Analysis of Landslide Susceptibility Using Monte

Carlo Simulation and GIS

Heon-Woo Lee, Hyuck-Jin Park, Ik Woo, and Jeong-Gi Um

Abstract
Since the landslide is one of the repeated geological hazards and causes a terrible loss of life
and properties in Korea, many different researches have been carried out to evaluate the
hazard and the susceptibility of landslide. The physical landslide model has been suggested
to evaluate the factor of safety in previous studies but the deterministic approach has been
utilized. However, applying the deterministic model in regional study area can be difficult
or impossible because of the difficulties in obtaining and processing of large spatial data
sets. With limited site investigation data, uncertainties were inevitably involved with.
Therefore, the probabilistic analysis method such as Monte Carlo simulation has been
utilized in this study. The GIS based infinite slope stability model has been used to evaluate
the probability of failure. The proposed approach has been applied to practical example.
The study area in Pyeongchang-gun, Gangwon-do has been selected since the area has been
experienced tremendous amount of landslide occurrence.

Keywords
Monte Carlo simulation
Probability of failure
Uncertainty

Introduction consequently slope instability is further exacerbated and

Landslides are one of the repeatedly occurring geologic
hazards experienced during the rainy season in Korea.
They cause the loss of about 23 human lives each year,
which accounts for nearly 25 % of the annual death due to
all natural disasters. Regardless of this, the development and
urbanization of mountainous areas is continuing, and
H.-W. Lee (*)
H.-J. Park
Department of Geoinformation Engineering, Sejong University,
98 Gunja-dong, Gwangjin-gu, Seoul, Republic of Korea
e-mail: sglee@uos.ac.kr
I. Woo
Department of Coastal Construction Engineering, Kunsan National
University, Gunsan, Republic of Korea
J.-G. Um
Department of Energy Resource Engineering, Pukyong National
University, Busan, Republic of Korea
the potential for landslides is increased.
Landslide occurrence is controlled by spatial and climatic
factors such as geology, geomorphology, vegetation, and
rainfall, and hence the prediction of, or susceptibility analysis
for, landslides is difficult. This is because an enormous
amount of spatial data must be acquired from the region
and then processed. Consequently, GIS (Geographic Infor-
mation Systems) has been widely and effectively used
in analyzing the spatial information relevant to landslides.
A great advantage of using GIS is that all spatial attributes
related to a particular area can be presented in the form
of a digitized map that can be saved, searched, modified,
and updated. It can also cover the general area, allowing
regional scale geographic and geologic data to be readily
analyzed. This capability has facilitated many studies
on landslide hazard assessment using GIS (Carrara et al.
1991; Gokceoglu and Aksoy 1996; Luzi and Floriana
1996; Guzzetti et al. 1999; Jibson et al. 2000; Lee and Min
2001; Gritzner et al. 2001; Lee and Choi 2003).

C. Margottini et al. (eds.), Landslide Science and Practice, Vol. 1, 371

DOI 10.1007/978-3-642-31325-7_49, # Springer-Verlag Berlin Heidelberg 2013
372 H.-W. Lee et al.

