of the uncer- for the hazard analysis and are of significant value in the conse-
tainty associated with the estimated risk and is capable of handling quence analysis stage. The risk estimation considers measures of
distributions that cover more than two orders of magnitude. the uncertainty in the input parameters and how these are carried
through the analysis to reflect uncertainty in the calculated risk
Keywords Rock falls . Uncertainty . Quantitative risk values.
assessment . Monte Carlo simulation
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
Track speed
No AxBxCxDxExF GxH
(40 km/h)
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.
Min
quantitatively through the analysis, would allow for a measure of
Sta an
nd
Me
Consequence
RISK
Vulnerability
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
Fig. 5 Monte Carlo simulation process to estimate the probability of a fatality and the risk to life of a crew member
Probability that the HDS is activated These adopted probabilities and their justification are presented in
The probability that the HDS is activated depends on the instabil- Table 3. Continuous upper and lower boundaries for the probabil-
ity volume, the spacing between wires, and the height of the HDS ity distributions that the HDS is activated are based on Table 3 and
relative to the slope geometry. An example of the HDS layout on are presented in Fig. 9.
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. Probability that a warning is issued and train speeds
Insufficient records include information on HDS activation, When a section of the HDS is activated, the nearest track circuit
whether the material was encountered blocking the track and the signal shows a track as occupied and activates a slow order. This
instability volume, for a reliable estimate of the HDS activation implies that the first train to encounter the blocked track receives a
probability. Based on the characteristics of the HDS and the slopes warning only if there is a track circuit signal between the activated
along the study area, we judged the HDS activation probabilities. HDS and the train. The section between track circuit signals is
0.6
er
Table 4 Input parameters used to estimate the probability of a warning (P[warn- 0.4
Low
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
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
0.6
s = standard
m+s
-s +s 0.4
m+2s
0.2
Risk m-2s
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 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
(Morgenstern 1995; Finlay and Fell 1997). The risks estimated in this
Canadian population (Statistics Canada 2010). (2) Data from Baecher and Christian
chapter are compared against widely used risk evaluation criteria. (2003). (3) Porter et al. (2009) suggestion that the incremental risk is low if it does
Even though these criteria were proposed for other locations and not exceed 0.2 % of the Canadian age-standardized risk of loss of life. (4) HSE
contexts, they are considered applicable for illustrative purposes. (2001). (5) ANCOLD (2003). (6) ERM (1998)
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 traces by the authors in the paper (Laperdin et al. 2014).
seven simultaneous debris flows going down the slopes of with analysis of the events consequences and photographs.
the Tunka Ridge in the vicinity of Arshan village, which is a In this article, we have compiled a more precise chronology
balneological and alpine resort (51 54 31 N, 102 25 44 E). of the events and have attempted to determine the main
The debris flows caused one life loss and several injuries, 112 causes of this natural phenomenon.
buildings were damaged, and 15 were completely destroyed.
The total volume of the transported deposits amounted to Study area
3106 m3. Debris flows formation began with the failure of
weak sediments in the hanging cirques. Similar phenomena Geographical and geomorphological setting
had not been recorded in the study area for over 40 years. Arshan village is situated at the foot of the southern slope of
The article presents a complete picture of the event and the Tunka Ridge, where the Kyngarga River enters the Tunka
analysis of geological, geomorphological, and meteorological Valley, at the altitude of 893 m a.s.l. (Fig. 1). In the village,
conditions for debris flows formation, for which extreme healing mineral and thermal waters discharge at the faults
local rainfall was the major cause. junction (Pinneker et al. 1968). The village is a balneological
and alpine resort which is visited annually by tens of thou-
Keywords Central Asia . Debris flow . Tunka Ridge . Mountain sands of people. The resident population of the village is
areas 2700 people, but tourists and visitors increase this figure
significantly, especially during the summer season. The vil-
Introduction lage housing is comprised of onetwo-storey houses and a
Worldwide debris flows cause catastrophic damage in moun- small number of threefour-storey buildings of the resorts
tain and foothill areas (Hrlimann et al. 2006; Yueping et al. and hotels; the constructions are mainly wooden with some
2011; Mateos et al. 2012; Jakob et al. 2013; Xiao et al. 2013; exceptions for brick.
Seplveda et al. 2014; Youssef et al. 2014; Bregoli et al. 2015). The Tunka Ridge is the eastern spur of the Eastern Sayan
Investigations of debris flows show that activation of debris Ridge. The maximum elevation of the Tunka Ridge is 3284 m
flow processes is often caused by abundant rainfall (Allen (Strelnikov Peak) with the altitude range of 2000 m. Within
et al. 2015; Martha et al. 2015). the case area, the maximum elevation reaches 2627 m a.s.l.
On 28 June 2014, at about 8 a.m. local time, as a conse- The ridge is bounded from the north and south by Main
quence of a heavy and prolonged rainfall in the vicinity of Sayan and Tunka faults, respectively (Fig. 1). The ridge is
the resort village of Arshan (Tunka District of the Republic eastwest orientated; its southern slope changes into pied-
of Buryatia), the Kyngarga River bursted its banks, and mont plain (Fig. 2). Within the study area, the ridge is cut
seven debris flows of various sizes simultaneously descended by deep valley of the Kyngarga River and numerous deep
the creek valleys of the Tunka Ridge (Laperdin et al. 2014). but short V-shaped shallow stream valleys. Those streams
Two of the seven streams came very close to the village and originate from glacial hanging cirques (Shetnikov 2001).
caused damage to buildings and infrastructure. According to
the chief directorate of the Ministry of Emergency Situations Geological conditions
of Russia for the Republic of Buryatia, the debris flows Geologically, the Tunka Ridge in the area of debris flows is
resulted in one life loss, several injuries, 112 buildings being composed of Archean rocks, lower Proterozoic complex
damaged, and 15 buildings being partially or completely (gneiss, gneiss-granites, schists) with thin marble inter-
destroyed. Moreover, the debris flow demolished the 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 ) .
Kyngarga River bridge, severely damaged local roads, and Bedrock differs in physical and mechanical properties, such
covered market gardens of local residents with stones and as abradability. Most resistant to water-rock flows are
sand. In addition, the debris flows partly affected the envi- granite-gneisses (their abradability is 0.160.18 cm 3 /cm 2 )
ronment: they changed active channels, destroyed large for- and gneisses (0.150.20 cm 3 /cm 2 ); marble are most prone
est areas, covered large areas with sandy debris material, and to abradability (6.688.72 cm 3 /cm 2) . Another important in-
the sediments dammed the mineral springs for a few days. dicator of the bedrock strength properties is weathering
Preliminary summary of the events was delivered hot on the rate; it varies from 0.11 cm/year for granites, up to
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/ According to the national park staff and locals, the maxi-
year for gneisses (Engineering geology of the Baikal region mum snow depth is 30 cm in the foothills and lowlands,
1968). Quaternary deposits are present by fluvioglacial sed- and it reaches 100140 cm in the mountains (in hanging
iments, alluvial terrace sediments, and proluvial-deluvial cirques). Perennial snow cover comes off in the third and
fan sediments (Fig. 3). The sediment material varies from sometimes in the second decade of April. In the mountains,
large boulders (over 2 m in diameter) to sand fraction. the snow cover remains for up to 180 days (data of
National Climatic Data Center). The overall picture of tem-
Climate perature, precipitation, and snow depth reflected in Fig. 4
Climatically, the area is characterized by considerable am- shows data for the Tunka weather station. Although
plitude of seasonal and daily temperatures and predomi- amount of precipitation and snow depth parameter values
nance of precipitation in the summer time. The average within the Arshan village area may differ from the values
annual temperature in Arshan village is 1.4 C. The recorded at the Tunka weather station, meteorological data
terrain features have impact on spatial distribution of pre- from this station has been used in this article in order to
cipitation. In the mountains, its amount is significantly present an overview of the general climatic characteristics
higher than in the lowland area (Ufimtsev and of the study area.
Shchetnikov 2003). According to Arshan meteorological sta-
tion (operated until 1997), which was located in the foothill Cryogenic conditions
area, the mean annual rainfall can reach 750 mm, which on In hanging cirques, the seasonal freeze-thaw layer is 2.0
average is 200 mm over the amount recorded at the Tunka 2.5 m or more (Laperdin and Trzhtsinskii 1976; Leshchikov
meteorological station located 20 km to the south in the 1978). Stability of the overlying weak sediments on the
valley (Meteorological monthly magazine 1972). In summer, slopes largely depends on the presence of the frozen layer
rainfall amount can reach up to 170 mm/day (Arshan me- and the regime of temperature distribution inside the layer
teorological station) (Meteorological monthly magazine in the annual cycle. Soil thawing begins in April and May
1972). The bulk of precipitation (about 70 % of mean and depends on the elevation, slope exposure, and vegeta-
annual rainfall) falls in the summer months from June to tion. In late Juneearly July, seasonally thawed layer does
August. For most of the year, precipitation falls as snow. not exceed 0.7 m on southern slopes and 0.4 m on
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
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
residents in the village. In the morning of 28 June, debris event (Klime and Vilmek 2011). The second, fifth, and
flows began to come down the brook valleys. The flows sixth debris flows had the greatest force and formed the
force can be judged from the fact that huge boulders of largest accumulative forms (alluvial fans). The most sig-
up to 5 m in diameter were moved down (Fig. 6); the nificant damage was caused by the fifth debris flow; it
debris flows damaged cars and buildings and washed out damaged the resort building and destroyed some local
the roads; the flows left erosional forms up to 1-m deep. houses (Fig. 8). The major parameters of the debris flow
The debris flow velocity varied from 4.80 to 13.07 m/s basins are presented in Table 1.
(Table 1). The traces on the trees and buildings suggest
that the highest level of debris flow reached more than Morphology and sedimentology
5 m in the area of the alluvial fans (Fig. 7). By studying Debris flows formation began with the failure of weak sedi-
scars and mud splashes on the trees, there is a way to ments in the hanging cirques at the height of 1640 to
determine the height of the debris flow level after the 1850 m a.s.l. Sediment failures in all debris flows basins,
except for the third, occurred in the frontal parts of the in the direction of the moraine rampart. Extreme water vol-
moraine ramparts. As a result of the extreme rainfall, hanging ume led to failure of weak sediments in the zone of transition
cirques accumulated large volumes of water. The existing from hanging cirque to the slope, where the inclination angle
permafrost prevented precipitation infiltration, thereby con- changes dramatically from 2 in the hanging cirque to 50 in
tributing to the formation of concentrated surface water flow 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)
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
coarse fraction,
alluvial fan of
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
sediments was up to 1 m.
Discussions
The length of
basin include
basin
flow
1965; Wilford et al. 2004). The first and second debris flow
a
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
Fig. 9 Failure of weak sediments on the bottom part of hanging cirque, which yellow linesthe surface of the hanging cirque, red linesthe surface of the
resulted in the debris flows beginning. The yellow dotted line denotes the line slope, blue linesthe horizontal surface. The red arrow marks the place of
of weak sediments failure. The figure shows the inclination angles of the surface: suprapermafrost water discharge. The photo was taken on 1 July 2014
and not clearly contoured. Different quantity of debris flow cirque bottoms (Fig. 9). The bulk of the solid component
material in two morphometrically equal debris flow basins of the debris flows includes deluvial and proluvial slope
is due to the fact that the seventh debris flow basin, as well deposits (Fig. 13). Their composition ranges from large
as the first one, was located at the edge of the rain front boulders (up to 5 m in diameter) to the sand fraction.
and got a smaller water volume. Deluvial and proluvial deposits provide coarse and fine
fraction for the solid component of the debris flow. An
Deposits analysis of the size distribution of the deluvial deposit
Solid part of the debris flows included proluvial-deluvial samples from the margins of debris flow paths demonstrat-
slope deposits, as well as glacial deposits of hanging ed that the presence of <1 mm fraction was 69 % (Fig. 14).
cirques. The input of bedrock in the debris flow is so small It is that particular size of fraction that combined with
that it cannot be considered as a source of the solid debris water forms a debris flow suspension effect and increases
material. Small amount of bedrock enter the flow due to the carrying capacity of a debris flow (Perov 2012; Ni et al.
the debris flow erosive action at the sites of bedrock expo- 2014).
sure in the propagation paths. Among the bedrocks, marble
outcrops are most affected by erosion. The hanging cirques Precipitation
of the Tunka range are filled mostly by glacial deposits, The extreme rainfall was responsible for the debris flows.
