Catena 140 (2016) 125–139

Contents lists available at ScienceDirect

Catena

journal homepage: www.elsevier.com/locate/catena

Evaluation of a combined spatial multi-criteria evaluation model and

deterministic model for landslide susceptibility mapping
A.M.S. Pradhan, Y.T. Kim ⁎
Dept. of Ocean Engineering, Pukyong National University, Busan 608-737, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: This study evaluated the application of a combined spatial multi-criteria evaluation model and deterministic
Received 5 September 2014 model for landslide susceptibility mapping in Deokjeok-ri Creek, located in the northeastern part of Korea. This
Received in revised form 26 November 2015 region has frequent shallow landslides often caused by intense rainfall on weathered granite soil slopes. This
Accepted 24 January 2016
study compared the predictive capability of two different models: a spatial multi-criteria evaluation (SMCE)
Available online 2 February 2016
model, which is a semi-quantitative model, and a shallow landslide stability (SHALSTAB) model, which is a
Keywords:
deterministic model used to produce shallow landslide susceptibility maps. For the SMCE model, input layers
Combined model of landslide causative factors (i.e., topographic, hydrological, soil, forest, and geological factors) were prepared
Deokjeok-ri Creek for pairwise comparison to obtain susceptibility weightage. For SHALSTAB, a digital elevation model was used
GIS to calculate slope and wetness indices. Field inventories were used to validate and combine the two models. A
Landslide susceptibility comparison of the susceptibility map obtained from the SMCE method with that obtained with the SHALSTAB
SHALSTAB method revealed that the total mismatch area between the two maps for all three susceptibility classes was
SMCE about 53%. Therefore, the two results were combined to improve the reliability of the susceptibility map. The
performance of the combined map was determined using the receiver operator curve (ROC). The area under
curve (AUC) revealed a success accuracy of 79.56%, and the predictive accuracy was 83.6%. These results
demonstrate that the combined model was more accurate than either individual model at delineating
landslide-prone areas of weathered granite soil slopes.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction economies. Landslide susceptibility assessment is a primary tool for

Concern about climate change has been increasing worldwide, and
localized high-intensity, short-duration rainstorms have been observed
in Asia. Korea is also experiencing changes in climate parameters,
including annual temperature and precipitation (Chung et al., 2004).
Data on the frequency of torrential rain have revealed a 25% increase
in torrential rain watches and a 60% increase in heavy rain warnings
over the last 20 years. South Korea has also been affected by extreme
rainfall events associated with typhoons: extreme rainstorms have
caused major damage via landslides and debris flows in the South
Korean mountains.
Typhoon Ewiniar in 2006 brought particularly heavy rains that
triggered many landslides, especially on soil-covered slopes over
weathered granite. Slope saturation due to infiltration of rainfall was
the primary cause of the landslides. In Korea, approximately 228 people
were killed by landslides from 2002 to 2011 (Pradhan and Kim, 2014).
Determining which areas are prone to landslides is important to pre-
serve human lives and avoid negative effects on regional and national
⁎ Corresponding author.
E-mail address: yuntkim@pknu.ac.kr (Y.T. Kim).
understanding the basic characteristics of slopes that are prone to
landslides, especially during extreme rainfall events (Dahal et al., 2012).
Researchers have developed various methods to assess landslide
susceptibility and landslide hazards and risk. Landslide susceptibility is
the likelihood of landslide occurrence in an area on the basis of local
terrain conditions (Brabb, 1984). To date, many techniques have been
developed and applied to produce landslide susceptibility maps. Soft
computing techniques such as artificial neural networks (Gomez and
Kavzoglu, 2005; Lee and Pradhan, 2007; Pradhan and Pirasteh, 2010)
and fuzzy approaches (Pourghasemi, 2008; Tangestani, 2009; Pradhan
and Lee, 2009; Akgün et al., 2012) have been used extensively for
landslide susceptibility assessment, and semi-quantitative, spatial,
multi-criteria evaluations have also been used for this purpose (Abella
and van Westen, 2007; Armas, 2011; Akgün et al., 2012; Neuhauser
et al., 2012).
Deterministic models have also been used extensively since the
1990s (Montgomery and Dietrich, 1994; Dietrich and Sitar, 1997;
Dietrich and Montgomery, 1998; Wilcock et al., 2003). Several
researchers have proposed different deterministic approaches based
on an infinite-slope stability model and rainfall infiltration models
(Montgomery and Dietrich, 1994; Dietrich et al., 1995; Terlien et al.,
1995; Crosta et al., 2003).

http://dx.doi.org/10.1016/j.catena.2016.01.022
0341-8162/© 2016 Elsevier B.V. All rights reserved.
126 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139

Most previous research has involved a GIS technique applied to an

individual susceptibility model or compared with maps resulting from
different susceptibility models. Recently, Goetz et al. (2011) used gener-
alized additive models (GAMs) to improve landslide susceptibility
models by combining the SHALSTAB model and a factor of safety (FS)
derived from an infinite-slope model with statistical and machine-
learning techniques; however, variables representing physically based
models did not significantly improve the empirical models, although
they may allow for better physical interpretation of the empirical
models. Choi et al. (2012) combined three landslide susceptibility
maps using frequency ratio (FR), logistic regression (LR), and artificial
neural network (ANN) models to make three different improved sus-
ceptibility maps, based on six landslide conditioning factors, and then
combined these maps. Prediction accuracy increased slightly as a result
of this combination. Rossi et al. (2010) highlighted that the combined
Fig. 2. Mean monthly rainfall in Inje area.
models resulted in a reduced number of errors and in less uncertain
predictions; the combination of landslide susceptibility zonation can
provide “optimal” susceptibility forecast.
Although comparative studies of multiple forecasts for landslide SMCE based models generally function under the principle that landslides
susceptibility assessment are available, a method for combining differ- are more likely to occur under similar ground conditions to previous
ent geographical forecasts into an optimal prediction is still to be events and such model of susceptibility does not usually take into account
established (Chen et al., 2014). Rossi et al. (2010) and Choi et al. triggering factors, such as earthquakes and precipitation (Dai et al., 2002;
(2012) adopted landslide susceptibility forecasting approaches similar Sidle and Ochiai, 2006). On the other hand, SHALSTAB model utilizes the
to those applied to determining optimal weather forecasting (Molteni physical properties and hydrological properties that control geomorpho-
et al., 1996, 2001; Marsigli et al., 2001). They used different models logical processes spatially. Various susceptibility mapping methods from
that made separate results, then evaluated these results independently different models such as quantitative or qualitative techniques have
and combined them into best performance optimal forecast. always advantages and disadvantages. In quantitative and qualitative
The main objective of this study was to identify landslide-prone techniques, critical rainfall as a thematic layer is always missing. The com-
areas with weathered granite soil using a combined approach involving bination of SMCE model with SHALSTAB can incorporate the rainfall trig-
different susceptibility models. It evaluated landslide susceptibility gering factor in semi-quantitative model. The main difference between
mapping using a semi-quantitative spatial multi-criteria evaluation the present study and previous approaches is its combination of a semi-
(SMCE) model and a shallow landslide stability model (SHALSTAB). quantitative SMCE and deterministic critical rainfall based models,
which helps to incorporate with critical rainfall scenario and improves
accuracy and reduces model variance. However these two models did
not depend on each other. This study has two strands: firstly, the suscep-
tibility class comparison and validation of two models, and secondly,
the susceptibility maps were subsequently combined into one map
using the normalized frequency ratio (NFR) method to assess the
maximum likelihood of future landslide occurrences.

