Habitat International 45 (2015) 114e125

Contents lists available at ScienceDirect

Habitat International
journal homepage: www.elsevier.com/locate/habitatint

GIS-based multicriteria evaluation with multiscale analysis to

characterize urban landslide susceptibility in data-scarce
environments
Suzana Dragi c a, *, Terence Lai a, 1, Shivanand Balram b, 2
cevi
a
Spatial Analysis and Modeling Laboratory, Department of Geography, Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6, Canada
b
Department of Geography, Simon Fraser University, 8888 University Drive, Burnaby, BC V5A 1S6, Canada

a r t i c l e i n f o a b s t r a c t

Article history: Landslides can have a severe negative impact on the socio-economic and environmental state of in-
Available online 28 July 2014 dividuals and their communities. Minimizing these impacts is dependent on the effective identification
of risk areas using a susceptibility analysis process. In such a process, output maps are generated to
Keywords: determine various levels of threat to human populations. However, the reliability of the process is
Fuzzy sets controlled by critical factors such as data availability and data quality. In data-scarce environments,
GIS
susceptibility analysis done at multiple interlocking geographic scales can provide a convergence of
Landslide susceptibility
evidence to reliably identify risk areas. In this study, multiscale analysis and fuzzy sets are combined with
Multicriteria evaluation
Multiscale
GIS-based multicriteria evaluation (MCE) to determine landslide susceptibility levels for areas of the
Validity Metro Vancouver region, British Columbia, Canada. Landslide-conditioning parameters are chosen based
on their relevance and effect on a particular scale of analysis. These parameters are derived for three
geographic scales using digital elevation models, drainage networks and road networks. An analytical
hierarchy process (AHP) analysis provides relative weights of importance to combine variables. The
landslide susceptibility analysis is done for regional, municipal and local scales at resolutions of 50 m,
10 m, and 1 m respectively. At each scale, susceptibility models are validated against real inventory data
using the seed cell area index (SCAI) method. The strong inverse correlation between the map classes and
the SCAI adds to confidence in the results. The developed approach can enable analysts in data-scarce
environments to reliably identify susceptible areas thereby improving hazard mitigation, emergency
services targeting, and overall community planning.
© 2014 Elsevier Ltd. All rights reserved.

Introduction economic costs include relocating communities, repairing physical

Landslides are described as the mass movements of slope-
forming materials comprising rocks, soils, artificial fill, or a com-
bination of these (Sidle & Ochiai, 2006). Causal factors such as slope
and erosion accumulate in an area and the onset of a rapid trigger
event releases the landslide. These trigger events include rainfall,
weathering, surface fractures, and earthquakes. The landslide
evolves swiftly and can have devastating impacts on the well-being
of humans and communities, especially in regions where urban
residential areas coincide with mountainous terrains. The
* Corresponding author. Tel.: þ1 778 782 4621; fax: þ1 778 782 5841
E-mail addresses: suzanad@sfu.ca (S. Dragi
cevi
c), tcl@sfu.ca (T. Lai), sbalram1@
sfu.ca (S. Balram).
1
Tel.: þ1 778 782 5628.
2 rapid development of Geographical Information Systems (GIS) that
Tel.: þ1 778 782 3702; fax: þ1 778 782 5841.
structures, and restoring water quality in streams and rivers (Yalcin,
2007).
The Joint Technical Committee on Landslides and Engineered
Slopes characterizes landslide susceptibility zoning as the spatial
distribution and classification of terrain units according to their
predisposition to result in landslides (Fell et al., 2008). Landslide
susceptibility assessments are important to engineers and city
planners because the susceptibility maps provide an additional tool
to support the selection of areas for development. In order to plan
for the adverse effects of landslides, since the 1970s landslide
susceptibility and hazard zoning techniques have been developed
by manually delineating susceptibility zones using aerial photo-
graphs (Blesius & Weirich, 2010; Brabb, Pampeyan, & Bonillia, 1972;
Drennon & Schleining, 1975). In recent decades, there was
improved progress in preparing hazard zoning maps because of the

