Computers & Geosciences 38 (2012) 2334

Contents lists available at ScienceDirect

Computers & Geosciences

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

An easy-to-use MATLAB program (MamLand) for the assessment of landslide

susceptibility using a Mamdani fuzzy algorithm$
A. Akgun a, E.A. Sezer b, H.A. Nefeslioglu c, C. Gokceoglu d,n, B. Pradhan e
a

Middle East Technical University, Mining Engineering Department, Ankara, Turkey

Hacettepe University, Computer Engineering Department, 06800 Beytepe, Ankara, Turkey
c
General Directorate of Mineral Research and Exploration, Department of Geological Research, 06520 Balgat, Ankara, Turkey
d
Hacettepe University, Geological Engineering Department, 06800 Beytepe, Ankara, Turkey
e
Institute of Advanced Technology, Spatial and Numerical Modelling Laboratory, University Putra Malaysia, 43400 Serdang, Malaysia
b

a r t i c l e i n f o

abstract

Article history:
Received 26 January 2011
Received in revised form
13 April 2011
Accepted 15 April 2011
Available online 6 May 2011

In this study, landslide susceptibility mapping using a completely expert opinion-based approach was
applied for the Sinop (northern Turkey) region and its close vicinity. For this purpose, an easy-to-use
program, MamLand, was developed for the construction of a Mamdani fuzzy inference system and
employed in MATLAB. Using this newly developed program, it is possible to construct a landslide
susceptibility map based on expert opinion. In this study, seven conditioning parameters characterising
topographical, geological, and environmental conditions were included in the FIS. A landslide inventory
dataset including 351 landslide locations was obtained for the study area. After completing the data
production stage of the study, the data were processed using a soft computing approach, i.e., a
Mamdani-type fuzzy inference system. In this system, only landslide conditioning data were assessed,
and landslide inventory data were not included in the assessment approach. Thus, a le depicting the
landslide susceptibility degrees for the study area was produced using the Mamdani FIS. These degrees
were then exported into a GIS environment, and a landslide susceptibility map was produced and
assessed in point of statistical interpretation. For this purpose, the obtained landslide susceptibility
map and the landslide inventory data were compared, and an area under curve (AUC) obtained from
receiver operating characteristics (ROC) assessment was carried out. From this assessment, the AUC
value was found to be 0.855, indicating that this landslide susceptibility map, which was produced in a
data-independent manner, was successful.
& 2011 Elsevier Ltd. All rights reserved.

Keywords:
Mamdani fuzzy inference system
Landslide susceptibility
Geographical Information Systems (GIS)
Sinop (Turkey)

1. Introduction
According to a review of the international landslide literature
(Gokceoglu and Sezer, 2009), the most frequently used terms
in this eld are landslide susceptibility, and Geographical
Information Systems, GIS. Several possible reasons mainly
include: (a) developments in computer techniques and Geographical Information Systems; (b) increasing awareness of the
socio-economic signicance of landslides; and (c) the complex
nature of landslides. Due to the complex nature of landslides,
several methods have been applied for the assessment of landslide susceptibility at a medium scale. In general, these methods
can be classied into three main groups: statistical (e.g., Can et al.,
2005; Gokceoglu et al., 2005; Lee, 2005; Clerici et al., 2006;

$
n

Code available from: http://www.iamg.org/CGEditor/index.htm.

Corresponding author. Tel.: 903122977735; fax: 903122992034.
E-mail address: cgokce@hacettepe.edu.tr (C. Gokceoglu).

0098-3004/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cageo.2011.04.012

Duman et al., 2006; Pradhan et al., 2006, 2008; Lee and Pradhan,
2006; Akgun and Bulut, 2007; Akgun et al., 2008; Tunusluoglu
et al., 2008; Lamelas et al., 2008; Gorum et al., 2008; Turer et al.,
2008; Nefeslioglu et al., 2008a; Nefeslioglu and Gokceoglu, 2011),
soft computing (e.g., Ercanoglu and Gokceoglu, 2002, 2004; Arora
et al., 2004; Lee et al., 2004; Ercanoglu, 2005; Lui et al., 2006;
Kanungo et al., 2006; Melchiorre et al., 2008; Nefeslioglu et al.,
2008b, 2010; Pradhan et al., 2009, 2010; Caniani et al., 2008;
Yilmaz, 2009; Bedarnik et al., 2010; Pradhan, 2010; Sezer et al.,
2011), and the combination of index maps (e.g., Anbalagan and
Singh, 1996; Gokceoglu and Aksoy, 1996; Turrini and Visintainer,
1998; Ayenew and Barbieri, 2005). In the literature of this eld, it
is possible to nd attempts to assess landslide susceptibility
based on expert opinion (Juang et al., 1992; Saboya et al., 2006;
Ercanoglu et al., 2008) without constructing an inference system.
However, all of the available methods for landslide assessment at
a medium scale (  1:25,000) are associated with uncertainties
arising from a lack of knowledge and high variability (Ercanoglu
and Gokceoglu, 2002). For these reasons, certain generalizations

24

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

and simplications are applied during data-driven analyses for

landslide susceptibility mapping at a medium scale. Additionally,
these data-driven methods do not require an expert opinion.
The data-driven methods include a training phase. After training,
testing, and validation stages are employed for assessing the quality
of the model. There are a number of disadvantages associated with
this type of process. First, for selection of the training data and the
modeling the system uses different training sets. Second, the results
produced are dependent on the dataset employed. Third, the
methods used, such as decision trees, support vector machines, or
articial neural networks, are black box approaches, and the user is
completely isolated from the learning stage. To increase the role of
expert opinion in such methods and to provide new insight related
to reusable model production for landslide susceptibility assessment, the Mamdani fuzzy inference system (FIS) appears to be a
potentially suitable approach.
FIS are capable of reducing uncertainties when solving complex problems, and the advantages of FIS are summarized by
Alvarez Grima (2000) as follows:
(a) FIS allows explicit expression of knowledge of the system via
fuzzy ifthen rules;
(b) FIS deals with the subjective uncertainty (fuzziness, vagueness, and imprecision) inherent in the way experts approach
their problems;
(c) FIS can combine numerical and categorical data; and
(d) FIS provides a sound mathematical basis.

