




































| 75  
 

         Geoplanning 
Vol 5, No. 1, 2018, 75-90                                                                                                                                                            Journal of Geomatics and Planning 

                                                                                                 E-ISSN: 2355-6544 
http://ejournal.undip.ac.id/index.php/geoplanning 

doi: 10.14710/geoplanning.5.1.75-90 

GIS-BASED LANDSLIDE SUSCEPTIBILITY ASSESSMENT AND FACTOR EFFECT 
ANALYSIS BY CERTAINTY FACTOR IN UPSTREAM OF JENEBERANG RIVER, 
INDONESIA  

P. F. Nurdina, T. Kubotab  

a Graduate School of Bio-Resources and Environmental Science, Kyushu University, Japan 
b Faculty of Agriculture, Kyushu University, 6-10-1 Hakozaki Higashi-Ku Fukuoka 812-8581, Japan 

 

Abstract: This study aimed to assess landslide susceptibility by employing certainty 

factors model (CF) to select the causative factors for landslide susceptibility mapping in 
Upstream of Jeneberang River, South Sulawesi. Indonesia. The landslide causative 
factors were: soil, slope angle, aspect, elevation, lithology, land use, distance to the 
river, drainage density, and precipitation. For validation purpose, landslide inventory 
map was randomly partition into two groups, 30% for the validation and 70% for the 
training. Landslide susceptibility maps were produced by logistic regression using 
original factor (all nine factors) and selected factor (four factors with positive CF 
value). The result of certainty factor analysis shows CF value is positive for elevation, 
land use, slope and drainage density. The accuracy of two landslide susceptibility maps 
were evaluated by calculating the area under the curve of Receiver Operating 
Characteristic (ROC) curves. The result shows the the success rate curve for nine factor 
map (80.2%)  is higher than four factor map (78%). But in case of closeness between 
success rate curve and predictive rate curve, certainty factors model has a closer 
distance. In this study, effect analysis studies show how the accuracy changes when 
the input factors are changed. 

 Copyright © 2018 GJGP-UNDIP  

This open access article is distributed under a  

Creative Commons Attribution (CC-BY-NC-SA) 4.0 International license. 

Nurdin, P. F., & Kubota, T. (2018). GIS-based Landslide Susceptibility Assessment by Certainty Factor and Logistic Regression in Upstream of 
Jeneberang River, Indonesia. Geoplanning: Journal of Geomatics and Planning, 5(1), 75-90. doi:10.14710/geoplanning.5.1.75-90 

1. INTRODUCTION 

A landslide occurs worldwide. However, the impact is greater in developing country. The spatial probability 
of landslide occurrence, also known as susceptibility (Brabb, 1985), is the probability that any given region 
will be affected by landslides, given a set of environmental conditions (Guzzetti et al., 2005). Many landslide 
causative factors have been considered in the literature for landslide susceptibility mapping, but it is not 
certain which factors produce the optimal result for an area under analysis. With the availability of 
increasing number of landslide causative factors, finding the best combination of factors has become an 
important research issue. Currently, there are no universal guidelines for the selection of landslide 
causative factors. Determining the causative factors are a difficult task,  Van Westen et al. (2003) stated 
that every study area has its particular set of factors, which triggering landslides. A factor can be a 
contributing one for landslide occurrence in an area but not in another one.  

Landslide causative factors could have different degrees of effects on the accuracy of landslide 
susceptibility maps that has been investigated by several types of research in the literature. Cuesta, 
Sánchez, & Garci’a (1999) assessed 209 landslide events from 1980 to 1994 in the Cantabrian Mountains in 
northwestern Spain and found that precipitation was the most influential causative factor. Moreiras 
(2005) considered lithology and slope as the most influential factors in landslide mapping based on the 
study area of the Rio Mendoza Valley in Argentina. Glenn et al. (2006) stated that topographic factors are 
highly influential parameters in landslide studies. They assessed the efficiency of laser scanning data 
(LiDAR)-derived topographic factors in characterizing landslide morphology and activity. For instance, 

OPEN ACCESS 

Article Info: 
Received: 31 July 2017 
in revised form: 10 Dec 2017 
Accepted: 20 March 2018 
Available Online:  30 April 2018 
 

Keywords:  
Landslide, Susceptibility Map, GIS, 
Certainty Factor, Logistic 
Regression  
 

Corresponding Author: 
Putri Fatimah Nurdin 
Kyushu University, Fukuoka, 
Japan, 
email: putrinurdin@kyudai.jp  
  

https://doi.org/10.14710/geoplanning.5.1.75-90
https://doi.org/10.14710/geoplanning.5.1.75-90
mailto:putrinurdin@kyudai.jp


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

76 | 
 

Costanzo et al. (2012) analyzed the relationships between a priori ranking of controlling factors and 
predictive performances. The results showed that slope angle, roughness, land use and topographical 
wetness index were main causative factors.  

On 26th March 2004, the huge collapse occurred in northern caldera wall of Bawakaraeng caldera (Elev. 
2,830 m). The collapsed area has caused a ridge including Mt. Sarongan (Elev. 2,514 m), and the collapsed 
mass volume was estimated more than 200 million m3. The collapsed mass has been running down as 
debris flow, it is reached the Bili-Bili reservoir and treated life time of the dam. The collapse was predicted 
to be caused by a combination of several factors; such as weak geological structures, steep high walls of the 
caldera and high rainfall intensity (Tsuchiya et al., 2004). The main portion of the collapse is two twin ridges 
of the caldera. These ridges were initially formed by the steep slope surface slipping down or creeping; the 
creep zone eventually expanded and settled in an unstable condition (CTI, 2006). Low cementation rocks, 
stock and second deposits of volcanic rocks are distributed at the base of Mt. Bawakaraeng forming a low 
strength base for the high walls of the caldera (CTI, 2006). 