which are composed of coarse material with an average Unfortunately, there is no operating weather station at
diameter of 1 m . I n additio n, as a result of rock present within a radius of 20 km from the Arshan village,
weathering, the hanging cirques and valleys accumulate and we cannot accurately estimate the amount of precipi-
large amounts of coluvial deposits. The quantity of glacial tation, which served as a trigger for the debris flows.
deposits of the cirques in debris flows is smaller than the However, we can compare the scale of natural events of 18
volume of deluvial deposits involved in the flow. Despite July 1962 and 27 July 1971 in Arshan. For those events, there
the large thickness of glacial deposits, they were not fully is available data about the rainfall volumes. In July 1971, a
involved in the debris flow for two reasons: firstly, cirque flash-flood occurred on the Kyngarga River. A water flow
glacial deposits are well drained, and secondly, the cirque swept away windmills, bridges, and houses (Makarov 2012).
bottoms are flat (821). Such conditions were unfavorable According to local witnesses of both events (1971 and 2014)
for the movement of the thick glacial deposits. The debris the flooding on the Kyngarga River in 1971 was stronger
flows partially involved the deposits of the front parts of than in 2014. According to the data from Arshan weather
frontal moraines, located on the bend from the cirque to station, 90.4 mm of rainfall was recorded that day
the ridge slope, where the slope is steeper than in the (Meteorological monthly magazine 1972). but the rainfall
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
Fig. 11 Morphological elements in the alluvial funs of debris flows: a lateral levee, b erosional incision
Geomorphological conditions in combination with the to the sixth debris flow basins and over the left upper
presence of weak sediments and meteorological situation tributaries of the Kyngarga River and affected only the
of the area favors hazardous gravitational processes, in- edge of the first and seventh catchment basins. Although
cluding debris flows. The terrain energy is favorable for there is lack of precise data on the rainfall amount that
debris flow occurrence in this region, Melton ratio for the day, based on the past events data (1962 and 1971), we can
debris flow basins ranges from 0.58 to 1.04. Such basins are conclude that the rainfall was minimum of at least
prone to formation of water flow of sufficient power for 90 mm/day. Extremely large amounts of rainfall created
transformation into a debris flow (Wilford et al. 2004). the conditions for watering the thickness of already
Debris flows of 28 June 2014 were triggered by intense thawed weak sediments, suspension of debris sediments
torrential rainfall. Precipitation had a local character, of different genesis, and lithology. The shift occurred
which is quite typical of the area. A 4.5-km-long rain down the frozen, inclined basement as a result of flood
front passed over the catchment area from the second and movement of avalanchine water flow; debris flow
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
Abstract This study presents a kinematic model of a translational with occasional beds of thin sandstone formations. The contact
slide in the Cidu section of the Formosan Freeway in Taiwan, between the two strata was identified as the primary sliding
including the associated kinematic processes and the geometry plane in the Cidu event (Fig. 2). It should be noted that the
of the deposition. A numerical simulation of the slide was per- strike of rock formations in this area is NE and that the dip
formed using the discrete element method. This kinematic model slopes dip toward the SE an angle of 1030. These rock slope
has been considered the distribution of bedding plane and joints, characteristics in conjunction with the fact that Freeway No. 3
engineering installations, and mechanical parameters in the devel- cuts through the dip slope increased the likelihood of a dip
opment of a numerical model capable of clarifying the kinematic slope failure (Wang et al. 2013).
characteristics of this translational slide with high velocities. When Following the incident, government agencies and researchers
the friction coefficient of each particle was set to 0.03, the predict- investigated the mechanisms associated with the dip slope slide
ed maximum velocity was 23.6 m/s and the debris reached the and augmented efforts to monitor the stability of dip slopes near
other side of the Formosan Freeway. The simulations demonstrat- national freeways, and established an early warning system.
ed that the three cars involved in the incident and the Formosan However, there remains considerable room for improvement,
Freeway itself were buried at 2 to 4 s during the event (the predicted particularly with regard to determining the velocity of sliding
average velocity was 10.2 m/s) and that the translational slide ceased masses and their sphere of influence. Extremely high velocity
its movement at 7.5 s, resulting in the formation of a slightly landslides, particularly rockslides that develop into debris ava-
fragmented depositional mass. lanches, are remarkable geological phenomena (Steven and Si-
mon 2006). Velocity is the most important parameter
Keywords Kinematic model . Translational slide . Numerical determining the destructive potential of landslides (Hungr
simulation . Discrete element method 2007), and terrain-related characteristics, such as mass separa-
tion, collisional interaction, and kinematic change from slope
Introduction mass sliding to particle flow, are important for estimating the
Numerous calamities throughout the metropolitan and mountain- sphere of influence in order to protect properties in the down-
ous areas of Taiwan have been caused by translational slides. Dip stream area (Lo et al. 2011, 2014). Consequently, this study
slopes, at which translational slides are more likely to occur, are employed a discrete element program (PFC3D 3.0) to construct
natural slopes (particularly of rock) in which the dip of the strata a numerical model of a translational slide in the Cidu section of
and the primary planes of weakness of the strata have the same the Formosan Freeway. Our goal was to compare the simulated
direction with a difference in angle of less than 15. The sudden deposits with those observed onsite, with the ultimate goal of
onset and velocity of translational slides (reaching 100 km/h or clarifying the kinematic processes occurring in each stage of the
28 m/s) can lead to severe disasters (Hung 2010). landslide. The study also conducted numerical simulations of the
At approximately 14:29 on April 25, 2010, a dip slope slide collapse in order to estimate the speed of the landslide, which
occurred at kilometer 3.1 in the Cidu section of the Formosan enabled us to explain the processes and characteristics of this
Freeway, an area referred to as Shi-gong-ge-shan. A substantial translational slide.
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 Methodology
vehicles passing through the area. The landslide occurred on a
sunny day with no recorded earth tremors, thus excluding the Principle of PFC3D
possibility that rainfall or an earthquake triggered the event Particle Flow Code in 3D (PFC3D) is a program developed by
(Wang et al. 2013). A topographical/geographical investigation Itasca in 1999 based on the discrete element method (Itasca
conducted by Lin et al. (2010) following the disaster revealed 2002). The PFC3D model is used mainly to simulate the move-
that, prior to the event, the slope in the area was distinctly ment and interactions of rigid spheres (ball elements). More
wedge-shaped with the tip protruding toward the northwest. complex behavior can be modeled by allowing the particles to
Near Shi-gong-ge-shan, the dip slope presented a flatiron feature be bonded together at their contact points, such that when the
in four underlying layers, as shown in Fig. 1, where A marks the inter-particle forces acting at any bond exceed the bond
topmost rock section, the gentle terrain of which may have strength, that bond is broken. In addition, the PFC3D model
formed after the dip slope slide. The Cidu landslide occurred also includes wall elements. The balls and walls interact with
in the bottommost rock section, marked by D in Fig. 1. Chen one another via the forces that arise at contacts. In addition,
et al. (2010) identified this section as a sandstone section (SS) at contacts may not form between two walls; thus, contacts are
the bottom of the Miocene Shihti Formation (St), the thickness either ball-to-ball or ball-to-wall (Itasca 2002). The PFC3D mod-
of which is approximately 15 to 20 m. The underlying stratum el is based on the explicit finite difference method of calculating
exposed by the landslide belongs to the Miocene Taliao Forma- changes in the system at each time step, when the positions,
tion (Tl), which consists lithologically of a thick layer of shale 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 Numerical modeling and a comparison of deposit formations
forcedisplacement law. This makes it possible to derive new
velocities and positions of the particles in accordance with Material parameters of the numerical model
Newtons second law of motion. This study used PFC3D to construct a numerical model of the dip
PFC is used to calculate contact stiffness under the assumption slope slide in the Cidu section of the Formosan Freeway. The
that the stiffness of the contact between two adjacent objects is advantage of this model is its ability to simulate the bonding and
treated as normal and shear springs in series. Normal stiffness (kn) separation of objects and also the substantial movements that
and shear stiffness (ks) are defined as follows: 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).
kA kB We used the results of mechanical experiments in conjunction
k 1
kA kB 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
where [A] and [B] represent the two objects in contact, and k could calibrate the conversion formula for the macro- and micro-
denotes normal stiffness (kn) or shear stiffness (ks). parameters and derive the micro-parameters of the simulated
In the slip model, given settings are used for the coefficient of materials.
friction. In the event that the slipping force at the interface be- To facilitate the simulation of energy dissipation in material
tween an element in contact exceeds the frictional resistance, collisions, we referred to the rebound coefficient obtained by Giani
slippage will occur, i.e., slippage occurs when the shear contact et al. (2004) in onsite tests (Table 1). This provided us with
force exceeds the friction and when the contact force is equal to damping parameter settings consistent with those found in the
the friction. The bonding model includes parallel bonds between field. The coefficient of kinetic friction is the key factor in the
particle elements, which combine element groups into other movement processes of dip slope landslides; however, obtaining
shapes. The bond will be broken when the external force applied such measurements can be difficult. Thus, this study compared the
to the elements exceeds the bonding forces. deposit formations that resulted from simulations using a variety
of coefficient settings and selected the setting that produced the plane) into the dip slope. Therefore, this study used ball ele-
result closest to the actual landslide deposit. ments with parallel bonds (setting a 60-t tensile strength and
an insertion into the dip slope of 20 for each ground anchor)
Full-scale numerical model to simulate the ground anchors and retaining wall at the toe of
1. 3D terrain: The study adopted a post-event (2010) 1 m1 m the dip slope (Fig. 4). Additionally, we fixed the bottom layer of
digital elevation model (DEM) for the source area and a pre- the particles so that the upper sliding mass could slide along
event (2009) 5 m5 m DEM for the deposit area. A total of the top of the bottom layer of the particles. Then, the parallel
25,342 wall elements were used for the construction of terrain, bonds will play the role of the ground anchors.
and the total length and width of the numerical model were set 4. Contact stiffness and bonding strength: Specimens obtained
at approximately 842 and 376 m, respectively. by on-site drilling (diameter of 10 cm and height of 25 cm)
2. Sliding mass: The sliding mass included 8552 ball elements were used to perform uniaxial compression tests to determine
with a radius of 1.5 m. In the source area, pre-landslide terrain the uniaxial compressive strength and elastic modulus. Com-
was constructed using wall elements with ball elements to fill parisons between actual and simulated compression tests were
the sliding mass. Once the ball elements were stable, bonding used to revise the conversion formula in order to derive the
strength was applied where the ball elements came into contact micro-parameters of the rock in the numerical simulation
with one another to form consolidated rock masses. Joints (Table 2).
were added to the rock mass, the contact surfaces of which 5. Damping coefficient: The method proposed by Giani (1992) for
possessed friction settings. the conversion of viscous damping parameters was used in an
3. Ground anchors: According to a summary report by the Tai- experiment to obtain preliminary settings for the onsite resti-
wan Geotechnical Society (2011), the total number of ground tution coefficient of rock collisions in the numerical simula-
anchors was 572, and each ground anchor had a strength of tion. The goal was to enhance consistency with the movement
60 t and a setting of 20 (angle with respect to a horizontal observed in the actual landslide.
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 throughout the collapse area: joint A-1, joint A-2, joint B, joint
Taiwan Geotechnical Society (a drilling and experiment report C, and joint D. Joint A-1 exhibited a dip slope bedding plane at
cited the residual shear strength of the wet shale as approxi- an attitude of N45 E/15 E (15 measurement points). The
mately 1417) were used to derive the initial settings of the attitudes of the four other joint sets were N44 E/45 E (three
friction coefficients. In the numerical simulation, the deposit measurement points), N67 E/60 E (two measurement
formations resulting from different friction coefficient settings points), N55 W/55 N (two measurement points), and N26
were compared to enable selection of the most suitable simu- W/50 S (three measurement points). The distribution of the
lation parameters. planes of weakness has a significant influence on dip slope
7. Joint setting: Field survey results from 22 representative mea- sliding, crumbling of the sliding mass, and the distribution of
surement points related to joint attitudes following the land- the deposit. We therefore included this information in the
slide (Fig. 5) revealed five sets of weak planes of weakness numerical model to increase the consistency of the simulation
(including the bedding plane and joints) distributed 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
Fig. 5 Orientation of bedding plane and joints at the Cidu section of the Formosan Freeway
Kinematic processes of the Cidu translational slide displacement of the dip slope extended to 0.51.7 m (some
In this study, the numerical simulation was divided into two main tension cracks were even found on the dip slope) and caused
parts. In the first part, we considered that the ground anchors were about 6 % of the ground anchors to fail (Fig. 9a).
located along the toe of the dip slope, then gradually reduced the 3. Reduction of ground anchor strength by 50 % (30 t): When the
ground anchors strength, and simulated the failure process. In the ground anchor strength was reduced by 50 %, there was about
second part, almost all of the ground anchors were considered to 89 % failure of ground anchors, causing the dip slope to lose
experience failure, causing translational sliding, and the dip slope support and a translational slide to begin (Fig. 9b). Field
movement process was simulated. Figure 9 presents the simulation investigations by the Taiwan Geotechnical Society performed
results of the first part with a friction coefficient of 0.03. The after the translational slide (April 2010) showed that only 58
simulation results of this first part are summarized in the follow- anchors remained in place and that 48 % of the remaining
ing points: anchors showed a fracture of their steel strands (Taiwan Geo-
technical Society 2011). These field investigation results are
1. Original strength of the ground anchors (60 t): The construc- similar to the results of the first part of the simulation. There-
tion of the anchor system was completed at the toe of the dip fore, the study used this model to perform the next step of
slope in 1998 (Taiwan Geotechnical Society 2011). In this kine- simulation (translational slide process).
