ORIGINAL PAPER
Application of weights-of-evidence and certainty
factor models and their comparison in landslide susceptibility
mapping at Haraz watershed, Iran
Hamid Reza Pourghasemi & Biswajeet Pradhan &
Candan Gokceoglu & Majid Mohammadi &
Hamid Reza Moradi
Received: 4 September 2011 / Accepted: 8 February 2012 / Published online: 9 March 2012
#
Saudi Society for Geosciences 2012
Abstract The main goal of this study was to investigate the
application of the weights-of-evidence and certainty factor
approaches for producing landslide susceptibility maps of a
landslide-prone area (Haraz) in Iran. For this purpose, the
input layers of t he landslid e conditi oning f actors were
prepared in the first stage. The landslide conditioning
factors considered for the study area were slope gradi ent,
slope aspect , altitude, lithology, land use, distance from
streams, distance from r oa ds, distance from faults, topo-
graphic wetness index, stream power index, stream transport
index and plan curvature. For validation of the produced
landslide susceptibility maps, the results of the analyses
were compared with the field-verified landslide locations.
Additionally, the receiver operating characteristic curves for
all the landslide susceptibility models were constructed and
the areas under the curves were calculated. The landslide
locations were used to validate results of the landslide
susceptibility maps. The verification results showed that
the weights-of-evidence model (79.87%) performed better
than certainty factor (72.02%) model with a standard error
of 0.0663 and 0.0756, respectively. According to the results of
the area under curve evaluation, the map produced by
weights-of-evidence exhibits satisfactory properties.
Keywords Landslide susceptibili ty
.
Weights of evidence
.
Certainty factor model
.
GIS
.
Remote sensing
.
Iran
Introduction
Landslides are one of the most catastrophic natural hazards
occurring in many areas of the world. Globally, they cause
hundreds of billions of dollars in damage and hundreds of
thousands of deaths and injuries each year (Aleotti and
Chowdhury 1999). Over the past 25 years, many govern-
ments and research ins titutions throughout the w orld have
invested considerable resources in assessing landslide
hazards and in attempting to produce maps portraying
their spatial distribution (Guzzetti et al. 1999). In spite of
improvements in recognition, mitigative measures, and
prediction and warning systems, landslide damage is s till
increasing worldwide (Schuster 1996). Losses resulting
from mass movements in Iran until the end of September
2007 have been estimated at 126,893 billion Iranian
Rials (about USD 12,700 million) using the 4,900 land-
slide database.
Similar approaches have been propose d by sev eral
investigators, including weights-of-evidence methods
H. R. Pourghasemi
:
M. Mohammadi
:
H. R. Moradi
Department of Watershed Management Engineering,
College of Natural Resources and Marine Sciences,
Tarbiat Modares University International Campus,
Noor, Iran
e-mail: hm_porghasemi@yahoo.com
B. Pradhan (*)
Institute of Advanced Technology, Spatial and Numerical
Modelling Laboratory, University Putra Malaysia,
43400 UPM Serdang, Selangor Darul Ehsan, Malaysia
e-mail: biswajeet24@gmail.com
B. Pradhan
e-mail: biswajeet@mailcity.com
C. Gokceoglu
Applied Geology Division, Department of Geological
Engineering, Engineering Faculty, Hacettepe University,
Ankara, Turkey
Arab J Geosci (2013) 6:2351–2365
DOI 10.1007/s12517-012-0532-7
(Bonham-Carter 1991; Lee et al. 2002 a ;Wuetal.2004;
Gokceoglu et al. 2005; Neuhäuser and Terhorst 2007;
Mathew et al. 2007; Bui et al. 2008; Zhu and Wang
2009;Regmietal.2010;OhandLee2011), wei ghting
factors (Çevik and Topal 2003), weighted linear combinations
of instability factors (Ayalew et al. 2004), landside nominal
risk fa ctors (Gupta a nd Joshi 19 90;Sahaetal.2005),
probabilistic models (Chung and Fabbri 2003, 2005;Lee
2004; Lee and Pradhan 2006, 2007; Akgun et al. 2011;
Pradhan et al. 2012), certainty factors (Binaghi et al.
1998), information values (Lin and Tung 2004;Sahaet
al. 2005), modified Bayesian estimation (Chung and Fabbri
1998) and data mining (Biswajeet and Saied 2010; Pradhan et
al. 2009, 2010a, b, c, d, e, 2011 Pradhan 2010a, b, c, 2011a, b
Pradhan and Lee 2010a, b; Pradhan and Buchroithner 2010;
Pradhan and Youssef 2010; Sezer et al. 2011; Oh and Pradhan
2011; Bui et al. 2011; Akgun et al.
2012). Understanding the
differences between the proposed approaches is not always
simple. The main differences are the rigour of the ap-
proach (Chung and Fabbri 1998) and the method used to
estimate the pri or probability of landslide occurrence. The
aim of the present study was to produce landslide susceptibil-
ity maps of the Haraz watershed in Iran by employing a
weights-of-evidence and certainty factor models.