The scope of this study aims to investigate the geo-environmental factors that contribute to landslide and 
assess the most significant causative factors to generate the landslide susceptibility map with better 
accuracy. The outline of this study is to select the most significant causative factors by certainty factor 
analysis, produce landslide susceptibility map using the selected and original causative factors by logistic 
regression and do the comparison between two models.  

2. DATA AND METHODS 

2.1. Study area 

The study area (Figure 1) is located in the upstream of Jeneberang River (Lengkese Sub-watershed). The 

study area is bounded by the latitude of 05°18 ′10″ and longitude 119°53′20″ with an area about 128.40 

km2 and most of the terrain is mountainous with highest peaks exceeding 2,795 m.  The study area located 

around 70 km east of capital city of South Sulawesi Province. Jeneberang River rises in Mt. Bawakaraeng, 

which has an elevation of 2,833 m above MSL and flows from the east to the west. Sulawesi Island has a 

tropical climate with characteristics of two seasons within a year, rainy seasons from November to May and 

a dry season from June to October. The precipitation is more than 700 mm in the month of February and 

rises to 900 mm in January (Tsuchiya et al., 2009). This area is a productive land but regularly experience a 

small to a big landslide, especially during the rainy season. Due to increasing of rainfall intensity, the 

probability of landslide occurrence, particularly shallow landslides increases and it is very sensitive to short 

lasting, high intensive rainfall (Hasnawir and Kubota, 2012). 

 

 

 

 

 

 

 

 

 

 

Figure 1. Hillshaded map of the study area and the landslide inventory 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 77  
 

 

Figure 2. Flow chart of the study. 

The lithological layer of the study area was created from geology map of Indonesia with scale 1:250,000. 
The land use layer was based on Indonesia Ministry of Forestry 2014. River and soil layers were based on 
BIG (Badan Informasi Geospasial / Geospatial Information Agency). Another factor such as mean annual 
rainfall data, collected from rainfall gauge stations was available around the study area. Landslide inventory 
of the area was detected by satellite image interpretation and verified by field investigation. 

2.1.1. Landslide inventory map 

According to Guzzetti et al. (1999), landslides which occurred in the past and present are keys to predicting 
landslides happening in future. However, due to the absence of historical records of landslides and their 
triggering factors, insufficient information and heterogeneity of subsurface conditions and lack of 
knowledge about their behavior make it very difficult to predict spatial and temporal probabilities of 
landslides. Indirect mapping methods use either statistical models or deterministic models to predict 
landslide prone areas, based on information obtained from the interrelation between landslide causative 
factors and the landslide distribution. Therefore landslide inventory is an essential component of landslide 
hazard zonation techniques. 

A total of 380 known landslides with a total of 9,901 pixels were prepared from field observations and 
remote sensing of the study area (Figure 1). Some of the archived landslide inventory databases were also 
used in the previous research to produce landslide hazard map. Landslides in the area include rotational 
slides and translational slides. For building the susceptibility map models, the landslide inventory was 
randomly partitioned into two groups: a training data set (70%, 266 landslides) and a validation data set 
(30%,  144 landslides). 

 

 

 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

78 | 
 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 3. Landslide site in study area 

2.2. Geospatial database of geo-environmental factors influencing landslides 

Data preparation is the first fundamental and important step for landslide susceptibility analysis. To 
mapping the potential landslide in sub-watershed Jeneberang, it first conducted studies on the factors that 
cause landslides. Factors that caused landslides are well known by experts or landslide researchers, but the 
main factor of avalanche between one watershed and others will vary due to the different of biophysical 
conditions. In this study, nine landslide causative factors were used, namely: elevation, slope aspect, slope 
angle, lithology, drainage density, distance to river, soil texture, mean annual rainfall and land cover or land 
use (Figure 4). Each category was divided into different classes by its value or feature.  

Digital elevation model (DEM), remotely sensed imagery and geological maps of the study area were used 
to create maps of the factors that were employed in subsequent stages. Digital Elevation Model (DEM) was 
the key to generate various topographic parameters related to landslide activity of the study area. With cell 
size 10 x 10 meter, elevation (<820 – 2,795 m), slope angle (0 - >45 degrees) and slope aspect layers have 
been extracted. The resolution and accuracy have a direct influence on the quality of these factors (Lee, 
2005). 

2.3. Probabilistic analysis 

The probabilistic analysis is performed using a methodology integrating the results into a spatial database 
using GIS. 

2.3.1. Certainty Factor Analysis 

The certainty factor (CF) model is a method for managing uncertainty in rule-based systems. Shortliffe and 
Buchanan (1975) developed the CF model in the mid-1970s for MYCIN, an expert system for the diagnosis 
and medical treatment. In this study, CF is applied to selecting the optimal causative factor related to 
landslide occurrence. Certainty factor can be calculated using the following functions:  

 

(1) 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 79  
 

Where PPa is the conditional probability of having a number of landslides event in a class of parameter and 
PPs is the prior probability of a total number of landslides in the study area. For each of the causative 
factors, the weights and contrast were calculated using the certainty factor method. The CF approach 
transforms each class into interval between -1 and 1, and it indicates a measure of belief and disbelief. A CF 
value of -1 indicates that an increasing uncertainty of landslide occurrence or the certainty of the 
hypothesis being true is very small, as compared with a high CF near to 1 means that decreasing uncertainty 
or the indication strongly supports the hypothesis as true. A value close to 0 means that there is no 
information of the landslide occurrence. The ppa and pps values were determined by overlaying each 
parameter layer with the landslide inventory layer in ArcGIS and landslides falling in each parameter class 
were determined. These values were used to determine the CF value of each classes. 