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- Figure 10 presents the simulation results of the second part
ment or deformation and the overall dip slope presented a (assume that almost all of the ground anchors have been
stable state. With long-time corrosion of the anchor head, the destroyed). The landslide kinematics were divided into the follow-
force of each ground anchor will be reduced gradually, even to ing time periods:
the point of failure. When we reduced the strength of the
ground anchors by 015 %, the dip slope still not incur any 1. 0.00.5 s: Almost instantaneously, the toe portion of the right
displacement or deformation and the overall dip slope pre- wing of the sliding mass (in the direction toward Taipei, near
sented a stable state. the side of the overpass) reached a velocity of 8.2 m/s, such that
2. Reduction of ground anchor strength by 30 % (42 t): When the the mean velocity of the entire sliding mass was 2.9 m/s
ground anchor strength was reduced by 30 %, the (Fig. 10a).
Table 3 Tendency of runout distance and run-up height with increasing value of the different parameters
Input parameters Value Runout distance Run-up height
Friction coefficient 0.010.3 Decreasing Increasing
Joint strength (MPa) 0.01100 Decreasing Increasing
Contact stiffness (N/m) 1e81e11 Increasing Decreasing
2. 0.52.0 s: During this period, the mean velocity of the topmost and it was not constrained by the effect of the ground anchors.
layer of the sliding mass exceeded 12.8 m/s, and the mean A few ground anchors may still have maintained support func-
velocity of the bottom layers remained steady at 9.6 m/s, which tionality at the toe of the dip slope, or the slip surface that
was sufficient to bury the three lanes of the freeway in the caused the velocity of the bottom layers was small relative to
Taipei-bound direction. The sliding mass had not yet crumbled, that of the topmost layer. Additionally, almost all of the ground
even at the time it collided with the vehicles in the first three anchors had lost support functionality at 0.52 s, even about
lanes of the highway, before swiftly pushing on toward the 13 % of the toe of the sliding mass appeared to be in a Bfloating
Keelung-bound lanes. In this stage, the load pressure of the state^ (this caused the contact area between the sliding mass
topmost layer was small relative to that of the bottom layers, and sliding surface to be reduced), and the static friction had
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
changed into dynamic friction of sliding mass movement, Evaluation of the Cidu translational slide modeling
which caused the velocity of the sliding mass to increase sud-
denly (Fig. 10b). The success of landslide modeling depends on the ability of the
3. 2.03.0 s: At this point, the toe of the right wing had reached model to maintain consistency with observed deposits while si-
the slope on the other side of the freeway and had begun to multaneously matching the reports of eyewitnesses and measure-
decelerating. This part of the sliding mass began to crumble ment data. In the current case, we sought to ensure agreement
along the weak plane of joint D, and some portions began to between simulation results and observations with regard to land-
settle onto the freeway. However, the main portion of the slide volume, the position of the collapse area, and the terrain of
sliding mass continued to the lanes on the far side (Fig. 10c). the slip surface. Agreement between other factors was made indi-
4. 3.05.0 s: During this time, the separated portions of the right rectly, e.g., choosing the friction coefficient for each slip surface,
toe stopped moving, leaving them scattered between the two the bond strength, and the contact stiffness to reproduce known
lanes. When the main portion of the sliding mass collided with patterns of deposition. Nevertheless, other consistencies that were
the slope on the far side of the freeway, it continued moving up not determined directly or indirectly were also observed:
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 1. Most of the proportions of materials that slid from the north-
decreased to 2.3 m/s (Fig. 10d). west onto the Formosan Freeway to the southeast were correct.
5. 5.07.5 s: At approximately the 5.0-s point, the main portion of In addition, the thickness distribution in the model closely
the sliding mass was climbing up the slope on the far side of matched the values observed throughout the deposit.
the freeway (Fig. 10e). Because of the slope barrier and influ- 2. Approximately 20 % of the sliding mass climbed the slope on
ence of increased friction, the velocity of the sliding mass the far side of the Formosan Freeway, approximately 35 % of
showed significant attenuation. The entire mass came to a the sliding mass remained on the freeway, and the remainder
complete stop by 7.5 s (Fig. 10f). stagnated on the slide surface. These values match the actual
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. et al. 2011). Therefore, succeeding landslide models also should
The rather more substantial deposits on the slip surface follow this mechanism to reconstruct the Cidu translational
formed a slightly fragmented deposit, providing evidence of slide event. The geomorphologic analysis revealed that the
the processes involved in the Cidu translational slide. source area of the slip surface was covered with rock masses
3. Combining geomorphologic analysis and field observation is and joints, which we represented using ball elements with joint
crucial to the construction of landslide models that accurately settings in the numerical model. Slip surfaces and irregular
represent actual landslides (Lo et al. 2011). Wang et al. (2013) terrain in the study area were represented using wall elements
suggested a mechanism for the Cidu translational slide, which in the construction of the terrain model. The low friction
indicated that the event was caused by hollowed-out dip slope coefficient reflects a reduction in the effective friction coeffi-
toes and infiltration of water into the surface that resulted in cient associated with the sliding surface due to pore fluid.
loss of strength in the clay-like material. This material, which is Conclusion
assumed to be the result of shearing during landslide move- This study used a discrete element method to simulate the kine-
ment and earlier deformation, forms a basically impermeable matic processes involved in the translational slide in the Cidu
layer at the base of the landslide material. As a result, water section of the Formosan Freeway. We included the distribution
pressure builds up over this layer and decreases the effective of planes of weakness, engineering installations (the ground an-
stress. This mechanism may have been the trigger initiating the chors and retaining wall), and mechanical parameters in the de-
Cidu translational slide. Following the initiation of movement, velopment of a numerical model capable of clarifying the
the clay-like material probably played an important role by kinematic characteristics of this translational slide with high ve-
forming a lubricating layer at the base of the landslide material locities. The results indicate that within 24 s from initiation, the
by maintaining high pore pressure during movement (Tsou mean velocity of the sliding mass reached 10.2 m/s, such that the
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
Abstract Typhoon Morakot brought extreme rainfall and initiated Numerous debris flows are initiated during the peak rainfall
numerous landslides and debris flows in southern Taiwan in intensity of a rainfall event in Taiwan (Chen et al. 2005; Huang
August of 2009. The purpose of this study is to identify the 2011). Shallow landslide-induced debris flows on vegetated slopes
extreme rainfall-induced landslide frequency-area distribution in usually occur only after significant antecedent rainfall and mod-
the Laonong River Basin in southern Taiwan and debris flow- erate to long duration rainfall in the USA (Baum and Godt 2010).
initiated conditions under rainfall. Results of the analysis show High intensity and short duration rains without antecedent rain-
that debris flows were initiated under high cumulative rainfall and fall trigger mostly debris flows in Eastern Pyrenees, Spain
long rainfall duration or high rainfall intensity. The relationship of (Corominas and Moya 1999). Separating antecedent rainfall and
mean rainfall intensity and duration threshold could reflect debris a rainfall event are required in order to estimate the critical rainfall
flow initiation characteristics under high rainfall intensity in short factors that initiate debris flows.
rainfall duration conditions. The relationship of cumulative rain- Debris flow warning by threshold rainfall could divide into
fall and duration threshold could reflect debris flow initiation various categorizes. The relationships between (mean) rainfall
characteristics under high cumulative rainfall in long rainfall du- intensity (Im or I) and duration (D) (Caine 1980; Cannon and Ellen
ration. Defining rainfall events by estimating rainfall parameters 1985; Wieczorek 1987; Keefer et al. 1987; Marchi et al. 2002; Chen
with different methodologies could reveal variations among inter- et al. 2005; Crosta and Frattini 2008; Guzzetti et al. 2008; Staley
mittent rainfall events for the benefit of issuing debris flow warn- et al. 2013), and intensity and (effective) cumulative rainfall (Re or
ings. The exponent of landslide frequency-area distribution R) (Chen et al. 2005; Chang and Chao 2006) are common climate
induced by Typhoon Morakot is lower than that induced by the factors used for constructing a predictive relationship for issuing
Chi-Chi earthquake. The lower exponent of landslide frequency- debris flow warnings. Other relationships included but not limited
area distribution can be attributed to the transportation and de- to the use of effective cumulative rainfall and duration (Chen et al.
position areas of debris flow that are included in the landslide 2005), rainfall intensity and antecedent rainfall (Corominas and
area. Climate change induced high rainfall intensity and long Moya 1999), and the daily precipitation and averaged rainfall of
duration of precipitation, for example, Typhoon Morakot brought raining days per year (Wilson 1997). A rainfall triggering
increased frequency of debris flow and created difficulty in issuing index (RTI) defined as the product of the rainfall intensity
warnings from rainfall monitoring. multiplied by the effective accumulated rainfall was used for
real-time debris flow monitoring (Jan and Lee 2004). Radar
Keywords Landslide . Power-law distribution . Climate product QPESUMS was tested to improve the resolution of
change . GIS spatial rainfall distribution for debris flow warning (Chen
et al. 2007a). The rainfall intensity per 10 min is used in
Introduction recent years for real-time monitoring in addition to the use
Climate change-induced abnormal rainfall has caused an increase of the hourly intensity (Floris et al. 2010).
in the frequency and risk of floods, landslides, and debris flows in Landslide and debris flows are growing as a threat to human
the world (Evans and Clague 1994; Joshi and Kumar 2006; Clarke beings due to global climate change. How could a watershed
and Rendell 2007; Petley et al. 2007; Hilker et al. 2009; Djalante initiate a debris flow under abnormal extreme rainfall conditions?
and Thomalla 2012; Mechler et al. 2014). Climate change and the It is an ongoing effort to predict landslide magnitude and frequen-
potential of extreme rainfall conditions may be a contributing cy distribution under extreme rainfall conditions (Chen 2013).
factor for landslides (Nadim et al. 2006). The extreme rainfall Landslide events show a power-law distribution (Stark and Hovius
events in the USA from 1999 to 2003, with daily precipitation total 2001; Guzzetti et al. 2002; Chen et al. 2007b; Van Den Eeckhaut
at one or more stations exceeding the 50-year recurrence amount et al. 2007; Chen 2009; 2012). The linear trend of the landslide
are generally consistent with previous studies of heavy precipita- frequency-area curve for larger landslides (the nearly straight line
tion and flash floods (Schumacher and Johnson 2006). Extreme portion) in power law (log-log plot) can be found through least
rainfall here is defined as rainfall intensity or cumulative rainfall squares regression as
exceeding the 100-year recurrent at a specified station for the
initiation characteristics of debris flow in Taiwan. Extreme log N A c logA S 1
rainfall events have increased flood risks in India
(Guhathakurta et al. 2011) and the frequency of flash floods, where N(A) is the number of landslides of area A, c is the slope of
cloud bursts, and/or glacial lake outburst floods in the Hima- the line defining the relationship, and S is the slope intercept. The
layan region (Joshi and Kumar 2006). Taiwan is expected to straight line with a slope of c has identical slopes in a double-
experience an increase in average annual maximum rainfall logarithmic plot with an exponent c.
from 322 mm in 19602008 to 371 mm in 20102099, due to The distribution curve of the power-law distribution in
climate change (Chiang and Chang 2011). frequency-area distribution exhibits an exponential gradient of
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
Oct.