Study area
The study area is located in the northern part of Iran, which
is one of the most landslide-prone areas in the country
(Pourghasemi 2008). The watershed area lies between
longitudes 52° 06′ 02″ Eand52°18′ 13″ Eandbetween
latitudes 35° 49′ 05″ N and 35° 57′ 39″ N. It is mountainous
and is located in the Alborz Folded geological zone (Fig. 1).
It covers two adj acen t 1:50 ,00 0 topog ra phic sh eets of th e
Army Geographic Institute of Iran and has an extent of
Fig. 1 Location map of the study area in Mazandaran province, Iran
2352 Arab J Geosci (2013) 6:2351–2365
about 114.5 km
2
. The main river in the study a rea is the
Haraz River. The temperature varies between 25°C in
winter and 36.5°C in summer. The mean annual rainfall
is around 500 mm, most of which falls between November
and January. The altitudes in the study area vary between
1,200 and 3,290 m. The slope angles of the area range
from 0° to as much as 70°. The majority of the area
(64.82%) is covered by moderate pasture. The other parts
of the study area are utilised for orchard and agricultural
(13.33%), residential (0.3%) and best pasture purposes
(21.55%).
Weights-of-evidence model
In recent years, many investigators (Bonham-Carter 1991;
Mathew et al. 2007; Neuhäuser and Terhorst 2007 ; Bui et al.
2008; Regmi et al. 2010; Pradhan et al. 2010c)have
experimented with methods t hat exploit, more or less
rigorously, Bay es’ conditi onal probability theorem. In
this framework, conditional probability is a measure of
the chance of a hypothesis being true or false given a
piece of evidence ( Gorsevski et al. 2003). For example,
Bayesian probabi list ic modelli ng is supplied for solving
problems of de cision-ma king under uncertainties. This
method is suitable for landslide susceptibility mapping
because its uncertainty is connected with landslide events
and their associations with the complex landscape
(Chung and Fabbri 1998; Gorsevski et al. 2003). Bayes’
theorem can be written as (Guzzetti 2005):
PAjBðÞ¼
PBjAðÞPð AÞ
PðBÞ
ð1Þ
So, the probability of phenomena B occurring given that
phenomena A has occurred, P(B|A), multiplied by the prob-
ability of phenomena A occurring, P(A), and divided by the
probability of phenomena B occurring, P(B).
In Eq. 1, P(A) is the “prior probability” (i.e. a reasonable
hypothesis on the probability of phenomena A), P(B) is the
“posterior probability” (i.e. the probability of B under all
possible outcomes for A), and P(A|B)isthe“probability”
(i.e. the conditional probability of A given B). In a best
Bayesian analysis , the prior probability has a minor effect
on the posterior probability, as most of the information
comes from the likelihood. When applied to landslide
susceptibility investigation, Bayes’ theorem is used to
select the probability t hat an area will improve slope
failures given the local environmental circumstances, as
indicated in Eq. 2 (Chung and Fabbri 1998):
PA
L
j V
0
ðrÞ
f
; V
1
ðrÞ ...; V
m
ðrÞgðÞ
¼
PV
0
ðrÞ
f
; V
1
ðrÞ ...; V
m
ðrÞgjA
L
ðÞPA
L
ðÞ
PV
0
ðrÞ; V
1
ðrÞ; ...; V
m
ðrÞðÞ
ð2Þ
where, A
L
denotes area of landslide in a mapping unit r
for which V
0
ðrÞ
f
; V
1
ðrÞ ...; V
m
ðrÞg is independent of
environmental conditions. Additionally, the mixture of
environme nt al conditi ons is special to the mapping unit
r.Equation2 showed that the probabili ty that a mapping
Fig. 2 The landslide inventory
map of the study area
Arab J Geosci (2013) 6:2351–2365 2353
unit r in t he study a rea wil l be influe nce d by a lands lide
which is equivalent to the probability of a landslide in
the study area, P(A
L
), multiplied by the probability of a
particular (unique) mixture of environmental factors given the
presence of a landslide, divided by the probability of the same
mixture of environmental factors in the whole study area. A
simple strategy is to acquire the three probabilities in the right-
hand side of Eq. 2 in a geographic information system (GIS)
from the related spatial densities. These probabilities can be
obtained as follows: (1) by dividing the entire A
L
in the study
area by the area of the mapping unit, for P(A
L
); (2) by dividing
the whole area of the unique condition unit by the extent of the
study area for PV
0
ðrÞ; V
1
ðrÞ; ...; V
m
ðrÞðÞ; and (3) by consid-
ering the percentage of the landslide area in the study area
Fig. 3 Lithology map of the
study area
Table 1 Description of
geological units of
the study area
No. Symbol Formation Lithology Geological age
A
Q
sc
– Scree Quaternary
Q
t
2
– Young terraces Quaternary
Q
t
1
– Old terraces Quaternary
B
Q
ag
– Agglomerate Quaternary
Q
ta
– Trachy andesitic lava flows Quaternary
Q
tu
– Ash tuff, lapilli tuff Quaternary
Q
b
– Olivine basalt Quaternary
C
K
tv
k
Karaj Green tuff, basaltic and limestone with gypsum and
conglomerate
Eocene
E
gy
k
Karaj Gypsum Eocene
D
PEz
Ziarat Limestone bearing nummulites and alveolina, conglomerate Paleocene
PEf
Fajan Conglomerate, agglomerate, some marl and limestone Paleocene
E
K2
– Biogenic and cherty limestone Late
Cretaceous
Kt
Tizkuh Orbitoline bearing limestone Late
Cretaceous
J1
Lar Massive to well bedded, cherty limestone Late Jurassic
Jd
Dalichai Well bedded, partly oolitic-detritic limestone, marly limestone Late Jurassic
JS
Shemshak Dark shale and sandstone with plant remains, coal Late Jurassic
TReL
Elika Thin bedded limestone Early Triassic
Pd
Dorud Cross-bedded, quartzitic sandstone Early Permian
2354 Arab J Geosci (2013) 6:2351–2365
characterised by the total area of the unique environmental
setting in Eq. 2,forPV
0
ðrÞ
f
; V
1
ðrÞ ...; V
m
ðrÞgjA
L
ðÞ.