http://dx.doi.org/10.1016/j.habitatint.2014.06.031
0197-3975/© 2014 Elsevier Ltd. All rights reserved.
 S. Dragicevic et al. / Habitat International 45 (2015) 114e125
greatly advanced the susceptibility mapping process in both effi-
ciency and accuracy (Weirich & Blesius, 2007). GIS facilitated large
volumes of data to be managed and quickly analyzed for use in
various landslide mapping studies (Yalcin & Bulut, 2007). Despite
the advances in GIS capabilities, landslide susceptibility studies are
still mostly being conducted at a single spatial scale. The objectives
of this study are to (i) develop a multiscale approach to model
landslides, (ii) implement the approach at regional, municipal and
local spatial scales, and (iii) test the approach using inventory data
from the Metro Vancouver area, District of North Vancouver (DNV),
Canada; and locally on the Berkley Escarpment in the DNV. The
spatial resolutions of the Metro Vancouver, DNV, and Berkley
Escarpment data are 50 m, 10 m, and 1 m respectively.
There are many techniques available to produce landslide sus-
ceptibility maps and these can be categorized generally into
quantitative methods, semi-quantitative methods, and qualitative
methods. Further, within each of these categories research efforts
are being made to use multiple approaches to improve the sus-
ceptibility map outputs.
Quantitative methods such as deterministic models (Gomes
et al., 2008; Gorsevski, Gessler, Boll, Elliot, & Foltz, 2006; Klimes,
2008; Luzi & Pergalani, 1996; Mergili, Schratz, Ostermann, &
Fellin, 2012; Santini, Grimaldi, Nardi, Petroselli, & Rulli, 2009;
Van Westen & Terlien, 1996; Wan, Lei, & Chou, 2012; Wu & Sidle,
1995) and probabilistic and statistical models (Ayalew &
Yamagishi, 2005; Dai & Lee, 2003; Ercanoglu, 2005; Kia et al.,
2012; Ohlmacher & Davis, 2003; Piacentini et al., 2012; Thapa,
2011; Xu, Xu, Lee et al., 2012; Yilmaz, 2009; Yilmaz & Keskin,
2009; Yilmaz, Topal, & Suzen, 2012) have minimal dependence
on human judgment and expert opinion to produce the output
maps. These quantitative techniques also require large volumes of
detailed data derived from laboratory tests and field surveys
making them highly unsuitable for regional scale studies (Van
Westen, Van Asch, & Soeters, 2006) or use in data-scarce envi-
ronments (Demoulin & Chung, 2007). These quantitative methods
have also been integrated with GIS and multicriteria evaluation
(MCE) techniques to better represent the spatial character of the
problem situation.
Heuristic methods, including both qualitative and semi-
quantitative, range from direct field mapping methods to com-
plex logical and computer-based systems that incorporate human
judgment and expert opinions (Castellanos Abella & Van Westen,
2008; Lai & Dragicevic, 2011; Lee & Choi, 2004; Pavel, Nelson, &
Fannin, 2011; Reis et al., 2012; Van Westen, Rengers, & Soeters,
2003; Wang, Guo, Fang, & Chang, 2012). A common heuristic
approach linking GIS and multicriteria evaluation (Jankowski, 1995)
uses expert opinions on multiple criteria with the resulting land-
slide maps categorized into zones of “very low”, “low”, “medium”,
“high”, and “very high” categories of susceptibility. A mixture of
criteria weighting tools such as the Analytical Hierarchy Process
(AHP) and criteria integration tools such as Weighted Linear
Combination (WLC) are used to combine the factors and generate
landslide susceptibility maps (Akgun & Bulut, 2007; Akgun & Turk,
2010; Gorsevski, Jankowski, & Gessler, 2006; Pourghasemi,
Pradhan, & Gokceoglu, 2012; Thanh & De Smedt, 2012; Wu &
Chen, 2009; Yalcin, 2008). Heuristic analysis is shown to be a
cost-effective approach for large study areas having limited acces-
sible data (Van Westen et al., 2006).
Recently, a combination of methods has been pursued. The
comparison of multiple susceptibility mapping methods is useful
because it provides an indication of the reliability range of the re-
sults (Lee et al., 2012; Regmi, Giardino, Vitek, & Dangol, 2010; Rossi,
Guzzetti, Reichenbach, Mondini, & Peruccacci, 2010; Sterlacchini,
Ballabio, Blahut, Masetti, & Sorichetta, 2011; Xu, Xu, Dai, & Saraf,
2012). However, limitations include the dependency on large
115
volumes of data and the inability of generalizing the comparison
results outside the data context.
In addition to data availability, scale is also an important factor.
Selecting the appropriate analysis scale is a challenge when pro-
ducing susceptibility maps because it is often a compromise be-
tween a desired scale and data availability. Nevertheless, landslide
susceptibility studies mostly generate a single map at a fixed scale
determined by data convenience. However, the scale of observation
will affect the analysis, outputs and interpretation. For example, at
a country level scale topography will explain the broad patterns of
slope, aspect, and flow accumulation but mask local finer scale
variations. As the scale changes, so do associated patterns of spatial
processes and this has implications for understanding any phe-
nomena and the applicability of methods and results from one scale
to another (Hay, Marceau, Dube, & Bouchard, 2001). Thus, under-
standing the behavior of a phenomenon at multiple scales is
imperative to determine the effect of scale on the spatial patterns
and processes (Wu, Jelinski, Luck, & Tueller, 2000).
The research literature has indicated only one analysis using a
multiscale MCE approach to evaluate landslides. The authors
developed a national landslide risk index map for Cuba with
additional analyses at provincial and municipal levels (Castellanos
Abella & Van Westen, 2007). One challenge was the lack of avail-
able data for the entire country and hence the authors were forced
to exclude deterministic landslide hazard assessment methods
from their analysis. They compromised by using MCE and AHP
methods to produce a qualitative landslide risk index. Further, with
this national level risk map they identified areas of high risk at the
provincial and municipal levels for additional statistical analysis
once relevant data becomes available. While the authors have
produced a useful representation of landslides at the national level,
it is clear the methods used and the scale of analysis were condi-
tioned on the available data.
In this study, we have integrated a multiscale analysis and fuzzy
sets into a GIS-based multicriteria evaluation (MCE) approach to
determine landslide susceptibilities in the Metro Vancouver region,
British Columbia, Canada. The focus is on shallow landslides trig-
gered by rainfall events typical of the study sites. This focus allows
for the generalization of the results to other study areas susceptible
to shallow landslides. The ability to model landslide susceptibility
across multiple scales allows various levels of decision makers to
identify susceptibility hotspots and effectively allocate resources
and services. The next sections give background to the theory,
outline the multiscale GIS-based MCE approach, and present results
from three scales of analysis. The implications of the work are then
discussed and conclusions stated.
Material and methods
GIS-based multicriteria evaluation
Geographic information systems (GIS) have evolved from pre-
forming functionalities such as geographic database management
to geovisualization analysis and are now able to provide advanced
scientific and mathematical analysis between multiple map layers
(Eastman, Jin, Kyem, & Toledano, 1995). Consequently, GIS is well-
suited and extensively applied to the design and development of
robust decision support systems capable of evaluating choices from
advanced spatial data analysis techniques at various scales of
analysis (Lai, 2011).
The goal of multicriteria evaluation (MCE) or multicriteria
analysis (MCA) is to assess choice possibilities when there are
multiple criteria and conflicting objectives (Carver, 1991;
Jankowski, 1995; Voogd, 1982). The MCE procedure integrates in-
formation from various standardized criteria to produce a single
116 S. Dragicevic et al. / Habitat International 45 (2015) 114e125
evaluation index. Criteria are the evidence being used and can be
either factors or constraints. A factor increases or decreases the
suitability of an alternative for the activity under consideration, and
a constraint limits the alternatives under consideration (Eastman
et al., 1995). The suitability levels of factors varies from Boolean
unsuitable to suitable (i.e. zeros and ones) or levels of certainty (i.e.
a continuous scale from zero to one) depending on the type of data
used and the subjectivity of each expert (Malczewski, 1999).
Fuzzy set theory addresses the uncertainty of data by assigning
the data elements to be members of a set along a continuous scale
rather than the common crisp binary set membership (Zadeh,
1965). Compared to linear scaling, fuzzy sets produce more real-
istic standardization since the fuzzy set membership represents a
specific relation between the criteria and possible outcomes (Lai,
2011; Wood & Dragicevic, 2007). In the case of Boolean overlay,
membership values change from 0 and 1 as there are crisp
boundaries between entities (Jiang & Eastman, 2000). After stan-
dardization, a rating matrix is developed to assess the relative
importance of each evaluation criteria with respect to the overall
problem objective (Voogd, 1982). Varying and prioritizing the
criteria allows for the generation of compromising alternatives and
ranking alternative outcomes by the various criteria (Malczewski,
2006). Varying the relative importance of criteria is based on
expert knowledge and facilitates sensitivity analysis and validation
of weights and rankings of alternatives (Feick & Hall, 2004;
Jankowski & Richard, 1994).
MCE approaches that use GIS explicitly provides significant
benefits in defining landslide susceptibility (Lai, 2011). Landslide
studies often use MCE techniques because the types of data avail-
able are commonly qualitative (from expert opinion) and quanti-
tative (from observed relationships between parameters and
landslides), therefore requiring a semi-quantitative method that
incorporates both types of data (Ayalew, Yamagishi, & Ugawa,
2004). These methods assign weights to the different parameters
affecting landslide susceptibility and many techniques exist for
determining weights (Akgun & Bulut, 2007). The most common
approach involves obtaining expert opinion to assigning weights in
the Analytical Hierarchy Process (AHP) and then combining
weights additively by weighted linear combination (WLC) to pro-
duce landslide susceptibility maps (Akgun, Dag, & Bulut, 2008;
Ayalew, Yamagishi, Marui, & Kanno, 2005; Komac, 2006; Yalcin &
Bulut, 2007). In data-scarce environments, the AHP is more useful
compared with data intensive alternatives such as logistic regres-
sion, artificial neural networks (ANN) and decision trees. In totality,
the combination of MCE and AHP to examine the convergence of
evidence at multiple scales is a novel approach for landslide sus-
ceptibility mapping (Lai, 2011).
Landslide-conditioning parameters
Landslides are generated from complex geophysical and social
processes. As such, many landslide-causing parameters are inte-
grated to improve the accuracy of the susceptibility mapping pro-
cess. However, in data-scarce environments it may not be possible
to obtain the required geological, lithological, weathering and soils
data that are typically used in the susceptibility mapping process.
Hence, surrogate variables derived from limited available data
provide an elegant solution to this challenge. In one example, the
commonly available digital elevation model (DEM) was used to
develop topographic stability indices for use in creating hazard and
susceptibility maps (Pack, Tarboton, Goodwin, & Prasad, 2005).
Correlation tests between the DEM derived data can further
simplify the data by reducing statistical redundancy without
compromising the results. For these reasons, and the fact the
landslides being studied are mostly triggered by rainfall events, we
selected parameters derived from topographic attributes, road
networks, and stream networks. The specific landslide-
conditioning parameters considered in this study were: slope
gradient, aspect gradient, stream power index (SPI), topographic
wetness index (TWI), elevation, distance to road networks, and
distance to stream networks.
Slope gradient and aspect gradient
Slope gradient is widely used in landslide susceptibility studies
since it is directly connected to the movement of landslide mate-
rials (Dai, Lee, Li, & Xu, 2001; Lee & Min, 2001; Van Westen et al.,
2003). Specifically, shear stresses on the slope material increases
with slope gradient and it is generally expected that landslides
occur on the steepest slopes. The slope for each raster cell is
calculated as a function of the cell resolution and neighboring cells
(Monmonier, 1982). Aspect gradient affects slope material in a in-
direct relationship because aspect determines the exposure of a
landscape to rainfall and solar radiation, and therefore the pro-
pensity of vegetation to grow, which in turn effects the soil stability
(Carrara et al., 1991). As a consequence, aspect gradient can be a
surrogate for data on soil attributes that may not be available. The
aspect for each raster cell is the direction in which the maximum
slope faces.
Stream power index
The stream power index (SPI) is widely used as a landslide
conditioning parameter (Conoscenti, Di Maggio, & Rotighano,
2008; Duman, Can, Gokceoglu, Nefeslioglu, & Sonmez, 2006;
Yilmaz, 2009). The index describes a measure of erosive power of
flowing water based on the assumption that runoff is directly
proportional to the upslope contributing area (Moore, Grayson, &
Ladson, 1991). In terms of each grid cell in the raster GIS, SPI is a
function of erosive power of runoff acting on each cell. SPI is
defined as:
SPI ¼ As  tan b (1)
where As ¼ specific catchment area (m2/m) and b ¼ slope gradient.
As As and b increase, the amount of water provided by upslope
areas and the flow velocity of water increase, thereby increasing SPI
and slope-erosion risk.
Topographic wetness index
The topographic wetness index (TWI) describes the effect of
topography on the location and size of saturated areas of runoff
generation (Moore et al., 1991). The TWI can be used as an esti-
mation of spatial patterns for soil moisture since topography con-
trols the hydrological conditions where surface runoff and
groundwater flow. TWI is defined as:
As
TWI ¼ ln (2)
tan b
where As ¼ specific catchment area (m2/m) and b ¼ slope gradient.
Steady state conditions and uniform soil properties are
assumed. The equation gives predicted values for areas of satura-
tion where As is large and b is small e typically these conditions
occur at converging slopes in the landscape combined with gentle
slope gradient areas.
Elevation
There is strong evidence that elevation is an indicator of land-
slide susceptibility (Duman et al., 2006; Gokceoglu, Sonmez,
Nefeslioglu, Duman, & Can, 2005; Pachauri & Pant, 1992). Land-
slides usually occur at intermediate elevations since slopes tend to
 S. Dragicevic et al. / Habitat International 45 (2015) 114e125 117