In the last decade, the Mamdani FIS has been used extensively
to solve complex and nonlinear problems of engineering geology.
An interesting and perhaps the most attractive characteristic of
fuzzy models compared with other methods commonly used in
geosciences, such as statistics, is that they are able to describe
such problems in a transparent way (Setnes et al., 1998).
A Mamdani FIS (Mamdani and Assilian, 1975) is a transparent
and expert opinion-based system. Although several studies using
this type of system (i.e., Den Hartog et al., 1997; Alvarez Grima and
Babuska, 1999; Gokceoglu, 2002; Kayabasi et al., 2003; Sonmez
et al., 2003, 2004; Nefeslioglu et al., 2003, 2006; Gokceoglu and
Zorlu, 2004; Gokceoglu et al., 2009a, b; Yagiz and Gokceoglu, 2010)
have been published related to various branches of engineering
geology, a Mamdani FIS has been applied for a limited number of
landslide susceptibility assessments to date (Lee and Juang, 1992;
Juang et al., 1992; Saboya et al., 2006). Lee and Juang (1992) and
Juang et al. (1992) proposed a quantitative assessment method to
evaluate the slope failure potential. Using fuzzy sets and linguistic
rules, they (Lee and Juang, 1992; Juang et al., 1992) assessed the
slope failure potential of relatively large areas in Hong Kong. More
than 10 years later, Saboya et al. (2006) used the Mamdani FIS for
assessment of failure susceptibility of soil slopes. Saboya et al.
(2006) constructed a Mamdani FIS by considering the opinions of
the experts. According to the authors (Saboya et al., 2006), the
conception of intelligent models by means of fuzzy logic in
geotechnical engineering is quite appropriate as long as in this
expertise area, the judgement, induction, deduction, and a sense of
importance are always present. Although useful for assessment of
landslide susceptibility, the Mamdani FIS has not been employed
commonly due to the difculties when applied in large areas
without some proper computer program. For this reason, the
purposes of the present study are to develop a MATLAB program
for the construction of a Mamdani FIS and to apply it for landslide
susceptibility evaluation at a medium scale. The present study
includes four main stages such as (i) development of MATLAB
program, (ii) data production; (iii) construction of the FIS
and landslide susceptibility mapping; and (iv) assessment of

the results. The major benets to be obtained from the study are
as follows:
(a) The Mamdani FIS can be constructed completely based on
expert opinion without an exhaustive data analysis process.
(b) An expert can reect his/her opinions on landslides in a
region in the model via fuzzy ifthen rules.
(c) The computational load is low, and results can be obtained in
a short time.
(d) Using the MATLAB program developed in this study, a
Mamdani FIS can be constructed, and production of the
degrees of landslide susceptibility can be easily obtained.

2. MATLAB program for Mamdani FIS

As noted above, several studies applying a Mamdani FIS have
been performed using MATLAB and published by a number of
researchers (i.e., Lee and Juang, 1992; Juang et al., 1992; Den
Hartog et al., 1997; Alvarez Grima and Babuska, 1999; Gokceoglu,
2002; Kayabasi et al., 2003; Sonmez et al., 2004; Saboya et al., 2006;
Nefeslioglu et al., 2006; Gokceoglu et al., 2009a; Yagiz and
Gokceoglu, 2010) in the engineering geology literature. According
to Setness et al. (1998), fuzzy models are described in the literature
as transparent and interpretable, but in fact, in many engineering
applications, they are used as a black-box tool. Various fuzzy
inference systems have been proposed. The Mamdani FIS (Fig. 1)
is one of these systems. The Mamdani FIS is perhaps the most
appealing fuzzy method to employ for engineering geology problems (Alvarez Grima, 2000).
The general characteristics of a Mamdani FIS were described
by Alvarez Grima (2000), and some general information on the
Mamdani FIS is provided in this section. A Mamdani FIS
is composed of the membership functions of input(s), fuzzy
ifthen linguistic rules, and output membership functions. The
general ifthen structure of the Mamdani algorithm is given as
Ri : if xi is Ail and. . .then y is Bi for i 1,2,. . .k,

where k is the numbers of rules; xi is the input variable (antecedent variable); and y is the output variable (consequent
variable).
The details of the composition of fuzzy relationships were given
by Ross (1995). Although many methods for the composition of
fuzzy relations (e.g., minmax, maxmax, minmin, maxmin, etc.)
are available in the literature, the maxmin and maxproduct
methods are the two most commonly used techniques (Ross,
1995). The basic form of a fuzzy composition process is given as
B A3R,
~

where A is the antecedent dened on universe X; B is the consequent

~
~
dened on universe Y; and R is the fuzzy relation characterising the
~
relation between specic inputs (X) and specic outputs (Y).
The calculation procedure of a Mamdani FIS can be given as
follows (Alvarez Grima, 2000):
(1) Compute the degree of fullment (aI) of the input for each
rule (i) by considering the degree of membership (m)

ai mAi1 X1 4mAi2 X2 4    4mAin Xn :

~

(2) For each rule, derive the output fuzzy set B using the
~i
minimum t-norm

mB0i y ai 4mB0i y:
~

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

25

Fig. 1. A generalized scheme of the Mamdani FIS structure.

(3) Aggregate the output fuzzy sets by taking the maximum:

mB0 max

i 1,2,...,k

mB0 y :

~i

The nal stage of the construction of a fuzzy inference system

is to select the defuzzication method. Aggregation of two or
more fuzzy output sets gives a new fuzzy set in the basic fuzzy
algorithm. In most cases, the result in the form of a fuzzy set is
converted into a crisp result by the defuzzication process
(Berkan and Trubatch, 1997). Due to its common use in practice
and computational simplicity, the center-of-gravity method
(Eq. (6)) is considered for use in the defuzzication process.
Z
Z
y Byydy
Bydy,
6
s

where s is the support for the fuzzy set B(y).

As stated previously, although the Mamdani FIS has been
applied to various engineering geology problems, it has not been
applied for landslide susceptibility assessment except for a few
studies (Lee and Juang, 1992; Juang et al., 1992; Saboya et al.,
2006). When considering the nature of landslide susceptibility
mapping at a medium scale, the appropriateness of the Mamdani
FIS for this problem is unargued. This is because landslide
susceptibility assessment is carried out using a very large
geographic dataset and several conditioning parameters. When
the number of parameters increases, the number of fuzzy sets and
ifthen rules also increases. Additionally, obtaining results for
each case is impossible. However, a simple and applicable tool for
the elimination of these problems has not previously been
reported. For this reason, the MATLAB program MamLand was
developed and applied to the case of Sinop (Northern Turkey).
Using MamLand, a Mamdani FIS can be generated, and a very
large dataset can be processed easily.
MamLand was developed as a MATLAB 7.8 (R2009a) (MATLAB,
2009) application. Thus, it requires MATLAB to run, as it is
developed for wrapping MATLAB functionalities by employing
useful user interfaces for the purpose of landslide susceptibility
map production. There are three MATLAB les in MamLand

(Fig. 3), including the main, modeling, and inference interfaces.

