Content uploaded by Chong Xu
Author content
All content in this area was uploaded by Chong Xu on Jul 19, 2015
Content may be subject to copyright.
217
Global View of Engineering Geology and the Environment – Wu & Qi (eds)
© 2013 Taylor & Francis Group, London, ISBN 978-1-138-00078-0
Earthquake-triggered landslide susceptibility mapping in the 2010 Yushu,
China earthquake struck area using artificial neural network model
Chong Xu
Key Laboratory of Active Tectonics and Volcano, Institute of Geology, China Earthquake Administration, China
ABSTRACT: The main purpose of this paper is to carry out earthquake-triggered landslide susceptibility
mapping in the 2010 Yushu earthquake struck area using optical remote sensing data, Geographical
Information System (GIS) technologies, and Artificial Neural Network (ANN) model. A total of 12
earthquake triggered landslides associated controlling parameters that were selected, including, elevation,
slope angle, slope aspect, slope curvature, slope position, distance from drainages, lithology, distance from
roads, Normalized Difference Vegetation Index (NDVI), co-seismic main surface fault-ruptures, Peak
Ground Acceleration (PGA) and distance from roads. Landslide susceptibility map was generated using
the training dataset, the 12 controlling parameters and an advanced ANN model. Weight of each param-
eter was determined by the back-propagation training method. Then the earthquake-triggered landslide
susceptibility indices were calculated using the trained back-propagation weights. The landslide suscepti-
bility map was compared with training dataset and testing dataset for obtaining success rate and predic-
tion rate of the model, respectively. The validation results show 81.274% of success rate and 79.915% of
prediction rate of satisfactory agreement between the constructing susceptibility map and the existing
landslides data.
Keywords: the 2010 Yushu earthquake, landslides, susceptibility, artificial neural network, GIS, remote
sensing
Vector Machine (SVM) (Brenning 2005; Yilmaz
2010; Xu & Xu 2012a; Xu et al. 2012a, 2012b).
Among these methods, for landslide susceptibility
mapping, ANN model is a relatively new approach
and has good performances in different countries
and study areas.
In this paper, inventory of landslides triggered
by the 2010 Yushu earthquake were prepared by
interpretation of aerial photographs and satel-
lite images in high resolution, and selecting field
surveys. Furthermore, 12 controlling parameters,
associated with earthquake triggered landslides,
were constructed using ArcGIS software for data
spatial analysis and manipulation. In addition,
ENVI software was used for the Artificial Neural
Network (ANN) model training. The validation
result shows a success rate of 81.274% and a pre-
diction rate of 79.915%.
2 STUDY AREA AND EARTHQUAKE
TRIGGERED LANDSLIDES
The study area is a rectangle area located in
Yushu County, Qinghai Province, China, approxi-
mately between 96°20′32.9″E and 97°10′8.9″E,
and 32°52′6.7″N and 33°19′47.9″N (Fig. 1).
1 INTRODUCTION
Large earthquake often triggers large number of
landslides that may cause tragic loss of life and eco-
nomic devastation (Keefer 1984; Xu et al. 2012a,
2013a). Earthquake triggered landslide suscepti-
bility mapping is very important because it could
provide planners with foreknowledge of dangerous
regions thereby helping with safe planning, disas-
ter management, and hazard mitigation.
There have been many publications about land-
slide susceptibility mapping using GIS technology
and remote sensing data, for example, many land-
slide susceptibility mapping studies were summa-
rized by Guzzetti et al. (1999), Brenning (2005),
and Chacon et al. (2006) etc. A number of differ-
ent models have been applied for landslide sus-
ceptibility mapping, for example, heuristic models
(Pandey et al. 2008; Wachal & Hudak 2000), con-
ditional probabilistic models (Ayalew & Yamagishi
2005; Saha et al. 2005; Dahal et al. 2008; Yilmaz &
Keskin 2009; Pradhan & Lee 2010; Pareek et al.
2010; Park 2010; Xu et al. 2012c, 2012d, 2012e,
2013b; Xu & Xu 2013), and soft computing tech-
niques, such as Artificial Neural Network (ANN)
(Sezer et al. 2011; Arora et al. 2004; Pradhan & Lee
2010; Yilmaz 2010; Xu et al. 2012b) and Support
218
On April 14, 2010 at 07:49 (Beijing time), a
catastrophic earthquake (Ms 7.1) struck the Yushu
county, Qinghai province, China. The region expe-
rienced a strong shaking during the earthquake.