be covered by a layer of thin colluvium that is prone to landslides

(Dai & Lee, 2002). Landslides are less likely to occur at very high
elevations because only rock exposures are present due to their
shear strength being much higher. In the lower elevations, slopes
are gentle and covered by thicker colluvium. The chance of land-
slides occurring is lower unless the water table rises to initiate
slope failure.

Distance to road networks

Gravel materials on roads are engineered and compacted to
withstand heavy loads making the surfaces flat and impermeable.
During heavy rainfall, road surfaces reduce infiltration causing
rapid overland flow and surface runoff (Horton, 1945). Roads can
have a major impact on surrounding areas as they allow water to
pool causing overload, undercut and saturated slopes (Guthrie,
2002). Roads also transmit vehicular vibrations as well as in-
crease the susceptibility for landslide initiation and intercept; they
store and produce sediment that influences downstream hill slopes
not necessarily near initiation sites (Wemple, Swanson, & Jones,
2001).

Distance to drainage networks

Landslides have been associated with proximity to drainage
networks because terrain modification caused by stream erosion
and undercutting of slope toes can influence landslide initiation
(Dai & Lee, 2002). Also, the presence of streams has the effect of
raising water tables in the surrounding areas of close proximity. It is
generally accepted that landslide frequency decreases as distance
from drainage networks increases.

Implementation

A unique multiscale GIS-based approach was developed using

MCE and fuzzy sets to combine the selected landslide parameters
and generate realistic output susceptibility maps. Each landslide-
conditioning parameter affects the landslide process and final Fig. 1. Overview of MCE components for each scale.

output maps so it is necessary to determine the relative importance

of each parameter. The parameters are standardized to a common
is the same except it is done at shorter distances. The same is true
measurement scale using fuzzy set membership functions (Zadeh,
for other topographic variables except elevation.
1965) and the Analytical Hierarchy Process (AHP), which calcu-
The use of elevation is suitable for a study area at a given scale if
lates the weight values for each parameter (Saaty, 1980). Final
the variation in altitude is sufficiently large. A small to medium
landslide susceptibility maps are then produced by combining the
scale study area would likely provide enough separation in land-
weighted standardized parameter maps using the weighted linear
slide classes between elevations. However, a large scale study
combination (WLC) method (Voogd, 1983). The above approach is
focusing on a hillslope or escarpment will not yield significant rise
applied successively to identify susceptible areas at increasing
scales and spatial resolutions (Fig. 1). This multiscale MCE approach
is applied to regional, municipal, and local scales at increasing
spatial resolutions of 50 m, 10 m, and 1 m respectively (Fig. 2).
These finer scales allow for the systematic identification of sus-
ceptibility ranging from the national level down to the community
level. The susceptibility maps produced at each of the spatial scales
can be used in comparative analysis because common important
parameters are used even though the scales are varying.

Defining scale-specific landslide-conditioning parameters

The set of landslide-conditioning parameters for each individual

scale was selected based on their relevance at the specific scale of
analysis. Topographic variables calculated from the DEM, such as
slope gradient, aspect gradient, SPI, and TWI can be used at all three
scales of analysis. These variables calculated at different spatial
resolutions represent the topography in each length of the cell size.
For example, slope gradient derived using a spatial resolution of
50 m is calculated by the average change in height over a cell length
of 50 m. For finer spatial resolutions, the slope gradient calculation Fig. 2. Interaction of MCE procedures between scale levels.
118 S. Dragicevic et al. / Habitat International 45 (2015) 114e125

and fall in elevation. Hence, the use of elevation as a landslide- Analytical hierarchy process
conditioning parameter is limited to scales that exhibit a certain
level of topographic variability. Distance away from roads is The number of parameters used in the susceptibility analysis can
another parameter possessing limitations due to spatial resolution. make it challenging to evaluate their importance and assign
A coarse resolution of 50 m will incorrectly represent a single lane weights appropriately. Analyzing and comparing two parameters at
road that typically spans 6 m across. A spatial resolution of 50 m a time in an organized manner helps to breakdown the assignment
would still be too large even for double-lanes in both directions that of weights into smaller entities. Performing multiple one-to-one
typically spans 12 m totally. For each landslide factor, consideration pairwise comparisons is a much simpler and intuitive process to
is given to whether its inclusion makes logical sense at a given scale determine a hierarchy of landslide influencing factors than trying to
and spatial resolution. evaluate all parameters as a whole. The technique used in this study
is the Analytical Hierarchy Process (AHP). AHP is useful since it has
the capability to handle both qualitative and quantitative criteria
Standardization of parameters using fuzzy sets
(Ayalew et al., 2004). The technique operates irrespective of the
data type because the fundamental input given by the user is an
Data standardization is needed prior to GIS-based landslide
expert judgment in answer to the question: How important is
susceptibility mapping to facilitate criteria combination. In addi-
parameter A compared to parameter B? The relative importance of
tion, standardizing the data to a common scale allows comparisons
parameters is translated to a 9 point continuous rating scale and
between fuzzy sets because it allows the degree of membership for
entered into a pairwise comparison matrix (Saaty, 1980) (Table 1).
data to be measured on the same continuous scale between 0 and 1
Ratings are given by 1, 3, 5, 7, and 9 starting with equal importance
(Zadeh, 1965). Traditionally, GIS applications have operated with
to increasing importance, with 2, 4, 6, and 8 as intermediate values.
crisp sets where spatial information captured from reality is rep-
Comparing a parameter with itself generates an equal rating of 1.
resented as discrete objects in space with a discrete definition.
Reciprocals represent the comparison of a pair of parameters in
When maps of landslides susceptibility are produced using infor-
reverse. For example, if parameter A was extremely more important
mation based on crisp sets with membership values being 0 or 1,
relative to parameter B a rating of 9 is given. The comparison in
the final outcome is identical to a Boolean overlay analysis. The
reverse of parameter B in relation to parameter A is given a rating of
fuzzification is also considered as a means to handle uncertainty
1/9. Once every possible pair of parameters has been given a rating,
arising from quantifying continuous scale data. Compared to linear
weights are calculated by matrix algebra using the eigenvector
scaling, fuzzy sets give a more realistic standardization output as
method to find the maximum or principal eigenvector of the matrix
using a fuzzy set membership represents a specific relation be-
(Eastman, 2009). The sum of weights must equal 1, which is a
tween the criteria and possible outcomes (Jiang & Eastman, 2000).
requirement of WLC in producing the final susceptibility map. The
The assumption of fuzzy set theory is that any transition be-
calculated weight values are the average of all possible ways of
tween categories is rarely a step function between non-
comparing the parameters and represent the importance of pa-
membership to membership. Rather, there is a gradual change in
rameters to each other (Malczewski, 1999). The ratings used in this
the degree of membership following a specified function. In this
study were obtained from the academic literature and expert
study, the landslide parameters are standardized from 0 to 1, where
knowledge contained in local landslide assessment reports.
0 represents the least susceptible areas and 1 represents the most
The calculation of the weights using a pairwise comparison
susceptible area. The specific fuzzy membership function used is
matrix may result in many variations, a degree of consistency is
the sigmoidal or s-shaped membership produced using a cosine
determined for the ratings. Saaty (1977) proposed an index of
function (Eastman, 2009). The function for each parameter is con-
consistency called the consistency ratio (CR). The CR is a measure of
structed by four control points (a, b, c, d) representing the inflection
the probability the matrix ratings were randomly generated. As a
points of the curve. Information from published literature, local
rule of thumb, a CR value of less than 10% is considered acceptable.
slope assessment reports, and expert knowledge were used to
Other values warrant further evaluation in the matrix ratings.
establish a relationship between parameters and landslides that
aided in the assignment of the fuzzy function control points (Akgun
& Bulut, 2007; BGC Engineering, 2006, 2010). The resulting curves Weighted linear combination
representing the degree of susceptibility, m(x), are characterized as
monotonically increasing, monotonically decreasing, and sym- In producing a final susceptibility map, the weighted linear
metric. The most commonly used and relevant membership func- combination (WLC) method is frequently used for aggregating
tion is the sigmoidal function given by: factor maps (Akgun et al., 2008). The WLC method used in this
P
study is expressed by S ¼ wi $xi, where S is the final susceptibility,
mðxÞ ¼ cos2 a (3) wi is the standardized weight of the factor i and xi is criterion score
of factor i. The standardized criteria maps created using fuzzy
where, in the case of a monotonically decreasing function:


xc p
a¼ (4) Table 1
dc 2 Nine point continuous rating scale.

Rating Comparison
when x < c, m(x) ¼ 1. In the case of a monotonically increasing
function: 1/9 Extremely Less important
1/7 Very strongly

 1/5 Strongly
xa p
a¼ 1 (5) 1/3 Moderately
ba 2 1 Equally
3 Moderately More important
when x > b, m(x) ¼ 1. 5 Strongly
The terms a, b, c, and d are control points that define the shape of 7 Very strongly
9 Extremely
the fuzzy functions.
 S. Dragicevic et al. / Habitat International 45 (2015) 114e125 119

membership functions are multiplied by the weight of each crite-
rion. The set of weights must sum to one in order for the resulting
susceptibility map to have the same range of values as the stan-
dardized factor maps. The weighted criteria maps are then com-
bined by summation to produce the susceptibility map. WLC is an
effective way to combine primary level weights (fuzzy standardized
parameters) with secondary-level weights (parameter importance)
to produce an overall result. Constraints in the form of qualitative
criteria, such as locations of water bodies, can be masked out of the
analysis. The main issue is to define the set of weights that repre-
sent the relative importance of each criterion in conditioning
landslides. AHP is used to assign weights in this study (Ayalew et al.,
2004).
Software and map validation
The GIS software used in this study were Idrisi Taiga (Eastman,
2009) and ArcGIS 9.3 (ESRI, 2009). Idrisi Taiga was used as the
analyzing platform for data manipulation and multicriteria evalu-
ation. ArcGIS was used for some analysis, visualization, and carto-
graphic output.
The validation technique seed cell area index (SCAI) is used to
evaluate the susceptibility maps at each scale (Suzen & Doyuran,
2004). It is calculated by normalizing the area percentages for
each landslide susceptibility class by the landslide cell percentages.
The collection of landslide cells within the susceptibility class is
also termed seed cells. The high susceptibility classes will have low
SCAI values and conversely the low susceptibility classes will have
high SCAI values. The SCAI has been used for susceptibility map
validation in many previous studies (Nefeslioglu, Duman, &
Durmaz, 2008; Suzen & Kaya, 2012; Yilmaz et al., 2012; Yilmaz,
2010).
Results and discussion
Multicriteria evaluation analysis
The parameters used in the analysis are: slope gradient, aspect
gradient, SPI, TWI, distance to drainage network, and elevation. The
control points for each parameter and the parameter ratings were
obtained from a recent survey of the literature employing similar
decision-making approaches (Table 2). These points were associ-
ated to a fuzzy membership for standardization prior to deter-
mining parameter weights. In AHP, each parameter is compared
with one another and given an individual rating. When each pair is
rated, parameter weights are calculated. The measured CR was at
2% meaning the computed weights are well below the acceptable
10% CR threshold. The weight value of slope gradient is highest,
followed by SPI, elevation, TWI, distance to drainage network, and
aspect gradient in decreasing order. The weighted parameter maps
are summated to produce a regional scale landslide susceptibility
map for the Metro Vancouver region. The map has a continuous
scale of numerical values in the range 0e1 making it difficult for
visually interpret. In order to reduce the visual map complexity, the
continuous values were categorized into equal intervals showing
degrees of susceptibility. The five categories were distinguished as
very low, low, medium, high, and very high (Fig. 3). The resulting
susceptibility map shows that the largest proportion of the study
area has low susceptibility occupying 37.53% of the entire area. The
very low, medium, and high susceptible classes make up 10.72%,
20.40%, and 23.72% respectively. The very high susceptibilities ac-
count for 7.64% of the study area. The point locations of past
landslides have been mapped and are used to perform the valida-
tion of this susceptibility map. It is desired that the high and very
high susceptibility classes to have very small SCAI. Conversely, low
and very low susceptibility classes should have higher SCAI. The
calculated SCAI values indicate the Berkley Escarpment map is
reliable as the high and very high susceptibly classes have very low
SCAI values, while the very low and low susceptibility classes are
relatively higher (Fig. 4).

In the sections that follow, the practical implementation and
resulting outputs of the developed GIS-based MCE and fuzzy set
approach at three different scales of analysis are presented.
Regional scale analysis: Metro Vancouver
Municipal scale analysis: District of North Vancouver
Study area
The District of North Vancouver (DNV) is located at the base of
the Coast Mountain Range just north of the City of Vancouver,

Study area Table 2

There have been approximately 200 reported incidences of Regional scale parameter standardization and pairwise comparison matrix.
landslides in the Metro Vancouver Region since 1909 (BGC
Parameter Membership Type Control points
Engineering, 2011). Many of these are weather related with the function
likely trigger being an accumulation of precipitation in preceding
Aspect Sigmoidal Monotonically c ¼ 90 , d ¼ 360
days or an onset of intense precipitation in a short period of time.
decreasing
Vancouver receives on average about 1300 mm of precipitation per Distance to Sigmoidal Monotonically c ¼ 50 m, d ¼ 1000 m
year. A large percentage of this precipitation, mostly rainfall, occurs drainage decreasing
from November to March, and landslides occurrences are strongly Elevation Sigmoidal Symmetric a ¼ 100 m, b ¼ 250 m,
c ¼ 750 m, d ¼ 800 m
correlated with these months.
Slope Sigmoidal Symmetric a ¼ 3 , b ¼ 20 , c ¼ 30 ,
d ¼ 40
Geospatial data Stream power Sigmoidal Monotonically a ¼ 0, b ¼ 300
index increasing
The data for the Metro Vancouver Region was derived from a
Topographic Sigmoidal Monotonically a ¼ 1, b ¼ 12
1:50,000 DEM available online from the GeoGratis Canada National wetness increasing
Topographic Data Base (NTDB) collection. The spatial resolution of index
the DEM is 50 m. Slope gradient, aspect gradient, SPI, and TWI were
Aspect Drainage Elevation Slope SPI TWI Weights
generated from the DEM. The drainage network at a 1:50,000 scale
Aspect 1 0.0448
was obtained from the GeoBase Canada National Hydro Network
Drainage 2 1 0.0688
(NHN) collection as vector lines and polygons. Also, a landslide Elevation 5 3 1 0.1706
inventory was compiled using various sources of information. Local Slope 7 5 3 1 0.4083
newspapers, engineering reports, scientific papers about landslide SPI 3 3 1 1/3 1 0.1772
incidences dating from as far back as 1909 have been documented TWI 3 2 1 1/3 1/2 1 0.1303

as point locations in the inventory (BGC Engineering, 2011). CR ¼ 0.02 < 0.1.
120 S. Dragicevic et al. / Habitat International 45 (2015) 114e125

Table 3
Municipal scale parameter standardization and pairwise comparison matrix.