Landslide hazard assessment methods can be divided into

two categories: quantitative and deterministic (or geotechnical).
The quantitative technique ranges from a stability scoring
system based on criteria such as slope, vegetation, parent
material, and geometry, to statistical models that link geo-
logic and geomorphologic attributes based on spatial corre-
lation (Chung et al. 1995; Chung and Fabbri 1998, 1999;
Dhakal et al. 2000). Most GIS techniques used in numerous
previous studies only consider the statistical relationships
between landslide occurrences and factors such as soil
type, land use, slope geometry, vegetation, and other Fig. 1 Infinite slope model
parameters, but not the failure mechanism. In other words,
the relationship between events and the environment has
been investigated without reference to a mechanical analy- As shown in Fig. 1, the groundwater level in an infinite
sis. In contrast, the geotechnical approach analyzes the slope is assumed to be parallel to the surface and to be
mechanical condition of slopes and evaluates their stability located at mz. In addition, the groundwater flow is parallel
through mathematical calculation (Montgomery and to the slip surface. If a slice of infinite slope mass that has
Dietrich 1994; Wu and Sidle 1995; Gorsevski 2002). In unit width is considered, as in Fig. 1, the normal (N) and
this approach, the physical properties of a slope are obtained shear stress (T) at the base of the slope element can be
and a mathematical model is used to evaluate the landslide given as:
susceptibility and hazard potential. The factor of safety (FS)
is calculated in this approach as an index of landslide sus- N ¼ W cos i ¼ ½ð1  mÞgt þ mgsat z cos i (1)
ceptibility. The geotechnical technique was used in this
study in order to overcome the limitation of the quantitative N
technique. However, the input parameters used in geotech- s¼ ¼ ½ð1  mÞgt þ mgsat z cos2 i (2)
sec i
nical technique should be obtained from very broad study
area with limited number of sampling and subsequently, the T ¼ W sin i ¼ ½ð1  mÞgt  mgsat z sin i (3)
uncertainties were involved. But the deterministic analysis
used only mean values for input parameters and conse- T
t¼ ¼ ½ð1  mÞgt þ mgsat z sin i cos i (4)
quently could not properly with uncertainties involved in sec i
input parameters Therefore, the probabilistic analysis was
proposed in this study to control the unceratinties effectively. From Fig. 1, the pore water pressure is:

u ¼ mz gw cos2 i (5)
Probabilistic Analysis of Landslide Hazard
Then the shear strength along the failure plane is:
Physical Landslide Model
S ¼ c0 þ ðs  uÞ tan f (6)
In order to utilize the geotechnical failure model and apply
the mechanical properties of soil to susceptibility analysis The factor of safety on the sliding surface is:
the infinite slope model is used. In this, the slope is assumed
to extend infinitely in its dip direction, and sliding is consid- s c0 þ ðs  uÞ tan f0
ered to occur along a plane parallel to the face of the slope. FS ¼ ¼ (7)
t t
This is the most usual approach in landslide analysis (Kamai
1991; Terlin 1996; Pack et al. 1998, 2001). When the lengths where i is the slope angle, gt is the total unit weight of soil, gw
of the landslide slip surfaces are large enough compared is the unit weight of water, gsat is the unit weight of saturated
with their depths, the infinite slope model may be accurately soil, z is the soil depth from the ground surface, m is the ratio
and successfully adapted. This model is appropriate for of the groundwater level to the soil depth, c0 is the effective
Korean landslides where the sliding surfaces are commonly cohesion, and ’0 is the effective friction angle.
located at shallow depth (NIDP 2000).
Analysis of Landslide Susceptibility Using Monte Carlo Simulation and GIS 373