 

2.3.2 The Multivariate Approach: Logistic Regression (LR) 

It is admitted that among the wide range of statistical methods proposed in landslide susceptibility 
mapping, Logistic Regression analysis has proven to be one of the most reliable approaches (Ayalew et al., 
2005). Logistic Regression analysis relates the probability of landslide occurrence (having values from 0 to 1) 
to the “logic” Z (where −1<Z <0 for higher odds of non-occurrence and 0<Z <1 for higher odds of 
occurrence). In the LR formula, the probability of landslide occurrence is expressed by: 

 

where P is the estimated probability of landslide occurrence and ranges from 0 to 1; Y is an indicator 
variable, X is the independent variables (landslide causative factors), X = (x0, x1, x2,. . . xn), x0 = 1; b is 
regression coefficient. To linearize the mentioned method as well as remove the 0/1 boundaries for the 
original dependent variable, the estimated P probability is transformed by the following formula: 

 

The alteration is referred to as the logit transformation. Theoretically, the logit transformation of binary 
data can ensure that the dependent variable is continuous and the logit transformation is boundless. 
Moreover, it can ensure that the probability surface will be continuous within the range [0, 1]. Using the 
logit transformations, the standard linear regression models can be obtained as follows: 

 

Here, b0 is the constant or intercept of the formula, b1, b2, . . . bn represents the slope coefficients of the 
independent parameters, x1, x2, . . . xn in the logistic regression and ε is the standard error. Multivariate 
regression analysis plays a central role in statistics that cause one of the most powerful and commonly used 
techniques (McCullagh and Nelder, 1989).  

 

(2) 

(3) 

(4) 

(5) 

 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

80 | 
 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 4. Thematic maps used in this study, (a) Elevation; (b) Slope; (c) Aspect; (d) Lithology; (e) Landuse; (f) 

Soil; (g) Distance to river; (h) Drainage density; (i) Precipitation. 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 81  
 

3. RESULTS AND DISCUSSION 

3.1. Physical Factor of Lanslide 

3.1.1. Elevation 

Elevation is commonly used to assess landslide susceptibility. The variation in elevation may be related to 
several environmental settings such as rainfall and vegetation variety. A digital elevation model (DEM) can 
categorize the local relief and locate points of maximum and minimum heights within the terrain. The 
elevation of the study area is 820 to 2,795 meters. Figure 3 shows that landslides mostly occurred at 820-
1,137 m (26%). One of the reason is the land use at those elevation is dominated by agriculture which 
mostly is paddy field. 

3.1.2. Slope angle 

Slope is often used to study landslide probability, and all studies into the probability of landslides consider 
slope (Dai et al., 2001; S. Lee and Talib, 2005). Highly sloped areas and cleared areas receive exposure to 
direct sunlight, which dries the soil and increases the chances of landslides. The slope angle is frequently 
considered to be one of the most influential factors for landslide modeling because it influences the shear 
forces acting on hill slopes (Dai et al., 2001). In the study area, landslides increased with increasing slope 
steepness. Most landslides occurred on slopes of 30-45 degrees (20%). 

3.1.3. Slope aspect 

The slope aspect describes the slope direction, and it identifies the downslope direction of the maximum 
rate of elevation change. Although the relationship between the aspect and the mass movement has been 
investigated for a long time, there is no general decision regarding the aspect/landslide relationship 
(Ercanoglu & Gokceoglu, 2004). However, the aspect is a significant factor in producing landslide 
susceptibility maps (Saro Lee et al., 2004). The slope aspect also plays an important role in exposing the 
topography to sunlight and drying winds, which control the soil moisture. This is an important factor in 
landslide studies (Magliulo et al., 2008). According to the number of pixels affected by slope failure, a 
southwest-facing slope (21,34%) ranks first followed by the north- (19,87%), northwest- (15,91%), and 
west-facing slopes (14,92%). Moreover, slopes facing south are more prone to landslides because they 
receive heavy rainfall during the monsoon season.  

3.1.4. Lithology 

Lithology is the most important parameter in this study of landslides because different lithology units have 
varying degrees of landslide vulnerability (Cuesta et al., 1999; Dai et al., 2001). The lithological units shown 
in the surface geologic maps were reclassified according to geology and development center. The result was 
a generalized geologic map. Finally, the map describes the distribution of six types of lithology:  

 TMC (Tertiary Miocene Camba): Marine sediment rocks vary with volcanic rocks, tuff sand vary with 
sandy tuff and clay stone; and have insertion marl, limestone, conglomerate, volcanic breccias, and 
coal. 

 QLVP, QLV, and QLVB (Quarter Lompobatang Volcanic):  Agglomerates, lava, breccias, lahar 
deposition and tuff. 

 TPBV (Tertiary Pliocene Baturape Cindako Volcanic): Lava and breccias, with insertion tuff and 
conglomerate.   