Parameter Membership Type Control points

function

Aspect Sigmoidal Monotonically c ¼ 90 , d ¼ 360

decreasing
Distance to Sigmoidal Monotonically c ¼ 50 m, d ¼ 1000 m
drainage decreasing
Distance to Sigmoidal Monotonically c ¼ 50 m, d ¼ 150 m
roads decreasing
Elevation Sigmoidal Symmetric a ¼ 5 m, b ¼ 200 m,
c ¼ 450 m, d ¼ 700 m
Slope Sigmoidal Symmetric a ¼ 3 , b ¼ 20 ,
c ¼ 30 , d ¼ 40
Stream power Sigmoidal Monotonically a ¼ 0, b ¼ 300
index increasing
Topographic Sigmoidal Monotonically a ¼ 1, b ¼ 12
wetness increasing
index

Aspect Drainage Road Elevation Slope SPI TWI Weights

Fig. 3. Landslide susceptibility for regional scale: case of Metro Vancouver Region. Aspect 1 0.0423
Drainage 2 1 0.0577
Road 2 3 1 0.1255
Elevation 5 3 2 1 0.1640
British Columbia. Several neighborhoods in the District are built Slope 7 5 2 3 1 0.3432
along the crests and bases of slopes and this may place residents at SPI 3 3 1 1 1/3 1 0.1510
risk. This problem is compound by some property owners who TWI 3 2 1 1 1/3 1/2 1 0.1163
extend their properties by placing fill on the tops of slopes. This CR ¼ 0.03 < 0.1.
action over steepens already unstable slopes. In addition, when
neighborhoods were developed prior to 1980, most storm water
was directed over the edge of escarpments instead of being diver-
procedure using fuzzy sets. Parameter weights were then calcu-
ted into the municipal storm drainage system (Eisbacher & Clague,
lated by AHP pairwise comparison. The measured CR was 3%, which
1981).
is below the acceptable threshold value. The weight value of slope
gradient is highest, followed by elevation, SPI, distance to road
Geospatial data network, TWI, distance to drainage network, and aspect gradient in
The data was provided by the DNV GIS Department. The data decreasing order (Table 3). The landslide susceptibility map for the
include: 10 m spatial resolution DEM, drainage networks, and road DNV was then produced by combining the weighted parameters.
networks. This data was used to calculate slope gradient, aspect The map results in a continuous scale of numerical values (i.e. 0e1)
gradient, SPI, TWI, and distances to drainage and roads. The land- and the susceptibilities categorized into equal intervals for better
slide inventory dataset is identical to that used in the regional scale visual interpretation. Five categories were distinguished as very
study describe earlier. low, low, medium, high, and very high (Fig. 5). From the suscepti-
bility map, the largest proportion of the study area has medium
susceptibility, occupying 31.47 % of the entire area. The very low
Multicriteria evaluation analysis
and low susceptibility classes make up 12.95% and 30.38%, while
The parameters used in the analysis are: slope gradient, aspect
21.20% is occupied by high susceptibility areas. The very high sus-
gradient, SPI, TWI, distance to drainage network, distance to road
ceptibilities account for 4.00% of the study area. The SCAI values
network, and elevation. The control points for each parameter are
shown in Table 3. These points were used in the standardization

Fig. 4. Densities of actual landslides among landslide susceptibility classes and

calculated SCAI values for the Metro Vancouver Region. Fig. 5. Landslide susceptibility for municipal scale: case of District of North Vancouver.
 S. Dragicevic et al. / Habitat International 45 (2015) 114e125 121

Table 4
Local scale parameter standardization and pairwise comparison matrix.

Parameter Membership Type Control points

function

Aspect Sigmoidal Symmetric a ¼ 180 , b ¼ 270 ,

c ¼ 270 , d ¼ 360
Distance Sigmoidal Monotonically c ¼ 5 m, d ¼ 100 m
to crest decreasing
Distance to Sigmoidal Monotonically c ¼ 50 m, d ¼ 1000 m
drainage decreasing
Distance Sigmoidal Monotonically c ¼ 40 m, d ¼ 100 m
to roads decreasing
Slope Sigmoidal Symmetric a ¼ 5 , b ¼ 20 , c ¼ 30 ,
d ¼ 40
Stream Sigmoidal Monotonically a ¼ 0, b ¼ 300
power increasing
index
Topographic Sigmoidal Monotonically c ¼ 2, d ¼ 4
wetness decreasing
Fig. 6. Densities of actual landslides among the landslide susceptibility classes and index
calculated SCAI values for the DNV.
Aspect Crest Drainage Road Slope SPI TWI Weights

Aspect 1 0.0461
Crest 3 1 0.1074
indicate that the generated map for the DNV is reliable since SCAI Drainage 1 1/3 1 0.0409
values are highest from the lowest susceptibilities to lowest in the Road 1 1/3 3 1 0.0572
highest susceptibilities (Fig. 6). Slope 7 3 7 7 1 0.3824
SPI 4 3 4 4 1/3 1 0.2105
TWI 3 3 3 3 1/3 1/2 1 0.1554
Local scale analysis: Berkley Escarpment
CR ¼ 0.04 < 0.1.

Study area Digitization, conversion, and interpolation of the contour intervals

The Berkley Escarpment is located in the DNV and runs along a produced a detailed DEM at a spatial resolution of 1 m. Slope
ridge east of the lower Seymour River area. Houses have been gradient, aspect gradient, SPI, TWI, and the distance to the slope
constructed at the top and bottom of the escarpment slopes. Some crest were calculated directly from the DEM. The stream network
houses have been built too close to the crest and others are built on
top of loose fill added during construction. These activities increase
the local slope making this particular residential area susceptible to
urban landslides in the past. Since the 1970s, six landslides have
originated from the crest of this escarpment, leading to loss of life
and causing extensive property damage (Fig. 7). The landslides that
have occurred were caused by periods of intense rainfall. The latest
failure occurred in 2005, which prompted the Provincial Govern-
ment of British Columbia and the DNV to order evacuations and
purchase at-risk homes.

Geospatial data
The data for the Berkley Escarpment were derived from a
topographic base map of scale 1:2000 consisting of 2 m elevation
contour intervals commissioned by the DNV on 6th March 2006.