2. Study area

Deokjeok-ri Creek in Inje County, Korea, which suffered a great deal

of damage following heavy rain in 2006, was selected as a suitable site
for evaluating landslide susceptibility models. Deokjeok-ri Creek is
located in the northeastern part of Korea, as shown in Fig. 1. Its area is
33.4 km2, and it is surrounded by steep mountains. The lithologies of
this area include Precambrian gneisses and Mesozoic granites.

Fig. 1. Location map of Deokjeok-ri. Fig. 3. Extreme rainfall on July 14 to 17 in Inje area.
 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139 127

Fig. 4. Picture with examples of different types of slope failure: a) Photo index. b, c) Bird's eye view of slope failure. d–f) Shallow slide and debris flow. g) Debris deposit.

The soils at the site are derived from weathered granite and exhibit a
wide range of conditions depending on the degree of weathering. They
range from soils close to weathered rock to those that contain fine sed-
iment grains, such as silt and clay. Most subsurface soils are typically in
an unsaturated state, especially in mountainous areas, but when rain in-
filtrates the ground, granite soils become partially or fully saturated.
This increases the fragility of the soil, as evidenced by grain breakage
observed in grain-size analysis (Lee and de Freitas, 1989), resulting in
significant loss of shear strength. Rainfall-induced landslides in this
area are usually shallow failures on weathered granite slopes, especially
on initially unsaturated slopes. These slope failures are triggered mainly
by infiltration of rainfall, which decreases matric suction and shear
strength (Kim et al., 2004; Rahardjo et al., 2005).
Typhoon Ewiniar, which made landfall in South Korea on 10 July
2006, helped bring record-breaking rainfall to the region. Fig. 2 shows
that the monthly precipitation in July 2006 was ~ 2.4 times greater
than the average monthly rainfall. Most of the residents in this area
live near the main channel of Deokjeok-ri Creek, and the extreme
rainfall in July 2006 resulted in 17 deaths and 12 missing in Inje County
(Lee and Yoo, 2009). All residential areas were swept away by landslides
and debris flows, which left survivors in the area isolated for three days
without assistance.
128 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139

Fig. 5. Landslide inventory map of Deokjeok-ri Creek.

In the wake of Typhoon Ewiniar, the Inje area was subjected to
extreme rainfall from 14 to 17 July 2006. This rainfall was caused by
two effects: a rain front from North Korea and water vapor that devel-
oped in China due to Typhoon Bilis (Bae, 2007). The total rainfall from
14 to 16 July 2006 was 402 mm. The maximum hourly rainfall intensity
on 15 July was 62 mm/h, as shown in Fig. 3, and the maximum rainfall
over the 24-h period was 192 mm.
Fig. 4 shows the different types of slope failures observed in the
study area: the most common failures were shallow landslides that
sometimes resulted in debris flows during periods of intense rainfall.
3. Method and materials
3.1. Landslide inventory map and soil parameters
A landslide-inventory map is used to identify the location of previ-
ous landslides, along with their types and the time of their occurrences
(Wieczorek, 1984; Einstein, 1988; Soeters and van Westen, 1996).
Landslide inventory maps can be prepared either by collecting historical
information of individual landslide events or by consulting satellite im-
agery and aerial photographs coupled with field surveys using GPS. To
produce a detailed and reliable landslide-inventory map of Deokjeok-
ri Creek, extensive field surveys and observations were performed in
the study area. A total of 748 landslides were mapped and a simple ran-
dom sampling technique was used for training (~90%) and validation
(~ 10%) data selection, as shown in Fig. 5. For validation of SHALSTAB
model, all 748 landslides were used.
Six disturbed and three undisturbed samples corresponding to
shallow landslide scars were taken. Their grain-size distribution, bulk
and dry densities, unsaturated triaxial shear strength, and internal
frictional angle were characterized using laboratory measurements.
The soil samples collected from the shallow landslides were residual
soil deposits formed by weathering of the bedrock. Table 1 lists a
geotechnical database, and Table 2 lists the results of triaxial shear
tests. The cohesion was 2.14 kPa, and the internal frictional angle was
35.61°. Six permeability measurements were conducted at different
depths using a 2800K1 Guelph Permeameter and the constant head
method. Table 3 lists the results of these measurements at the depths
of 20, 40, and 60 cm. The residual deposits were highly permeable,
and the average in situ permeability was 1.87 × 10−5 m/s.
Soil depth is an important factor in assessment of landslide instabil-
ity, especially in infinite-slope stability analysis. It is often highly
variable and difficult to predict. Moreover, soil depth surveys are

Table 1 Table 2
Soil parameters of the Deokjeok-ri Creek area. Triaxial shear strength and frictional angle.

Properties Average Rock type Location Effective frictional angle Cohesion

Moisture content (%) 9.92 ϕ′ c' (kPa)
Specific gravity 2.59 Granite Deokjeok-ri 35.61 2.14
Unit weight (kN/m3) 16.44
Density (kg/m3) 1.68
Dry unit weight (kN/m3) 14.98
Void ratio 0.736 Table 3
Porosity (%) 42.40 The in situ permeability at different depths in the Deokjeok-ri Creek area.
Coeff. of uniformity 6.88
Depth (cm) In situ permeability (m/s) Average permeability (m/s)
Coeff. of curvature 1.36
Deokjeok-ri Creek
USCS SP-SW
Gravel (%) 38.06 20 4.18 × 10−5 1.83 × 10−5 1.87 × 10−5
Sand (%) 58.89 40 1.40 × 10−5 1.56 × 10−5
Silt and clay (%) 3.05 60 1.30 × 10−5 9.48 × 10−6
 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139 129