 QAC (Quarte Aluvium): gravel, sand, clay, mud and coral limestone. 
The relationship between landslide occurrence and lithological condition showed that QLV has the highest 
percentage of landslides (63,88%) of the six other lithology classes. QLV is a common volcanic sediment 
formation in South Sulawesi.   

 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

82 | 
 

Figure 5. The spatial relationship between landslide occurrence and causative factors. Landslide 
occurrence correlates strongly with environmental factors that might trigger its mechanism. 

 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 83  
 

3.1.5. Soil 

The physical properties of soil are often used for parameter analysis of landslides via a probabilistic 

approach to soil texture. Soil texture can affect the other physical soil properties such as water infiltration, 

porosity, and permeability of water and power to pass groundwater. In Indonesia, soil is classified via the 

United States Department of Agriculture system, the Food and Agriculture Organization of the United 

Nations (FAO), and the Centre of Soil and Agroclimatic Research. Soil in the study area is from the Andosol 

family and is divided into three types: Dystrandepth, Dystropepts, and Tropaquepts. Landslides are mostly 

accumulated on Dystrandepts soil type (77,25%). This might be related to the location of Dystrandepts 

deposition, which are mostly found at higher altitudes. This class accounts for the largest proportion of the 

areas. 

3.1.6. Drainage density 

Drainage density is the total stream length per unit area of a river basin. Hasegawa et al. (2009) noticed that 
if precipitation increased, then an area with a higher drainage density is more often prone to a shallow 
landslide. A large-scale landslide is frequent in areas with less drainage density.  In this study area, the 
0.004-0.008 km subclass has the highest landslide percentage (26%). 

3.1.7. Distance to river 

Rivers play a major role in modifying the terrain by incising different rocks (Meten et al., 2015). Runoff plays 
an important role and is a triggering factor in landslides. According to Meten et al. (2015), rivers have a 
significant role in facilitating landslides. The analysis assessed the influence of distance to river and drainage 
density on landslide. In the case of the relationship between landslide occurrence and distance to river, 
subclasses of 0 – 60 have the highest landslide percentage (70%). Gully erosion along the river may initiate 
landslides. Areas closer to the river network have more erosive forces that erode the base of the slope to a 
greater degree.  

3.1.8. Precipitation 

Rainfall is the principal climatic variable that influences landslide distribution. It is affected by topography, 
elevation, and vegetation—factors that are all interrelated. Mountainous areas cause the air currents to rise 
and cool resulting in increased precipitation with elevation (Walker and Shiels, 2013). The climate of the 
study area is tropical humid, and precipitation varies with elevation. The average annual precipitation is 
between 3,100 – 3,800 mm. The rainfall data area was obtained from two weather stations: Lengkese 
station and Malino station. Malino has the highest landslide percentage (79%) followed by Lengkese area.  

3.1.9. Land use 

Land use also plays an important role in the stability of the slope. The land covered by forest regulates 
continuous water flow. Water regularly infiltrates this area whereas the cultivated land affects the slope 
stability due to saturation of the covered soil. Land use in the study area is mainly occupied by dry land 
agriculture, mixed garden, forested area, paddy field, and savannah. Paddy field covered mostly located in 
the lowland and river floodplain. The landslide area was mostly in agricultural (27,57%) and secondary 
forest (27,18%). 

3.2. Factors selection using Certainty Factor (CF) 

The landslide distribution for each class is expressed by the number of occurring pixels and was used to 
calculate CF values. Table 1 shows the result of Z value of each causative factor. Based on the Certainty 
Factor method, four causative factors were detected with high influence to slope instability in the study 
area: land use (z value: 0.82), elevation (z value: 0.56), drainage density (z value: 0.25), and slope angle (z 
value: 0.30). These four factors have positive relationships with landslide occurrence. Therefore, these four 
factors were selected for further process to create an optimized landslide susceptibility map. The highest Z 
value is land use. Landslides especially correspond to the primary forest subclass (CF value: 0.80).  

 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

84 | 
 

Table 1.  The relationship between landslide occurrence and causative factors 

Theme Class Pixels in class Landslides CF Z 
SOIL Dytstrandepts 508,608.00 7,649.00 0.49 (0.72) 

 Dystropepts 732,874.00 2,128.00 -0.63  
 Tropaquepts 42,449.00 124.00 -0.62  

LITHOLOGY camba formation (TMC) 65,868.00 195.00 -0.62 (1.00) 
 parasitic eruption (QLVP) 199,491.00 565.00 -0.63  
 lompobattang  volcanic rock (QLV) 572,811.00 6,325.00 0.30  
 baturape volcanic rock (TPBV) 31,343.00 129.00 -0.47  
 aluvium deposition (QAC) 27,945.00 0.00 -1.00  
 breksi (QLVB) 386,473.00 2,687.00 -0.10  

LANDUSE Primary Forest 15,192.00 582.00 0.80 0.82 
 Secondary Forest, forest plantation 279,523.00 2,691.00 0.20  
 Brush & Savana 165,513.00 2,458.00 0.48  
 Open Land 50,810.00 499.00 0.22  
 Agriculture, plantation & settlement 539,506.00 2,730.00 -0.35  
 Paddy Field 201,138.00 343.00 -0.78  
 Water & sand 32,249.00 598.00 0.59  

ASPECT North 233,598.00 1,967.00 0.08 (0.21) 
 Northeast 118,658.00 998.00 0.08  
 East 50,658.00 349.00 -0.11  
 Southeast east 56,486.00 370.00 -0.15  
 Southeast 146,194.00 1,052.00 -0.07  
 South west 198,444.00 2,113.00 0.28  
 West 224,796.00 1,477.00 -0.15  
 Northwest 255,097.00 1,575.00 -0.20  