Fig. 7. Locations of past landslide events on the Berkley Escarpment. Fig. 8. Landslide susceptibility for local scale: case of Berkley Escarpment in DNV.
122 S. Dragicevic et al. / Habitat International 45 (2015) 114e125
Fig. 9. Densities of actual landslides among the landslide susceptibility classes and
calculated SCAI values for the Berkley Escarpment.
and road network for the area were provided by the DNV in the
form of maps at a scale of 1:2000. The areal extents of the land-
slides were also obtained from DNV maps at the same scale.
Multicriteria evaluation analysis
The parameters used in the analysis are: slope gradient, aspect
gradient, SPI, TWI, distance to drainage network, distance to road
network, and distance to the escarpment crest. The elevation
Fig. 10. Comparison of susceptibilities generated under various spatial scales and resolutions. a) Regional scale with 50 m; b) municipal scale with 10 m; and c) local scale with 1 m.
parameter was removed at this local scale since variation in the
elevation was very low, being less than 100 m variation between
maximum and minimum. Distance from the crest was only used at
this scale because this area was localized at a single escarpment,
allowing the precise delineation of the crest line to be straightfor-
ward. The control points for each parameter are adapted from those
of the regional and municipal studies (Table 4). These points were
used to relate parameter values to their fuzzy membership. The
assignment of weights was achieved by pairwise comparison using
the AHP method and the individual ratings and parameter weights
are presented in Table 4. The measured CR was at 4% meaning the
computed weights are acceptable. The weight value of slope
gradient is highest, followed by SPI, TWI, distance to escarpment
crest, distance to road network, aspect gradient, and distance to
drainage network, in decreasing order. The weighted parameter
maps are summated to produce a local scale landslide susceptibility
map for the Berkley Escarpment. Continuous values of suscepti-
bility were then categorized into equal intervals producing the
following five classes: very low, low, medium, high, and very high
(Fig. 8). From the susceptibility map, the largest proportion of the
study area has very low susceptibility, occupying 54.34% of the
entire area. The low, medium, and high susceptible classes make up
19.87%, 12.01%, and 12.16%, respectively. The very high suscepti-
bilities account for 1.62% of the study area. The locations of past
landslides and their runouts have been mapped and are used to
perform the validation of this susceptibility map. The calculated
SCAI values show that the generated map for the Berkley Escarp-
ment is reliable as the high and very high susceptibly classes have
 S. Dragicevic et al. / Habitat International 45 (2015) 114e125
very low SCAI values, whereas the very low and low susceptibility
classes are relatively much higher (Fig. 9).
Map comparisons
The SCAI values given in Figs 4, 6, and 9 indicate the generated
maps are reliable since the high and very high susceptibility classes
have very low SCAI values, whereas the SCAI values of the very low
and low susceptibility classes are very high. Further, these results
indicate the evaluation technique across multiple scales produced
consistent results. This provides strong reason to perform multi-
scale analyses to confirm the robustness and persistence of findings
across scales. The spatial resolution of data is also an important
factor when performing analyses. Fig. 10 shows the comparison
between resolutions of 1 m, 10 m, and 50 m over the same area. At
the coarse scale of 50 m, broad patterns of susceptibility emerge
showing the escarpment area as a major hotspot. At the medium
scale of 10 m, more detail emerges to show variations in high
susceptibility within the escarpment area. Finally, at the local scale
of 1 m, specific locations are now highlighted. This multiscale
approach provides multiple benefits. It enables confirmation of the
results by overlaying maps of different resolutions to obtain
consensus areas of susceptibilities. In addition, it allows planners to
use the maps separately to support global, regional or local tar-
geting and mitigation efforts. All the maps confirm at various levels
of detail that the escarpment poses the highest landslide risk and
flat residential areas east of the escarpment has low landslide risk.
Conclusions
The developed GIS-based fuzzy multicriteria evaluation and
multiscale approach was implemented on the Metro Vancouver
Region, the DNV, and the Berkley Escarpment as a novel and
comprehensive means to analyze landslide susceptibility at mul-
tiple spatial scales and resolutions. The need for the developed
approach stems from cases where landslide risk situations exist but
fine scale data or traditional data such as land cover, soil types,
geology and lithology does not exist. Multiscale data derived from
the commonly available DEM provides a reliable surrogate. The
MCE utilizing Analytical Hierarchy Process (AHP) and weighted
linear combination (WLC) methods were applied successively to
study areas with cell resolutions of 50 m, 10 m, and 1 m. Careful
consideration was given to the definition of relevant parameters
across different scales since the size of each grid cell is a limiting
factor for certain parameters. The developed approach produced
three landslide susceptibility maps at different spatial resolutions
with each susceptibility map validated using the seed cell area in-
dex (SCAI) technique. The maps are found to be reliable and
internally consistent between and within scales. This makes them
suitable for comparative analysis. Moreover, the approach can
accommodate qualitative appreciation of environmental factors as
defined by expert judgment. The multiscale GIS-based MCE
approach can be useful for planning landslide mitigation studies
beginning at a regional scale and proceeding to finer scales because
the results at varying scales will have different impacts on the
decision-making process. The developed approach can enable an-
alysts in data-scarce environments to reliably identify susceptible
areas thereby improving hazard mitigation, emergency services
targeting, and overall community planning.
Acknowledgments
This study was supported by a Natural Sciences and Engineering
Research Council (NSERC) of Canada Discovery Grant awarded to
the first author. The District of North Vancouver (DNV) GIS
123
Department provided elevation base maps and contour maps. We
thank Mark Leir and Neil Ripley for providing the landslide in-
ventory of the Metro Vancouver Region.
References
Akgun, A., & Bulut, F. (2007). GIS-based landslide susceptibility for Arsin-Yomra
(Trabzon, North Turkey) region. Environmental Geology, 51, 1377e1387.
Akgun, A., Dag, S., & Bulut, F. (2008). Landslide susceptibility mapping for a
landslide-prone area (Findikli, NE of Turkey) by likelihood-frequency ratio and
weighted linear combination models. Environmental Geology, 54, 1127e1143.
Akgun, A., & Turk, N. (2010). Landslide susceptibility mapping for Ayvalik (Western
Turkey) and its vicinity by multicriteria decision analysis. Environmental Earth
Sciences, 61, 595e611.
Ayalew, L., & Yamagishi, H. (2005). The application of GIS-based logistic regression
for landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central
Japan. Geomorphology, 65, 15e31.
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, 432e445.
Ayalew, L., Yamagishi, H., & Ugawa, N. (2004). Landslide susceptibility mapping
using GIS-based weighted linear combination, the case in Tsugawa area of
Agano River, Niigata Prefecture, Japan. Landslides, 1, 73e81.
BGC Engineering. (2006). District of North Vancouver, Berkley Landslide risk
management: phase 2 assessment of risk control options. In Editor (Ed.), Book
District of North Vancouver, Berkley Landslide risk management: Phase 2 assess-
ment of risk control options. Publisther.