The main user interface enables the user to choose a modeling or
inference section.
The modeling interface of the modeling module of MamLand is
designed from the bottom up to form an FIS model (Fig. 2).
The main input for this module is an Excel format le
including the planned model properties. Considering the commonness and readability of Excel les, a template is designed in
Excel le format to carry the description of the model by the user
(Fig. 3a). This template is thought of as a deliverable part of
MamLand, but it can also be constructed easily by a user because
of its simplicity. After completing the necessary information for
the planned model in the Excel le template, it is supplied as the
input to the modeling module using the names of the les that
will be used for the storage of the model and the rules draft. The
storage of the model is a process that is generally employed, but
the storage of a rules draft is a special requirement for the
purpose of MamLand. A number of parameters are used in
the landslide susceptibility assessment process, and as a result,
the number of rules can be very high. For example, 192 rules are
used in this study, and there are only 2 membership functions
used for most of the parameters. It is known that the increment in
the number of fuzzy sets directly triggers the increment in the
number of rules. Additionally, an expert can express his/her
knowledge with linguistic terms that are used in the model. By
considering all of these practices, the conditional portion of the
rules is automatically produced in linguistic form and stored in a
comma-separated text le for the purpose of compatibility with
various applications (Fig. 3b). After an expert user constructs the
consequent portions of the rules, he/she can import the rules,
which are stored in an Excel le (Fig. 3c). Meanwhile, if an expert
user wants to add, delete, or modify the conditional portions of
the rules, he/she can do these activities easily because the
produced matrix is fully compatible with the MATLAB rule matrix.
After completing the rules for the model, all of the information
related to the model is stored, and the model becomes ready to
use. Thus, the rst user can describe the model in an easy manner
and transfer all relevant information to MATLAB with a minor
effort. A second user (or the rst user) has a comfortable
environment in which to describe the rules because the main
portion of the rules is automatically produced using the experts

26

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

Fig. 2. MamLand: interfaces, inputs, and outputs of the MamLand.

Fig. 3. (a) Template le section for model description, (b) le section of Automatic generated rules draft, and (c) le section of completed rules.

style. The expert lls in the consequent portions of the rules and
uploads them easily, instead of constructing all of the rules
individually from the MATLAB FIS GUI. When the numbers of
rules increase, the importance of this ability increases.
The inference interface of the inference module of MamLand is
highly simplied, as there is no complex process involved for the
users. Additionally, the users of this module do not need to be
experts on landslides. In the inference stage, the users select the
model to be used for inference; then they select the data to be
used; and nally, they specify the output le to be used for the
storage of the degrees of landslide susceptibility. After completing
these stages, the users can trigger the inference with the relevant
button on the interface, and they receive a message about the
status of the inference. After the inference is complete, if the users
wish to observe the saturation of the output values, MamLand
provides this information via a bar chart. The reason for displaying the saturation of the output values is to provide fast feedback

for the users. In this graph, the output range is divided into the
number of output fuzzy sets, and intervals are justied according
to the ranges of these fuzzy sets. However, they are used only to
divide the output range into meaningful intervals, and they do not
directly symbolize the output fuzzy sets. In this chart, the users
can see the number of the outputs located in each interval. This
saturation can reect information about the rule organization of
the model or the number of cases included in the eld data.

3. Study area
The susceptibility assessment in this study is performed in
Sinop city and its near vicinity in the northernmost part of the
Black Sea region of Turkey, located 421060 000 and 411450 000 in the
north and 341570 000 and 351150 000 in the east (Fig. 4) and occupying
an area of approximately 392 km2.

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

27

Rainfall is the main source of water in the study area, and it is

the most important reason for triggering landslides. According to
meteorological data for the period 19702009, the study area
receives an average annual rainfall of 672 mm. Most of the rainfall
occurs during October, with a mean value of 91.3 mm (General
Directory of Meteorological Services of Turkey, 2010). Topographical and morphological determinations using a digital elevation
model (DEM) produced by interpolating the contour lines of a
topographic sheet at a scale of 1:25,000 (Fig. 5) showed that
topographical elevations range from 0 to 1040 m, and the slope
angle reaches 881 in some locations in the area. Whereas the
northern part of the area exhibits a gently sloping topography, the
slopes in the southern region are steep, due to lithological
characteristics.
The geological map used in this study had a scale of 1:100,000,
and it was prepared by the General Directorate of Mineral
Research and Explorations (2008) (Fig. 6). In the study region,
various lithological units crop out ranging from the Campanian
lower Maastrichtian to the Quaternary. Since the late
Campanian, the basin has been affected by compressional tectonic
ysch-type sediments of the Cankurtaran formation and its

Fig. 5. Digital elevation model (DEM) of the study area.

Fig. 4. Location of the study area.

volcanic component, which are the products of arc magmatism

deposition (Gedik et al., 1981). Following the deposition of the
Cankurtaran formation, during the late MaastrichtianPaleocene,
clayey limestone, marl (Akveren formation) in late Maastrichtian
Paleocene sandstone, marl (Atbasi formation) in Lutatian siltstone, and sandstoneclaystone (Kusuri formation) were deposited (Gedik et al., 1981). This sedimentary sequence from bottom
to top indicates regressive development in general. The sandstone
beds concentrated at the lower levels of the Kusuri formation
have been designated an Ayancik member. These tectonic and
volcanic activities in the north have caused the stratigraphy
around the Sinop Peninsula to differ somewhat from that of
neighboring areas. In the late CampanianMaastrichtian period,
the Hamsaros formation, associated with volcanic activity, and
the Karaada formation, consisting of shallow reefal limestone,
were deposited (Gedik et al., 1981). After the Middle Eocene, the
area became land. Around Sinop and Gerze, the Sinop formation
from the Miocene, which exhibits various facies of shore, tidal,
lagoonal, and deltaic environments, the Sarkum formation of the
late Plioceneearly Pleistocene, consisting of meandering river
and oodplain sediments, and the Kale formation of the late
Pleistocene, consisting of shore sediments, were deposited (Barka
et al., 1985). The Holocene is represented by alluvial and beach
sediments deposited in the northern parts of the study area.
The study area is one of the most landslide-prone areas of
Turkey, and it has been subjected to a number of landslides.
According to the classication proposed by Cruden and Varnes
(1996), the type of these landslides is generally active deep-seated
rotational, and they usually occur on slopes above rivers and creeks.
Additionally, shoreline landslides are also common in the area.

28

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

Fig. 6. Simplied lithology map of the study area (General Directorate of Mineral
Research and Explorations, 2008).

The occurrence of landslides is mainly controlled by the lithology

and slope gradient in the area. Surface-water erosion and undercutting of slopes by both surface water and cliff erosion are also
considered as other important landslide-controlling and triggering
factors. Due to the weathering of lithological units in the area, the
occurrence of clay is common, and the presence of clays also
contributes to the occurrence of landslides.