ELEVATION <820 281,570.00 1,005.00 -0.54 0.56 
 820 - 1,137 323,730.00 2,594.00 0.04  
 1,137 - 1,451 235,103.00 1,615.00 -0.11  
 1,451 - 1,845 278,558.00 2,223.00 0.03  
 1,845 - 2,299 129,473.00 1,592.00 0.38  
 2,299 - 2,795 35,497.00 872.00 0.69  

DISTANCE TO 
RIVER 

0 – 60 818,631.00 6,944.00 0.09 (1.00) 

 60 – 120 384,836.00 2,254.00 -0.24  
 120 – 200 78,739.00 697.00 0.13  
 200 – 250 1,579.00 6.00 -0.51  
 250 -350 146.00 0.00 -1.00  

DRAINAGE 
DENSITY 

0 — 0.004 172,338.00 905.00 -0.32 0.25 

 0.004 – 0.008 750,490.00 4,384.00 -0.24  
 0.008 – 0.012 319,030.00 4,110.00 0.40  
 0.012 – 0.019 42,073.00 502.00 0.36  

SLOPE ANGLE 0 – 5 115,433.00 157.00 -0.82 1.00 
 5 – 10 235,161.00 424.00 -0.77  
 10 – 15 248,266.00 811.00 -0.58  
 15 – 20 209,183.00 1,104.00 -0.32  
 20 – 25 158,647.00 1,293.00 0.05  
 25 – 30 111,267.00 1,578.00 0.46  
 30 – 35 74,494.00 1,373.00 0.59  
 35 – 45 82,741.00 1,972.00 0.68  
 >45 48,739.00 1,189.00 0.69  

Precipitation malino 947,840.00 7,807.00 0.06 (0.14) 
 lengkese 336,091.00 2,094.00 -0.19  

 

 

 

 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 85  
 

Most landslide cases seen here occurred on natural slopes where vegetation grows perennially. Elevation 
plays a dominant role in landslide occurrence in the study area. The CF value is positive for intermediate 
elevation (820-1,137 m) and increased further above 1,451 m. From 2,299 m – 2,795 m, dystrandepts 
dominate the soil, and landslides are common (CF value: 0,691). It is widely accepted that the slope angle 
directly influences slope instability. Landslide probability increases from 20 -25 degrees, and it increases 
further and corresponds to slope angle. The increase in slope angle is affected by the shear stress in the soil 
or unconsolidated material increases. Gentle slopes generally have a lower frequency of landslides than 
steep slopes because of the lower shear stresses seen in low gradients (Lee & Talib, 2005). In the study 
area, most landslides are associated with slope angles greater than 45 degrees (CF value: 0,69). 

The drainage density negatively impacts landslide susceptibility due to their abrasive forces along the base 
of the slope. In the study area, the subclass of 0.08-0.012 km/km2 has the highest CF value of 0.40 where 
most shallow landslides are observed. Costanzo et al. (2012) identified the factors based on the ranks 
associated with the factor’s expected contribution to the predictive skill of a multivariable model. 
Approaches adopting discriminant analysis and logistic regression on the forward selection of variables, 
however fail when most of the variables are statistically significant. Although this method includes less 
computation, it requires to categorize the data into landslide and non-landslide groups which is rather 
exhausted. The proposed model using CF eliminated these limitations because it used only landslide pixels 
in the computation, and hence is very fast. Prior definition of hazard classes is not required in CF approach 
and it also supplies advantage of rendering the definition of susceptible classes transparent. Moreover, the 
proposed model is a relatively straightforward method that allows the causative factors to be ranked 
according to their certainty values in the range between -1 to 1. It is assumed that positive CF values have a 
high correlation with the landslide occurrence, and vice versa. 

3.3 Landslide susceptibility mapping using logistic regression   

In this study, a logistic regression model was developed using an equal proportion of landslide and non-
landslide pixels over ten iterations and all non-landslide data as a comparison. The constant and coefficient 
of independent variables were provided by logistic regression analysis using SPSS. Landslide data were 
randomly selected by SPSS based on the number of balanced proportion of non-landslide pixels,. Hence, 
this study proposes to examine ten iterations to get optimal results and a sense of fairness as shown in 
Tables 2 and 3. By applying a logistic regression model, the landslide occurrence probability was measured. 
If the values are closer to unity, then landslides are more likely to occur. 

Table 2. Iteration for all causative factor models 

Iteration Data Equal 
Drainage 
density 

Distance 
to river 

Soil Landuse Lithology Elevation Aspect Precipitation Slope Constant ROC 

Iteration 01 B 0.69 1.34 0.87 0.28 -0.42 0.10 0.55 1.44 0.59 -6.06 0.80 
Iteration 02 B 0.67 1.03 0.80 0.31 -0.36 0.12 0.74 1.40 0.62 -5.94 0.80 
Iteration 03 B 0.60 1.22 0.87 0.31 -0.39 0.08 0.64 1.04 0.64 -6.00 0.80 
Iteration 04 B 0.66 1.18 0.89 0.32 -0.37 0.05 0.63 1.47 0.62 -6.10 0.80 
Iteration 05 B 0.62 1.09 0.84 0.30 -0.45 0.13 0.71 1.65 0.60 -6.10 0.79 
Iteration 06 B 0.67 1.18 0.83 0.30 -0.39 0.12 0.59 1.47 0.62 -6.04 0.80 
Iteration 07 B 0.67 1.08 0.86 0.28 -0.41 0.12 0.74 1.55 0.60 -6.11 0.80 
Iteration 08 B 0.68 1.06 0.89 0.29 -0.46 0.15 0.72 1.49 0.60 -6.06 0.80 
Iteration 09 B 0.69 1.18 0.83 0.31 -0.37 0.12 0.69 1.53 0.60 -6.21 0.80 