Landslides triggered by the earthquake were inter-
preted from high resolution aerial photographs
and multi-source satellite imageries pre- and
post-earthquake, verified by selected field checking.
A total of 2036 landslides were interpreted (Xu
et al. 2012f, 2013c) in the study area (Fig. 1).
Both training and testing data are needed in
landslide susceptibility mapping using Artificial
Neural Network (ANN) model. There is no exact
mathematical rule for selecting the training and
testing data. In this study, the 2036 landslides were
randomly partitioned into two subsets: a training
dataset, which contains 80% (1628 landslides), was
used for training the model; and a testing dataset
20% (408 landslides) was used for the purpose of
model testing. Figure 1 also shows the distribution
map of training and testing landslides. In addition,
it is necessary to obtain satisfactory sample data
representing absence of landslide occurrences to
fit the ANN model. As there 1628 landslide points
representing the presence of landslide occurrence,
a same number of 1628 points are sampled ran-
domly from the region of stable slopes in the study
area during the earthquake. Therefore, the training
dataset contains 1628 positive samples and 1628
negative samples, a total of 3256 samples.
3 ARTIFICIAL NEURAL NETWORK
MODEL
ANN has the ability to handle imprecise and fuzzy
data they can work with continuous, categorical
and binary data without violating any assump-
tions. As assessment of probability for landslide
occurrence is performed through the forecast of
future events from experience of past landslides, it
may be considered an ideal application for ANN
models (Yilmaz 2010; Xu et al. 2012b).
The Back-Propagation (BP) training algorithm
is the most frequently used neural networks model,
and it is an instructive training model (Yilmaz 2010;
Pradhan & Lee 2010; Xu et al. 2012b). It is accepted
that the most useful neural networks in prediction
and decision algorithm are BP and Radial Basis
Function (RBF) networks. In this paper, BP algo-
rithm is used. The BP training algorithm is trained
by using a set of samples of associated input and
output values. The purpose of artificial neural net-
works is to build a model of the data-generating
process, so that the networks can generalize and pre-
dict outputs from inputs that it has not previously
seen (Pradhan & Lee 2010; Xu et al. 2012b). BP
algorithm created by generalizing the Widrow-Hoff
learning rule to multiple layer networks and nonlin-
ear differentiable transfer function is used (Yilmaz
2010; Xu et al. 2012b). This learning algorithm is
a multi-layered neural network, which consists of
an input layer, several hidden layers and an out-
put layer (Fig. 2). All of those layers may contain
multiple nodes (Yilmaz 2010; Xu et al. 2012b). The
hidden and input layer neurons process their inputs
by multiplying each input by a corresponding
weight, summing the product, and then processing
the sum using a nonlinear transfer function to pro-
duce a result. An artificial neural network “learns”
by adjusting the weights between the neurons in
response to the errors between the actual output
values and the target output values. At the end of
this training phase, the neural network provides a
model that should be able to predict a target value
from a given input value (Pradhan & Lee 2010;
Xu et al. 2012b).
There are two stages involved in using neural
networks for multi-source classification: (a) the
training stage, in which the internal weights are
adjusted; and (b) the classifying stage. Typically,
the BP algorithm trains the networks until some
targeted minimal error is achieved between the
desired and actual values of the networks. Once the
training stage is complete, the network is used as a
Figure 1. Landslides triggered by the Yushu earthquake,
training and testing data for ANN modeling.
Figure 2. Three tiered architecture of neural network.
219
feed-forward structure to produce a classification
for the entire data (Xu et al. 2011).
A neural network consists of a number of inter-
connected nodes. Each node is a simple processing
element that responds to the weighted inputs, it
receives from other nodes. The arrangement of the
nodes is referred to as the networks architecture
(Xu et al. 2012b).
The networks used in this study consisted of
three layers (Fig. 2). The first layer is the input
layer, where the nodes were the elements of a fea-
ture vector. The second layer is the internal or “hid-
den” layer. The third layer is the output layer that
presents the output data. Each node in the hidden
layer is interconnected to nodes in both the preced-
ing and following layers by weight connections.