Probabilistic Approach Table 1 Distribution of the slope failure probability for various
groundwater levels when COV is 20 %
The strength parameters of soil, such as cohesion and fric- m (%)
tion angle are prerequisite to evaluate the factor of safety Probability of failure (%) 0.0 0.5 0.75 0.9
from the infinite slope model. However, true values of the 01 67.0 51.5 43.8 39.5
cohesion and friction angle cannot be obtained from field 15 14.1 13.0 11.4 8.0
investigation or laboratory test due to the extensive size of 5  10 4.2 6.4 5.9 4.3
10  20 3.9 6.9 7.4 6.2
the study area and the limitation of the sampling numbers.
20  30 2.0 4.1 4.9 5.0
Therefore, the uncertainties are involved in the procedure of
30  50 2.5 5.1 6.9 8.0
determination for cohesion and friction angle. Consequently
50  70 1.7 3.4 4.8 6.8
the cohesion and friction angle were considered as random
70  90 1.8 3.3 4.8 6.4
variables in this study and subsequently the probabilistic 90  100 2.8 6.6 10.1 15.8
approaches were implemeneted.
Monte Carlo simulation approach was used as the prob-
abilistic approaches in this study to analyze the probabilis-
tic properties of random variables and evaluate the Landslide Location Detection and Spatial
probability of slope failure. The Monte Carlo simulation Database Construction
is known as the most complete probabilistic analysis since
all the random variables are represented by their probability The accurate detection and location of landslides is impor-
density function and the probability of failure as the result tant for their susceptibility analysis. A field survey is widely
of reliability analysis is represented by the probability considered the most accurate detection method in the land-
density function. The advantages of Monte Carlo simu- slide inventory process. However, using a field survey as an
lation are that it is relatively easy to implement on a incipient step in the data collection process is difficult, time
computer and it can deal with a wide range of functions, consuming, and costly, especially in mountainous areas
including those that cannot be expressed conveniently in an where access is limited or even impossible. Remote sensing
explicit form (Baecher and Christian 2003). Therefore, in is therefore employed to overcome the lack of field data.
this study, Monte Carlo simulation was used in the evalua- For this work, aerial photographs taken before and after
tion of the probability of slope failure. In the simulation landslides were used to detect event locations. Field surveys
procedure, random numbers between zero and one were were used to verify landslide locations that were indicated
generated from a uniform distribution. Then, using the by a study of aerial photographs. The slides were determined
cumulative distribution of the probability density function by a comparison of the sequential images then verified by
for each variable, a unique value for each variable was fieldwork. A photo database was constructed through ortho-
selected randomly. The group of randomly selected rectification and by merging many images. Recent events
parameters was combined with the fixed input data to were seen as a break in the forest canopy, bare soil, or other
generate a single random value for the factor of safety. geomorphologic characteristics typical of landslide scars;
This process was repeated many times to generate a suffi- for example, head and side scarps, flow tracks, and soil and
cient number of different factors of safety. debris deposited below the scar. Approximately 40 mono-
chromatic aerial photographs were used to map recent
landslides and to assemble a database to assess their surface
Study Area area and number within the study region. A total of 483
landslides were mapped.
The Pyeongchang-gun area of Korea was selected as a suit- In order to analyze geomorphologic attribute such as
able test site for the proposed approach. On July 2006, this slope angle, which is indispensable parameter in the geo-
locality experienced a large number of landslides following technical technique, digitized information is required. To
heavy rainfall of 227 mm in 1 day. The bedrock of the study achieve this, contour and survey base points with an eleva-
area consists mainly of biotite granite. Aerial photographs tion value read from the 1:5,000 scale topographic maps
and 1:5,000 scale digital topographic maps were examined were extracted, and the digital elevation model (DEM) was
to locate the landslides. In addition, 1:5,000 scale topo- made with a 5 m resolution. Using the DEM, the slope angle
graphic maps, 1:25,000 scale soil maps, and 1:50,000 scale was calculated.
geologic maps were examined. In order to accurately evalu- The 1:25,000 scale soil maps were used to acquire the
ate landslide susceptibility, the landslide activity was located characteristics of the soil, such as the topographic map, the
both by using the aerial photographs and digital topographic soil texture map, and the effective soil depth. The topo-
maps and by conducting a field survey. graphic map was based on the geomorphology of soil
374 H.-W. Lee et al.

Fig. 2 Failure probability distribution when COV is 0.1. (latitude 37 360 12.1600 N, Longitude 128 310 53.8100 E)