Table 4 pairwise comparison judgments, or the judgment is considered

Scale of relative importance suggested by Saaty (1977). 100% consistent.
Intensity of Definition Explanation Saaty (2000) also randomly generated reciprocal matrices using the
importance scales 1/9, 1/8,…, 1,… 8, 9 to evaluate a random consistency index (RI):
1 Equal importance Two activities contribute Table 5 lists the standard values of RI.
equally to objective Saaty (1977) introduced a consistency ratio (CR), which is a
3 Weak importance of one Experience and judgment comparison between the consistency index and the RI:
over another slightly favor one activity
over another
CI
5 Essential or strong Experience and judgment CR ¼ ; ð2Þ
importance strongly favor one activity RI
over another
7 Demonstrated importance An activity is strongly where RI is a random index and depends on the order of the matrix. If
favored and its dominance
the threshold of CR is achieved (CR b 0.1), the weights of each row of
demonstrated in practice
9 Absolute importance The evidence favoring one the matrices are calculated.
activity over another is the An open source GIS software, Integrated Land and Water Information
highest possible System (ILWIS) was used for pairwise comparison of the factors. With a
order of affirmation
pairwise comparison matrix, each class of a CF is compared to all other
2, 4, 6, 8 Intermediate values When compromise is needed
between the two classes of same CF to evaluate whether they are equally significant, or
adjacent judgments whether one of them is somewhat more significant or better than the
1/9 1/8 1/7 1/6 1/5 1/4 1/3 1/2 1 2 3 4 5 6 7 8 9 other. The method then creates a matrix containing the pairwise com-
Less important ⟺ More important parison judgments for the criteria, from which a priority vector is derived
of relative weights for these elements (the principal eigenvector of the
matrix). Moreover, due to the fact that more information is retrieved
from the decision maker, the method can deliver an inconsistency mea-
time-consuming, and soil depth is often difficult to measure in even sure. The main disadvantage of this method is that subjective preference
small basins like Deokjeok-ri. In this study, soil depth was estimated in the ranking of factors may differ from one expert to another (Kayastha
through field observations of 50 landslide scars. The value adopted for et al., 2012). To avoid this limitation, the intensity of importance was
the entire basin was considered to be equal to the soil depth at which judged using landslide frequency in each classes of CFs. This measure
slope failure occurred, i.e., an average depth of 1 m. can be used to verify in what measure the judgments supplied are con-
sistent (Sharifi and Retsios, 2003). AHP is especially designed to assess
3.2. Spatial multi-criteria evaluation weights within a hierarchical structure of the criteria. However, due to
the fast-growing number of pairwise comparisons it is not sensible to
Spatial multi-criteria evaluation (SMCE) is a technique that assists use the method for a large set of criteria (ITC 2001).
stakeholders make decisions with respect to a particular goal (in this
case landslide susceptibility). The input is a set of maps that are spatial 3.2.1. Estimation of terrain parameters for SMCE
representations of the criteria, which are grouped, standardized, and The main steps for mapping landslide susceptibility involved data
weighted in a criteria tree. The theoretical background for SMCE is collection and construction of a spatial database in the GIS environment.
based on the analytic hierarchical process (AHP) developed by Saaty The landslide-inducing factors considered in this study are essentially
(1980), which is used to determine the weight of each criterion and the intrinsic causes for which pertinent data could be collected
analyze the relative importance of the criteria. SMCE can be defined as from available resources and fieldworks. These data were collected
a decision aid and mathematical tool allowing comparison of different based on their relevance and availability. In this study, 15 causative
alternatives according to many criteria, often conflicting, in order to factors (CFs) were used to identify the terrain characteristics of land-
guide decision-makers toward a judicious choice. slide occurrence. Causative factors were classified into five groups. The
In order to determine relative weights, AHP is used to compare first group included topographic factors, consisting of aspect, slope,
factors using a scale (intensity of importance) from 1 to 9 if the elevation and curvature. The second group included hydrological factors
factors have a direct relationship and a scale from 1/2 to 1/9 if the such as distance to drainage, drainage density, TWI, SPI, STI, and SDC.
factors have an inverse relationship, as shown in Table 4 (Saaty, The third group included soil factors, consisting of soil type and soil
1977). Another appealing feature of the AHP is its ability to evaluate depth. The fourth group included forest factors, consisting forest type
pairwise rating inconsistency. Saaty (2000) proved that for a consis- and timber age. The fifth group included geological factors.
tent reciprocal matrix, the largest eigenvalue λMax is equal to the
number of comparisons n. A measure of consistency, called the 3.2.1.1. Topographic factors. A digital elevation model (DEM) of the study
consistency index (CI), is defined as follows: area was prepared using light detection and ranging (LIDAR) data, and
geomorphological thematic data layers were prepared from the DEM.
λMax Aspect is related to parameters such as precipitation and sunlight
CI ¼ ; ð1Þ
n−1 exposure (Ercanoglu et al., 2004). In this study, the aspect of the slopes
of the terrain surface was divided into nine classes (Fig. 6a).
where CI gives information about logical consistency among If a slope angle is steep, the slope is at increased risk for landslide
pairwise comparison judgements in a perfect pairwise comparison occurrence. Slope angle affects the velocity of both surface and subsur-
case. When CI = 0, there is no logical inconsistency among the face flow and hence soil water content, soil formation, and erosion

Table 5
Random consistency index (RI) (Saaty, 1980, 2000).

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

RI 0 0 0.58 0.9 1.12 1.24 1.32 1.41 1.45 1.49 1.51 1.53 1.56 1.57 1.59
130 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139
Fig. 6. Topographic factors: a) Aspect, b) slope, c) elevation and d) curvature.
potential (Cooke and Doornkamp, 1990). The entire slope range in this
study was divided into five categories: 0–15°, 15–25°, 25–35°, 35–45°,
and N 45° (Fig. 6b).
Elevation is well known to influence a large number of biophysi-
cal parameters and anthropogenic activities. These conditions, in
turn, are likely to affect slope stability and generate slope failures
(Vivas, 1992). In the study area, elevation ranged from 195 to
1225 m. Elevation values were divided into six categories using
intervals of 200 m (Fig. 6c).
Slope shape acts as a concentrating or scattering factor, and
has a strong influence on slope stability in steep terrain and is a
primary determinant of subsurface water in the landscape. There
are three basic slope units: (1) divergent or convex, (2) planar or
straight, and (3) convergent or concave. Generally, convex slopes
are more stable because they disperse runoff more equally down
the slope. Concave slopes are considered to be potentially unstable
because they concentrate water at the lowest point and contribute
to the buildup of adverse hydrostatic pressure (Stocking, 1972)
(Fig. 6d).
3.2.1.2. Hydrological factors. Infiltration of rainfall into soil and runoff
are significant controlling factors of landslide occurrence, especially
for rainfall-induced landslides on weathered granite soil slopes. In
watersheds, landslides frequently occur on stream banks.
The proximity of a slope to a drainage structure is another important
factor for instability. In this study, five different buffer zones were
prepared from stream maps using the inverse distance-weighted inter-
polation method: (i) 25 m, (ii) 25–50 m, (iii) 50–75 m, (iv) 75–100 m,
and (v) N100 m from the drainage (Fig. 7a).
Drainage density is the ratio of the total length of the stream
to the area of the drainage basin. When drainage densities are
higher, infiltration is lower and movement of surface flow is faster
(Pachauri et al., 1998; Cevik and Topal, 2003). In this study, drainage
density was divided into four classes: b5 m/m 2 , 5–10 m/m 2 ,
10–15 m/m2, and N15 m/m2 (Fig. 7b).
The stream power index (SPI) (Fig. 7c) is a measure of the erosive
power of water flow based on the assumption that discharge is
proportional to the specific catchment area (Moore et al., 1991).
Higher SPI values lead to an increased risk of slope erosion (Moore
et al., 1991).
A dimensionless sediment transport index (STI) was derived by con-
sidering the transport capacity limiting sediment flux and catchment
evolution erosion theories (Moore and Wilson, 1992). In this study,
STI was divided into four classes (Fig. 7d).
The topographic wetness index (TWI) is an indicator of the spatial
distribution of soil moisture because groundwater flow often follows
surface topography. In the present study, TWI was divided into three
classes (Fig. 7e).
Soil drainage refers to the frequency and duration of periods when
the soil is saturated with water. In the present study, soil drainage
characteristics (SDC) were classified into four categories: poorly
drained, somewhat poorly drained, moderately drained, and well
drained (Fig. 7f).
3.2.1.3. Soil factors. Soil properties influence the rate of water move-
ment and the capacity of the soil to hold water. Under unsaturated
conditions, fine-textured soils tend to hold more water than
coarse-textured soils (Sidle et al., 1985). Based on soil classification
guidelines from the United States Department of Agriculture
(USDA, 1993), the Deokjeok-ri Creek area is composed of five soil
types, shown in Fig. 8a. Most soil types at the study site were sandy
loam, consisting of forest soils. The downstream soil was silty clay
 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139 131

Fig. 7. Hydrological factors: a) Distance to drainage, b) drainage density, c) stream power index, d) sediment transport index, e) topographic wetness index and f) soil-drain character.

loam, and this area was mostly cropland. The northern and southern Soil depth influences the shear stress and shear strength of the slope
edges of the watershed were rocky due to high elevations and and is also associated with landslide volume. Generally, the deeper the
steep slopes. soil, the larger the moving mass will be. In this study, relative soil

Fig. 8. Soil factors: a) Soil type and b) soil depth.