June
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 Month
intensity-duration relationship and their frequency-area dis-
tribution under extreme rainfall conditions. Fig. 2 Monthly average rainfall distribution at Jiasian rain gauge station in
southern Taiwan from year 1951 to 2010 (source WRA 2010)
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 the Jiasian rain gauge station (Fig. 2). The basin received
area had been documented by numerous studies (Weng et al. 2011; 2300 mm of extreme rainfall during the Typhoon Morakot
Wu et al. 2011). The basins elevations span from 27 m in the plains event in Taiwan.
to 3941 m in the mountain areas and is divided into 76 sub- Debris flow hazards were initiated in the study area when
basins (Fig. 1). Typhoon Kalmegi landed on Taiwan in 2008. Kalmegi caused
The annual average rainfall in the study area was the deaths of seven people and two missing in the study area
3360 mm per year from 1982 to 2010 (WRA 2010). The (NFA 2008). Debris flows initiated by Typhoon Morakot in
maximum rainfall intensity was 106 mm/h, and the cumula- 2009 caused the deaths of 43 people in the study area (source
tive rainfall was up to 2244 mm from August 712, 2009 at SWCB 2008, 2009).
Alishan rain
gauge station
Jiasian rain
gauge station
Fig. 1 Site location of the Laonong River Basin and Typhoon Morakot that brought rainfall distribution in Taiwan on from 710 August 2009
f AL dN c =dAL 2
Fig. 3 Rain gauge stations, landslides, and debris flows induced by Typhoon
Morakot and geologic conditions in the Laonong River Basin (source CGS 2000)
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
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 dN c =dAL bA
L 3
main geologic formation is argillite, slate, and phyllite in the basin
(Fig. 3; source CGS 2000). Landslides were located in argillite, slate, where b is a constant and is the exponent of the fit line, and its
and phyllite in the middle area. Weak geological conditions, steep value is equal to c+1.0 (Malamud et al. 2004).
slopes, high cumulative rainfall, and earthquakes cause numerous The definition of critical rainfall is the total amount of
debris flow and landslide disasters in the basin (Jan et al. 2011). 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
intensity to the time of the triggering of the landslide (Govi The critical rainfall intensity (Ic) is defined as the rainfall
and Sorzana 1980). A rainfall event starts as intensity over intensity on the time of debris flow initiated. The mean rainfall
4 mm/h and ends as intensity lower than 4 mm/h lasting for intensity, Im (mm/h), is defined as
6 h. The effective cumulative rainfall, Re (mm), is defined as
the sum of current daily rainfall (d0) and antecedent rainfall I m Re =D 5
(Jan and Lee 2004; Jan et al. 2006):
where D is rainfall duration via (average of) the nearest rain gauge
Re d0 1 d 1 2 d 2 14 d 14 ; t 0:5t=T 4
station during the initiation of debris flow.
The 61 debris flow cases in middle Taiwan listed by Chen
where t is empirical attenuation coefficient, dt (mm) is daily et al. (2005), 14 Kalmegi-induced cases, and 38 Morakot-
rainfall of t day, and T is half-life (used 1 day herein). induced cases in southern Taiwan documented by SWCB
Fig. 5 Effective cumulative rainfall and duration (Ac-D) for initiation of debris flows
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 Due to climate change, Taiwan is expected to experience increased
than Kalmegi (Fig. 9). rainfall in the flooding seasons and decreased rainfall during dry seasons
The debris flows sourced from Chen et al. (2005) were located in 20202039 (IPCC 2012). Rainfall intensity will increase in summer,
in middle Taiwan with some basins affected by the Chi-Chi earth- and extreme rainfall events will frequently occur (TCCIP 2011). It shows
quake in 1999 that were initiated under the lowest critical rainfall that high rainfall intensity or high cumulative rainfall is expected and
intensity. It also means that earthquake shaking causes the debris prone to initiate landslide and debris flows during the climate change.
flow to be initiated under lower critical rainfall intensity for the Different rainfall parameters reflect different rainfall characteristics. A
earthquake-induced landslides upstream before the initiation of prediction model to issue a warning of debris flow should reflect the
debris flow. The Ic-D threshold line (Ic=38.86D0.343) is identical to threshold characteristics of rainfall, for example, under high cumulative
the results of Chen et al. (2005) for the new added data (Kalmegi rainfall and long duration in an intermittent rainfall (e.g., Morakot) or in
and Morakot) located in southern Taiwan. The result shows a high rainfall intensity and short duration in a rolling rainfall (e.g.,
higher threshold than the available results (Caine 1980; Ceriani Kalmegi). The relationship of Ac-D (Fig. 5) reflects debris flow initiation
et al. 1992; Calcaterra et al. 2000) and comparable to the result of characteristics under high cumulative rainfall and long duration. The
Guadagno (1991). Typhoon Kalmegi showed the high mean rainfall relationship of Ic-D reveals the effects of earthquake-induced low critical
intensity under short rainfall duration (Im-D) to initiate debris rainfall intensity to initiate debris flows (Fig. 9). The relationship of Im-D
flows (Fig. 10). It shows that the Im-D threshold could improve reflects debris flow initiation characteristics under high rainfall intensity
the uncertainty of the threshold for debris flow, for example, when and low duration conditions and eliminates the effects of an earthquake
affected by earthquake, under low critical rainfall intensity. (Fig. 10). Defining rainfall events by estimating rainfall parameters with
Fig. 9 Critical rainfall intensity and duration (Ic-D) to initiate debris flows
different methodologies could reveal variations among intermittent The frequency-area distribution for the Chi-Chi earthquake-
rainfall events for the benefit of issuing debris flow warnings. induced larger landslides can be expressed as
ln f 2:06lnA 5:79r 2 93%ChiChi EQ 8
Characteristics of debris flow frequency-area distribution
The frequency-area distribution of landslides and debris flows can
be expressed as (Fig. 11) The distribution of debris flow (ADF) and landslides (AL) areas
(AT=ADF+AL) has a larger naked area under the same frequency (fT=fL)
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 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
The interpretations of naked areas from remote sensing images exponent in the landslide frequency-area distribution curve
were separated into two categories for debris flow source areas and (Fig. 12).
landslides and debris flows including source, transportation, and The concept of critical states (barely stable) could be extended to
deposition areas. The debris flow source areas are overlapped, but landslides in a basin and its potential to debris flow to present the
the two categories are drawing their distribution separately. Its current state of a basin during rainfall (Chen 2012) and to sediment
purpose is to show the different area interpretation from images budgets. The sediment budget is an estimation of volumes (or rates of
(including debris flow transport and deposition or excluded) change) for sediments entering (source from landslide and soil erosion)
changing the power-law exponent distribution. Although, it is hard and leaving a basin (Bowen and Inman 1966). A basin is in sediment
to differentiate the source and transport area of debris flows. We balance (equilibrium state) as the residual is equal to zero in the
identified some debris flow cases, and the result exhibit an in- sediment budget equation (Kraus and Rosati 1999).
creased exponent distribution for instance. If we can well separate If we divide a basin state under landslide into three categories for the
all the dataset, then the exponent of the distribution will become exponent of landslide frequency-area distribution equal to (critical
higher. The least square best fit equation for debris flow source state), over and smaller than , then landslide in a basin having an
areas and landslides can be expressed as exponent of frequency-area distribution equal to is defined as the
critical state herein. This is similar to the critical exponent for cumulative
ln f 1:39lnA 5:07r 2 89% 7 frequency-area distribution with exponent c=1.0 in the conceptual
Debris flow source areas and land slides in Laonong basin sandpile model (Bak et al. 1987, 1988) and =2.0 in noncumulative form
(c=1.0) (Guzzetti et al. 2002; Malamud et al. 2004). The exponent of in the study area on 4 March 2008 (ML=5.2) and others (Weng et al.
coseismic landslide frequency-area distribution of the Chi-Chi earth- 2011).
quake is 2.06 in noncumulative form and 1.06 for the cumulative form, The exponent value varied with the triggering mechanism and types
which is explained by the critical exponent. The coseismic landslides of natural disasters (Hergarten 2003; Van Den Eeckhaut et al. 2007).
during the Chi-Chi earthquake are primarily shallow landslides (Lee Landslide frequency-area distribution in an unstable state basin shows
et al. 2008). It shows that more small areas of landslides and a lower an exponent close to 1.0 in Chushui creek (Chen et al. 2007b), Dachia
exponent occurred during Typhoon Morakot in the study area than the (Chen 2009), and in the Lushan basins (Chen 2012) in Taiwan. The
Chi-Chi earthquake-induced (ML=7.3) in the middle of Taiwan. Al- exponent in the study basin is affected by the occurrence of the numer-
though, the higher landslide ratio (the sum of landslide areas in a basin ous debris flows. It is sourced from the fact that the naked area
divided by the basin area) could have been affected by the earthquakes 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
Log(Frequency)
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
Landslides (2016) 13:165172 Massimo Melillo I Maria Teresa Brunetti I Silvia Peruccacci I Stefano Luigi Gariano I
DOI 10.1007/s10346-015-0630-1 Fausto Guzzetti
Received: 5 May 2015
Accepted: 10 August 2015
Published online: 26 August 2015 Rainfall thresholds for the possible landslide occurrence
Springer-Verlag Berlin Heidelberg 2015
in Sicily (Southern Italy) based on the automatic
reconstruction of rainfall events
Abstract Review of the literature on the reconstruction of the landslide information, (ii) the methods used to reconstruct the
rainfall responsible for slope failures reveals that criteria for the rainfall conditions responsible for the failures, and (iii) the
identification of rainfall events are lacking or somewhat subjective. methods used to determine the thresholds. A specific source of
To overcome this problem, we developed an algorithm for the uncertainty lays in the characterization of the rainfall event re-
objective and reproducible reconstruction of rainfall events and sponsible for the landslides (Guzzetti et al. 2007). Despite their
of rainfall conditions responsible for landslides. The algorithm importance, criteria for the objective and reproducible
consists of three distinct modules for (i) the reconstruction of reconstruction of rainfall conditions that triggered landslides are
distinct rainfall events, in terms of duration (D, in h) and cumu- still lacking, poorly formalized, or ambiguous. In order to
lated event rainfall (E, in mm), (ii) the identification of multiple overcome this problem, Melillo et al. (2015) proposed an algorithm
ED rainfall conditions responsible for the documented landslides, to define automatically rainfall events from standard rainfall re-
and (iii) the definition of critical rainfall thresholds for possible cords, with an objective and reproducible procedure. The algo-
landslide occurrences. The algorithm uses pre-defined parameters rithm was implemented using the R open-source software for
to account for different seasonal and climatic settings. We ap- advanced statistical computing and graphics (http://www.r-
plied the algorithm in Sicily, southern Italy, using rainfall mea- project.org, release 2.15.2).
surements obtained from a network of 169 rain gauges, and In this work, we use a new release of the algorithm that allows
information on 229 rainfall-induced landslides occurred between calculating reproducible rainfall thresholds from multiple ED rain-
July 2002 and December 2012. The algorithm identified 29,270 fall conditions that have resulted in landslides. We tested the
rainfall events and reconstructed 472 ED rainfall conditions as algorithm in Sicily, southern Italy.
possible triggers of the observed landslides. The algorithm
exploited the multiple rainfall conditions to define objective Rainfall and landslide data
and reproducible empirical rainfall thresholds for the possible Rainfall data consist of hourly rainfall measurements collected
initiation of landslide in Sicily. The calculated thresholds may be in the period between July 2002 and December 2012 by a
implemented in an operational early warning system for shallow network of 169 rain gauges managed by the Sistema
landslide forecasting. Informativo Agrometeorologico Siciliano (Agro-meteorological
Information Service of Sicily) (Fig. 1). For each rain gauge,
Keywords Algorithm . Landslide . Rainfall . Rainfall data are stored in a worksheet, where the header contains
event . Thresholds information on rain gauge name, code, geographical coordi-
nates, and elevation. Each row in the worksheet reports the
Introduction date (day, month, and year), the time (h), and the hourly
Rainfall is the primary trigger of landslides in Italy. In large areas, rainfall (in mm).
the prediction of rainfall-induced landslides relies on the defini- We used the catalogue of rainfall events with landslides
tion of empirical rainfall thresholds. Generally, rainfall thresholds described in Gariano et al. (2015). The catalogue lists 229 rainfall
are the lower bounds to empirical distributions of rainfall condi- events that have triggered 265 shallow landslides in Sicily (red
tions that have resulted in landslides, including rainfall intensity, I, dots in Fig. 2) in the period from July 2002 to December 2012.