Fig. 3 Failure probability distribution when COV is 0.2. (latitude 37 360 12.1600 N, Longitude 128 310 53.8100 E)
Analysis of Landslide Susceptibility Using Monte Carlo Simulation and GIS
Fig. 4 Failure probability distribution when COV is 0.3. (latitude 37 360 12.1600 N, Longitude 128 310 53.8100 E)
Table 2 Percentage of landslide cells in hazard class
m (%) 0.0 (%) 0.5 (%) 0.75 (%)
COV ¼ 10 17.1 34.0 46.8
COV ¼ 20 27.5 49.2 60.2
COV ¼ 30 48.4 63.9 68.9
distribution, and the soil texture map was based on the soil
size distribution. The effective thickness was evaluated from
the depth to bedrock. In addition, the geologic map was used
to obtain the contributing lithology.
Analysis of Random Properties for Input
Parameters and Monte Carlo Simulation
In the geotechnical model, the mechanical input parameters
such as the shear strength of the soil, soil thickness, and the
groundwater level are the requisite parameters. However,
certain input factors such as cohesion and the friction angle
of soil can be obtained only by laboratory testing and it is
logistically difficult to acquire this information across the
wide area of soils in the region. Consequently, the represen-
tative values of unit weight, cohesion, and internal friction
angle for each soil type were determined from the research
results of the Korea Highway Corporation (2009). In this
375
procedure, the uncertainties were involved in cohesion and
0.9 (%) friction angle and therefore, they were considered as random
59.2 variables. Then the probabilistic analysis approach was car-
67.5 ried out. In order to carry out the probabilistic analysis, the
72.3 probabilistic properties (that is, mean, standard deviation,
and probability density function) of random variables should
be determined. However, due to limited amount of data
which can be obtained, they were estimated on the basis of
judgment, experience, or results published by others (Hoek
2007). From the previous studies (Mostyn and Li 1993;
Nilsen 2000; Pathak and Nilsen 2004; Park et al. 2005) the
cohesion and friction angle were found to be normally
distributed. The mean values of strength parameters for
each soil type were estimated from the results of the previous
research. However, the standard deviation of each random
variable could not be obtained. Therefore, coefficients of
variation (COV) for cohesion and friction angle were used.
The coefficients of variation (COV) of cohesion and friction
angle are known as 10 % in many previous researches
(Schulz 1975; Cherubini 2000). However, since the cohesion
and friction angle used in this study were estimated from the
previous research, the uncertainties could affect the strength
parameters more seriously. Therefore, the variances of the
strength parameters could be larger than expected, so the
larger COV (that is, 20 % and 30 %) was used to analyse in
376
Fig. 5 Factor of safety distribution when m ¼ 0.9. (latitude 37 360 12.1600 N, Longitude 128 310 53.8100 E)
Table 3 Results of the deterministic analysis
0.0 0.5 0.75
m (%) (%) (%) (%)
% of landslide cells in hazard 10.2 27.0 40.3
class
this study. That is, COV whose range were varied from 10 %
to 30 %, were used in the probabilistic analysis in order to
evaluate the effect of the uncertainties.
Using the probabilistic properties of random variables
and other input parameters, the Monte Carlo simulation
procedure was implemented in MATLAB software. The
Monte Carlo simulation includes the following steps.
1. Generate the random variables such as cohesion and
friction angle using random number generator and cumu-
lative distribution function for cohesion and friction
angle.
2. Calculate the factor of safety for a pixel using the
generated random variables (cohesion and friction
angle). After N repeated calculation, the probability dis-
tribution curve for the factor of safety is obtained. Then
the probability of failure for the cell is evaluated.
H.-W. Lee et al.
3. Repeated the above calculation for all other pixels and
0.9 produce the slope failure probability map.
(%) In addition, in order to evaluate the effect of the ground-
57.4 water to the stability of slope, the groundwater level (m in