132 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139

Fig. 9. Forest factors: a) Forest type and b) timber age.

Fig. 10. Geological factor: a) Granite exposure and b) geological map.

depth was classified into the categories of shallow, moderate, or deep, as
depicted in Fig. 8b.
3.2.1.4. Forest factors. The Deokjeok-ri Creek area has four main forest
types: broadleaved forest (mostly Mongolian Oak), Japanese larch,
Japanese red pine, and Korean pine, along with non-forested areas
consisting mainly of cultivated land and soil deposits (Fig. 9a).
When an entire area is covered with vegetation, dramatic changes
in the size or age of the vegetation can help distinguish shear zones.
Disrupted sediment generally results in the growth of trees of similar
age and hence size (McCalpin, 1974). Fig. 9b shows different categories
of timber age in the study area.
3.2.1.5. Geological factors. Landslides are greatly affected by the litholog-
ical properties of a land surface. Because different lithological units have
different landslide susceptibility values, it is very important to consider
individual unit contributions when mapping susceptibility. As described
earlier, the lithology of this area consists of Precambrian gneisses and
Mesozoic granites, as shown in Fig. 10.
3.3. Shallow landslide stability (SHALSTAB) method
SHALSTAB is a well-known physical-based model developed by
Montgomery and Dietrich in 1994 (Montgomery and Dietrich, 1994).
The model performs an infinite-slope stability analysis assuming
steady-state conditions. Their approach is based on the assumption

Fig. 11. Procedure used to combine two models.

 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139 133

Table 6
The weight value for each group and causative factor using pair-wise comparison
for SMCE.

Group Causative factors Resulting normalized weight values

Topographic Elevation 0.066

Slope 0.55
Aspect 0.069
Curvature 0.314
Consistency ratio (CR) = 0.0670
Hydrologic Distance to drainage 0.538
Drainage density 0.042
TWI 0.106
SPI 0.125
STI 0.149
SDC 0.041
Consistency ratio (CR) = 0.0958
Soil Soil type 0.125
Soil depth 0.875
Fig. 13. Terrain area percent occupied by different susceptibility classes.
Consistency ratio (CR) = 0
Forest Forest type 0.833
Timber age 0.167 where a is the drainage area or contribution area, b is the width of the
Consistency ratio (CR) = 0 outflow boundary or topographic contour length, Q (m h−1) is the
Geology Geology 1
critical rainfall necessary to trigger landslides, T (m2 h− 1) is the
Consistency ratio (CR) = 0
Topographic 0.318 transmissivity of the soil (defined as the product of the soil thickness
Hydrologic 0.281 and saturated hydraulic conductivity), and θ (°) is the local slope.
Soil 0.104 SHALSTAB predicts the critical rainfall necessary for slope failure
Forest 0.147 throughout a study area (Montgomery and Dietrich, 1994). The main
Geology 0.151
Consistency ratio (CR) = 0.0871
equation that is solved in SHALSTAB to compute the critical rainfall
(Q) for each grid cell in the unit mapping is as follows:
 0 
that the specific upslope area is a surrogate measure of subsurface Tb sinθ c ρS tanθ
Q¼ þ 1− ; ð4Þ
flow at any point in the landscape. This assumption is based on a set a ρw gz cos2 θ tanϕ0 ρw tanϕ0
of hypotheses: (1) uniform upslope recharge rate; (2) equilibrium
condition of the upslope catchment response; and (3) lateral flow where ρw (kg/m3) is the density of water, g (m/s2) is the acceleration of
parallel to the surface slope and therefore unit contour length equiva- gravity, z (m) is the soil thickness, ρs (N/m2) is the soil bulk density, ϕ'
lent to the unit flow width (Chirico et al., 2003). SHALSTAB applies (°) is the soil friction angle, and c′ (N/m2) is cohesion.
the TOPOG hydrologic model (O'Loughlin, 1986), which calculates The predictive index of this model is expressed in mm day−1 of crit-
the relative wetness (or relative soil saturation) w based on the ical rainfall, so low values indicate that low-intensity, short-duration
steady-state saturated water flow parallel to the slope plane: rainfall may trigger shallow landslides, whereas high values indicate
that more intense storms are necessary to generate slope instability
Q a Actual runoff at given steady state rainfall (Montgomery and Dietrich, 1994). The assumption of SHALSTAB is
w¼ ¼ ; ð3Þ
T b sinθ Soil profile saturation that when the layers of soil are at an angle to the horizontal, the force

Fig. 12. Landslide susceptibility map using SMCE.