vs. rainfall duration, D, (e.g., Caine 1980; Aleotti 2004; Guzzetti In particular, 204 rainfall events induced a single landslide, 18
et al. 2007, 2008; Cannon et al. 2008; Martelloni et al. 2012; Staley rainfall events induced two landslides, 5 rainfall events induced
et al. 2012) and cumulated event rainfall, E, vs. rainfall duration, D, three landslides, and 2 rainfall events induced five landslides, for
(e.g., Innes 1983; Cannon and Ellen 1985; Wieczorek 1987; Crosta a total of 265 landslides. In the subsequent analysis, we use only
1998; Kanji et al. 2003; Vennari et al. 2014). In other cases, rainfall the first triggered landslide of each rainfall event (a total of 229
thresholds are defined in order to separate the rainfall conditions landslides).
that have and have not resulted in landslides (e.g., Jibson 1989; Each record in the catalogue includes (i) event identification
Crozier 1997; Corominas and Moya 1999; Marchi et al. 2002; Berti (code number, reference source), (ii) landslide localization (place,
et al. 2012). Most of the empirical rainfall thresholds available in geographic accuracy), (iii) landslide type (if available from the
the literature are defined using nonobjective and poorly reproduc- source of information), and (iv) temporal information of the
ible methods (Guzzetti et al. 2007, 2008). Moreover, rainfall failure (day, month, year, time, date, temporal accuracy). We
thresholds are affected by uncertainties that limit their use in defined three classes of geographic accuracy and three classes of
landslide warning systems (Aleotti 2004; Guzzetti et al. 2008; temporal accuracy, depending on the type and quality of the
Bach-Kirschbaum et al. 2012; Gariano et al. 2015). Sources of information (Gariano et al. 2015). As an example, the first class
uncertainty include (i) the availability and quality of rainfall and of geographic accuracy was attributed to landslides located with an
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 specify the landslide type, the majority of the failures was classified
includes landslides for which the exact time of occurrence is as generic shallow landslide (SL).
known. Figure 3 shows the number and percentage of each type
of landslides in the catalogue according to Cruden and Varnes Method
(1996). Classification is based primarily on the landslide descrip- We applied the automatic procedure proposed by Melillo
tion given in the information sources. Since many reports failed to 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
rainfall measurements. The algorithm exploits a continuous record New release of the algorithm
of rainfall measurements to determine the duration D (in h) and the In the new release, the algorithm allows to select the representative
cumulated rainfall E (in mm) of distinct rainfall events. To single out rain gauge in the neighborhood of each landslide. As stated above,
rainfall events from rainfall series, the algorithm requires a set of using information on the date/time of landslide occurrence, the
input parameters varying with the season and the climate. In partic- algorithm reconstructs the duration D (in h) and the cumulated
ular, a minimum dry period (i.e., a period without rainfall or with a rainfall E (in mm) for multiple rainfall conditions responsible for
negligible amount of rainfall) between two consecutive rainfall pe- the initiation of the failures. All the possible rainfall conditions
riods is defined. The length of the dry period may vary, depending on associated with landslides are then used to determine empirical
seasonal and climatic conditions. In particular, in Sicily, where cli- rainfall thresholds. In the previous release, only one rainfall con-
mate is Mediterranean (mild and wet winters, hot and dry summers), dition arbitrarily chosen was used to define the thresholds.
we separated the rainfall events selecting a dry period of 96 h for the A bootstrapping statistical technique (Peruccacci et al. 2012)
wet season from November to March, and of 48 h for the dry season and a frequentist method (Brunetti et al. 2010) were included in
from April to October. the new release of the algorithm. This improvement allows the
For each of the 229 landslides, the algorithm reconstructed a definition of thresholds and their associated uncertainties for
variable number of equally (multiplicity) probable rainfall condi- different exceedance probabilities. The general form of the thresh-
tions. Overall, 472 multiple rainfall conditions were associated old curves is a power law:
with the 229 landslides. In the following, we describe how we
improved the algorithm. E D 1
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
where is a scaling constant (the intercept), is the shape synthetic k series allows calculating the mean value of and
parameter (that defines the slope of the power law curve), and parameters and their uncertainty and .
and represent the uncertainties of and , respectively. The use of the algorithm to determine the rainfall conditions
More in detail, the bootstrapping technique generates k series of m responsible for landslides allows to avoid the previous manual
randomly selected events from a data set of n events (which reconstruction and to decrease the uncertainty associated with
correspond to all the possible rainfall conditions associated with the definition of the rainfall events. Moreover, the subjectivity
each landslide calculated by the algorithm). The m events were inherent to the selection of the (D,E) pair is overcome by the use
selected with a probability of selection inversely proportional to of all the multiple rainfall conditions able to trigger the individual
their multiplicity. As an example, events with multiplicity equal to landslide.
2 have probability of selection equal to 0.5. Analysis of the
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.
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
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
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
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
Figure 5 shows the two-dimensional frequency density distri- Using the kriging method (Bivand et al. 2008), we interpo-
bution (Kernel Density Estimation, KDE2d function, Bivand and lated the cumulated rainfall obtained from the rain gauge net-
Gebhardt 2000) of the reconstructed events (a) in the (D,E) loga- work, and we calculated the probability density for each
rithmic plane and (b) in geographic coordinates (longitude, lati- category (Fig. 6). BLow^ events are mostly localized in the
tude). In Fig. 5a, the highest density values are in the area southeastern part of the region, while Bmoderate-heavy^ events
delimited by 40 h<D<150 h and 15 mm<E<50 mm. In Fig. 5b, the are concentrated in the northwestern area of the island. Heavy-
highest frequency of rainfall events is located in the northeastern torrential and torrential events are mainly localized in the
area of the island. Eastern side of the region.
According to the classification proposed by Alpert et al. (2002),
we subdivided the reconstructed rainfall events in six categories of Rainfall thresholds
daily rainfall, from Blight^ to Btorrential^ (Table 2). In order to In order to define rainfall thresholds, we used data collected in the
assign each event to a specific category, we adopted the following period between July 2002 and December 2012, described in Rain-
procedure. For events lasting up to 24 h, we assigned a category on fall and landslide data. Table 3 shows statistics of the 472 rainfall
the basis of the cumulated event rainfall. For events longer than conditions. The rainfall duration D varies from 2 to 605 h, with an
24 h, we calculated the maximum value of cumulated rainfall in average value of 89 h, and cumulated event rainfall E from 5.6 to
24 h using a moving window and assigned the corresponding 659.6 mm, with an average value of 84.1 mm. Half of the ED
category. Almost 40 % of the events are classified as Blight- rainfall conditions have a mean rainfall intensity greater than
moderate,^ and only 3.8 % as Bheavy-torrential^ or Btorrential.^ 2.31 mm/h.
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 parameters with the accumulated deviatoric plastic strains, which
the first impounding of a nearby reservoir is investigated through is a major contribution to progressive failure analysis of slopes.
the analyses of the shear strength reduction behavior of slip zone Troncone (2005) presented the results of a numerical study on
soil. In view of the experimental observations, an empirical strain- a landslide that was triggered by deep excavations at the slope
dependent soil model is established and is then incorporated in toe and further extended this analysis into a three-
finite element analyses. The numerical analysis results show that dimensional problem (Troncone et al. 2014). The elasto-
the failing sections progressively develop due to the soil strength viscoplastic model was adopted to account for the pro-
declines from peak toward residual, and the shear zone propagates nounced strain-softening behavior of the soil in the landslide.
within the front slope. It is demonstrated in the numerical results Other analyses of progressive failure based on elastic-plastic
that the toe weighting measure has a significant effect on strain-softening constitutive models had been reported in lit-
restraining the shear displacements of the soils and preventing erature. For instance, Chai et al. (Chai and Caeter 2009)
the progressive failure of the landslide. The field observations proposed an improved model to simulate the failure of an
further confirmed the stability condition of the reinforced embankment using the Modified Cam Clay model. Through a
landslide. similar approach, Conte et al. (2010) investigated slope failure
cases trigged by weathering. More recently, Samaneh (2013)
Keywords Slip zone . Residual shear proposed a numerical method that can be used to evaluate
strength . Landslide . Strain-softening . Slope stability . Finite the post-failure large deformation of slopes. In addition to
element analysis these numerical studies, the residual strengths of different
types of slip zone soils have been investigated over the past
Introduction few decades (Skempton 1985; Wen et al. 2007; Chen and Liu
The stability analysis of natural and engineering slopes is a com- 2014). The comparison between the residual strengths obtain-
plex problem, especially when the residual shear strength behavior ed in laboratory tests and those measured in the field has
of slip zone soils has to be taken into account. Owing to the been made to gain a better understanding of the reactivation
strength reduction phenomenon of the soil with increasing shear of ancient landslides (Mesri and Nejan 2012).
strain, the progressive slope failure process is characterized by a In this study, the stability analysis of an ancient landslide
nonuniform mobilization of the shear strength along the potential during the first filling of a reservoir is studied considering the
slip surface. Many factors contribute to this process, one of which strength reduction characteristics of the slip zone soil. In view of
is the toe slipping triggered by submerging or trimming. It is well the experimental observations in laboratory, an empirical relation-
known that an ancient landslide, generally characterized by ample ship between shear strength parameters and shear displacement is
dimensions and existing slip zones, can remain stable or meta- established and is then incorporated into a plastic criterion to
stable for centuries on the river bank. However, the landslide will account for the strain-dependent behavior of the slip zone soil.
be probably reactivated by the first impounding of newly built Based on the proposed model, finite element analyses are per-
reservoirs in the vicinity (Wang et al. 2008). Extensive studies have formed to study the progressive failure of this landslide. The
been performed to investigate the main instability mechanism of numerical modeling results indicate that toe weighting is an effec-
bank slopes, most of which considered the hydromechanical in- tive measure to prevent the reactivation of the landslide.
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). Landslide features
However, the literature gives little information on how shear zones The typical geological condition of the ancient landslide is shown
propagate when the slope toe is gradually submerged in water. On in Fig. 1a. This landslide is situated on the Wujiang River, a
the other hand, during the first filling of a reservoir, the shear tributary of the Beijiang River, 300 km northern of Guangzhou
strength of the soil mass involved significantly reduces from peak City, China (Fig. 1b). It has a steep front slope, and the total
toward residual due to the strain-softening effect. This may result volume of the landslide is about 2.4 million m3. The aerial view
in nonuniform interruptions of the equilibrium state of bank of the site before construction of the reservoir is shown in Fig. 1c.
slopes, which has not been well addressed. Therefore, an in- The subsoil is essentially composed of sericite quartz sandstones
depth understanding of the progressive failure process induced and slates. Drilling and other geological evidence revealed a
by the submergence of the slide toe is essential for enhancing the preexisting slip zone within the landslide, which is composed by
capability of landslide mitigation and prevention. clay with sand and silty clay. The thickness of the slip zone is
In the past few decades, numerous quantitative studies have between 30 and 150 cm (see Fig. 1a).
been carried out for estimating the progressive failure and post- According to engineering experience, during the first
failure deformation of slopes in soils with strain-softening behav- impounding of the reservoir, the steep front slope will be fully
ior. Potts et al. (1990, 1997) proposed an approach to simulate the submerged and become locally unstable. Will the toe instability
strain-softening properties of brittle soils by reducing the strength 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 density =2.05 g/cm3. According to the Chinese National Standard
by toe slipping must be carefully evaluated in detail before a safe (GB 50021-2001 2001), the soil is classified as clay containing sand
design of the reservoir can be made. with low liquid limit (CSLL). Based on the results of laboratory
shear tests (Chen and Liu 2014), the relationship between the
Strain-dependent soil model Mohr-Coulomb (MC) shear strength parameters and the shear
A series of laboratory tests were conducted to study the physical displacement was obtained (Fig. 2). It is seen that both the
and mechanical behavior of the slip zone soil. The undisturbed soil cohesion c and friction angle were dependent on shear
specimens were collected from trial pits where the slip zone forms displacement. In particular, the post-peak cohesion of the soil
an outcropping (see point A in Fig. 1a). The grain size distribution sharply decreased to zero when the shear displacement
of the soil is 30.7 % clay (<5 m), 40.6 % silt (510 m), 15.0 % fine accumulated.
sand (75250 m), and 13.7 % mid-sized sand with gravel The strength reduction characteristic of the slip zone soil is of
(>250 m). The soil has an initial water content w=16.0 % and a great importance for slope stability analysis. Assuming that the
20 20
(o)
(Conte et al. 2010). In plane strain condition, it is es p16
15 15 q
10 c 10
x y 2 2x 2y 2xy , where x and y are normal
strains in the x and y directions, respectively, and xy is shear
5 5
strain. In this study, the strain thresholds esp and esr are
0 0 experimentally determined using laboratory test results.
d (mm)
Numerical analysis method of the landslide
Fig. 2 Relationship between shear strength parameters and shear displacement
Analysis model
The numerical analyses of the landslide were carried out
using the finite element software ANSYS. The typical cross-
shear displacement occurs at the slope toe due to environmental
section of the landslide shown in Fig. 1a was selected for the
factors, such as water level fluctuation or toe excavation, the shear
finite element analyses. Figure 4 shows the finite element
strength of the soil involved will gradually reduce with accumu-
mesh and boundary conditions used in the analyses. The
lated shear displacement until it reaches a residual state. This
whole landslide consists of the slope mass, the slip zone,
behavior leads to a progressive failure process, which starts with
and the bed rock, which are denoted as upper soil, slip zone
a section of the soil of large shear displacements, and then prop-
soil, and lower soil, respectively. Because the behavior of the
agates within the whole shear zone due to stress redistribution
slip zone soil depends on both the strain-softening effect and
effect. In finite element analyses, if the strength reduction behavior
the size of the elements in which the collapse occurs, a
of soil can be truly modeled, the progressive failure of landslide
technique of adaptive mesh refinement was adopted for the
can be numerically simulated with high reliability.
red zone in Fig. 4, which means that the mesh will be locally
Following Potts et al. (1990), the strength reduction behav-
refined to improve the calculation precision. The main pa-
ior of soil can be described by the strain-softening function in
rameters used in the numerical analyses are listed in Table 1.
terms of both effective and total stresses, as shown in Fig. 3.