Iteration 010 B 0.66 1.12 0.84 0.26 -0.46 0.18 0.56 1.38 0.60 -5.77 0.79 

 

Table 3. Iteration for selected causative factor models 

Iteration Data Equal 
Drainage 
density 

Landuse Elevation Slope Constant ROC 

Iteration 01 B 0.74 0.43 0.33 0.67 -2.62 0.77 

Iteration 02 B 0.73 0.45 0.34 0.70 -2.65 0.78 

Iteration 03 B 0.67 0.46 0.33 0.72 -2.63 0.78 

Iteration 04 B 0.72 0.48 0.31 0.70 -2.66 0.78 

Iteration 05 B 0.66 0.44 0.33 0.69 -2.55 0.77 

Iteration 06 B 0.85 0.62 0.39 0.71 -3.00 0.77 

Iteration 07 B 0.71 0.43 0.35 0.68 -2.61 0.77 

Iteration 08 B 0.72 0.44 0.41 0.70 -2.69 0.78 

Iteration 09 B 0.74 0.46 0.34 0.68 -2.67 0.77 

Iteration 010 B 0.72 0.40 0.41 0.68 -2.65 0.77 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

86 | 
 

  

(6) 

 

Based on logistic regression analysis for all causative factors (Table 2), precipitation has the highest 

coefficient of 1.47 which means that in the study area precipitation plays the important rules in triggering 

the landslide. Meanwhile, Table 3 shows that slope plays the most important rules than three others 

causative factors. In this case, precipitation was excluded because it was not selected by certainty factors 

analysis (Z= -0.14). The natural break method or Jenks optimization is also known as the goodness of 

variance fit (GVF) method. It has been used widely especially by planners. It determines the best 

arrangement of values for the different classes. This method maximizes values between classes and reduces 

the variance within classes. The five classes include very low, low, moderate, high, and very high values 

describing the level of landslide susceptibility in the study area. Figure 4 (a) shows the landslide 

susceptibility map using nine causative factors. The LSM model for nine causative factors was obtained 

using the coefficient values as the equation 6 :  

 Z = 0.05 (Elevation) + 0.62 (Slope) + 0.63 (Aspect) + (-0.37) (Lithology) + 0.32 (Landuse) + 0.89 (Soil) + 1.18 
(Distance to river) + 0.66 (Drainage density) + 1.47 (Precipitation) – 6.10     ……………………………………. 
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 
 

Figure 6.  (a) Landslide susceptibility map using nine causative factors generate by logistic regression. (b) 

Landslide susceptibility map using four causative factors generate by logistic regression. (c) Enlarged image 

of LSM using nine causative factors. (d) Enlarge image of LSM using four causative factors. 

Based on the certainty factor, further LR analysis was conducted using the four highest impact causative 

factors of landslide occurrences. Optimization was conducted to gain insight into whether the accuracy of 

the landslide susceptibility map can be increased. The LSM model using four causative factors was obtained 

using the coefficient values as the equation 7. 

Z = 0.05 (Elevation) + 0.62 (Slope) + 0.32 (Land use) + 0.66 (Drainage density)-2.63  ……………………………  (7) 
 

(a) (b) 

(c) (d) 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 87  
 

Finally, the regression coefficients of predictors were imported to generate the landslide susceptibility map 

in GIS as shown in Figure 6 (a) and (b). The enlarged images in Figure 6 (c) and (d) facilitate comparison 

between these two maps. The distribution of medium to high susceptibility areas are much more wide 

spread in the map of nine causative factors than the four-factor map which is more specific to some 

location. 

3.4. Accuracy assessment of susceptibility maps  

Landslide susceptibility maps without validation are less meaningful (Chung and Fabbri, 2012). To validate 

the landslide susceptibility maps, landslides in the study area were divided into two parts based on random 

partitions. These partitions divided the area into two groups: prediction (training) and validation (testing). 

The ROC curve is a graphical representation of the trade-off between the false negative and false positive 

rates for every possible cut-off value. The area under curve (AUC) is a useful indicator to validate the 

prediction performance of the model. The success rate curve describes how well the model and controlling 

factor predict landslides (Chung and Fabbri, 2003). Accuracy is evaluated by the area under the ROC curve. 

The AUC value lies between 0.5 to 1. An area of 1 represents an excellent classifier and an area of 0.5 

represents a worthless classifier.  

In this study, both the training data (70% of 380 landslide polygons) and validation (the remaining 30% of 

380 landslide polygons) datasets were selected to assess the models. The training data was used for the 

LSM success rate, and the validation data was used for prediction. The success rate and prediction rate can 

be obtained by comparing the landslide susceptibility results at known landslide locations. In SPSS software, 

the AUC of the success rate was derived by linking the landslide index in logistic regression model and the 

CF model after using landslide data for training. Subsequently, the AUC of predictive rate was obtained 

using landslide data for validation. 