134 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139
Fig. 14. Percentage of landslide occurrences in each susceptibility class.
of gravity causes a downslope flow component. This lateral flow occurs
with any amount of rain and for both unsaturated and saturated soil
(Montgomery and Dietrich, 1994, Barling et al., 1994). One drawback
of this model is that it does not account for the uncertainty of input
parameters as in SINMAP (Pack and Tarboton, 2004) model by giving
the option of identifying an upper and lower range of values for
each variable.
3.4. Combined model
In this study to produce optimal landslide susceptibility zonation
map that obtained from SMCE and SHALSTAB models, a simple nor-
malized frequency ratio (NFR) model was applied as a bivariate
probability analysis. The susceptibility maps obtained from the
SMCE and SHALSTAB methods were considered as new input factors
to develop a combined landslide susceptibility map and using the
detected landslide locations and the obtained susceptibility maps,
the normalized frequency ratio (NFR) model was applied as a bivariate
probability analysis, as shown in Fig. 11.
Fig. 15. Landslide susceptibility map using SHALSTAB.
The frequency ratio model has some advantage of simplicity; more
importantly, inputs, and calculation process are readily understood
(Lee and Pradhan, 2007). Using NFR, any number of different types of
information can be combined into a single map, as long as such informa-
tion can be quantified. Normalizing the data provides that each CF has
the same range of values for the area to be investigated. It is a common
practice to measure how two variables simultaneously change to deter-
mine whether the variables are related to one another (Pradhan et al.,
2012). With regard to NFR, it is a common practice to assume that land-
slide occurrence is determined by landslide-related factors, and that
future landslides will occur under the same conditions as past landslides
(Lee and Pradhan, 2006). Based on these ratio values, each category was
assigned an attribute number and then was rescaled in the range of
0.1–0.9 using the Max–Min normalization procedure as follows:
v − MinðvÞ
v0 ¼ ðU − LÞ þ L; ð5Þ
MaxðvÞ − MinðvÞ
where v′ is the normalized data matrix, v is the original data matrix, and
U and L are the upper and lower normalization boundaries, respectively.
v can be calculated as the ratio between area of landslide to the area
of the domain of class. v′ represents an approximation to the normal
distribution of the probability.
4. Result and discussion
4.1. Landslide susceptibility mapping using SMCE
The SMCE model was evaluated using open source integrated land
and water information system (ILWIS) (Nijmeijer et al., 2001), a GIS
environment. All comparisons were based on a pairwise method.
Table 6 lists the resulting normalized weight values of the 15 CFs and
overall weight values. The resulting consistency ratio (CR) for the
pairwise comparison matrix for the 15 CF dataset layers was 0.0871,
indicating that the comparisons of characteristics were perfectly
consistent and that the relative weights were appropriate for use in
the susceptibility mapping.
 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139
Fig. 16. Terrain area percent occupied by different susceptibility classes.
In order to produce a landslide susceptibility map with the SMCE,
the weights corresponding to parameters were multiplied by the
relevant parameter maps and then all the weighted parameters
were summed. The resulting weight index map showed the spatial
distribution of landslide susceptibility. Then, the weighted map
was divided into three susceptibility classes — low, moderate, and
high — to aid visual interpretation as presented in Fig. 12. To obtain
the classified susceptibility map, the distribution of the total weight
value was analyzed, and the values of the natural break were used.
According to the map, 41.7% of the study area has low susceptibility
and approximately 27.3% of the total area has high susceptibility as
shown in Fig. 13.
Fig. 14 presents the percentages of landslide occurrence for both the
training and validation data under the different susceptibility classes.
Overall, 46.67% of the training data landslides and 42.92% of the valida-
tion data landslides were in the high susceptibility class. The moderate
susceptibility class included about 40% of the training data landslides
and 32.79% of the validation data landslides, and the low susceptibility
class included 13.33% of the training data landslides and 24.29% of the
validation data landslides.
4.2. Landslide susceptibility using SHALSTAB
A susceptibility map was obtained using the SHALSTAB model,
shown in Fig. 15, includes seven classes that vary from unconditionally
unstable to unconditionally stable. Fig. 16 presents the landslide suscep-
tibility distribution calculated using SHALSTAB: A total of 2.4% of the
area was categorized as being unconditionally unstable, 5.88% of the
Fig. 17. Percentage of landslide occurrences in each susceptibility class.
135
area had critical rainfall b50 mm/day, 6.91% had critical rainfall of
50–100 mm/day, 11.87% of the area was categorized as unstable with
100–200 mm/day of critical rainfall, and 15.69% was delineated as
unstable with 200–400 mm/day. Similarly, 22.49% of the area was
identified as unstable with critical rainfall N400 mm/day. Also, 34.75%
of the area of the creek was found to be unconditionally stable.
The landslide density observed in each of the categories calculated
using SHALSTAB is depicted in Fig. 17. The SHALSTAB approach to insta-
bility modeling was quite successful in describing slope failure in the
study area, identifying 15.37% of 748 observed landslides in uncondi-
tionally unstable areas, 27.67% in regions with critical rainfall of
b50 mm/day, 18.98% in areas with critical rainfall of 50–100 mm/day,
18.45% in areas with critical rainfall of 100–200 mm/day, 11.9% in
areas with critical rainfall of 200–400 mm/day, and 4.81% of landslides
in areas with critical rainfall of N400 mm/day. The landslide density
was greater in areas of greater instability. The fact that 2.81% of
landslides occurred in areas that were identified as unconditionally
stable may be considered as a measure of the error in the model.
4.3. Validation and comparison of models
Validation is a fundamental step in the development of a landslide
susceptibility map. The performance of the two models can be
determined using the receiver operator curve (ROC) technique. The
ROC is a useful method for representing the quality of deterministic
and probabilistic detection and forecast systems (Swets, 1988). The sus-
ceptibility index (derived from SMCE and SHALSTAB) was reclassified
into two classes using cutoff values. A contingency table was derived
for the specific cutoff by intersecting the susceptibility classes and
presence/absence landslides. By changing the value of cutoff, it is possi-
ble to obtain different contingency tables that correspond to different
points on the ROC curve (SafeLand Deliverable D2.8, 2011). When
classifying a grid from the unstable map, four outcomes are possible
(shown in Table 7): if landslides were observed in a cell that was calcu-
lated to be unstable, this was considered a true positive (TP); if land-
slides were not observed in an area that was calculated to be unstable,
it was considered a false positive (FP); if landslides were observed
in an area that was calculated to be stable, it was considered a false
negative (FN); and if no landslides were observed in an area that was
calculated to be stable, it was considered a true negative (TN). The
ROC was calculated by plotting the fraction of positive outcomes that
were correctly identified (i.e., the TNs) versus the fraction of positive
outcomes that were not correctly identified (i.e., the FNs). The area
under the ROC (AUC) can be used as a metric to quantify the overall
performance of the model such that the larger the area, the better the
performance of the model.
For classifying the accuracy, the following ranking were considered
for the accuracy test given by Hosmer and Lemeshow (2000): 0.90–1
(excellent), 0.80–0.90 (good), 0.70–0.80 (fair), 0.60–0.70 (poor),
and 0.50–0.60 (fail).
Success rate results were obtained using the landslide grid cells in
the training dataset. Fig. 18 shows the success rate curves of the two
landslide susceptibility maps obtained from the SMCE and SHALSTAB
methods. For SMCE, the area under a curve is 0.6823, which means
that the overall success rate of the landslide susceptibility zonation
map is 68.23% and the success rate is poor. Then, the areas under the
Table 7
Parameters for the ROC curve calculation (modified from Swets, 1988).
Landslide pixel Area without
landslide pixel
Landslide occurrence based on model True positive (TP) False positive (FP)
Landslide free area based on model False negative (FN) True negative (TN)
136 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139
Fig. 18. ROC curves of SMCE and SHALSTAB models.
prediction-rate curves (AUC) were further estimated using validation
data, and the landslide susceptibility map obtained by SMCE had a pre-
dictive rate of 75.14% and the prediction result is fair. For SHALSTAB, all
landslide data were used for the modeling of the susceptibility map: its
calculated success rate was 61% which means success rate is poor.
To check the effective spatial distribution of the susceptibility zones,
the map produced by SHALSTAB was again reclassified as low (stable),
moderate (rainfall 0–100 mm/day), or high (greater than 100 mm/day,
because a total of 192 mm of rain fell on the day of the event) suscepti-
bility classes, as shown in Fig. 19. The areas occupied by the different
level-of-susceptibility classes of the reclassified SHALSTAB and SMCE
are almost the same, as depicted in Fig. 20. Next, the two maps
were overlaid in GIS, and the respective percentage coverage for each
susceptibility class was evaluated.
Fig. 19. Landslide susceptibility map obtained by reclassification of SHALSTAB.
Fig. 20. Comparison between area occupied by SMCE and reclassified SHALSTAB models.
A comparison of the landslide susceptibility maps using the
landslide-density analysis and success-rate-curve analysis indicated
that these two methods produce reliable results. However, these analy-
ses did not assess the spatial agreement between the maps. Fig. 21
shows details of the two susceptibility maps for an area where
landslides were observed. Although most of the observed landslide
area falls in high susceptibility zones in both maps, the spatial agree-
ment between the respective susceptibility classes is not perfect.
Although the landslide-density and success-rate analyses seemed to in-
dicate that the models produce similar results, they may actually yield
considerable mismatch between predicted susceptibility values. It is
therefore necessary to compute the spatial agreement areas between
the two landslide susceptibility maps in order to evaluate their consis-
tency (Kayastha et al., 2012).
The spatial agreement area between the two landslide suscepti-
bility maps, expressed as a percentage of the total area, was deter-
mined as the total area with the same landslide susceptibility
zonation on both maps (van Westen et al., 2003; Gupta et al., 2008;
Kayastha et al., 2012). Table 8 lists the spatial agreement areas
between the susceptibility maps obtained from SMCE and SHALSTAB.
 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139 137