Using the elasto-plastic failure criterion coupled with the em-
In this figure, p and cp are MC peak strength parameters,
pirical strength-strain relationship given in Fig. 3, a strain-
and r and cr are residual strength parameters, respectively.
dependent elasto-plastic soil model is derived. This model can be
When the strain level esesp, the cohesion c and friction
regarded as an extension of the elastic-perfectly plastic model. In
angle maintain the peak value. The strength reduction
this model, the soil is assumed to obey the elasto-plastic failure
phenomenon begins when es>esp, following a linear decreas-
criterion with an added strain-softening feature. The approach in
ing function in terms of c and . When the strain level
combination with the MC criterion has been successfully used to
reaches esr, the strength parameters become constant again.
characterize the behavior of various strain-softening clays in the
The application of the piecewise tri-linear function to nu-
past few years (Potts et al. 1990; Troncone 2005; Kovacevic et al.
merical analysis faces two challenges. One problem is how to
2013). In the ANSYS software, the MC failure surface is normally
define the shear strain, which is not homogeneously distrib-
substituted by the Drucker-Prager (DP) failure cone, which is
uted in the soil sample during laboratory testing. The other
circumscribed to the MC hexagon pyramid. From the point of
problem lies in determining the strain thresholds esp and esr,
view of numerical efficiency, this will bring much convenience to
which should be done based on abundant experimental data.
numerical calculation.
The Drucker-Prager yield criterion F can be written as follows:
p
F I 1 J2 k 1
(%)
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
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)
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)
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
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 stabilization scheme. The landslide is designed to be rein-
surface develops progressively, wherein the average soil strength is forced by backfilling rock masses at the slope toe, which is
less than the peak strength. It is evident that a potential slip can called Btoe weighting.^ Figure 8b shows the simulated lateral
take place if no reinforcement measure is adopted. That is, the displacements of Zel=154.6 m, in which the Bcumulative^ in-
reservoir impounding causes the slope toe to slide and the slope dicates the accumulated displacements at the end of the
body to deform, which may lead to progressive failure because of analysis, and the Bincremental^ means the incremental dis-
the strain dependent behavior of the slip zone soil. placements as the reservoir water level rises from 135.1 to
154.6 m. The plastic strains at the end of the analysis are
Reinforcement measure shown in Fig. 8c.
In order to improve the stability of the bank slope, reinforc- The numerical results indicate that, after constructing a
ing the slope toe should be considered during the design berm at the toe, the slope deformation is significantly de-
phase of the reservoir. Figure 8a illustrates the slope creased. The deformation pattern is different from that of
a) b)
c) d)
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)
the non-toe reinforcement case. The upper landslide body and softening cases. According to the above analyses, the
tends to move downward due to the softening of the slip stability of the landslide can be guaranteed once the sub-
zone soil. However, the displacements of the front slope are merged part of the slope is stabilized after reservoir
effectively restrained. As a consequence, the landslide has a impoundment.
relatively uniform distribution of deformation, and the maxi- A few additional explanations can further verify the nu-
mum accumulated displacement is only 64 mm, which occurs merical modeling results. Figure 9 shows the slope toe after
at the middle height of the slope (Fig. 8b). The equivalent toe weighting. After the reservoir construction work, some
plastic strain occurs in the upstream along the slip zone field measurements were taken in the landslide.
(Fig. 8c), and the maximum strain is only 2.9 %. As the Unfortunately, the detailed deformation data during the first
plastic strain is well below the threshold esp, there is no reservoir impoundment have not been recorded. When the
difference in the simulation results between the nonsoftening reservoir water level rose from 135.0 to 150.0 m (the current
a)
b)
Cumulative Incremental
c)
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
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
Abstract Landslide impacts on infrastructure and society in the 2012). Like elsewhere in the world (e.g., Walkinshaw 1992; Hearn
Federal Republic of Germany are associated with damage costs of et al. 2008; Winter et al. 2014), annual landslide losses at country
about US$300 million on annual average. Despite the large overall level peak in the transportation sector (roads, railways, etc.), as
losses due to widespread landslide activity, there is a lack of indicated by damage costs of US$7080 million for Germanys
historical impact assessments, not just for Germanys low moun- highway system (~39,400 km) alone (Klose 2015; see also Krauter
tain areas but those of entire Central Europe as well. This paper is et al. 2012). More than 6 % of the countrys territory shows
a collection of three case studies from Germany that seek a better significant levels of landslide hazard (based on Dikau and Glade
understanding of landslide impacts and their economic relevance 2003), wherefore costs per capita reach a total of US$3.7 per year
at local and regional level. The first case study investigates damage (Table 1).
types and mitigation measures at a representative landslide site in Landslide impacts are mainly addressed by two types of studies,
ways that support to gain insight into historical hazard interac- which are (i) impact assessments, either as cost surveys (ex post)
tions with land use practices. This case history is followed by a case or risk analyses (ex ante) (cf. Corominas et al. 2014; Meyer et al.
study dealing with fiscal cost impacts of landslide damages for an 2013), and (ii) vulnerability assessments, i.e., the analysis of the
example city and the highway system of the Lower Saxon Uplands, exposure and sensitivity to landslide damage, most often from a
NW Germany. In addition to a cost-burden analysis for affected technical and retrospective (ex post) point of view (cf. Alexander
public budgets, an overview of the principles of disaster financing 2005; Papathoma-Khle et al. 2011). The two approaches of impact
in landslide practice is given. The third case study is focused on assessment are always applied separately, with landslide databases
the conflicts of urban development in hazard areas, with an eco- and Geographic Information Systems (GIS), however, playing a
nomic approach to balancing safety and public welfare interests. major role in both of them (cf. Chacn et al. 2006; Van Den
Each case study is based on historical data sets extracted from Eeckhaut and Hervs 2012; Van Westen 2013). As systematically
Germanys national landslide database. This paper presents three compiled data archives of past and current landslide events
different case studies that in combination are a first step towards (Hervs and Bobrowsky 2009; Hervs 2013), todays landslide
assessing landslide impacts in integrated perspective. databases provide a foundation for studying hazard impacts, es-
pecially by means of GIS and geospatial analysis (cf. Dikau et al.
Keywords Landslide impacts . Germany . Case 1996; Harp et al. 2011; Guzzetti et al. 2012). The key to understand-
studies . Mitigation . Damage costs 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
Introduction historical studies occupy center stage in impact assessment,
The nature of landslide impact in western and central European gaining strongly in importance through progress in web-based
countries such as the Federal Republic of Germany (abbr. data mining and archiving (e.g., Battistini et al. 2013; Baum et al.
Germany) is still poorly understood and the severity heavily 2014; Taylor et al. 2015). Nonetheless, there has been a shift from
underestimated (cf. Spizzichino et al. 2010, 2013; Gibson et al. cost survey to risk analysis in recent years, with a focus on
2013; Klose et al. 2014a). In contrast to Bglobal landslide hotspots^ correlating costs with the physical vulnerability of elements at risk
(Himalayan Arc, Central America, etc.), where regional clusters of (see above). Empirical economics yet suggests significant uncer-
triggered landslides often result in catastrophic disasters (e.g., tainty of such values at risk (cf. Van Westen et al. 2006; Galli and
Dilley et al. 2005; Nadim et al. 2006; Petley 2012), landslides in Guzzetti 2007), wherefore some few studies turn the attention on
Germany and adjacent low mountain areas are local phenomena historical losses again, trying to integrate cost survey into risk
with limited areal extension (cf. Bednarczyk 2009; Damm and analysis (e.g., Crovelli and Coe 2009; Wills et al. 2014; Klose et al.
Klose 2015). The associated damages vary spatially in frequency, 2015a).
timing, and intensity and thus are difficult to summarize and Today, global landslide impact is estimated at US$20 billion in
quantify, either in Germany or elsewhere (e.g., Blchl and Braun overall losses annually (Table 1), given the available national cost
2005; Vranken et al. 2013; Bowman 2015; Klose et al. 2015a). Even estimates and the assumption that losses are widely comparable
though costly, high-magnitude landslides are reported on a regular between industrialized countries (Schuster 1996). These reference
basis (Figs. 1 and 2), the nations landslide problem is rather driven costs amount to 17 % of the 19802013 annual average global
by a large number of small, less destructive damage events and natural disaster losses (US$121 billion, Munich Re 2014) and with
widespread insidious losses, which rarely receive broader attention that provide a first idea of the economic relevance of landslide
(e.g., Terhorst and Kreja 2009; Damm and Klose 2014, 2015). From impact worldwide. The true damage costs in terms of the financial
a national perspective, these separate, widely dispersed landslide burdens on society and public budgets, however, are poorly
impacts, however, add up to annual direct costs of roughly US$300 reflected in these benchmarks, either globally or nationally (e.g.,
million (Klose 2015), exceeding the losses of 100-year events of Fleming and Taylor 1980; Olshansky and Rogers 1987; Olshansky
other hazards, especially winter storms and river floods, within a 1996; Kjekstad and Highland 2009). For a better understanding of
few years or decades (e.g., Klawa and Ulbrich 2003; Kron et al. 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 (cf. Anderson and Holcombe 2013) but large parts of the developed
regional data sets are required, including ones that (i) distinguish world as well (cf. Fuchs 2009; Klose et al. 2014a). Although
public and private losses (e.g., Taylor and Brabb 1972; Shearer et al. influencing the severity of landslide impact, some of the most
1983; Creasey 1988; Godt 1999), (ii) inform on disaster financing, important societal risk factors are still too narrowly considered
insurance, or budgetary impacts (e.g., Risley 1993; Olshansky 1996; in recent studies, including (i) human activity is often a root cause
Paus 2005), and (iii) provide insights into cost factors and rela- of landslides, not only by predisposing or triggering them (e.g.,
tionships to landslide types, magnitudes, and mitigation (e.g., Huang and Chan 2004; Nadim et al. 2011) but also as a result of
Smith 1958; Quinn et al. 2011; Bowman 2015). In terms of archive inadequate (low-cost) mitigation, (ii) the level of tolerable or
studies, case histories, or cost surveys, these approaches support acceptable risk, a measure driving a large part of landslide costs
analyzing the complexity of historical damage costs and thus in industrialized countries, is highly variable, differing between
create added value for impact assessment in integrated perspective individuals, public or private organizations, and societies, with its
(e.g., Wasowski 1998; Cornforth 2005; Van Den Eeckhaut et al. nature being to change over time (cf. Harmsworth and Raynor
2010). 2005; Winter and Bromhead 2012; Scolobig et al. 2014), and (iii)
The types and impacts of urban landslide damages are complex decision makers are faced with the difficult task of finding the
(cf. Alexander 1989; Glade and Crozier 2005; Schuster and right balance in landslide mitigation, since being required to
Highland 2007; Petley 2009), with hazard and risk being closely evaluate complex information (e.g., Howell et al. 1999; Michaels
interlinked with the technical and financial coping capacity 2005), while considering best practices and economic rationales
(resilience) at community level, not just in developing countries (cf. Crozier 2005). The complex societal dimensions of landslide
risk along with basic geophysical factors make it necessary to turn discussion in Section 5 puts the key findings of these historical,
towards integrated impact assessments (e.g., Casale and Margottini economic, and societal analyses into a broader perspective. The
1999). First attempts have recently been undertaken to combine cost final section concludes with outlining the importance of integrated
survey with risk analysis (e.g., Crovelli and Coe 2009; Wills et al. 2014; approaches to impact assessment in the future. Note that all
Klose et al. 2015a); a systematic integration of approaches, however, landslide losses in this study are given in US dollar and 2013
is still at the very beginning, wherefore individual case studies 2014 values, if not specified otherwise. Losses published in previ-
continue to play a major role (cf. Klose 2015). ous studies are adjusted for inflation by using the US consumer
Our knowledge about landslide impacts is still based on very price index (CPI-U 08-2013 or CPI-U 05-2014).