 

 

  

 

 

 

 

 

 

 
 

 

 

 

Figure 7. The area under the curve (AUC) represents A) success rate and B) prediction rate curve using four 

causative factors. C) Success rate and D) prediction rate curve using nine causative factors. 

B) 

Prediction rate curve = 

70,6% 
Success rate curve = 

78% 

A) 

Predictive rate curve = 

66,9% 

D) 

Success rate curve 

= 80% 

C) 

https://doi.org/10.14710/geoplanning.5.1.75-90


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

88 | 
 

The ROC curves for all nine causative factors or four causative factors are shown in the Figure 6. The AUC of 

the success rate curve and the predictive rate curve for nine causative factors are 0.80 and 0.67, 

respectively. It means that the accuracy of the nine causative factors is 80% for training and 66.9% for the 

validation. In the case of the four causative factors model, the success rate and predictive rate curve return 

accuracies were 78% and 70%, respectively, with an AUC of 0.78 and 0.70. This proves that the nine 

causative factor models are better at explaining the cause of landslide occurrences than the four causative 

factors model.  

The AUC curve determined with the validation data set should be approximately equal to the AUC curve 
determined with the training dataset, but it is generally lower than the success curve because the landslide 
data about validation areas are not used for modeling (Ngadisih et al., 2013). It is interesting that the 
similarity of the success rate and the predictive rate values of the four factors models is closer than the 
nine-factor model. Meten et al. (2015) stated that the proximity of success rate and predictive rate values 
are also important because it shows how the logistic regression helps predict landslides. 

4 CONCLUSION 

This study shows the selection of optimum causative factors to build an effective landslide susceptibility 

map. Four out of nine causative factor was selected by using a certainty factor analysis (elevation, slope, 

landuse and drainage density). Higher prediction accuracy was obtained from the landslide susceptibility 

map based on a combination of nine causative factors. The result shows that decreasing the number of 

causative factors may not always result in higher prediction accuracy. For instance, the combination of nine 

causative factors showed a higher success rate (80%) than the combination of four landslide factors 

(success rate 78%). This proved that the landslide susceptibility map from nine causative factors is quite 

acceptable and should have a greater degree of influence in causing landslides.  

There are some limitations and assumptions in this method. The limitations are related to the landslide 
inventory data. These data did not include the total number of the landslide events within the study area. 
Furthermore, the output LS map presents only the predicted spatial distribution of landslides and not their 
temporal probability. Despite these limitations, the produced landslide susceptibility map could be useful to 
the community and local officials. It could help design future land-use plans and implementation of 
developments. 

5 ACKNOWLEDGMENTS 

The first author would like to thank Indonesia Endowment Fund for Education (LPDP) for the scholarship 
grant to pursue the Ph.D. study.  

6 REFERENCES 

Ayalew, L., Yamagishi, H., Marui, H., & Kanno, T. (2005). Landslides in Sado Island of Japan: Part {II}. {GIS}-
based susceptibility mapping with comparisons of results from two methods and verifications. 
Engineering Geology, 81(4), 432–445. [Crossref] 

Brabb, E. E. (1985). Innovative approaches to landslide hazard and risk mapping. In International Landslide 
Symposium Proceedings, Toronto, Canada (Vol. 1, pp. 17–22). [Crossref] 

Chung, C.-J. F., & Fabbri, A. G. (2003). Validation of Spatial Prediction Models for Landslide Hazard Mapping. 
Natural Hazards, 30(3), 451–472. [Crossref] 

Chung, C.-J. F., & Fabbri, A. G. (2012). Systematic Procedures of Landslide Hazard Mapping for Risk 
Assessment Using Spatial Prediction Models. In Landslide Hazard and Risk (pp. 139–174). John Wiley 
& Sons, Ltd. [Crossref] 

Costanzo, D., Rotigliano, E., Irigaray, C., Jiménez-Perálvarez, J. D., & Chacón, J. (2012). Factors selection in 
landslide susceptibility modelling on large scale following the gis matrix method: application to the 
river Beiro basin (Spain). Natural Hazards and Earth System Sciences, 12(2), 327–340. [Crossef] 

https://doi.org/10.14710/geoplanning.5.1.75-90
https://doi.org/10.1016/j.enggeo.2005.08.004
https://doi.org/10.1016/0148-9062(87)91363-5
https://doi.org/10.1023/b:nhaz.0000007172.62651.2b
https://doi.org/10.1002/9780470012659.ch4
https://doi.org/10.5194/nhess-12-327-2012


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

 | 89  
 

CTI Engineering. (2006). Report on Urgent Survey for Consulting Engineering Services of Bawakaraeng 
Urgent Sediment Control Project, Ministry of Public Works, Indonesia. 

Cuesta, M. J. D., Sánchez, M. J., & Garci’a, A. R. (1999). Press archives as temporal records of landslides in 
the North of Spain: relationships between rainfall and instability slope events. Geomorphology, 30(1–
2), 125–132. [Crossef] 

Ercanoglu, M., & Gokceoglu, C. (2004). Use of fuzzy relations to produce landslide susceptibility map of a 
landslide prone area (West Black Sea Region, Turkey). Engineering Geology, 75(3–4), 229–250. 
[Crossef] 

Glenn, N. F., Streutker, D. R., Chadwick, D. J., Thackray, G. D., & Dorsch, S. J. (2006). Analysis of {LiDAR}-
derived topographic information for characterizing and differentiating landslide morphology and 
activity. Geomorphology, 73(1–2), 131–148. [Crossef] 

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. [Crossef] 

Guzzetti, F., Reichenbach, P., Cardinali, M., Galli, M., & Ardizzone, F. (2005). Probabilistic landslide hazard 
assessment at the basin scale. Geomorphology, 72(1–4), 272–299. [Crossef] 

Hasegawa, S., Yamanaka, M., Mimura, T., Dahal, R. K., & Nonomura, A. (2009). Drainage density as 
rainfallinduced landslides susceptibility index. In International Seminar on Hazard Management for 
Sustainable Development in Kathmandu, Nepal (pp. 72–75). 