4 DATA ANALYSIS
A number of 12 thematic maps of landslide con-
trolling parameters were constructed, including
topographic factors such as elevation, slope angle,
slope aspect, slope curvature, topographic position,
and distance from drainages; geologic factors such
as lithology and distance from all faults; seismic
factors such as Peak Ground Acceleration (PGA)
and distance from main surface ruptures; vegeta-
tion factor such as NDVI map; human activity fac-
tor such distance from roads.
A Digital Elevation Model (DEM) was first
constructed using the 1:50000scale topographic
maps. Contour lines, spot heights, and drainages
lines were digitized using ArcGIS software. The
DEM involved 15358 rows and 10493 columns, i.e.
58212015 cells (5 m × 5 m) enclosing a rectangle
of NWW-SEE direction, about 1455.3 km2. The
slope angle, slope aspect, slope curvature, and top-
ographic position (Weiss 2006) layers were calcu-
lated using the 5 m × 5 m DEM. The drainages and
roads maps were extracted from the 1:50000 scale
topographic maps in vector format respectively.
Then the drainage buffer and road buffer were
both calculated in 100 m intervals. The drainage
and road buffer maps were converted into a raster
grid with 5 m × 5 m cells for application of the
artificial neural network model.
The geological map in a scale of 1:200000 from
China Geological Survey provides lithology and
faults information of the study area. The distance
from faults map was calculated in 200 m intervals.
The lithology map and faults buffer map were
converted into raster format with 5 m × 5 m reso-
lution for the following analysis. Table 1 shows
geologic units and their descriptions in the study
area.
The Normalized Difference Vegetation Index
(NDVI) map was obtained from ETM+ satellite
images. The NDVI value was calculated by using
the common formula NDVI = (IR − R)/(IR + R).
The NDVI value denotes areas of vegetation in
an image. The presence of dense green vegetation
implies high NDVI values. Sparse vegetation, on
the other hand, implies low NDVI values.
There is a strong correlation between distance
from co-seismic surface fault-ruptures and earth-
quake-triggered landslides (Xu & Xu 2012b). The
surface fault-ruptures buffer was calculated in
500 m intervals. Then the surface fault-ruptures
buffer map was converted into a raster grid with
5 m × 5 m cells for consequent used. USGS (US
Geological Survey, 2010) derived Peak Ground
Acceleration (PGA) map from ground motion
amplitudes recorded on seismic sensors. A regional
contour map of PGA from USGS indicated a range
of PGA values of 0.12 to 0.38 g in the study area.
The PGA map in vector format was converted into
raster format in 5 m × 5 m resolution.
Classifications of the 12 landslide associated
controlling parameters used in this study are shown
in Table 2. Frequency ratio was used for calculating
the rating of the relative importance of each factor
class to landslide occurrence (Fig. 3). Equation 1
was used for frequency ration calculation:
FR = (N-lsi/N-lstotal)/(N-gridi/N-gridtotal) (1)
where FR is the frequency ratio of certain class
in certain factor; N-lsi represents the landslide
number of the class; N-lstotal represents the
total landslides number in the study area (2036
Table 1. Geologic units and their descriptions in the
study area.
No.
Geologic
unit Description of lithology
1. Q4hLake silt, peat deposits
2. Q4al-pl Alluvium, fluvial deposits, gravel
3. N Quartz sandstone, breccia
4. T3bt Limestone, quartz sandstone,
siltstone, coarse sandstone, slate
5. T3kn3Quartz sandstone, siltstone, slate,
mica schist
6. T3kn2Limestone, amphibolite schist,
marble, metamorphic
conglomerate
7. T3kn1Feldspathic sandstone, siltstone,
slate, limestone, phyllite
8. T2jl2Limestone
9. T2jl1Slate, limestone, siltstone, quartz
sandstone, conglomerate
10. C-P Quartz sandstone, sandstone
11. Magmatic
rocks
Quartz diorite, granite, diabase
220
landslides); N-gridi represents the grid number
of the class; N-gridtotal represents the total grid
number of the study area (58212015 grid cells).
The frequency ratio value could assess the effect
of certain class on earthquake triggered landslide
occurrence. Take slope angle as example, the cor-
relation between slope angle and frequency ratio
value is shown in Figure 3B. It clearly shows that
the higher slope angle value, the higher frequency
ratio value, in other words, the higher landslide
susceptibility probability.