limited loss data. The aim of the study was therefore to review and
analyze landslide impacts in Germany from a historical and socio- Study area
economic perspective. This paper is organized as follows. The The case studies presented in this investigation refer to (i) the
second section gives a brief overview of the case study areas of highway system (i.e., BBundesstraen^ equal to US routes) in the
this research work. In the third section, the methods used and the Lower Saxon Uplands, NW Germany, operated by the Lower
data material are presented in detail. The focus of the fourth Saxony Department of Transportation (NLStBV), and (ii) the city
section is on three different case studies of landslide impact. A of Hann. Mnden (~20,000 inhabitants), an example city located
in the south of the Lower Saxon Uplands (Figs. 3 and 4). This 2005; Klose et al. 2012a). The most frequent types of landslides
region in the northwestern part of the Central European Uplands damaging highways are shallow or deep-seated rotational slides in
constitutes a low mountain area with moderate topographic relief soil material and embankment fills (e.g., Gidde 2012; Klose et al.
and elevations ranging from 50 to 950 m a.s.l. The entire Lower 2015a). Furthermore, small to medium large rockfalls and planar
Saxon Uplands cover an area of 7,400 km2 and are moderately rockslides are widespread in this region, especially at large cut
populated (~170 persons per km2), except for some few regional slopes or below steep rock outcrops (e.g., Gnther et al. 2004;
population centers (Gttingen, Hildesheim, etc.) with ~100,000 Schnborn and Heucke 2012). Landslide activity within the city
inhabitants (Destatis 2015). A large part of this region is charac- limits of Hann. Mnden is also characterized by slow-moving
terized by hilly to mountainous relief with deeply incised river or landslides as well as dormant and periodically reactivated slide
mountain valleys and steep cuesta scarps (e.g., Reicherter et al. masses of late modern age (Fig. 4). These urban landslides extend
2008; Figs. 3 and 4). in part over more than 10,000 m2 and often affect highly developed
In the study area, landslides are spatially clustered according to hillsides (cf. Damm 2000; Damm and Pflum 2004). Landslide
relief intensity, land use patterns, and the areas geologic structure, volumes, sizes, and velocities vary greatly in the Lower Saxon
especially the dominant soil or bedrock types and properties (cf. Uplands (Klose 2015), but since large slope failures rarely show
Varga et al. 2006; Damm et al. 2009, 2010; Klose et al. 2014b). high velocities (e.g., THW 1957; Meyer 2005), landslide magnitudes
Throughout the Lower Saxon Uplands, there are 77 km of high- range from low to moderate intensity in most cases (e.g., Schunke
ways (6 % of total network length) that show significant hazard 1971; Tilch 1999; Klose et al. 2014b).
exposure, with clusters of areas at risk along highway sections
located below steep and deeply weathered valley slopes and on Methods and data material
hillsides covered by moisture-sensitive periglacial sediments
(Klose et al. 2014b, 2015a). The city of Hann. Mnden is built to Landslide database
a large extent on mountainous terrain and large slopes above a 300 The present study makes use of historical data sets extracted from
m deep valley basin. This urban area is prone to landslides due to a Germanys national landslide database (Damm and Klose 2014,
combination of predisposing factors such as local topography and 2015). This data pool of ~5,000 landslide events covers the entire
relief, cohesive soils, and disintegrated bedrock, as well as histor- country and a time period of about the past 150 to 200 years, with
ical population pressures and slope disturbance (cf. Damm 2000, the oldest landslide, however, being recorded as early as 1137. A
2005; Varga et al. 2006). Thus, roughly 10 % of the citys total area top-down approach to data collection enables to condense data-
are located in potential hazard zones (Fig. 3b), with major impli- base coverage in case study areas such as the Lower Saxon
cations to all types of urban infrastructures (roads, sewer systems, Uplands. The collection of data at regional level is focused on
etc.). As often having reached their expected lifespans, these age- adding detail to the data pool by targeted archive studies, whereby
ing infrastructures are particularly vulnerable to landslides, where- most sources are now available in digital format, often compiled in
by tight maintenance budgets hamper to mitigate deterioration historical catalogs, online records, and project management data-
and improve asset life (cf. Damm and Pflum 2004; Maurischat and bases (cf. Damm and Klose 2014, 2015; Klose 2015). To access this
Klose 2014). information, database development takes advantage of a continu-
A wide range of landslide types and processes occurs on an ously growing data network, especially with partners like the
annual basis in the Lower Saxon Uplands (e.g., Ackermann 1953; NLStBV or the municipal office of Hann. Mnden. As regards
Mortensen 1960; Schunke 1971; Meyer 2005; Bense et al. 2011; Klose knowledge exchange with the NLStBV, a systematic transfer of
et al. 2014b), most often triggered by heavy rainfall and/or con- data from the Project Information and Management System
struction during prolonged wet periods (e.g., Tilch 1999; Damm (PRIMAS) has been established, providing access to data sets on
current damage events and all landslide-related business processes However, as compared with database development, expert inter-
(cf. Klose et al. 2015a; Jaeger and Klose 2015). Alternatively, the views for the purpose of historical or socioeconomic case studies,
science-practice cooperation with Hann. Mndens municipal of- especially at local and regional level, are organized in a slightly
fice is based on many years of archive works and expert interviews, different manner. Instead of widely standardized damage or cost
which enabled compiling a widely complete landslide record since surveys (e.g., Walkinshaw 1992; Blchl and Braun 2005; Highland
the mid-nineteenth century (cf. Damm 2000; Damm and Pflum 2006), semi-structured interviews that leave room for an open and
2004; Maurischat and Klose 2014). lively discussion occupy center stage, including options for multi-
ple interview waves after rethinking of the survey results (see also
Expert interviews Vranken et al. 2013). This requires strategic partnerships with
The involvement of practitioners by conducting expert interviews experts from landslide practice (e.g., National Research Council
to collect first-hand loss information is a critical step for both 2004), particularly with practitioners who provide practical in-
impact assessment and the development of landslide databases sights into how landslide damage is managed and financed, either
(e.g., Guzzetti et al. 2012; Van Den Eeckhaut and Hervs 2012). from geotechnical or economic and legal perspective (e.g.,
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 for hazard mapping, this principle also holds true for impact and
interviews that underpin the case studies presented here were cost assessments. Case histories of landslide impact illustrate the
conducted with experts in planning, controlling, and finance from scientific importance of historical data sets and their application in
the municipal office of Hann. Mnden as well as the NLStBV hazard studies (e.g., Glade 2001; Calcaterra et al. 2003; Raka et al.
headquarters based in Hannover and the relevant regional high- 2014, 2015). The main purpose of such historical studies is to track
way offices. This investigation mainly refers to the expertise of and analyze the interactions between land use practices, landslide
interviewees affiliated with the NLStBV Regional Office activity, and hazard mitigation at specific locations of infrastruc-
Gandersheim, which is the highway office that is responsible for ture over time (cf. Klose 2015). More specifically, case histories are
maintaining large parts of the highway system in the Lower Saxon focused on analyzing the temporal exposure and vulnerability of
Uplands. urban infrastructures, the changing risk perception of decision
makers, and the resilience to damages at local level, wherefore
Case histories of landslide impact historical landslide databases play a key role in their development.
Historical information on past and current landslide damages is Best practices in compiling landslide databases useful for histori-
critical for achieving a better understanding of landslide impacts cal impact assessments include, among others, traditional archive
in the future. As originally proposed by Varnes and IAEG (1984) works, the search of web and press archives, and data mining of
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)
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)
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)
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)
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 perspective. According to this rule of thumb applied in landslide
but also over periods of decades (cf. Damm 2000; Damm and practice at local level, urban planners do not necessarily feel
Pflum 2004). obliged to fully protect all types of capital assets at any given costs.
The consequence is a (non-official!) rule of thumb indicating However, if there is a serious risk of loss of life, landslide hazards
that isolated private property losses occurring in periods of 30 are usually perceived as publicly unacceptable, and planners are
50 years are publicly tolerable from an overall economic forced to take direct actions. This inevitably implies a strict
Landslides (2016) 13:201210 Maricar L. Rabonza I Raquel P. Felix I Alfredo Mahar Francisco A. Lagmay I Rodrigo Narod C. Eco I
DOI 10.1007/s10346-015-0626-x Iris Jill G. Ortiz I Dakila T. Aquino
Received: 20 October 2014
Accepted: 4 August 2015
Published online: 1 September 2015 Shallow landslide susceptibility mapping using high-
Springer-Verlag Berlin Heidelberg 2015
resolution topography for areas devastated by super
typhoon Haiyan
Abstract Super typhoon Haiyan, considered as one of the most permeability, and land cover (DAmato Avanzi et al. 2004; Hong
powerful storms recorded in 2013, devastated the central Philip- et al. 2007; Liao et al 2010; Wieczorek and Sarmiento 1988). Usu-
pines region on 8 November 2013 with damage amounting to more ally, the triggering mechanism for shallow landslides on
than USD 2 billion. Hardest hit is the province of Leyte which is destabilized slopes usually begins with excessive rainfall. It is
located in central Philippines. Rehabilitation of the areas that were characterized by infiltration into the soil leading to the buildup
devastated requires detailed hazard maps as a basis for well- of pore pressure, reduced soil suction value, reduced shear
planned reconstruction. Along with severe wind, storm surge, strength, and increased shear stress, weakening slope stability
and flood hazard maps, detailed landslide susceptibility maps for (Antronico and Gull 2000; Campus et al. 2001; DAmato Avanzi
the cities and municipalities of Leyte (7246.7 km2) province are et al. 2003; Giannecchini 2006; Van Asch et al. 1999). This could
necessary. In order to rapidly assess and delineate areas suscepti- cause surface sliding due to the textural-granulometric disconti-
ble to rainfall-induced shallow landslides, Stability INdex MAP- nuity within the soil-bedrock interface, causing drastic changes in
ping (SINMAP) software was used over a 5-m Interferometric the infiltration rate (Wieczorek 1987). Slope instability brought
Synthetic Aperture Radar (InSAR)-derived digital terrain model about by the complex interaction of interrelated factors becomes
(DTM) grid. Topographic, soil strength, and hydrologic parame- a major concern when they lie in close proximity to people and
ters were used for each pixel of a given DTM grid to compute for property (Soeters and Van Westen 1996).
the corresponding factor of safety. The landslide maps generated Conventionally, landslide susceptibility assessments are con-
using SINMAP are highly consistent with the landslide inventory ducted through field surveys to account for multiple relevant
derived from high-resolution satellite imagery from 2002 to 2014 controlling factors (Hong et al. 2007). However, the zonation of
with a detection percentage of 97.5 % and missing factor of 0.025. shallow landslide potential for provinces with large areas may be
These demonstrate that SINMAP performs well despite the lack of difficult in the Philippines, especially given the difficulty of
an extensive geotechnical and hydrological database in the study accessing mountainous areas and the cost of ground monitoring
area. The detailed landslide susceptibility classification is useful to networks. Thus, remote sensing may be the only practical means
identify safe and unsafe areas for reconstruction and rehabilitation to obtain data for landslide susceptibility mapping (Catani et al.
efforts. These maps complement the debris flow and structurally 2005; Hong et al. 2007; Nadim et al. 2006). High-resolution satel-
controlled landslide hazard maps that are also being prepared for lite remote sensing products can provide detailed provincial-wide
rebuilding Haiyans devastated areas. shallow landslide susceptibility maps. Studies conducted by Fabbri
et al. (2003) show that the layers extracted solely from a digital
Keywords Landslide . Natural hazard . SINMAP . Susceptibility elevation model (DEM) generated more accurate results than
map . Spatial analyses . Philippines layers derived from geological data, surficial materials, and land
use, indicating that topography is the primary determinant of
Introduction landslide location (Coe et al. 2004; Hong et al. 2007).