4. Data production
Among the parameters included in this model, a reliable
landslide inventory describing the type, activity, and spatial
distribution of all landslides is essential for landslide susceptibility assessment, as it is the fundamental component of the
assessments. (Soeters and Van Westen, 1996). In Turkey, a landslide inventory project at a national scale has been conducted by
the Geological Research Department of the General Directorate of
Mineral Research and Exploration, MTA. When preparing the
landslide inventory, to identify landslides, vertical black and
white aerial photographs at a medium scale of 1:35,000 from
1955 to 1956 were used (Nefeslioglu et al., 2010). A landslide
database of the study area is included in this national inventory
data, and 351 landslide locations were taken into account in the
study region. Because the modeling approach employed in this
study is independent of landslide data, the landslide inventory
was only used for the validation process at the end of the study. In
previous studies, different sampling methodologies have been
applied in using landslide data for landslide susceptibility

mapping (Nefeslioglu et al., 2008a; Yilmaz, 2010a). In these

studies, it was suggested that accumulation/depletion zones or
rupture zones of the sliding masses should not be considered as
an assessment parameter in either the modeling or the validation
stages. For this purpose, initially considering a spatial resolution
of 25  25 m2, buffer zones were drawn over 25 m around the
landslide polygons. Then, only the main scarp portions of these
buffered polygons, which are recognized as the landslide occurrence area, were distinguished for use in the validation stage of
the predicted results.
To produce a landslide susceptibility map for the study area
using the Mamdani FIS, seven landslide conditioning parameters,
including altitude, lithology, slope gradient, curvature, a normalized difference vegetation index (NDVI), a stream power index
(SPI), and a topographical wetness index (TWI), were used (Fig. 7),
and all of these data were produced in raster format with a pixel
size of 25  25 m2 to be compatible with the spatial resolution.
The lithology map was rst digitized using the 1:100,000 scale
published map, and then the obtained vector data were converted
into raster data. After obtaining the digital lithology data, the
relationships between the lithological units and the occurrence of
landslides were examined. On close inspection of Table 1, it can
clearly be seen that many of the landslides in the study region
(76.5%) occurred in three formations: the Sinop formation, the
Atbasi formation, and the Akveren formation. The other lithological units associated with landslide occurrence, the Kusuri
formation (13.66%) and the Cankurtaran formation (4.25%), were
also evaluated as having landslide occurrence potential.
Topographical parameters including altitude, slope gradient,
curvature, SPI, and TWI were produced from the DEM of the study
area (Fig. 6). Initially, the DEM was constructed by implementation of 1:25,000 scale topographical map contours using ArcGIS
9.3 software (ArcGIS Version 9.3, 2008). Altitude was obtained
from the DEM as grid-type data. Altitude was evaluated as a good
indicator for landslide susceptibility and has been used by many
researchers (Pachauri and Pant, 1992; Ercanoglu and Gokceoglu,
2002; Nefeslioglu et al., 2010). For this reason, in this study,
altitude was considered as a landslide-controlling parameter.
According to the relationship between altitude and landslide
occurrence, the mean altitude value on the grid cells with landslides was found to be 954.95 m ( 7190.39) (Table 2).
It is clear that the slope gradient is one of the most important
topographical parameters controlling landslide occurrence, and
for this reason, these data are taken into consideration in almost
all landslide susceptibility, hazard, and risk assessment studies
(i.e., Atkinson and Massari, 1998; Can et al., 2005; Nefeslioglu
et al., 2008a, 2010; Yilmaz, 2010b; Akgun and Turk, 2010;
Pradhan et al., 2010), and the slope gradient is also considered
as a landslide-controlling parameter in the present study. Based
on the descriptive statistics given in Table 2, it can be seen that
landslides in the area typically occur on gentle slopes. Topography is the main factor controlling the spatial variation of slope
stabilities. It affects the spatial distribution of soil moisture, and
groundwater ow often follows surface topography (Rodhe and
Seibert, 1999; Zinko et al., 2005; Yilmaz, 2010b). Topographic
indices have, therefore, been used to describe spatial soil moisture
patterns (Moore et al., 1991). One of these indices is the TWI
developed by Beven and Kirkby (1979) within a runoff model. To
calculate TWI values, Moore et al. (1991) proposed the equation
TWI lnAs =tan b,

7
2

where As is the specic catchment area (m /m) and b is the slope

gradient (in degrees). According to Wood et al. (1990), the
variation in topographical components is often much greater than
the local variability in soil transmissivity, and Eq. (7) can be used
to calculate the TWI. By comparing the obtained TWI values with

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

29

Fig. 7. Landslide conditioning parameters used in this study (A) altitude; (B) slope gradient; (C) curvature; (D) normalized difference vegetation index (NDVI); (E) stream
power index (SPI); and (F) topographical wetness index (TWI)).

Table 1
The distribution of the geological formations with respect to landslide in the study area.
Formation

Alluvium
Beach deposits
Kale formation
Sarkum formation
Sinop formation
Kusuri formation
Ayanck member
Atbas- formation
Karaada limestone
Akveren formation
Hamsaros formation
Cankurtaran formation

Symbol

Qal
QpcQpk
TplQs
Tms
Tek
Teka
Tpea
Ktpk
Ktpa
Kh
Kc

Grid cells with landslides

All grid cells

Frequency ratio (a/b)

Frequency

% (a)

Frequency

% (b)

0
94
246
3607
27373
12339
0
24326
0
17381
0
3841

0
0.10
0.27
4.00
30.31
13.66
0
26.94
0.00
19.25
0.00
4.25

72935
1513
3348
56231
133197
94246
1333
117063
88
107343
20588
18905

11.64
0.24
0.53
8.97
21.25
15.03
0.21
18.68
0.01
17.13
3.28
3.03

the landslide occurrence (Table 2), it can clearly be seen that the
mean TWI values were commonly calculated on the landslide
bodies. This can be assumed to be a result of the availability of

0.00
0.41
0.50
0.44
1.42
0.90
0.00
1.44
0.00
1.12
0.00
1.40

lithological units with a relatively high permeability and low

surface runoff. The other secondary derivative of the DEM is the
SPI. This index is used to describe potential ow erosion and

30

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

Table 2
General descriptive statistics of topographical variables with respect to landslides.
Data

Variable

Min.

Max.

Grid cells with landslides (89,207)

Altitude (m)
Slope gradient (1)
Curvature
NDVI
SPI
TWI

0.82
0.00
 0.62
 0.59
0.00
2.72

954.94
62.74
0.54
0.72
37056.96
15.06

226.22
12.21
 0.18
0.22
310.47
6.30

190.39
7.35
0.08
0.18
859.90
1.39

Grid cells without landslides (626,271)

Altitude (m)
Slope gradient (1)
Curvature
NDVI
SPI
TWI

0.00
0.00
 0.91
 0.72
0
2.20

1040
87.94
1.02
0.77
41551.45
17.35

204.22
10.60
15  10  6
0.26
177.13
6.43

210.51
9.31
0.20
0.30
643.43
1.75

related landscape processes (Moore et al., 1991). The SPI is

calculated from the formula
SPI As  tan b,

Mean

SD

very close to the mean value obtained for grid cells with landslides. Hence, it can be concluded that the landslides in the study
area occur where the vegetation cover density is in the moderate
class, representing shrub and grassland elds in the region.