BGC Engineering. (2010). District of North Vancouver, Landslide risk summary. In
Editor (Ed.), Book District of North Vancouver, landslide risk summary. Publisther.
BGC Engineering. (2011). Metro Vancouver Landslide inventory [shapefile]. In Edi-
tor (Ed.), Book metro Vancouver Landslide inventory [shapefile]. Publisther.
Blesius, L., & Weirich, F. (2010). Shallow Landslide susceptibility mapping using
stereo air photos and thematic maps. Cartography and Geographic Information
Science, 37, 105e118.
Brabb, E. E., Pampeyan, E. H., & Bonillia, M. G. (1972). Landslide susceptibility in San
Mateo County, California. In Editor (Ed.), Book Landslide susceptibility in San
Mateo County, California. Publisther. Map MF-360.
Carrara, A., Cardinali, M., Detti, R., Guzzetti, F., Pasqui, V., & Reichenbach, P. (1991).
GIS techniques and statistical models in evaluating landslide hazard. Earth
Surface Processes and Landforms, 16, 427e445.
Carver, S. J. (1991). Integrating multi-criteria evaluation with geographical infor-
mation systems. International Journal of Geographical Information Systems, 5,
321e339.
Castellanos Abella, E. A., & Van Westen, C. J. (2007). Generation of a landslide risk
index map for Cuba using spatial multi-criteria evaluation. Landslides, 4,
311e325.
Castellanos Abella, E. A., & Van Westen, C. J. (2008). Qualitative landslide suscep-
tibility assessment by multicriteria analysis: a case study from San Antonio del
sur, Guantanamo, Cuba. Geomorphology, 94, 453e466.
Conoscenti, C., Di Maggio, C., & Rotighano, E. (2008). GIS analysis to assess landslide
susceptibility in a fluvial basin of NW Sicily (Italy). Geomorphology, 94,
325e339.
Dai, F. C., & Lee, C. F. (2002). Landslide characteristics and slope instability modeling
using GIS, Lantau Island, Hong Kong. Geomorphology, 42, 213e228.
Dai, F. C., & Lee, C. F. (2003). A spatiotemporal probabilistic modelling of storm-
induced shallow landsliding using aerial photographs and logistic regression.
Earth Surface Processes and Landforms, 28, 527e545.
Dai, F. C., Lee, C. F., Li, J., & Xu, Z. W. (2001). Assessment of landslide susceptibility on
the natural terrain of Lantau Island, Hong Kong. Environmental Geology, 40,
381e391.
Demoulin, A., & Chung, C. J. F. (2007). Mapping landslide susceptibility from small
datasets: a case study in the Pays de Herve (E Belgium). Geomorphology, 89,
391e404.
Drennon, C. B., & Schleining, W. G. (1975). Landslide hazard mapping on a shoe-
string. Journal of the Surveying and Mapping Division, 101, 107e114.
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, 241e256.
Eastman, J. R. (2009). IDRISI Taiga. In Editor (Ed.), Book IDRISI Taiga. Publisther.
Eastman, J. R., Jin, W. G., Kyem, P. A. K., & Toledano, J. (1995). Raster procedures for
multi-criteria/multi-objective decisions. Photogrammetric Engineering and
Remote Sensing, 61, 539e547.
Eisbacher, G. H., & Clague, J. J. (1981). Urban landslides in the vicinity of Vancouver,
British Columbia, with special reference to the December 1979 rainstorm. Ca-
nadian Geotechnical Journal, 18, 205e216.
Ercanoglu, M. (2005). Landslide susceptibility assessment of SE Bartin (West Black
Sea region, Turkey) by artificial neural networks. Natural Hazards and Earth
System Sciences, 5, 979e992.
ESRI. (2009). ArcGIS 9.3. In Editor (Ed.), Book ArcGIS 9.3. Publisther.
Feick, R. D., & Hall, G. B. (2004). A method for examining the spatial dimension of
multi-criteria weight sensitivity. International Journal of Geographical Informa-
tion Science, 18, 815e840.
124 S. Dragicevic et al. / Habitat International 45 (2015) 114e125
Fell, R., Corominas, J., Bonnard, C., Cascini, L., Leroi, E., & Savage, W. Z. (2008).
Guidelines for landslide susceptibility, hazard and risk zoning for land-use
planning. Engineering Geology, 102, 99e111.
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, 65e83.
Gomes, R. A. T., Guimaraes, R. F., Carvalho, O. A., Fernandes, N. F., Vargas, E. A., &
Martins, E. S. (2008). Identification of the affected areas by mass movement
through a physically based model of landslide hazard combined with an
empirical model of debris flow. Natural Hazards, 45, 197e209.
Gorsevski, P. V., Gessler, P. E., Boll, J., Elliot, W. J., & Foltz, R. B. (2006). Spatially and
temporally distributed modeling of landslide susceptibility. Geomorphology, 80,
178e198.
Gorsevski, P. V., Jankowski, P., & Gessler, P. E. (2006). An heuristic approach for
mapping landslide hazard by integrating fuzzy logic with analytic hierarchy
process. Control and Cybernetics, 35, 121e146.
Guthrie, R. H. (2002). The effects of logging on frequency and distribution of
landslides in three watersheds on Vancouver Island, British Columbia. Geo-
morphology, 43, 273e292.
Hay, G. J., Marceau, D. J., Dube, P., & Bouchard, A. (2001). A multiscale framework for
landscape analysis: object-specific analysis and upscaling. Landscape Ecology, 16,
471e490.
Horton, R. E. (1945). Erosional development of streams and their drainage basins;
hydrophysical approach to quantitative morphology. Geological Society of
America Bulletin, 56, 275e370.
Jankowski, P. (1995). Integrating geographical information systems and multiple
criteria decision-making methods. International Journal of Geographical Infor-
mation Systems, 9, 251e273.
Jankowski, P., & Richard, L. (1994). Integration of GIS-based suitability analysis and
multicriteria evaluation in a spatial decision support system for route selection.
Environment and Planning B: Planning and Design, 21, 323e340.
Jiang, H., & Eastman, J. R. (2000). Application of fuzzy measures in multi-criteria
evaluation in GIS. International Journal of Geographical Information Science, 14,
173e184.
Kia, M. B., Pirasteh, S., Pradhan, B., Mahmud, A. R., Sulaiman, W. N. A., & Moradi, A.
(2012). An artificial neural network model for flood simulation using GIS: Johor
River Basin, Malaysia. Environmental Earth Sciences, 67, 251e264.
Klimes, J. (2008). Use of the deterministic approach for the landslide susceptibility
mapping, Vsetinske vrchy Highland, Czechia. Geografie, 113, 48e60.
Komac, M. (2006). A landslide susceptibility model using the analytical hierarchy
process method and multivariate statistics in penialpine Slovenia. Geo-
morphology, 74, 17e28.
Lai, T. (2011). Modelling spatial dynamics of landslides: Integration of GIS-based
cellular automata and multicriteria evaluation methods. M.Sc. Thesis. Depart-
ment of Geography. Simon Fraser University, Canada.
Lai, T., & Dragicevic, S. (2011). Development of an urban landslide cellular automata
model: a case study of North Vancouver, Canada. Earth Science Informatics, 4, 69e80.
Lee, M. J., Choi, J. W., Oh, H. J., Won, J. S., Park, I., & Lee, S. (2012). Ensemble-based
landslide susceptibility maps in Jinbu area, Korea. Environmental Earth Sciences,
67, 23e37.
Lee, S., & Choi, J. (2004). Landslide susceptibility mapping using GIS and the weight-
of-evidence model. International Journal of Geographical Information Science, 18,
789e814.
Lee, S., & Min, K. (2001). Statistical analysis of landslide susceptibility at Yongin,
Korea. Environmental Geology, 40, 1095e1113.
Luzi, L., & Pergalani, F. (1996). Applications of statistical and GIS techniques to slope
instability zonation (1:50.000 Fabriano geological map sheet). Soil Dynamics
and Earthquake Engineering, 15, 83e94.
Malczewski, J. (1999). GIS and multicriteria decision analysis. Wiley.
Malczewski, J. (2006). GIS-based multicriteria decision analysis: a survey of the
literature. International Journal of Geographical Information Science, 20, 703e726.
Mergili, M., Schratz, K., Ostermann, A., & Fellin, W. (2012). Physically-based
modelling of granular flows with Open Source GIS. Natural Hazards and Earth
System Sciences, 12, 187e200.
Monmonier, M. (1982). Computer-assisted cartography: Principles and prospects.
Prentice-Hall, Inc.
Moore, I. D., Grayson, R. B., & Ladson, A. R. (1991). Digital terrain modelling: a re-
view of hydrological, geomorphological, and biological applications. Hydrolog-