5 LANDSLIDE SUSCEPTIBILITY
ANALYSIS USING THE ANN MODEL
ENVI 4.7 software was used for training and test-
ing the neural networks. A three-layer feed-for-
ward networks that consists of an input layer, one
hidden layer, and one output layer was used as a
networks structure. A total of 3256 grid cells (1628
landslides and 1628 non-landslides) were selected
as training sites, and factors were then adjusted as
below (Xu et al. 2012b):
Activation function: Logistic
Training threshold contribution: 0.9 (default)
Training rate: 0.2 (default)
Training Momentum factors: 0.9 (default)
Training root square error (RMS) exit criteria:
0.1 (default)
Table 2. Impact factors of earthquake-triggered landslide and their classes.
Factors Classes of factors
Elevation/m 1: <3800; 2: 3800–4000; 3: 4000–4200; 4: 4200–4400; 5: 4400–4600; 6: 4600–4800;
7: 4800–5000; 8: >5000
Slope angle/° 1: 0–5; 2: 5–10; 3: 10–15; 4: 15–20; 5: 20–25; 6: 25–30; 7: 30–35; 8: 35–40; 9: >40
Slope aspect 1: Flat; 2: N; 3: NE; 4: E; 5: SE; 6: S; 7: SW; 8: W; 9: NW
Slope curvature/m−11: <−1; 2: −1∼−0.1; 3: −0.1∼−0.05; 4: −0.05∼−0.01; 5: −0.01∼−0.005; 6: −0.005∼−0;
7: 0∼0.005; 8: 0.005∼0.01; 9: 0.01∼0.05; 10: 0.05∼0.1; 11: 0.1∼1; 12: >1
Topographic position 1: Ridge; 2: Upper slope; 3: Middle slope; 4: Flat slope; 5: Lower slope; 6: Valley
Distance from surface rupture/m 1: 0–500; 2: 500–1000; 3: 1000–1500; 4: 1500–2000; 5: 2000–2500; 6: 2500–3000;
7: 3000–3500; 8: 3500–4000; 9: 4000–4500; 10: 4500–5000; 11: 5000–5500;
12: 5500–6000; 13: 6000–6500; 14: 6500–7000; 15: 7000–7500; 16: 7500–8000;
17: 8000–8500; 18: 8500–9000; 19: 9000–9500; 20: 9500–10000; 21: >10000
PGA/g 1: >0.38; 2: 0.34–0.38; 3: 0.30–0.34; 4: 0.26–0.30; 5: 0.22–0.26; 6: 0.18–0.22;
7: 0.14–0.18; 8: 0.14
Distance from roads/m 1: <100; 2: 100–200; 3: 200–300; 4: 300–400; 5: 400–500; 6: 500–600; 7: 600–700;
8: 700–800; 9: 800–900; 10: 900–1000; 11: >1000
NDVI 1: <0; 2: 0–0.05; 3: 0.05–0.1; 4: 0.1–0.15; 5: 0.15–0.2; 6: 0.2–0.25; 7: 0.25–0.3;
8: 0.3–0.35; 9: 0.35–0.4; 10: 0.4–0.45; 11: 0.45–0.5; 12: 0.5–0.55; 13: 0.55–0.6;
14: >0.6
Distance from drainages/m 1: <100; 2: 100–200; 3: 200–300; 4: 300–400; 5: 400–500; 6: 500–600; 7: 600–700;
8: 700–800; 9: 800–900; 10: 900–1000; 11: >1000
Lithology See Table 1
Distance from all faults/m 1: <200; 2: 200–400; 3: 400–600; 4: 600–800; 5: 800–1000; 6: 1000–1200;
7: 1200–1400; 8: 1400–1600; 9: 1600–1800; 10: 1800–2000; 11: >2000
Figure 3. Frequency ratio of factors relate to respective
training and testing datasets. (A): Elevation; (B): Slope
angle; (C): Slope aspect; (D): Slope curvature; (E): Topo-
graphic position; (F): Distance from main surface rup-
tures; (G): PGA; (H): Distance from roads; (I): NDVI;
(J): Distance from drainages; (K): Lithology; (L) Dis-
tance from all faults.
221
Number of hidden layers: 1 (default)
Number of training iterations: 1000 (default).