Fig. 21. Example showing a mismatch between the predicted landslide susceptibility.

For the three susceptibility classes, the perfect agreement area
was 46.73% of the total area of the creek, and the total mismatch
was 53.27%. The specific susceptibility probability classes of the
SMCE map did not agree totally with the corresponding classes of
the SHALSTAB map. The final susceptibility maps from the SMCE
and SHALSTAB methods produced relatively similar results for
the low susceptibility zone, but less agreement for the moderate
susceptibility zone. This result occurred because the SMCE is based
on AHP and the deterministic method is based on hydro-geotechnical
properties of soil.
4.4. Combining the SHALSTAB and SMCE models
To improve the reliability of the susceptibility map, the results of the
two models were combined. Table 9 lists the calculated normalized
weightage, the high classes have high weightage. The landslide suscep-
tibility map obtained by combing these two methods was categorized
into low, moderate, and high susceptibility zones using natural break
of weight value distribution. Fig. 22 presents this resulting landslide
susceptibility map: 35.8% of the area was categorized as having low
susceptibility, 34.5% was categorized as moderately susceptible, and
the remaining 29.7% was considered to have high susceptibility. Analy-
sis of the landslide distribution with respect to the susceptibility classes
was an important part of this study, so a landslide density analysis was
performed on the three susceptibility classes. An ideal landslide suscep-
tibility map has landslide density values that increase from a lower to a
higher susceptibility class. Fig. 23 plots the landslide densities for the
three landslide susceptibility classes of the combined model. For the
landslides used as training data, landslide density gradually increased
from lower to higher susceptibility classes. About 82.2% of the training
data landslides were located in the high and moderate susceptibility
classes, and 74.7% of the validation data landslides were identified as
having the same class as that projected by the model, demonstrating
Table 8
Spatially agreed area between landslide susceptibility maps obtained from SHALSTAB
and SMCE.
the reliability of the combined map. Of the training data landslides,
about 18.1% were in low susceptibility areas; 25.3% of the validation
data landslides fell within this class.
The success rate result was obtained by comparing the landslide
training data with the combined susceptibility map, as shown in
Fig. 24. The area under curve (AUC) yielded a success accuracy of
79.6% which means success rate is fair. Because the success rate method
used the training landslide data that had already been used to build the
landslide models, the success rate is not a suitable method for assessing
the prediction capability of the models (Bui et al., 2011). However, this
method is useful for determining the performance of the models. Thus,
the combined map was validated using the prediction rate curve. The
prediction result was obtained by comparing the landslide validation
data with the combined susceptibility map. The AUC was calculated as
83.6%, indicating that the prediction rate is good. This result indicates
that the susceptibility map obtained by the combined approach
was reasonably accurate at predicting the landslide susceptibility of
Deokjeok-ri Creek. The combined approach yielded much higher
accuracy than either individual model.
5. Conclusions
To predict future shallow landslides at Deokjeok-ri Creek, landslide
susceptibility mapping was performed using the SMCE and SHALSTAB
methods. The SMCE was applied and a landslide susceptibility map
was prepared using the result of a combination of 15 causative factors
responsible for landslide susceptibility. In the SMCE, the results indicat-
ed that the area identified as zones of high susceptibility accounted for
27.3% of the total hillslope area. About 86.67% of the training data
landslides and 75.71% of the validation data landslides occurred in
high and moderate susceptibility zones, respectively.
Seven different instability classes based on the Q parameter were
calculated in SHALSTAB. Approximately 34.75% of the entire study
area was calculated as unconditionally stable. When the map from the
SHALSTAB method was crossed with the map obtained from SMCE,
46.73% of the total area was in perfect agreement. Because the SMCE is

Susceptibility classes Susceptibility classes Area

based on SMCE based on SHALSTAB percentage Table 9
Low Low 24.16a Calculation of normalized weightage to combine two models.
Moderate 11.63
Input Class Area % of Area % of Frequency Normalized
High 5.91
parameter class landslide ratio (FR) weightage
Moderate Low 11.89
Moderate 10.29a SMCE Low 41.70 24.07132 0.577285 0.10
High 8.81 Moderate 30.99 34.02675 1.097944 0.54
High Low 6.43 High 27.31 41.90193 1.534239 0.90
Moderate 8.61 SHALSTAB Low 42.67 28.08321 0.658166 0.10
High 12.28a Moderate 30.59 34.17533 1.116976 0.59
a High 26.73 37.74146 1.411698 0.90
Perfectly agreed area between two models.
138 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139

Fig. 22. Combined landslide susceptibility map.

based on AHP and the SHALSTAB is based on hydro-geotechnical Acknowledgments

properties of soil, a total mismatch of 53.27% resulted for the three
susceptibility classes. The two results obtained from SMCE and This research was supported by the Public Welfare and Safety
SHALSTAB were integrated using the frequency ratio method to im- Research Program through the National Research Foundation of Korea
prove the prediction performance of the susceptibility map. About (NRF), funded by the Ministry of Science, ICT, and Future Planning
82.2% of the training data landslides were located in the high and (grant No. 2012M3A2A1050977), a grant (13SCIPS04) from the Smart
moderate susceptibility classes, and 74.7% of the validation data Civil Infrastructure Research Program funded by the Ministry of Land,
landslides were identified as having the same class as that projected Infrastructure and Transport (MOLIT) of Korea government and Korea
by the model, demonstrating the reliability of the combined map. Agency for Infrastructure Technology Advancement (KAIA) and the
The area under curve (AUC) gave a success accuracy of 79.6%. The Brain Korea 21 Plus (BK 21 Plus).
combined map was validated using the prediction rate curve. The
prediction result was obtained by comparing the landslide validation
data with the susceptibility map, and the AUC was calculated as
83.6%, that indicates the prediction result is good. This study demon-
strated that the combined model produces more accurate results
for predicting landslides in landslide-prone areas with weathered
granite soil slopes than either individual model.