ical Processes, 5, 3e30.
Nefeslioglu, H. A., Duman, T. Y., & Durmaz, S. (2008). Landslide susceptibility
mapping for a part of tectonic Kelkit Valley (Eastern Black Sea region of Turkey).
Geomorphology, 94, 401e418.
Ohlmacher, G. C., & Davis, J. C. (2003). Using multiple logistic regression and GIS
technology to predict landslide hazard in northeast Kansas, USA. Engineering
Geology, 69, 331e343.
Pachauri, A. K., & Pant, M. (1992). Landslide hazard mapping based on geological
attributes. Engineering Geology, 32, 81e100.
Pack, R. T., Tarboton, D. G., Goodwin, C. N., & Prasad, A. (2005). SINMAP 2. A stability
index approach to terrain stability hazard mapping, technical description and
users guide for version 2.0. In Editor (Ed.), Book SINMAP 2. A stability index
approach to terrain stability hazard mapping, technical description and users guide
for version 2.0. Publisther.
Pavel, M., Nelson, J. D., & Fannin, R. J. (2011). An analysis of landslide susceptibility
zonation using a subjective geomorphic mapping and existing landslides.
Computers & Geosciences, 37, 554e566.
Piacentini, D., Troiani, F., Soldati, M., Notarnicola, C., Savelli, D., Schneiderbauer, S., &
Strada, C. (2012). Statistical analysis for assessing shallow-landslide suscepti-
bility in South Tyrol (south-eastern Alps, Italy). Geomorphology, 151, 196e206.
Pourghasemi, H. R., Pradhan, B., & Gokceoglu, C. (2012). Application of fuzzy logic
and analytical hierarchy process (AHP) to landslide susceptibility mapping at
Haraz watershed, Iran. Natural Hazards, 63, 965e996.
Regmi, N. R., Giardino, J. R., Vitek, J. D., & Dangol, V. (2010). Mapping landslide
hazards in Western Nepal: comparing qualitative and quantitative approaches.
Environmental & Engineering Geoscience, 16, 127e142.
Reis, S., Yalcin, A., Atasoy, M., Nisanci, R., Bayrak, T., Erduran, M., Sancar, C., &
Ekercin, S. (2012). Remote sensing and GIS-based landslide susceptibility
mapping using frequency ratio and analytical hierarchy methods in Rize
province (NE Turkey). Environmental Earth Sciences, 66, 2063e2073.
Rossi, M., Guzzetti, F., Reichenbach, P., Mondini, A. C., & Peruccacci, S. (2010).
Optimal landslide susceptibility zonation based on multiple forecasts. Geo-
morphology, 114, 129e142.
Saaty, T. L. (1977). Scaling method for priorities in hierarchical structures. Journal of
Mathematical Psychology, 15, 234e281.
Saaty, T. L. (1980). The analytical hierarchy process: Planning, priority setting, resource
allocation. McGraw-Hill.
Santini, M., Grimaldi, S., Nardi, F., Petroselli, A., & Rulli, M. C. (2009). Pre-processing
algorithms and landslide modelling on remotely sensed DEMs. Geomorphology,
113, 110e125.
Sidle, R. C., & Ochiai, H. (2006). Landslides: Processes, prediction, and land use.
American Geophysical Union.
Sterlacchini, S., Ballabio, C., Blahut, J., Masetti, M., & Sorichetta, A. (2011). Spatial
agreement of predicted patterns in landslide susceptibility maps. Geo-
morphology, 125, 51e61.
Suzen, M. L., & Doyuran, V. (2004). A comparison of the GIS based landslide sus-
ceptibility assessment methods: multivariate versus bivariate. Environmental
Geology, 45, 665e679.
Suzen, M. L., & Kaya, B. S. (2012). Evaluation of environmental parameters in logistic
regression models for landslide susceptibility mapping. International Journal of
Digital Earth, 5, 338e355.
Thanh, L. N., & De Smedt, F. (2012). Application of an analytical hierarchical process
approach for landslide susceptibility mapping in A Luoi district, Thua Thien Hue
Province, Vietnam. Environmental Earth Sciences, 66, 1739e1752.
Thapa, P. B. (2011). Landslide susceptibility modelling in the central Nepal Lesser
Himalaya. Zeitschrift Der Deutschen Gesellschaft Fur Geowissenschaften, 162,
405e420.
Van Westen, C. J., Rengers, N., & Soeters, R. (2003). Use of geomorphological in-
formation in indirect landslide susceptibility assessment. Natural Hazards, 30,
399e419.
Van Westen, C. J., & Terlien, M. T. J. (1996). An approach towards deterministic
landslide hazard analysis in GIS. A case study from Manizales (Colombia). Earth
Surface Processes and Landforms, 21, 853e868.
Van Westen, C. J., Van Asch, T. W. J., & Soeters, R. (2006). Landslide hazard and risk
zonation e why is it still so difficult? Bulletin of Engineering Geology and the
Environment, 65, 167e184.
Voogd, H. (1982). Multicriteria evaluation with mixed qualitative and quantitative
data. Environment and Planning B: Planning and Design, 9, 221e236.
Voogd, H. (1983). Multicriteria evaluation for urban and regional planning. Pion.
Wan, S., Lei, T. C., & Chou, T. Y. (2012). A landslide expert system: image classifi-
cation through integration of data mining approaches for multi-category
analysis. International Journal of Geographical Information Science, 26, 747e770.
Wang, W. D., Guo, J., Fang, L. G., & Chang, X. S. (2012). A subjective and objective
integrated weighting method for landslides susceptibility mapping based on
GIS. Environmental Earth Sciences, 65, 1705e1714.
Weirich, F., & Blesius, L. (2007). Comparison of satellite and air photo based land-
slide susceptibility maps. Geomorphology, 87, 352e364.
Wemple, B. C., Swanson, F. J., & Jones, J. A. (2001). Forest roads and geomorphic
process interactions, Cascade Range, Oregon. Earth Surface Processes and Land-
forms, 26, 191e204.
Wood, L. J., & Dragicevic, S. (2007). GIS-Based multicriteria evaluation and fuzzy sets
to identify priority sites for marine protection. Biodiversity and Conservation, 16,
2539e2558.
Wu, C. H., & Chen, S. C. (2009). Determining landslide susceptibility in Central
Taiwan from rainfall and six site factors using the analytical hierarchy process
method. Geomorphology, 112, 190e204.
Wu, J. G., Jelinski, D. E., Luck, M., & Tueller, P. T. (2000). Multiscale analysis of
landscape heterogeneity: scale variance and pattern metrics. Geographic Infor-
mation Science, 6, 6e19.
Wu, W. M., & Sidle, R. C. (1995). A distributed slope stability model for steep
forested basins. Water Resources Research, 31, 2097e2110.
Xu, C., Xu, X. W., Dai, F. C., & Saraf, A. K. (2012). Comparison of different models for
susceptibility mapping of earthquake triggered landslides related with the
2008 Wenchuan earthquake in China. Computers & Geosciences, 46, 317e329.
Xu, C., Xu, X. W., Lee, Y. H., Tan, X. B., Yu, G. H., & Dai, F. C. (2012). The 2010 Yushu
earthquake triggered landslide hazard mapping using GIS and weight of evi-
dence modeling. Environmental Earth Sciences, 66, 1603e1616.
Yalcin, A. (2007). Environmental impacts of landslides: a case study from East Black
Sea Region, Turkey. Environmental Engineering Science, 24, 821e833.
Yalcin, A. (2008). GIS-based landslide susceptibility mapping using analytical hi-
erarchy process and bivariate statistics in Ardesen (Turkey): comparisons of
results and confirmations. Catena, 72, 1e12.
 S. Dragicevic et al. / Habitat International 45 (2015) 114e125
Yalcin, A., & Bulut, F. (2007). Landslide susceptibility mapping using GIS and digital
photogrammetric techniques: a case study from Ardesen (NE-Turkey). Natural
Hazards, 41, 201e226.
Yilmaz, C., Topal, T., & Suzen, M. L. (2012). GIS-based landslide susceptibility
mapping using bivariate statistical analysis in Devrek (Zonguldak-Turkey).
Environmental Earth Sciences, 65, 2161e2178.
Yilmaz, I. (2009). Landslide susceptibility mapping using frequency ratio, logistic
regression, artificial neural networks and their comparison: a case study from
Kat landslides (Tokat-Turkey). Computers & Geosciences, 35, 1125e1138.
125
Yilmaz, I. (2010). The effect of the sampling strategies on the landslide susceptibility
mapping by conditional probability and artificial neural networks. Environ-
mental Earth Sciences, 60, 505e519.
Yilmaz, I., & Keskin, I. (2009). GIS based statistical and physical approaches to
landslide susceptibility mapping (Sebinkarahisar, Turkey). Bulletin of Engineer-
ing Geology and the Environment, 68, 459e471.
Zadeh, L. A. (1965). Fuzzy sets. Information and Control, 8, 338e353.