where As is the specic catchment area (m /m) and b is the slope

gradient in degrees. As the specic catchment area and gradient
increase, the amount of water contributed by upslope areas and
the velocity of water ow also increase.
Nefeslioglu et al. (2010) emphasized that maximum SPI values
are calculated in drainage channels, and for this reason, the values
of the SPI for grid cells with landslides should not be expected to
be high. According to the statistical results given in Table 2, this
expectation is also reasonable for this study. This is because while
the differences between the maximum and the minimum SPI
values for the grid cells with landslides are high, the mean values
for the grid cells with landslides are relatively low. Thus, it can be
concluded that the SPI parameter constitutes a vital landslideconditioning parameter in the study area.
The next type of DEM-derived topographical data used in this
study is for curvature. The curvature represents the morphology
of the topography (Wilson and Gallant, 2000). Whereas a positive
curvature indicates that the surface is upwardly convex at that
terrain surface, a negative curvature indicates that the surface is
upwardly concave at the terrain surface. A value of zero denes a
straight terrain surface. The minimum, maximum, and mean
curvature values obtained for the grid cells with landslides were
0.62, 0.54, and  0.18, respectively (Table 2). According to these
results, it can be concluded that many landslides occurred at
concave terrain surfaces in the study area. The curvature, SPI, and
TWI data were produced using DIGEM 2.0 software, which was
developed for digital terrain analysis by Conrad (2002), and then
exported into ArcGIS 9.3 software to be used for model
construction.
In this study, the NDVI was used as an environmental parameter. The NDVI is a measure of surface reectance and gives a
quantitative estimate of vegetation growth and biomass (Hall
et al., 1995). Very low values of the NDVI (0.1 and below)
correspond to barren areas, sand, or snow. Moderate values
(0.20.3) represent shrub and grasslands, whereas high values
(0.60.8) indicate temperate and tropical rainforests (Weier and
Herring, 2005). The NDVI was calculated from the formula
NDVI IR2R=IR R,

where IR is the infrared portion of the electromagnetic spectrum

and R is the red portion of the electromagnetic spectrum. Based
on the statistical values given in Table 2, the minimum and
maximum NDVI values for grid cells without landslides are 0.72
and 0.77, respectively, with a mean of 0.26. The mean NDVI value
for grid cells with landslides is calculated as 0.22, and this value is

5. Landslide susceptibility mapping, and validation of the

model
In this study, a Mamdani FIS is constructed. The Mamdani FIS for
the assessment of landslide susceptibility for Sinop (Northern
Turkey) includes a total of 7 inputs: altitude, lithology, slope,
curvature, NDVI, SPI, and TWI (Fig. 7). Except for lithology, all of
the inputs are constructed using two membership functions (Fig. 8).
Lithology is formed by three crisp membership functions. A total of
12 different lithology types are cropped out in the study area. These
lithologies are reclassied under three classes in the present study.
When applying these classications, the landslide density and the
behavior of the lithology type in opposing to landsliding are
considered. The Sinop formation, Kusuri formation, Atbasi formation, and Akveren formation are assigned to the high membership
function because the landslide densities in these formations vary
between 13.66% and 30.31%. The second class, moderate, includes
two formations, the Sarikum formation and the Cankurtaran formation. The landslide densities of these formations are 4.00% and 4.25%,
respectively. The other formations are not associated with landslides
or have a negligible landslide density (o1.00%). To minimize the
uncertainty, a 50% overlap is applied between the fuzzy sets for each
input parameter, and triangular membership functions are used for
each fuzzy set. To provide a generalization for the study area, the
minimum number of the fuzzy sets is considered. This also affects
the number of the ifthen rules. The output includes ve fuzzy sets
in the form of triangular membership functions (Fig. 9). One of the
main parts of a Mamdani FIS is the fuzzy ifthen rules. In this study,
a total 192 ifthen rules are used. The ifthen rules are described
only using expert opinion. When describing the ifthen rules, the
following considerations obtained from general knowledge of landslides and the eld observations are used:
(a) Landslides in the study area occur in association with four
lithologies (classied as high). For this reason, if the rule
includes high in the lithology input, the output is accepted
as high or very high.
(b) The other landslide conditioning factors are considered
equally. If three inputs are positive and three inputs are
negative for landslide occurrence, the output is assigned as
low.
(c) If four inputs are positive and two inputs are negative for
landslide occurrence, the output is accepted as moderate.

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

31

Fig. 8. Mamdani FIS application scheme, followed in this study.

(d) If ve inputs are positive and one input is negative for

landslide occurrence, the output is accepted as high.
(e) If six inputs are positive for landslide occurrence, the output
is accepted as very high.
(f) If the lithology is low, the output is accepted as very low.

Fig. 9. Landslide susceptibility index map obtained by Mamdani-type FIS

methodology.

After modeling the Mamdani FIS and producing the susceptibility values using MamLand, the susceptibility values were
stored as a text le. Then, this le was exported into ArcGIS
9.3 software as point data containing output database information. After exporting the data into point-type data, they were
converted into a raster le type so that they could be evaluated,
and a landslide susceptibility map was obtained (Fig. 9).
For visual interpretation of this map, the necessity of classifying data into categorical susceptibility classes arose. Four data
classiers for this purpose are reported in the literature: equal
interval, standard deviation, natural break, and quantile classication (Ayalew et al., 2005; Akgun et al., 2008). When considering
all of these classiers, the distribution of the data should be taken
into account because class intervals change based on the chosen
classier (Ayalew et al., 2005). For example, if the data distribution is close to normal, equal interval or standard deviation
classiers should be used. If the data distribution has a positive
or negative skewness, the quantile or natural break distribution
classiers could be chosen. In this study, before choosing the best
data classier for the obtained data, the data distribution histogram was taken into consideration (Fig. 10). Then, all of the
classiers noted above were applied to the data. The equal
interval and standard deviation classiers were found to be
generally unsuccessful due to the fact that the data distribution
skewed positively. Additionally, the natural break classier
classied the data into categorical susceptibility classes that were
inferior to the quantile classier. Therefore, the quantile data
classication approach was chosen, and the landslide susceptibility index map was classied into ve susceptibility classes:
very low, low, moderate, high, and very high (Fig. 11).

32

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

Fig. 10. Histogram of the numerical predicted indices.

Fig. 12. AUC assessment of the Mamdani-type FIS-based landslide susceptibility map.

by varying the decision threshold (Brenning, 2005) In the AUC

method, the AUC, with values ranging from 0.5 to 1.0, is used to
assess the accuracy of the constructed model. The AUC denes the
quality of the probabilistic model by describing its ability to
reliably predict an occurrence or nonoccurrence event (Remondo
et al., 2003; Nandi and Shakoor, 2009). An ideal model presents an
AUC value close to 1.0, whereas a value close to 0.5 indicates
inaccuracy in the model (Fawcett, 2006). The AUC value of the
ROC curve for the Mamdani FIS model was found to be 0.855
(Fig. 12).

6. Conclusions

Fig. 11. Landslide susceptibility map obtained by reclassication of the landslide

susceptibility index map.