Fig. 23. Area occupied by susceptibility classes and densities of landslide. Fig. 24. Accuracy assessment of combined model.
 A.M.S. Pradhan, Y.T. Kim / Catena 140 (2016) 125–139
References
Abella, E.A.C., van Westen, C.J., 2007. Generation of a landslide risk index map for Cuba
using spatial multi-criteria evaluation. Landslides 4, 311–325.
Akgün, A., Sezer, E.A., Nefeslioglu, H.A., Gokceoglu, C., Pradhan, B., 2012. An easy-to-use
MATLAB program (MamLand) for the assessment of landslide susceptibility using a
mamdani fuzzy algorithm. Comput. Geosci. 38 (1), 23–34.
Armas, I., 2011. An analytic multicriteria hierarchical approach to assess landslide
vulnerability, case study: Cornu village, subcarpathian Prahova Valley/Romania. Z.
Geomorphol. 55, 209–229.
Bae, S., 2007. Cause analysis of 2006 concentrated heavy rain which occurred in Inje-gun.
Korean Assoc. Reg. Geog. 13 (4), 396–408.
Barling, R.D., Moore, I.D., Grayson, R.B., 1994. A quasi-dynamic wetness index for
characterizing the spatial distribution of zones of surface saturation and soil water
content. Water Resour. Res. 30 (4), 1029–1044.
Brabb, E.E., 1984. Innovative approaches to landslide hazard mapping. Proceedings 4th
International Symposium on Landslides, Toronto 1 pp. 307–324.
Bui, D.T., Pradhan, B., Lofman, O., Revhaug, I., Dick, O.B., 2011. Landslide susceptibility
mapping at Hoa Binh province (Vietnam) using an adaptive neuro fuzzy inference
system and GIS. Comput. Geosci. http://dx.doi.org/10.1016/j.cageo.2011.10.031
(article on-line first available).
Cevik, E., Topal, T., 2003. GIS-based landslide susceptibility mapping for a problematic seg-
ment of the natural gas pipeline, Hendek (Turkey). Environ. Geol. 44 (8), 949–962.
Chen, G., Meng, X., Tan, L., Zhang, F., Qiao, L., 2014. Comparison and combination of
different models for optimal landslide susceptibility zonation.quarterly. J. Eng. Geol.
Hydrogeol. 47 (4), 283–306.
Chirico, G.B., Grauspm, R.B., Western, A.W., 2003. On the computation of the
quasi-dynamic wetness index with multiple-flow-direction algorithm. Water
Resour. Res. 39 (5), 1115. http://dx.doi.org/10.1029/2002WR001754.
Choi, J., Oh, H.J., Lee, H.J., Lee, C., Lee, S., 2012. Combining landslide susceptibility maps
obtained from frequency ratio, logistic regression, and artificial neural network
models using ASTER images and GIS. Eng. Geol. 124, 12–23.
Chung, Y.S., Yoon, M.B., Kim, H.S., 2004. On climate variations and changes observed in
South Korea. Clim. Chang. 66 (1–2), 151–161.
Cooke, R.U., Doornkamp, J.C., 1990. Geomorphology in Environmental Management:
A New Introduction. second ed. Claren don Press, Oxford (xxiv +410 pp.).
Crosta, G.B., Dal Negro, P., Frattini, P., 2003. Soil slips and debris flows on terraced slopes.
Nat. Hazards Earth Syst. Sci. 3, 31–42. http://dx.doi.org/10.5194/nhess-3-31-2003.
Dahal, R.K., Hasegawa, S., Bhandary, N.P., Poudel, P.P., Nonomura, Y.R., 2012. A replication
of landslide hazard mapping at catchment scale. Geomatics Nat. Hazard. Risk 3,
161–192. http://dx.doi.org/10.1080/19475705.2011.629007.
Dai, F.C., Lee, C.F., Ngai, Y.Y., 2002. Landslide risk assessment and management: an
overview. Eng. Geol. 64 (1), 65–87.
Dietrich, W.E., Montgomery, D.R., 1998. SHALSTAB: A Digital Terrain Model for Mapping
Shallow Landslide Potential. http://socrates.berkeley.edu/~geomorph/shalstab/.
Dietrich, W.E., Sitar, N., 1997. Geoscience and geotechnical engineering aspects of debris-
flow hazard assessment, invited overview paper. In: Chen, C.-L. (Ed.), Debris-Flow
Hazards Mitigation: Mechanics, Prediction, and AssessmentProceedings of the First
Int. Conf.ASCE, San Francisco, pp. 656–676.
Dietrich, W.E., Reiss, R., Hsu, M., Montgomery, D.R., 1995. A process-based model for
colluvial soil depth and shallow landsliding using digital elevation data. Hydrol.
Process. 9, 383–400.
Einstein, H.H., 1988. Special lecture: landslide risk assessment procedure. Proceedings of
5th Symposium on Landslides, Lausanne vol. 2 pp. 1075–1090 (July 1988).
Ercanoglu, M., Gokceoglu, C., Van Aseh, W., 2004. Landslide susceptibility zoning north of
Yenice (NW Turkey) by multivariate statistical techniques. Nat. Hazards 32, 1–32.
Goetz, J.N., Guthrie, R.H., Brenning, A., 2011. Integrating physical and empirical landslide
susceptibility models using generalized additive models. Geomorphology 129,
376–386. http://dx.doi.org/10.1016/j.geomorph.2011.03.001.
Gomez, H., Kavzoglu, T., 2005. Assessment of shallow landslide susceptibility using
artificial neural networks in Jabonosa River Basin, Venezuela. Eng. Geol. 7, 11–27.
Gupta, R.P., Kanungo, D.P., Arora, M.K., Sarkar, S., 2008. Approaches for comparative
evaluation of raster GIS-based landslide susceptibility zonation maps. Int. J. Appl.
Earth Obs. Geoinf. 10, 330–341.
Hosmer, D., Lemeshow, S., 2000. Applied Logistic Regression. second ed. Wiley,
New York, p. 392.
Kayastha, P., Dhital, M.R., de Smedt, F., 2012. Evaluation of the consistency of landslide
susceptibility mapping: a case study from the Kankai watershed in east Nepal.
Landslides 10 (6), 785–799.
Kim, J., Jeong, S., Park, S., Sharma, J., 2004. Influence of rainfall induced wetting on the
stability of slopes in weathered soils. Eng. Geol. 75 (3–4), 251–262.
Lee, S.G., de Freitas, M.H., 1989. A revision of the description and classification of weathered
granite and its application to granites in Korea. Q. J. Eng. Geol. 22, 31–48.
Lee, S., Pradhan, B., 2006. Probabilistic landslide risk mapping at Penang Island, Malaysia.
J. Earth Syst. Sci. 115 (6), 661–672.
Lee, S., Pradhan, B., 2007. Landslide hazard mapping at Selangor, Malaysia, using frequen-
cy ratio and logistic regression models. Landslides 4 (1), 33–41.
Lee, C., Yoo, N., 2009. A study on debris flow landslide disasters and restoration at Inje in
Kangwon Province, Korea. Korean Soc. Hazard Mitigation 9 (1), 99–105.
Marsigli, C., Montani, A., Nerozzi, F., Paccagnella, T., Tibaldi, S., Molteni, F., Buizza, R., 2001.
A strategy for high resolution ensemble prediction. Part II: Limited area experiments
in four Alpine flood events. Quart. J. Meteorol. Soc. 127, 2095–2116.