infinite slope model) was varied from 0 % to 90 % in the
probabilistic analysis.
Evaluation of Slope Failure Probability
Table 1 shows the distributions of the probability of slope
failure for the different groundwater level. As groundwater
level rises due to rainfall, the area with high failure proba-
bility is increased and the area with low failure probability is
decreased. This means the study area is becoming unstable,
as expected. Figures 2, 3 and 4 show the calculated proba-
bility of slope failure at 90 % groundwater level for the cases
of 10 %, 20 % and 30 % COV, respectively.
In order to verify the accuracy of the probabilistic analy-
sis, the hazard class (or unstable cell) was determined.
According to previous research (Priest and Brown 1983),
10 % of failure probability is evaluated as unstable.
Analysis of Landslide Susceptibility Using Monte Carlo Simulation and GIS
Therefore, the cell whose probability of failure is larger than
10 % is designated as unstable.
Table 2 shows the percentage of landslide cells in hazard
class. As can be observed in Table 2, the percentage of
landslide cells in hazard class is increased with COV.
When the groundwater level is 90 %, the accuracy (per-
centage of landslide cells in hazard class) is 59.2 %, 67.5 %
and 72.3 % at 10 % COV, 20 % COV and 30 % COV,
respectively. That is, as COV is increased, the accuracy of
the analysis is increased.
In order to compare the results of the probabilistic analysis
with the deterministic analysis, the factor of safety was
evaluated using same input parameters used in the probabilis-
tic analysis. The result shows in Fig. 5 and Table 3. The
percentage of landslide cells in hazard class in the determin-
istic analysis is 57.4 % at 90 % groundwater level. But this
value is lower than the results of the probabilistic analysis,
which show 59.2 %, 67.5 % and 72.3 % at 10 % COV, 20 %
COV and 30 % COV, respectively. That is, the accuracy of the
deterministic analysis is lower than the one of the probabilistic
analysis. This is because the deterministic analysis used the
single fixed mean values for cohesion and friction angle and,
subsequently the variance (or uncertainties) of the data was not
considered in the analysis. Consequently the probabilistic
analysis which effectively deals with the uncertainties shows
better outcome in the prediction of landslide hazard.
Results
Many different researches have been used to evaluate the
susceptibility of landslide since the landslide is one of the
repeated geohazards. The previous studies were mainly
focused on the statistical relationship between causative
factors and landslide occurrence. However, those approaches
were not considered failure mechanisms and engineering
properties of geological materials. Therefore in this study
the physical landslide model has been suggested and the
geotechnical input parameters were evaluated. However,
uncertainties were inevitably involved with geotechnical
input parameters since large spatial data set for input
parameters are difficult to obtain from large study area.
Therefore, Monte Carlo simulation, one of the most com-
monly used probabilistic analysis is used probabilistic anal-
ysis, is used. For this, cohesion and friction angle were
considered as random variable and their mean values were
obtained from soil map and previous researches. In order to
evaluate the effect of the groundwater, the groundwater
levels were varied from 0 % to 90 %. Furthermore, COV
values in the uncertain parameters were varied from 10 % to
30 % to evaluate the uncertainties. The results showed that
the probabilistic analysis method showed more accurate than
the deterministic analysis.
377
Acknowledgments This work was supported by the National
Research Foundation of Korea (NRF) grant funded by the Korea
government (MEST) (No. 2010–0021314) and the project to educate
GIS experts.
References
Baecher GB, Christian JT (2003) Reliability and statistics in geotech-
nical engineering. Wiley, New Jersey, 605
Carrara A, Cardinali M, Detti R, Guzzetti F, Pasqui V, Reichenbach P
(1991) GIS techniques and statistical models in evaluating landslide
hazard. Earth Surf Proc Land 20:427–445