On 8 November 2013, typhoon Haiyan (Typhoon BYolanda^ in the It is also important to develop accurate hazard maps
Philippines) reached its peak intensity and swept through the pinpointing the locations of areas vulnerable to landslides for
central Philippines with 1-min sustained winds of 315 km/h at rehabilitation work. The identification of potential landslides and
1800 UTC (JTWC 2013). Powerful winds, heavy rains, and storm the delineation of unstable and suitable areas for development can
surges triggered widespread flooding and landslides, particularly help ensure public safety through improved land use and disaster
affecting the provinces of Leyte (Fig. 1). Typhoon Haiyan left 6300 management planning, management of residential and recreation-
fatalities, 28,689 injured, and 1061 missing, as well as damage al areas, and infrastructural development (AGS 2007; Sarkar and
amounting to over USD 2 billion (NDRRMC 2014). Kanugo 2004). In the case of areas heavily devastated by Haiyan,
Although strong winds and storm surges of up to 56 m reliable, detailed, and readily accessible landslide susceptibility
(Lagmay et al. 2015) were the primary causes of damage in Leyte, maps are critical for well-planned rehabilitation and
landslides were also a significant factor. Shallow landslides are reconstruction.
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 Current approach to landslide susceptibility mapping in the Philippines
complex terrain exposed to frequent heavy rainfall (Liao et al. Prior to this work, government agencies in the Philippines imple-
2010). Relevant shallow landslide-controlling factors include local ment a qualitative method for landslide susceptibility mapping.
conditions such as slope materials and slope surface morphology, This involves the use of landslide inventories, geologic maps, and
soil properties, topography, groundwater, lithology, bedrock geomorphologic information to identify possible sites of failure.
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 wherein rapid assessment is highly critical to conduct well-
opinion of specialists on the subject. The experts use a checklist planned reconstruction and rehabilitation.
to give subjective weights for the factors (e.g., slope angle, slope A sample of maps made by Mines and Geosciences Bureau -
weight, material type, nearness to fault, degree of gully erosion, Department of Environment and Natural Resources (MGB-DENR)
vegetative cover, presence of road cuts on hillsides, etc.) and then Philippines is shown in Fig. 2. These belong to the current national
derive a cumulative influence factor representing the degree of maps for landslide susceptibility. In the context of disaster manage-
hazard (Table 1). Other than the slope gradient, there are no other ment, hazard maps of 1:100,000 to 1:50,000 are not suitable for localized
quantitative bases for the hazard level assigned to an area (MGB emergency response. Current enhancement efforts by the MGB include
2013). Thus, the designating weights are highly subjective. producing hazard maps at a finer scale of 1:10,000 (Fig. 2b).
To date, quantitative and deterministic methods for landslide The same scenario occurs in all the available susceptibility
susceptibility assessment has not been used on a nationwide scale maps in the country. The major goal is to improve these existing
in the Philippines. The recent availability of high-resolution topo- government maps of the Philippines to enhance post-disaster
graphic models for the country has made this possible. Quantita- rehabilitation using newly available high-resolution terrain data
tive procedures are advantageous in rehabilitation purposes 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
of evacuation and relocation to prevent repeat of disasters within hydrology models that involve modelling of groundwater and
the area in the future. vadose zone flows are GEOtop-FS (Rigon et al. 2006), TOPMODEL
Here, we implement a widely used method to characterize shallow (Beven and Kirkby 1979), dSLAM (Wu and Sidle 1995), and
landslides that can be employed in the nationwide mapping effort as SHETRAN (Burton and Bathurst 1998). SINMAP has been widely
a rapid shallow landslide assessment method. Using the typhoon- used and tested in several landslide-prone areas because of its
and landslide-prone Leyte province as a case study, available high- applicability, minimum requirement for input while allowing
resolution digital elevation data is utilized to generate shallow land- multi-calibration of the studied area, and model modification in
slide susceptibility maps. Unlike delineating hazard zones using smaller areas (Calcaterra et al. 2004; Deb and El-Kadi 2009; Lan
historical data, maps generated using a deterministic slope stability et al. 2004; Meisina and Scarabelli 2007; Morrissey et al. 2001;
geographic information system (GIS) model incorporates pixel- Zaitchik and van Es 2003; Zaitchik et al. 2003). It also adds the
specific soil strength parameters and hydraulic characteristics. Also, capability of representing results into a probabilistic framework.
it takes full advantage of the fact that landslide source areas are, in The reliability and degree of SINMAP model has been tested by
general, strongly controlled by shallow groundwater flow conver- Zizioli et al. (2013) by using receiver operating characteristic (ROC)
gence accompanied by the reduction of shear strength and increased analysis as compared to other widely used physically based
pore water pressure (Pack et al. 2001, 2005). models: SHALSTAB, TRIGRS, and SLIP. The values in the ROC
The most recent approaches to deterministic slope stability curve (sensitivity against one specificity) represent the ability of a
research involves diverse hydrology models using geotechnics at model to correctly classify between negative and positive observa-
different levels of complexity (Connell et al. 2001; Dietrich et al. tions. The global accuracy obtained for the SINMAP model
1992, 1993, 1995, 2001; Grayson et al. 1992a, b; Lan et al. 2004; expressed by the area under the curve (AUC) was equal to 0.79.
Montgomery and Dietrich 1994; Pack et al. 1998, 2001, 2005; Wu The global accuracy for SHALSTAB, TRIGRS, and SLIP was 0.78,
and Sidle 1995). Successful physically based stability model exam- 0.79, and 0.78 respectively. The values demonstrate that the four
ples are SHALSTAB (shallow slope stability model; Montgomery models have similar degrees of success for prediction of shallow
and Dietrich 1994), SINMAP (Stability Index Mapping; Pack et al. landslide source areas. Considering this, SINMAP model was se-
2005; Tarboton 1997), TRIGRS (Transient Rainfall Infiltration and lected for the case study due to limitation on geotechnical data or
Grid-Based Regional Slope-Stability Model; Baum et al. 2008), and soil strength parameters of Leyte province. With SINMAP, the
SLIP (Shallow Landslide Instability Prediction; Montrasio 2000; spatial variability of soil strength parameters can be expressed by
Montrasio and Valentino 2008). Other examples of advanced a range of values.
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
example of the precipitation that can produce shallow translation- assigned polygons were saved as vector files and analyzed in ArcGIS.
al movement). This range matches the maximum accumulated The landslide inventory statistics including number of landslide,
rainfall of 440 mm for a 72-h period or 147 mm/day on average total surface area damage, and density per locality were computed
(MapAction 2013). using the satellite images of Leyte province.
The transmissivity rate (m2/h) was calculated using the equa- Visual comparison was done with the SINMAP model and the
tion, T=Kb, where K is the hydraulic conductivity of the soil and b landslide inventory. Accuracy assessment is performed by calcu-
is the soil depth in meters. In the context of shallow landslide lation of detection percentage and miss factor.
movement, soil thickness of 1.5 m is assumed to be uniform in the
simulation. Data on hydraulic conductivity is obtained from a Results
database of soils in the Philippines (Hamazaki et al. 1990). Field
and laboratory descriptions and test results, along with literature Visual comparison of model and landslide inventory
values for soil properties given in Hammond et al. (1992) were also The area distribution for low, medium, and high susceptibility to
used to constrain reasonable ranges of soil input parameters for shallow landslides generated by SINMAP is shown in Fig. 6a.
the stability index modelling. Unsafe areas from shallow landslides were found to be 31 % of
the total land area of Leyte. Safe areas from shallow landslides
Validation comprise 69 %. The previous 1:10,000 landslide susceptibility map
The validation process of the generated SINMAP model is sum- by MGB (Fig. 2b) presents noticeably less stable areas compared to
marized on Fig. 5. Landslide inventory map consisting of 363 the SINMAP model (54 %) as shown in Fig. 6b.
landslides was used to validate the SINMAP model for Leyte Initially, visual comparison was done by manual overlaying of
province. The inventory utilized remote sensing techniques using the landslide inventory over the SINMAP model and MGB map.
various satellite images that have imagery dates from 2002 to 2014. The number of landslide polygons that fall within each suscepti-
Satellite imagery includes SPOT of 5 to 10 m spatial resolution, bility class is tabulated (Tables 3 and 4). Among the 363 landslide
QuickBird of 0.6 to 2.4 m, and WorldView of 0.5 m. These images polygons in Leyte, 9 were within the stable zones of the SINMAP
were inspected using manual detection method to identify land- model (Table 3). Majority of the polygons fell within the medium
slide events in Google Earth, freeware program, depending on its (66.1 %) susceptibility zones. For the MGB map, most polygons fell
availability. within the high and very high susceptibility zones (Table 4).
Satellite images were loaded and inspected in Google Earth.
Detected landslide events were delineated with polygons by outlining Accuracy assessment
the visible disturbed body. Each identified landslide was classified The number and extent of mapped landslides are critical to disas-
based on Cruden and Varnes classification (Varnes 1958, 1978). The ter relief and deterministic landslide assessment (Yange et al.
types of landslides indentified were mainly slides and flows. The 2013). Therefore, in this procedure, the susceptibility zones defined
by the SINMAP model were compared with the manually delin- The SINMAP results showed that the DP for the study area was
eated landslide inventory. This accuracy assessment procedure is 97.5 % with MF of 0.025. These values demonstrate that the
also done to the MGB map to compare its accuracy with the SINMAP approach performs well despite the lack of an extensive
SINMAP model. geotechnical and hydrological database in the study area. This also
By visual comparison, the landslide polygons were either suggests a consistency in the accuracy of the InSAR-derived DTM
marked as true positive or false negative using the following and the high-resolution imagery used for the landslide polygon
category definitions: 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
& True positive (TP): a landslide event occurred at the area of qualitative methods involving historical landslide data and rele-
instability defined by SINMAP vant geomorphological information. Both susceptibility mapping
& False negative (FN): a landslide event occurred outside the area methods establish high accuracy when compared to the landslide
of instability defined by SINMAP inventory; however, they differ much in the amount of stable areas
they define.
Table 3 Summary of landslide polygons captured by SINMAP model for Leyte province
Stable Low Medium High Total
No. of landslides 9 43 240 71 363
% of slides 2.5 11.8 66.1 19.6 100
especially for the rehabilitation work of Haiyan-affected commu- It is critical to provide rapid assessment and delineation of
nities because it is imperative that rebuilding be carefully planned areas susceptible to rainfall-induced shallow landslides for well-
to prevent future disasters. planned reconstruction and rehabilitation. The method of shallow
The landslides that were not captured by the SINMAP simula- landslide susceptibility mapping we implement in the Philippines
tions for Leyte could be due to a different failure mechanism not addresses the need for rapidly generated landslide hazard maps.
considered by the infinite slope stability model. For instance, These shallow landslide hazard maps have been made freely avail-
SINMAP does not consider structural controls for slope instability. able to different relief and rehabilitation agencies in typhoon
These landslides occur much deeper and are controlled primarily by Haiyan-ravaged areas. The shallow landslide maps that were pro-
fracture planes that pervade in mountainous terrain. Structurally vided are suited to the requirements of the rehabilitation and
controlled landslide maps are also being prepared under the new reconstruction process. It can still be and should be improved
Nationwide Hazards Mapping Program called Project NOAH where for better accuracy. More detailed geological and geotechnical
high-resolution topographic maps derived from Light Detection and assessments are necessary in the future. One area of improvement
Ranging (LiDAR) and InSAR are used. Along with the identification can be the replacement of InSAR-derived DTM data with more
of alluvial fans and simulation of debris flows, the SINMAP and the accurate LiDAR topographic data.
structurally controlled landslide hazard maps will constitute the
detailed hazard maps at 1:10,000 scale for landslides in Haiyan- Conclusion
devastated areas. These maps are available online for free in the The landslide map generated using SINMAP are found to be highly
Project NOAH portal (http://noah.dost.gov.ph). 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