The validation of predicted data is one of the most important

parts of a probability-based map production process (Remondo
et al., 2003). In the landslide susceptibility literature, although
several methods are available for this, the area under the curve
obtained from the ROC (receiver operating characteristics) plot is
the most preferred and applicable type of statistical assessment
for this purpose (Lee, 2005; Yilmaz, 2010a, b; Akgun and Turk,
2010; Nefeslioglu et al., 2010). ROC (Zweig and Champbell, 1993)
plots should be drawn. These curves are obtained by plotting all
combinations of sensitivities (on the y-axis) and proportions of
false-negatives (1-specity; on the x-axis) that may be obtained

A great number of landslide susceptibility assessment papers

have been published in landslide literature. Several different
approaches for the assessment of landslide susceptibility have
been employed. Almost all of these approaches are data driven.
Although the Mamdani FIS is an expert-based system, it had
previously been applied to a limited number of landslide susceptibility assessment studies (Lee and Juang, 1992; Juang et al., 1992;
Saboya et al., 2006). However, it is a suitable system for landslide
susceptibility assessment because it has a considerable capacity
to model complex and nonlinear systems. The main reason for its
lack of use has been the absence of a suitable computer program
in which to implement it. It is possible to construct a Mamdani FIS
using MATLAB. However, it is not possible to run a model
including a very large geographic dataset using existing MATLAB
software. To eliminate this problem, the MATLAB program
MamLand was developed in this study. This program will assist
landslide experts in assessing regional landslide susceptibility by
directly employing expert opinion. MamLand is designed to be
easy to use.
Sinop (Northern Turkey), which is a landslide-prone area, was
considered as a case study area. The landslides identied in the
study area are generally of an active deep-seated rotational type,
and they usually occur on gentle slopes. Shoreline landslides are
also observed in the area. Considering the main characteristics of
the landslides and the eld observations, a total of 7 inputs were
selected for the model: altitude, lithology, slope gradient, curvature, NDVI, SPI, and TWI. Except for lithology, all of the inputs are

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

formed by two membership functions, while lithology includes

three input membership functions. To minimize the uncertainty, a
50% overlap is applied between the membership functions. The
output, the landslide susceptibility degree, is formed by ve
membership functions (Fig. 9). The FIS constructed for the study
area has a total of 192 ifthen rules. The ifthen rules are
described only using expert opinion. The AUC value of the ROC
curve for the Mamdani FIS model constructed in the present study
was found to be 0.855. Finally, we conclude that this methodology
will present new opportunities for landslide researchers, as they
can employ their knowledge in this model via membership
functions and ifthen rules.

Appendix A. Supplementary material

Supplementary data associated with this article can be found
in the online version at doi:10.1016/j.cageo.2011.04.012.

References
Akgun, A., Bulut, F., 2007. GIS-based landslide susceptibility for Arsin-Yomra
(Trabzon, North Turkey) region. Environmental Geology 51, 13771387.
Akgun, A., Dag, S., Bulut, F., 2008. Landslide susceptibility mapping for a landslideprone area (Findikli, NE of Turkey) by likelihood frequency ratio and weighted
linear combination models. Environmental Geology 54 (6), 11271143.
Akgun, A., Turk, N., 2010. Landslide susceptibility mapping for Ayvalik (Western
Turkey) and its vicinity by multicriteria decision analysis. Environmental Earth
Sciences 61 (3), 595611.
Alvarez Grima, M., 2000. Neuro-Fuzzy Modeling in Engineering Geology. A.A.
Balkema, Rotterdam, 244pp.
Alvarez Grima, M., Babuska, R., 1999. Fuzzy model for the prediction of unconned
compressive strength of rock samples. International Journal of Rock Mechanics
and Mining Science 36 (3), 339349.
Anbalagan, R., Singh, B., 1996. Landslide hazard and risk assessment mapping of
mountainous terrainsa case study from Kumaun Himalaya, India. Engineering Geology 43 (4), 237246.
ArcGIS (Version 9.3), 2008. Integrated GIS Software. ESRI, CA.
Arora, M.K., Gupta, A.S.D., Gupta, R.P., 2004. An articial neural network approach
for landslide hazard zonation in the Bhagirathi (Ganga) Valley, Himalayas.
International Journal of Remote Sensing 25 (3), 559572.
Atkinson, P.M., Massari, R., 1998. Generalized linear modelling of susceptibility to
landsliding in the central Appennines, Italy. Computers and Geosciences 24
(4), 373385.
Ayalew, L., Yamagishi, H., Marui, H., Kanno, T., 2005. Landslide in Sado Island of
Japan: part II. GIS-based susceptibility mapping with comparison of results
from two methods and verications. Engineering Geology 81, 432445.
Ayenew, T., Barbieri, G., 2005. Inventory of landslides and susceptibility mapping
in the Dessie area, northern Ethiopia. Engineering Geology 77 (12), 115.
Barka, A., Sutcu, Y.F., Gedik, I., Tekin, T.F., Arel, E., Ozdemir, M, Erkal, T., 1985. Sinop
Nukleer Enerji Santraline Iliskin Jeolojik Arastirmalar Sonuc Raporu: MTA
Derleme Rapor No. 7963. Ankara, Turkey, 50pp, unpublished.
Bednarik, M., Magulova, B., Matys, M., Marschalko, M., 2010. Landslide susceptibility assessment of the KralovanyLiptovsky Mikulas railway case study.
Physics and Chemistry of the Earth, Parts A/B/C 35 (35), 162171.
Berkan, R.C., Trubatch, S.L., 1997. Fuzzy System Design Principles, Building Fuzzy
IfThen Rule Bases. The Institute of Electrical and Electronics Engineers, New
York, 496pp.
Beven, K.J., Kirkby, M.J., 1979. A physically based, variable contributing area model
of basin hydrology. Hydrological Sciences Bulletin 24, 4369.
Brenning, A., 2005. Spatial prediction models for landslide hazards: review,
comparison and evaluation. Natural Hazards and Earth System Sciences 5,
853862.
Can, T., Nefeslioglu, H.A., Gokceoglu, C., Sonmez, H., Duman, T.Y., 2005. Susceptibility assessment of shallow earthows triggered by heavy rainfall at three
subcatchments by logistic regression analyses. Geomorphology 72, 250271.
Caniani, D., Pascale, S., Sdao, F., Sole, A., 2008. Neural networks and landslide
susceptibility: a case study of the urban area of Potenza. Natural Hazards 45,
5572.
Clerici, A., Perego, S., Tellini, C., Vescovi, P., 2006. A GIS-based automated
procedure for landslide susceptibility mapping by the conditional analysis
method: the Baganza valley case study (Italian Northern Apennines). Environmental Geology 50, 941961.
Conrad, O., 2002. Digitales Gelande-modell (DiGeM) Terrain Analysis Software.
/http://www.geogr.unigoettingen.de/pg/saga/digem/S (accessed 18.04.06.).
Cruden, D.M., Varnes, D.J., 1996. Landslide types and processes. In: Turner, A.K.,
Shuster, R.L. (Eds.), Landslides: Investigation and Mitigation. Transportation
Research Board Special Report 247. pp. 3675.