Cherubini C (2000) Relibability evaluation of shallow foundation
bearing capacity on c, phi soils. Can Geotech J 37:264–269
Chung CF, Fabbri AG (1998) Three Bayesian prediction models for
landslide hazard. In: Proceedings of the international association for
mathematical geology, Italy, pp 204–211
Chung CF, Fabbri AG (1999) Probabilistic prediction models for land-
slide hazard mapping. Photogramm Eng Rem Sens 65:1389–1399
Chung CF, Fabbri AG, van Westen CJ (1995) Multivariate regression
analysis for landslide hazard zonations. In: Carrara A, Guzzetti F
(eds) Geographic information system in assessing natural hazards.
Kluwer, Dordrecht, pp 107–133
Dhakal AS, Amada T, Aniya M (2000) Landslide hazard mapping and
its evaluation using GIS: an investigation of sampling schemes for a
grid cell-based quantitative method. Photogramm Eng Rem Sens
66:981–989
Gokceoglu C, Aksoy H (1996) Landslide susceptibility mapping of the
slopes in the residual soils of the Mengen region by deterministic
stability and image processing techniques. Eng Geol 44:147–161
Gorsevski PV (2002) Landslide hazard modeling using GIS. Ph.D.
thesis, University of Idaho, 240p
Gritzner ML, Marcus WA, Aspinall R, Custer SG (2001) Assessing
landslide potential using GIS, soil wetness modeling and topo-
graphic attributes. Geomorphology 37:149–165
Guzzetti F, Carrara A, Cardinali M, Reichenbach P (1999) Landslide
hazard evaluation; a review of current techniques and their applica-
tion in a multi-scale study, Central Italy. Geomorphology
31:181–216
Hoek E (2007) Rock engineering; Course note by Evert Hoek [Online].
http://www.rockeng.utoronto.ca/Hoekcorner.htm
Jibson RW, Harp LW, Michael JA (2000) A method for producing
digital probabilistic seismic landslide hazard maps. Eng Geol
58:271–289
Kamai T (1991) Slope stability assessment by using GIS. Science and
Technology Agency of Japan, in Japanese
Korea Highway Corporation (2009) Road design cuideline, p 472
Lee S, Choi U (2003) Development of GIS-based geological hazard
information system and its application for landslide analysis in
Korea. Geosci J 7:243–252
Lee S, Min K (2001) Statistical analysis of landslide susceptibility at
Yongin, Korea. Environ Geol 40:1095–1113
Luzi I, Floriana P (1996) Application of statistical and GIS techniques
to slope instability zonation. Soil Dyn Earthq Eng 15:83–94
Montgomery DR, Dietrich WE (1994) A physically based model for the
topographic control on shallow landsliding. Water Resour Res
30:1153–1171
Mostyn GR, Li KS (1993) Probabilistic slope analysis – state of play.
In: Proceeding of conference on probabilistic method in geotechnical
engineering, pp 89–109
NIDP (National Institute for Disaster Prevention) (2000) Fundamental
issues for landslide hazard avoidance or mitigation plans. Research
report, p 276 in Korean
378
Nilsen B (2000) New trend in rock slope stability analysis. Bull Eng
Geol Environ 58:173–178
Pack TT, Tarboton DG, Goodwin CN (1998) The SINMAP approach to
terrain stability mapping. In: Proceedings of the 8th congress of the
international association of engineering geology, Vancouver,
pp 21–25
Pack TT, Tarboton DG, Goodwin CN, (2001) Assessing terrain stability
in a GIS using SINMAP. In: Proceedings of the 15th annual GIS
conference, GIS 2001
Park HJ, West TR, Woo I (2005) Probabilistic analysis of rock slope
stability and random properties of discontinuity parameters, Inter-
state Highway 40. Eng Geol 79:230–250
H.-W. Lee et al.
Pathak D, Nilsen B (2004) Probabilistic rock slope stability analysis for
himalayan condition. Bull Eng Geol Environ 63:25–32
Priest SD, Brown ET (1983) Probabilistic rock slope stability of
variable rock slopes. Trans Inst Min Metall, 92p
Schultz E (1975) The general significance of statistics for civil engineers.
In: Proceedings of 2nd international conference on application of
statistics and probability in soil and structural engineering, Aachen
Terlin MTJ (1996) Modeling spatial and temporal variations in rainfall
triggered landslide. International Institute for Aerospace Survey and
Earth Science, Publication no. 32
Wu W, Sidle RC (1995) A distributed slope stability model for steep
forested basins. Water Resour Res 31:2097–2110