After the networks goal was reached, the whole
study area was fed into the networks in order to
estimate the landslide susceptibility probability
value. The set of susceptibility values obtained
in each grid were then converted into raster file,
and landslide susceptibility probability value map
completed. The landslide susceptibility probabil-
ity value ranges from 0 to 1 with 0 indicating no
chance of a landslides and 1 certainty. The result
from the ANN model shows that the landslide
susceptibility probability values in the study area
range from 0.0005 to 0.9989.
6 VALIDATION OF THE MODEL
The training and testing datasets were used to
evaluate the validity of the landslide susceptibility
evaluation result as success rate and prediction rate
respectively. To obtain the success and prediction
rate curves for landslide susceptibility probability
map, the calculated landslide susceptibility prob-
ability values of all cells were sorted in descending
order. Then the order cell values were categorized
into 100 classes using “quantile” method in Arc-
GIS, with 1% cumulative intervals, and classified
landslide susceptibility index maps were also pre-
pared with the slicing operation in GIS software of
ArcGIS. This map was crossed with the landslide
training and testing datasets respectively. Then
the success rate curve and the predictive accuracy
curve were created from the cross table values. The
rate curves were created and their areas under the
curve explain how well the model and controlling
factors predict the landslide. Therefore, the area
under the curve can assess the model validation
qualitatively. The rate verification results appear as
two lines (Fig. 4).
The validation result showed satisfactory
agreement between the susceptibility map and the
existing landslides distribution data. These curves
are measured of goodness of fit. As shown in
Figure 4, a total area equal to one denotes perfect
prediction accuracy; whereas an area equals to or
less than 0.5 showed that the model was invalid. In
this study, areas under the curve of success rate and
prediction rate are respective 0.81274 and 0.79915,
and meaning that the success and prediction rate
are 81.274% and 79.915% respectively, and thus
the model was valid.
As already mentioned, landslide susceptibil-
ity probability values in the study area are ranged
from 0.0005 to 0.9989. For providing classified
landslides susceptibility map, four breakpoints
such as 0.2, 0.4, 0.6, and 0.8 are selected to divide
the landslide susceptibility probability value into
five ranges, including, 0.0005 to 0.2, 0.2 to 0.4, 0.4
to 0.6, 0.6 to 0.8, and 0.8 to 0.9989. These classes
represent very low, low, moderate, high, very high
susceptibility respectively. Figure 5 shows the clas-
sified susceptibility map of the study area.
7 DISCUSSION AND CONCLUSIONS
In this paper, a soft computing technique approach
to evaluating the susceptibility area of earthquake
triggered landslides using GIS technology and
remote sensing images is presented. There are
many published papers about landslide suscepti-
bility mapping using different methods based on
GIS technology, especially in the last twenty years.
And much work has been undertaken to solve the
deficiencies and difficulties in the susceptibility
evaluation and create a procedure for preparing
landslide susceptibility map which is both sim-
ple and with a high degree of accuracy. However,
researches about statistical methods applied in
Figure 4. Area Under Curve (AUC) represents the
success rate and prediction rate.
Figure 5. Landslide susceptibility map produced from
ANN model.
222
earthquake triggered landslide susceptibility
mapping are relatively rare (e.g. Lee & Evangelista
2006; Pareek et al. 2010; Lin & Tung 2004; Lee
et al. 2008; Xu et al. 2012a, 2012b).
The landslide inventory is naturally event-based.
It is not possible to use a multi-temporal landslide
inventory as the traditional landslide susceptibility
analysis. Therefore, the inventory of landslides trig-
gered by the earthquake was randomly partitioned
into two subsets, training dataset and testing data-
set. From the application of ANN model, the land-
slide susceptibility map is constructed. The results
of verification show 81.274% of success rate and
79.915% of prediction rate. The verification results
are of high values. It shows that the ANN model
can be used as a precise tool in the earthquake trig-
gered landslide susceptibility mapping when a suf-
ficient number of data are available.
This paper may provide useful references and
suggestions to engineers and planners involved in
land use planning, seismic hazard mitigation and
infrastructure planning, and of great help to engi-
neers and planners for choosing suitable locations
to implement developments in earthquake struck
area. Furthermore, it also adds an extra value to
the literature of earthquake triggered landslide
susceptibility mapping using soft computing
technique (e.g. ANN model), GIS technique, and
remote sensing data.