33

Den Hartog, M.H., Babuska, R., Deketh, H.J.R., Alvarez Grima, M., Verhoef, P.N.W.,
Verbruggen, H.B., 1997. Knowledge-based fuzzy model for performance
prediction of a rock cutting trencher. International Journal of Approximate
Reasoning 16, 4366.
Duman, T.Y., Can, T., Gokceoglu, C., Nefeslioglu, H.A., Sonmez, H., 2006. Application
of logistic regression for landslide susceptibility zoning of Cekmece Area,
Istanbul, Turkey. Environmental Geology 51, 241256.
Ercanoglu, M., 2005. Landslide susceptibility assessment of SE Bartin (West Black
Sea region, Turkey) by articial neural networks. Natural Hazards and Earth
System Sciences 5, 979992.
Ercanoglu, M., Gokceoglu, C., 2002. Assessment of landslide susceptibility for a
landslide-prone area (north of Yenice, NW Turkey) by fuzzy approach.
Environmental Geology 41, 720730.
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, 229250.
Ercanoglu, M., Kasmer, O., Temiz, N., 2008. Adaptation and comparison of expert
opinion to analytical hierarchy process for landslide susceptibility mapping.
Bulletin of Engineering Geology and the Environment 67, 565578.
Fawcett, T., 2006. An introduction to ROC analysis. Pattern Recognition Letters 27,
861874.
Gedik, A., Ozbudak, N., Korkmaz, S., Iztan, H., Agridag, D.S., 1981. Sinop havzasinin
jeolojisi ve petrol olanaklari ile ilgili on sonuclar. In: TJK, 35. Bilim ve Teknik
Kurultayi, Bildiri Ozetleri, Ankara. pp. 3536.
General Directory of Meteorological Services of Turkey, 2010. /http://www.dmi.
gov.trS (accessed 25.12.10.).
General Directorate of Mineral Research and Explorations, 2008. 1:100,000 Scale
Geology Maps of Turkey, Sinop D33-E33 Sheets. MTA Publications, Ankara.
Gokceoglu, C., 2002. A fuzzy triangular chart to predict the uniaxial compressive
strength of the Ankara agglomerates from their petrographic composition.
Engineering Geology 66 (12), 3951.
Gokceoglu, C., Aksoy, H., 1996. Landslide susceptibility mapping of the slopes in
the residual soils of the Mengen region (Turkey) by deterministic stability
analyses and image processing techniques. Engineering Geology 44, 147161.
Gokceoglu, C., Sezer, E., 2009. A statistical assessment on international landslide
literature (19452008). Landslides 6, 345351.
Gokceoglu, C., Sonmez, H., Nefeslioglu, H.A., Duman, T.Y., Can, T., 2005. The 17
March 2005 Kuzulu landslide (Sivas, Turkey) and landslide-susceptibility map
of its near vicinity. Engineering Geology 81, 6583.
Gokceoglu, C., Sonmez, H., Zorlu, K., 2009a. Estimating the uniaxial compressive
strength of some clay bearing rocks selected from Turkey by nonlinear
multivariable regression and rule-based fuzzy models. Expert Systems 26
(2), 176190.
Gokceoglu, C., Zorlu, K., 2004. A fuzzy model to predict the uniaxial compressive
strength and the modulus of elasticity of a problematic rock. Engineering
Applications of Articial Intelligence 17 (1), 6172.
Gokceoglu, C., Zorlu, K., Ceryan, S., Nefeslioglu, H.A., 2009b. A comparative study
on indirect determination of degree of weathering of granites from some
physical and strength parameters by two soft computing techniques. Materials
Characterization 60, 13171327.
Gorum, T., Gonencgil, B., Gokceoglu, C., Nefeslioglu, H.A., 2008. Implementation of
reconstructed geomorphologic units in landslide susceptibility mapping: the
Melen Gorge (NW Turkey). Natural Hazards 46 (3), 323351.
Hall, F.G., Towhshend, J.R., Engman, E.T., 1995. Status of remote sensing algorithms
for estimation of land surface state parameters. Remote Sensing of Environment 51, 138156.
Juang, C.H., Lee, D.H., Sheu, C., 1992. Mapping slope failure potential using fuzzy
sets. Journal of Geotechnical Engineering ASCE 118, 475494.
Kanungo, D.P., Arora, M.K., Sarkar, S., Gupta, R.P., 2006. A comparative study of
conventional, ANN black box, fuzzy and combined neural and fuzzy weighting
procedures for landslide susceptibility zonation in Darjeeling Himalayas.
Engineering Geology 85 (34), 347366.
Kayabasi, A., Gokceoglu, C., Ercanoglu, M., 2003. Estimating the deformation
modulus of rock masses: a comparative study. International Journal of Rock
Mechanics and Mining Science 40 (1), 5563.
Lamelas, M.T., Marinoni, O., Hoppe, A., De La Riva, J., 2008. Doline probability map
using logistic regression and GIS technology in the central Ebro Basin (Spain).
Environmental Geology 54, 963977.
Lee, D.H., Juang, C.H., 1992. Evaluation of failure potential in mudstone slopes
using fuzzy sets. ASCE Geotechnical Special Publication 31, Stability and
Performance of Slopes and Embankment-II, vol. 2; pp. 11371151.
Lee, S., 2005. Application of logistic regression model and its validation for
landslide susceptibility mapping using GIS and remote sensing data. International Journal of Remote Sensing 26, 14771491.
Lee, S., Pradhan, B., 2006. Probabilistic landslide hazards and risk mapping on
Penang Island, Malaysia. Earth System Science 115 (6), 661672.
Lee, S., Ryu, J.H., Won, J.S., Park, H., 2004. Determination and application of the
weights for landslide susceptibility mapping using an articial neural network. Engineering Geology 71, 289302.
Lui, Y., Guo, H.C., Zou, R., Wang, L.J., 2006. Neural network modelling for regional
hazard assessment of debris ow in Lake Qionghai Watershed, China.
Environmental Geology 49, 968976.
Mamdani, E.H., Assilian, S., 1975. An experiment in linguistic synthesis with a
fuzzy logic controller. International Journal of ManMachine Studies 7 (1),
113.
MATLAB, 2009. Users Guide Version 7.8, R2009a. MathWorks Co., USA.