HASNAWIR, & KUBOTA, T. (2012). Rainfall Threshold for Shallow Landslides in Kelara Watershed, Indonesia. 
International Journal of Erosion Control Engineering, 5(1), 86–92. [Crossef] 

Lee, C. F., Li, J., Xu, Z. W., & Dai, F. C. (2001). Assessment of landslide susceptibility on the natural terrain of 
Lantau Island, Hong Kong. Environmental Geology, 40(3), 381–391. [Crossef] 

Lee, S., Ryu, J.-H., Won, J.-S., & Park, H.-J. (2004). Determination and application of the weights for landslide 
susceptibility mapping using an artificial neural network. Engineering Geology, 71(3–4), 289–302. 
[Crossef] 

Lee, S., & Talib, J. A. (2005). Probabilistic landslide susceptibility and factor effect analysis. Environmental 
Geology, 47(7), 982–990. [Crossef] 

Magliulo, P., Lisio, A. Di, Russo, F., & Zelano, A. (2008). Geomorphology and landslide susceptibility 
assessment using {GIS} and bivariate statistics: a case study in southern Italy. Natural Hazards, 47(3), 
411–435. [Crossef] 

McCullagh, P., & Nelder, J. A. (1989). An outline of generalized linear models. In Generalized Linear Models 
(pp. 21–47). Springer {US}. [Crossef] 

Meten, M., PrakashBhandary, N., & Yatabe, R. (2015). Effect of Landslide Factor Combinations on the 
Prediction Accuracy of Landslide Susceptibility Maps in the Blue Nile Gorge of Central Ethiopia. 
Geoenvironmental Disasters, 2(1). [Crossef] 

Moreiras, S. M. (2005). Landslide susceptibility zonation in the Rio Mendoza Valley, Argentina. 
Geomorphology, 66(1–4), 345–357. [Crossef] 

Ngadisih, Yatabe, R., Bhandary, N. P., & Dahal, R. K. (2013). Integration of statistical and heuristic 
approaches for landslide risk analysis: a case of volcanic mountains in West Java Province, Indonesia. 
Georisk: Assessment and Management of Risk for Engineered Systems and Geohazards, 8(1), 29–47. 
[Crossef] 

Shortliffe, E. H., & Buchanan, B. G. (1975). A model of inexact reasoning in medicine. Mathematical 
Biosciences, 23(3–4), 351–379. [Crossef] 

Tsuchiya, S., Koga, S., Sasahara, K., Matsui, M., Nakahiro, M., Watanabe, H., … Yoshida, K. (2004). 
Reconnaissance of the gigantic landslide occurred on Mt. Bawakaraeng in the south Sulawesi state of 
Indonesia and unstable debris sedimentation (prompt report). Journal of the Japan Society of Erosion 
Control Engineering, 57(3), 40–46. 

Tsuchiya, S., Sasahara, K., Shuin, S., & Ozono, S. (2009). The large-scale landslide on the flank of caldera in 
South Sulawesi, Indonesia. Landslides, 6(1), 83–88. [Crossef] 

Van Westen, C. J., Rengers, N., & Soeters, R. (2003). Use of Geomorphological Information in Indirect 
Landslide Susceptibility Assessment. Natural Hazards, 30(3), 399–419. [Crossef] 

https://doi.org/10.14710/geoplanning.5.1.75-90
https://doi.org/10.1016/s0169-555x(99)00049-5
https://doi.org/10.1016/j.enggeo.2004.06.001
https://doi.org/10.1016/j.geomorph.2005.07.006
https://doi.org/10.1016/s0169-555x(99)00078-1
https://doi.org/10.1016/j.geomorph.2005.06.002
https://doi.org/10.13101/ijece.5.86
https://doi.org/10.1007/s002540000163
https://doi.org/10.1016/s0013-7952(03)00142-x
https://doi.org/10.1007/s00254-005-1228-z
https://doi.org/10.1007/s11069-008-9230-x
https://doi.org/10.1007/978-1-4899-3242-6_2
https://doi.org/10.1186/s40677-015-0016-7
https://doi.org/10.1016/j.geomorph.2004.09.019
https://doi.org/10.1080/17499518.2013.826030
https://doi.org/10.1016/0025-5564(75)90047-4
https://doi.org/10.1007/s10346-009-0143-x
https://doi.org/10.1023/b:nhaz.0000007097.42735.9e


Nurdin and Kubota / Geoplanning: Journal of Geomatics and Planning, Vol 5, No 1, 2018, 75-90 
doi: 10.14710/geoplanning.5.1.75-90 

 

90 | 
 

Walker, L. R., & Shiels, A. B. (2013). Large scales and future directions for landslide ecology. In Landslide 
Ecology (pp. 227–240). Cambridge University Press. [Crossref]  

 
 

https://doi.org/10.14710/geoplanning.5.1.75-90
https://doi.org/10.1017/cbo9780511978685.008