ACKNOWLEDGMENTS
This research is supported by the National Science
Foundation of China (Grant No. 41202235)
REFERENCES
Arora, M.K., Das Gupta, A.S., Gupta, R.P. 2004. An
artificial neural networks approach for landslide
hazard zonation in the Bhagirathi (Ganga) Valley,
Himalayas. International Journal of Remote Sensing
25(3): 559–572.
Ayalew, L. & Yamagishi, H. 2005. The application of
GIS-based logistic regression for landslide suscep-
tibility mapping in the Kakuda-Yahiko Mountains,
Central Japan. Geomorphology 65(1–2): 15–31.
Brenning, A. 2005. Spatial prediction models for landslide
hazards: review, comparison and evaluation. Natural
Hazards and Earth System Sciences 5(6): 853–862.
Chacon, J., Irigaray, C., Fernandez, T., Hamdouni, R.E.
2006. Engineering geology maps: landslides and geo-
graphical information systems. Bulletin of Engineering
Geology and the Environment 65(4): 341–411.
Dahal RK, Hasegawa S, Nonomura A, et al. 2008.
GIS-based weights-of-evidence modelling of rainfall-
induced landslides in small catchments for landslide
susceptibility mapping. Environmental Geology 54(2):
311–324.
Guzzetti, F., Carrara, A., Cardinali, M., Reichenbach, P.
1999. Landslide hazard evaluation: a review of current
techniques and their application in a multi-scale study,
Central Italy. Geomorphology 31(1–4): 181–216.
Keefer, D.K. 1984. Landslides caused by earthquakes.
Geological Society of America Bulletin 95(4):
406–421.
Lee, S. 2004. Application of Likelihood Ratio and Logis-
tic Regression Models to Landslide Susceptibility
Mapping Using GIS. Environmental Management
34(2): 223–232.
Lee, C.T., Huang, C.C., Lee, J.F., Pan, K.L., Lin, M.L.,
Dong, J.J. 2008. Statistical approach to earthquake-
induced landslide susceptibility. Engineering Geology
100(1–2): 43–58.
Lee, S. & Evangelista, D.G. 2006. Earthquake-induced
landslide-susceptibility mapping using an artificial
neural network. Natural Hazards and Earth System
Sciences 6(5): 687–695.
Lin, M.L. & Tung, C.C. 2004. A GIS-based potential
analysis of the landslides induced by the Chi-Chi
earthquake. Engineering Geology 71(1–2): 63–77.
Pandey, A., Dabral, P.P., Chowdary, V.M., Yadav, N.K.
2008. Landslide Hazard Zonation using Remote
Sensing and GIS: a case study of Dikrong river basin,
Arunachal Pradesh, India. Environmental Geology
54(7): 1517–1529.
Pareek, N., Sharma, M.L., Arora, M.K. 2010. Impact of
seismic factors on landslide susceptibility zonation:
a case study in part of Indian Himalayas. Landslides
7(2): 191–201.
Park, N.W. 2010. Application of Dempster-Shafer theory
of evidence to GIS-based landslide susceptibility
analysis. Environmental Earth Sciences 62(2): 367–376.
Pradhan, B. & Lee, S. 2010. Landslide susceptibility
assessment and factor effect analysis: backpropaga-
tion artificial neural networks and their comparison
with frequency ratio and bivariate logistic regression
modelling. Environmental Modelling and Software
25(6): 747–759.
Saha AK, Gupta RP, Sarkar I, Arora, M.K., Csaplovics, E.
2005. An approach for GIS-based statistical landslide
susceptibility zonation—with a case study in the
Himalayas. Landslides 2(1): 61–69.
Sezer, E.A., Pradhan, B., Gokceoglu, C. 2011. Manifestation
of an adaptive neuro-fuzzy model on landslide suscep-
tibility mapping: Klang valley, Malaysia. Expert Sys-
tems with Applications 38(7): 8208–8219.
US Geological Survey. 2010. Advanced National Seismic
System (ANSS), Shake Map, Global Region, Maps
of Ground Shaking and Intensity for Event 2010 yr,
Southern Qinghai, China, http://earthquake.usgs.gov/
eqcenter/shakemap.