34

A. Akgun et al. / Computers & Geosciences 38 (2012) 2334

Melchiorre, C., Matteucci, M., Azzoni, A., Zanchi, A., 2008. Articial neural networks and cluster analysis in landslide susceptibility zonation. Geomorphology 94, 379400.
Moore, I.D., Grayson, R.B., Ladson, A.R., 1991. Digital terrain modeling: a review of
hydrological, geomorphological and biological applications. Hydrological Processes 13 (4), 305320.
Nandi, A., Shakoor, A., 2009. A GIS-based landslide susceptibility evaluation using
bivariate and multivariate statistical analyses. Engineering Geology 110,
1120.
Nefeslioglu, H.A., Duman, T.Y., Durmaz, S., 2008a. Landslide susceptibility mapping
for a part of tectonic Kelkit Valley (Eastern Black Sea region of Turkey).
Geomorphology 94 (34), 401418.
Nefeslioglu, H.A., Gokceoglu, C., 2011. Probabilistic risk assessment in medium
scale for rainfall induced earthows: Catakli catchment area (Cayeli, Rize,
Turkey). Mathematical Problems in Engineering Article ID 280431.
Nefeslioglu, H.A., Gokceoglu, C., Sonmez, H., 2008b. An assessment on the use of
logistic regression and articial neural networks with different sampling
strategies for the preparation of landslide susceptibility maps. Engineering
Geology 97, 171191.
Nefeslioglu, H.A., Gokceoglu, C., Sonmez, H., 2003. A Mamdani model to predict the
weighted joint density. Lecture Notes in Articial Intelligence 2773,
10521057.
Nefeslioglu, H.A., Gokceoglu, C., Sonmez, H., 2006. Indirect determination of block
size by empirical and fuzzy models: Supren (Eskisehir, Turkey) Marbles.
Engineering Geology 85, 251269.
Nefeslioglu, H.A., Sezer, E., Gokceoglu, C., Bozkir, A.S., Duman, T.Y., 2010. Assessment of landslide susceptibility by decision trees in the metropolitan area of
Istanbul, Turkey. Mathematical Problems in Engineering, 115. doi:10.1155/
2010/901095.
Pachauri, A.K., Pant, M., 1992. Landslide hazard mapping based on geological
attributes. Engineering Geology 32, 81100.
Pradhan, B., 2010. Remote sensing and GIS-based landslide hazard analysis and
cross-validation using multivariate logistic regression model on three test
areas in Malaysia. Advances in Space Research 45 (10), 12441256.
Pradhan, B., Lee, S., Buchroithner, M.F., 2009. Use of geospatial data for the
development of fuzzy algebraic operators to landslide hazard mapping: a case
study in Malaysia. Applied Geomatics 1, 315.
Pradhan, B., Lee, S., Mansor, S., Buchroithner, M.F., Jallaluddin, N., Khujaimah, Z.,
2008. Utilization of optical remote sensing data and geographic information
system tools for regional landslide hazard analysis by using binomial logistic
regression model. Applied Remote Sensing 2, 111.
Pradhan, B., Sezer, E.A., Gokceoglu, C., Buchroithner, M.F., 2010. Landslide susceptibility mapping by neuro-fuzzy approach in a landslide prone area (Cameron
Highland, Malaysia). IEEE Transactions on Geoscience and Remote Sensing 48
(12), 41644177.
Pradhan, B., Singh, R.P., Buchroithner, M.F., 2006. Estimation of stress and its use in
evaluation of landslide prone regions using remote sensing data. Advances in
Space Research 37, 698709.
Remondo, J., Gonzalez, A., Diaz de Teran, J.R., Cendrero, A., Fabbri, A., Cheng, C.F.,
2003. Validation of landslide susceptibility maps: examples and applications
from a case study in Northern Spain. Natural Hazards 30 (3), 437449.
Rodhe, A., Seibert, J., 1999. Wetland occurrence in relation to topography: a test of
topographic indices as moisture indicators. Agricultural and Forest Meteorology 9899, 325340.

Ross, T.J., 1995. Fuzzy Logic with Engineering Applications. McGraw-Hill Inc., New
York 579pp.
Saboya Jr., F., Alves, M.G., Pinto, W.D., 2006. Assessment of failure susceptibility of
soil slopes using fuzzy logic. Engineering Geology 86, 211224.
Setnes, M., Babuska, R., Verbruggen, H.B., 1998. Rule-based modeling: precision
and tranparency. IEEE Transactions on Systems Man and Cybernetics, Part C
28, 165169.
Sezer, E.A., Pradhan, B., Gokceoglu, C., 2011. Manifestation of an adaptive neurofuzzy model on landslide susceptibility mapping: Klang valley, Malaysia.
Expert Systems with Applications 38 (7), 82088219.
Soeters, R., Van Westen, C.J., 1996. Slope instability recognition analysis and
zonation. In: Turner, K.T., Schuster, R.L. (Eds.), Landslides: Investigation and
Mitigation. Transportation Research Board National Research Council,
Washington, DC, pp. 129177 Special Report No. 247.
Sonmez, H., Gokceoglu, C., Ulusay, R., 2003. An application of fuzzy sets to the GSI
system in rock engineering. Engineering Applications of Articial Intelligence
16 (3), 251269.
Sonmez, H., Tuncay, E., Gokceoglu, C., 2004. Models to predict the uniaxial
compressive strength and the modulus of elasticity for Ankara agglomerates.
International Journal of Rock Mechanics and Mining Science 41 (5), 717729.
Tunusluoglu, M.C., Gokceoglu, C., Nefeslioglu, H.A., Sonmez, H., 2008. Extraction of
potential debris source areas by logistic regression technique: a case study
from Barla, Besparmak and Kapi mountains (NW Taurids, Turkey). Environmental Geology 54, 922.
Turer, D., Nefeslioglu, H.A., Zorlu, K., Gokceoglu, C., 2008. Assessment of geo
environmental problems of the Zonguldak province NW Turkey. Environmental Geology 55 (5), 10011014.
Turrini, M.C., Visintainer, P., 1998. Proposal of a method to dene areas of
landslide hazard and application to an area of the Dolomites, Italy. Engineering
Geology 50, 255265.
Weier, J., Herring, D., 2005. Measuring Vegetation (NDVI and EVI). Earth Observatory Library of NASA. /http://earthobservatory.nasa.gov/Library/MeasuringVegetationS.
Wilson, J.P., Gallant, J.C., 2000. Terrain Analysis: Principles and Applications. John
Wiley and Sons, New York, 469pp.
Wood, E.F., Sivapalan, M., Beven, K.J., 1990. Similarity and scale catchment storm
response. Reviews in Geophysics 28, 118.
Yagiz, S., Gokceoglu, C., 2010. Application of fuzzy inference system and nonlinear
regression models for predicting rock brittleness. Expert Systems with
Applications 37, 22652272.
Yilmaz, I., 2009. Landslide susceptibility mapping using frequency ratio, logistic
regression, articial neural networks and their comparison: a case study from
Kat landslides (Tokat-Turkey). Computers and Geosciences 35 (6), 11251138.
Yilmaz, I., 2010a. The effect of the sampling strategies on the landslide susceptibility mapping by conditional probability (CP) and articial neural network
(ANN). Environmental Earth Sciences 60, 505519.
Yilmaz, I., 2010b. Comparison of landslide susceptibility mapping methodologies
for Koyulhisar, Turkey: conditional probability, logistic regression, articial
neural networks, and support vector machine. Environmental Earth Sciences
61, 821836.
Zinko, U., Seibert, J., Dynesius, M., Nilsson, C., 2005. Plant species numbers
predicted by a topography based groundwater-ow index. Ecosystems 8,
430441.
Zweig, M.H., Campbell, G., 1993. Receiver-operating characteristic (ROC) plots.
Clinical Chemistry 39, 561577.