An advantage of Bayesian probabilistic modelling is the
possibility of incorporating uncertainty into the susceptibil-
ity model and considering expert knowledge explicitly
(Chung and Fabbri 1998).
Certainty factor model
Among the commonly used GIS analysis models for
landslide s usceptibility, certainty factor (CF) model has
been widely considered and experimentally investigated
in the literature ( Chung and Fabbri 1993; Binaghi et al.
1998;LuziandPergalani1999; Lan et al. 2004;
Kanungo et al. 2011). The CF approach is one of the
possible proposed favourabilit y functions to handle the
problem of combination of different data layers and the
heterogeneity and uncertainty of the input data. The main
difference is the bivariate model with other models of
how to combine the maps. Thus, the maps classifieds and
then weight of each pixel is obtained using Eq. 3:
CF ¼
PP
a
PP
s
PP
a
1PP
s
ðÞ
if PP
a
PP
s
PP
a
PP
s
PP
s
1PP
a
ðÞ
if PP
a
< PP
s
8
<
:
ð3Þ
where, PP
a
is the conditional probability of landslide
event occurring in class a and PP
s
is the prior probability
of total num ber of landslide events in the study area A.
With the use of the CF model, each class or area is
assigned a value that varies within the interval [−1, 1].
A positive value means a growth in the certainty of the
landslide occurren ce, whereas a ne ga tive value coincides
with a decrease in the certainty of landslide occurrence.
A value close to 0 means that there is not enough
information about the variable and thus, it is difficult to
give information about the certainty of landslide occurrence.
The CF values are calculated for all condition factors by
overlaying and calculating the landslide frequency as given
then the CF values of all parameters in 12 landslide condi-
tioning factors are determined using Eq. 3. Next, the CF
values of the landslide conditioning factor are pairwise com-
bined using the CF combination rule. A combination of two
CF values, X and Y from two different layers of information is
aCFvalueZ obtained as follows (Chung and Fabbri 1993;
Binaghi et al. 1998; Luzi and Pergalani 1999):
Z ¼
X þ Y XY X ; Y 0
X þY
1min Xjj; YjjðÞ
X ; Y opposite sign
X þ Y þ XY X ; Y < 0
8
<
:
ð4Þ
The pairwise combination by using the inte gration rule of
Eq. 4 is performed repeatedly until all the CF layers are
combined to obtain the landslide susceptibility.
Thematic data preparation
Various thematic data layers representing landslide condition-
ing factors, such as slope gradient, slope aspect, altitude, lithol-
ogy, land use, distance to faults, distance to streams, distance to
roads, topographic wetness index (TWI), stream power index
(SPI), stream transport index (STI) and plan curvature were
prepared. These factors fall under the category of preparatory
factors, which make the area susceptible to movement without
actually initiating a landslide; thus, these factors are considered
to be responsible for the occurrence of landslides in the regions
for which pertinent data can be collected from available resour-
ces and from the field. The triggering factors, such as rainfall
and earthquake, set the movement off by shifting the slope
from a marginally stable to an actively unstable area. Further-
more, the attributes of the ground in terms of landslide suscep-
tibility were considered in the present study. Since, past data on
triggering factors such as rainfall and earthquakes in relation to
landslide occurre nces were not available. Consequently, these
factors were not considered in this study.
Landslide inventory map
The mapping of existing landslides is essential for
studying the relationship between the landslide distribu-
tion and the conditioning factors. To produce a detailed
and reliable landslide inventory map, extensive field
surveys and observations were performed in the study
area. A total of 78 landslides were identified and
mapped in the study area by evaluating aerial photos
in 1:25,000 scale and by field survey (Fig. 2). The
modes of failure for the landslides identified in the
Fig. 4 Land use map of the study area
Arab J Geosci (2013) 6:2351–2365 2355