Wachal, D.J. & Hudak, P.F. 2000. Mapping landslide sus-
ceptibility in Travis County, Texas, USA. GeoJournal
51(3): 245–253.
Weiss, A.D. 2006. Topographic position and landforms
analysis. http://www.jennessent.com/downloads/tpi-
poster-tnc_18 × 22.pdf.
Xu, C. & Xu, X.W. 2012a. Spatial prediction models
for seismic landslides based on support vector
machine and varied kernel functions: A case study of
the 14 April 2010 Yushu earthquake in China. Chinese
Journal of Geophysics 55(6): 666–679.
223
Xu, C. & Xu, X.W. 2012b. Comment on “Spatial
distribution analysis of landslides triggered by
2008.5.12
Wenchuan Earthquake, China” by
Shengwen Qi, Qiang Xu, Hengxing Lan, Bing Zhang,
Jianyou Liu [Engineering Geology 116 (2010) 95–108].
Engineering Geology 2012, 133–134: 40–42.
Xu C. & Xu X.W. 2013. Controlling parameter analy-
ses and hazard mapping for earthquake triggered-
landslides: an example from a square region in Beichuan
County, Sichuan Province, China. Arabian Journal of
Geosciences. doi:10.1007/s12517-012-0646-y.
Xu, C., Dai, F.C., Xu, X.W., Lee, Y.H. 2012a. GIS-based
support vector machine modeling of earthquake-
triggered landslide susceptibility in the Jianjiang River
watershed, China. Geomorphology 145–146: 70–80.
Xu, C., Xu, X.W., Dai, F.C., Saraf, A.K. 2012b.
Comparison of different models for susceptibility
mapping of earthquake triggered landslides related
with the 2008 Wenchuan earthquake in China.
Computers & Geosciences 46: 317–329.
Xu, C., Xu, X.W., Lee, Y.H., Tan, X.B., Yu, G.H.,
Dai, F.C. 2012c. The 2010 Yushu earthquake trig-
gered landslide hazard mapping using GIS and weight
of evidence modeling. Environmental Earth Sciences
66(6): 1603–1616.
Xu, C. Xu, X.W., Yu, G.H. 2012d. Earthquake triggered
landslide hazard mapping and validation related with
the 2010 Port-au-Prince, Haiti earthquake. Disaster
Advances 2012, 5(4): 1297–1304.
Xu, C., Xu, X.W., Dai, F.C., Xiao, J.Z., Tan, X.B.,
Yuan, R.M. 2012e. Landslide hazard mapping using
GIS and weight of evidence model in Qingshui river
watershed of 2008 Wenchuan earthquake struck
region. Journal of Earth Science 23(1): 97–120.
Xu, C., Xu, X.W., Yu, G.H. 2012f. Study on the
characteristics, mechanism, and spatial distribution
of Yushu earthquake triggered landslides. Seismology
and Geology 34(1): 47–62. (in Chinese).
Xu, C., Xu, X.W., Yao, X., Dai, F.C. 2013a. Three
(nearly) complete inventories of landslides triggered
by the May 12, 2008 Wenchuan Mw 7.9 earthquake of
China and their spatial distribution statistical analysis.
Landslides. doi:10.1007/s10346-013-0404-6.
Xu, C., Xu, X.W., Yao, Q., Wang, Y.Y. 2013b. GIS-based
bivariate statistical modeling for earthquake- triggered
landslides susceptibility mapping related to the
2008 Wenchuan earthquake, China. Quarterly Jour-
nal of Engineering Geology and Hydrogeology 46(2):
221–236.
Xu, C., Xu, X.W., Yu, G.H. 2013c. Landslides triggered
by slipping-fault-generated earthquake on a plateau:
an example of the 14 April 2010, Ms 7.1, Yushu,
China earthquake. Landslides. doi:10.1007/s10346-
012-0340-x.
Yilmaz, I. 2010. Comparison of landslide susceptibil-
ity mapping methodologies for Koyulhisar, Turkey:
conditional probability, logistic regression, artifi-
cial neural networks, and support vector machine.
Environmental Earth Sciences 61(4): 821–836.
Yilmaz, I. & Keskin, I. 2009. GIS based statistical and
physical approaches to landslide susceptibility map-
ping (Sebinkarahisar, Turkey). Bulletin of Engineering
Geology and the Environment 68(4): 459–471.