McCalpin, J., 1974. Preliminary Age Classification of Landslides for Inventory Mapping,
21st Annual Symposium on Engineering Geology and Soils Engineering. Proceedings.
University of Idaho, Moscow, Idaho, USA, pp. 99–111.
139
Molteni, F., Buizza, R., Palmer, T.N., Petroliagis, T., 1996. The ECMWF ensemble prediction
system: methodology and validation. Q. J. R. Meteorol. Soc. 122 (529), 73–119.
Molteni, F., Buizza, R., Marsigli, C., Montani, A., Nerozzi, F., Paccagnella, T., 2001.
A strategy for high-resolution ensemble prediction. Part I: definition of represen-
tative members and global model experiments. Quart. J. Roy. Meteor. Soc. 127,
2069–2094.
Montgomery, D.R., Dietrich, W.E., 1994. A physically based model for the topographic
control on shallow landsliding. Water Resour. Res. 30, 1153–1171.
Moore, I.D., Wilson, J.P., 1992. Length–slope factors for the Revised Universal Soil Loss
Equation: simplified method of estimation. J. Soil Water Conserv. 47, 423–428.
Moore, I.D., Grayson, R.B., Ladson, A.R., 1991. Digital terrain modeling: a review
of hydrological, geomorphological, and biological applications. Hydrol. Process.
5, 3–30.
Neuhauser, B., Damm, B., Terhorst, B., 2012. GIS-based assessment of landslide suscepti-
bility on the base of the weights-of-evidence model. Landslides 9 (4), 511–528.
Nijmeijer, R., de Haas, A., Dost, R.J.J., Budde, P.E., 2001. Ilwis 3.0 Academic User's Guide.
ITC, Enschede.
O'Loughlin, E.M., 1986. Prediction of surface saturation zones in natural catchments by
topographic analysis. Water Resour. Res. 22 (5), 794–804.
Pachauri, A.K., Gupta, P.V., Chander, R., 1998. Landslide zoning in a part of the Garhwal
Himalayas. Environ. Geol. 36 (3–4), 325–334.
Pack, R., Tarboton, D., 2004. Stability index mapping (SINMAP) applied to the prediction
of shallow translational landsliding. Geophys Res Abstracts 6, 05122.
Pourghasemi, H.R., 2008. Landslide Hazard Assessment Using Fuzzy Logic (Case Study:
A Part of Haraz Watershed) M.Sc. Thesis Tarbiat Modarres University International
Campus, Iran (92 pp.).
Pradhan, A.M.S., Kim, Y.T., 2014. Relative effect method of landslide susceptibility
zonation in weathered granite soil: a case study in Deokjeok-ri Creek, South Korea.
Nat. Hazard 72, 1189–1217.
Pradhan, B., Lee, S., 2009. Landslide risk analysis using an artificial neural network model
focusing on different training sites. Int. J. Phys. Sci. 3 (11), 1–15.
Pradhan, B., Pirasteh, S., 2010. Comparison between prediction capabilities of neural
network and fuzzy logic techniques for landslide susceptibility mapping. Disaster
Adv. 3 (3), 26–34.
Pradhan, A.M.S., Dawadi, A., Kin, Y.T., 2012. Use of different bivariate statistical landslide
susceptibility methods: a case study of Kulekhani watershed, Nepal. J. Nepal Geol.
Soc. 45, 1–12.
Rahardjo, H., Lee, T., Leong, E.C., Rezaur, R.B., 2005. Response of a residual soil slope to
rainfall. Can. Geotech. J. 42 (2), 340–351.
Rossi, M., Guzzetti, F., Reichenbach, P., Mondini, A.C., Peruccacci, S., 2010. Optimal
landslide susceptibility zonation based on multiple forecasts. Geomorphology 114
(3), 129–142.
Saaty, T.L., 1977. A scaling method for priorities in hierarchical structures. J. Math. Psychol.
15, 234–281.
Saaty, T., 1980. The Analytical Hierarchy Process. McGraw-Hill, New York.
Saaty, T.L., 2000. second ed. The Fundamentals of Decision Making and Priority Theory
with the Analytic Hierarchy Process vol VI. RWS Publications, Pittsburg (478 pp.).
SafeLand: Deliverable 2.8, 2011. Recommended Procedures for Validating Landslide
Hazard and Risk Models and Maps.
Sharifi, M.A., Retsios, V., 2003. Site selection for waste disposal through spatial multiple
criteria decision analysis. Proceedings of the 3rd International Conference on Decision
Support for Telecommunications and Information Society DSTIS, 4–6 September
2003, Warsaw, Poland (15 pp. — Includes a case study).
Sidle, R.C., Ochiai, H., 2006. Landslides: Processes, Prediction, and Land Use vol. 18.
American Geophysical Union.
Sidle, R.C., Pearce, A.J., Loughlin, C.L.O., 1985. Hillslope Stability and Land-Use. American
Geophysical Union, Washington, DC, USA (125 pp.).
Soeters, R., Van Westen, C.J., 1996. Slope instability recognition, analysis and zonation. In:
Turner, A.K., Schuster, R.L. (Eds.), Landslides, Investigation and Mitigation,
Transportation Research Board, National Research Council, Special Report 247.
National Academy Press, Washington D.C., U.S.A., pp. 129–177.
Soil Survey Staff, 1993. Soil survey manual. Soil Conservation Service. US Department of
Agriculture Handbook 18 (Archived from the original on 2006-02-14. Retrieved
2006-07-02).
Stocking, M.A., 1972. Relief analysis and soil erosion in Rhodesia using multivariate
techniques. Z. Geomorphol. 16, 432–443.
Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science 240, 1285–1293.
Tangestani, M.H., 2009. A comparative study of Dempster–Shafer and fuzzy models for
landslide susceptibility mapping using a GIS: an experience from Zagros Mountains,
SW Iran. J. Asian Earth Sci. 35 (1), 66–73.
Terlien, M.T.J., Van Westen, C.J., Van Asch, T., 1995. Deterministic modelling in
GIS-based landslide hazard assessment. In: Carrara, A., Guzzetti, F. (Eds.),
Geographical Information Systems in Assessing Natural Hazards. Kluwer,
Dordrecht, pp. 57–77.
van Westen, C.J., Rengers, N., Soeters, R., 2003. Use of geomorphological information in
indirect landslide susceptibility assessment. Nat. Hazards 30, 399–419.
Vivas, L., 1992. Los Andes Venezolanos. Academia Nacional de la Historia, Caracas (300 pp.).
Wieczorek, G.F., 1984. Preparing a detailed landslide-inventory map for hazard evaluation
and reduction. Bull. Assoc. Eng. Geol. 21 (3), 337–342.
Wilcock, P.R., Schmidt, J.C., Wolman, M.G., Dietrich, W.E., Dominick, D.W., Doyle, M.W.,
Gordon, E.G., Iverson, R.M., Montgomery, D.R., Pierson, T.C., Schilling, S.P., Wilson,
R.C., 2003. When models meet managers: examples from geomorphology. In:
Wilcock, P.R., Iverson, R.M. (Eds.), Prediction in Geomorphology, American
Geophysical Union Monogragh Series 135, pp. 27–40.