Feasibility assessment of multi-criteria decision making and

quantitative landslide susceptibility methods: A case study of Mao-

Maram Manipur.
K Khusulio1, Rohan Kumar2
1
Research scholar, Lovely Professional University, Phagwara, Punjab
2
Asst. professor, Lovely Professional University, Phagwara, Punjab
1
[email protected] [email protected]

Abstract
The present work aimed to assess the feasibility of three methods namely, analytical hierarchy
process, frequency ratio and binary logistic regression that are used for landslide susceptibility
analysis. Twelve causal factors such as slope, relative relief, road buffer, fault/fold and thrust
(FFT) buffer, topographic wetness index (TWI), aspect, stream power index (SPI), drainage
buffer, profile curvature, land used/land cover (LULC), soil and geology as well as a landslide
inventory were retrieved using multi-source spatial data. Correlation between landslides and
causative factors was estimated using frequency ratio on a geospatial platform with 234 landslide
locations which were extracted using field visits and remote sensing imageries and found to be
reasonably influencing landslides in the study region. These factors were then used in all three
landslide susceptibility models for the calculation and generation of a landslide susceptibility
map (LSM). LSM of the study region Mao-Maram (tehsil) of Manipur was generated with five
classes very low, low, medium, high, and very high. Cumulative percentage curve and receiver
operating characteristic (ROC) approaches were used for validation of the model performance.
Cumulative percentage curve area under curve (AUC) accuracy for frequency ratio, analytical
hierarchy process and binary logistic regression were found to be 77.24%, 73.22%, and 80.32%
respectively. ROC curve under AUC for frequency ratio 88.7%, analytical hierarchy process
77.4%, and binary logistic regression 90%. A landslide susceptibility map derived from every
three models can be helpful for city or village planners and can be used for planning,
development, mitigation, and disaster prevention.
Keywords. Landslide; GIS; Frequency ratio; Analytical hierarchy process; Binary logistic
regression.

1. Introduction
In the last four decades or so, the world has seen a variety of disasters such as earthquakes,
landslides, volcanic eruptions, floods, hurricanes, tornadoes, blizzards, tsunamis, cyclones and
wildfires that are occurring frequently. India is particularly vulnerable to floods, earthquakes,
drought, cyclones, glacial lake outburst flood (GLOF), cloudbursts, heatwave, flashflood and
landslides (https://ndma.gov.in/). Landslides are the most frequent disaster occurring during the
months of raining season in a large part of India. Himalayan states including J&K, Uttarakhand,
Himachal Pradesh, Sikkim, Arunachala Pradesh, Tripura, Manipur, Nagaland, Meghalaya and
part of Assam; Peninsular India states including Maharashtra, Karnataka, Kerala, Tamil Nadu
and part of Madhya Pradesh are frequently witnessing landslide events. Landslides are also
reported from states like West Bengal, Jharkhand and Orissa. The impact of landslide disasters is
amplified in the regions where the topography is rugged and human settlements are remotely
established (Kumar and Anbalagan 2016; Shano et al. 2020). Landslide events have been
worsened by anthropogenic activities such as the construction of infrastructure in hilly terrain
(dams, tunnels, roads, canals, townships, and so on). Landslides inflict severe damage to life,
property and geo-environmental conditions mainly in a hilly region. Landslide can be defined as
outward or downward movement of soil, rock, debris, mud, etc., along the slope they include
topples, rock falls, rock slides, rock slump, debris slide, debris flow, mudflow, etc., (Varnes
1978). Landslides are caused by several different causes, including geological, geomorphologic,
and climatic reasons. Normally the most significant components are geomorphology (slope
gradient, aspect, and relative relief), bedrock geology (structure, lithology and degree of
weathering) soil (depth, permeability, structure, and porosity), hydrologic conditions and land
use and land cover. Utilizing scientific inquiry, we may analyze and anticipate landslide-prone
places, and so limit landslide-induced devastation by applying proper mitigating measures
(Kundu et al. 2013; Galve et al. 2016; Pourghasemi et al. 2020). Landslide hazard zonation
(LHZ) can be defined as the general sense of division of land surface into an area and zoning of
such area according to a degree of actual and potential hazard from other mass movement or
landslide on the slope (Varnes 1984). The present study area Manipur is a part of Purvanchal
Himalayan hill which is highly prone to earthquakes (zone-5). The vulnerable nature of the
terrain in combination with high precipitation with the rapid development of infrastructure and
construction of the heavy structure in a hilly region of Manipur is resulting in increased landslide
activities. In a recent landslide incident in Manipur, at least 14 people were killed and 72 others
disappeared after a massive landslide occurred at a railway construction site in the Noney district
on June 29, 2022. Several territorial army personnel, railroad employees, and locals were
impacted by this tragic incident, and the landslide also blocked the Ijei river, creating a reservoir
that may overwhelm nearby homes (https://www.republicworld.com/). Although the landslide
situation in Manipur is severe, no effective slope stability assessment has been used in the state
of Manipur, India, even though 90% of Manipur is extremely prone to landslide activity every
year. Mao-Maram in the state of Manipur is one of the locations that is regularly impacted by
rainfall-induced landslides, and it has yet to be explored. Landslides in this area block the
national highways and other highways for days making it difficult for transportation and
travelling. Local people living in this region have faced numerous problems due to landslides
such as damage to houses, agricultural activities, farmlands and also shifting of colonies due to
creak in the residences caused due to landslides. As a result, evaluation in this region and
landslide susceptibility mapping may give helpful information in reducing disaster losses and act
as a guideline for long-term land-use planning.
Identification of current landslides, evaluation of existing causal variables, and creation of
landslide susceptibility maps are all required for landslide mitigation in areas that are already
failing or prone to falling (Wubalem 2021). The goal of the landslide susceptibility map is not
only to determine which variables have the greatest impact on landslides in the area but also to
estimate the proportional contribution of each component to slope collapse. It is also critical to
establish a link between the causes and landslides to predict future landslide hazards (Chen and
Wang 2007). Landslide susceptibility mapping was challenging a few years ago owing to a lack
of remote sensing techniques and advances in GIS technologies. However, with the progress of
computers, remote sensing, and GIS, creating a landslide susceptibility map is now simple
(Pradhan 2013; Wubalem 2021). Landslide susceptibility maps may be created using a variety of
quantitative and qualitative methodologies. Bivariate regression analysis, multivariate regression
analysis, logistic regression analysis, fuzzy logic, and artificial neural networks are examples of
quantitative methodologies employed by researchers (Das and Lepcha 2019). Many researchers
have used many models and compared them to see which one is the most accurate. Quantitative
models for landslide susceptibility evaluations may be split into three categories: probabilistic,
deterministic, and statistical models (Peethambaran et al. 2020). Deterministic models are suited
to site-specific techniques due to the need for extensive geological, geomorphological and
geotechnical data relevant to slope stability, probabilistic models are based on the observed
correlation between landslide causal factors (CF) and landslide inventory whereas Statistical
models are more objective and are commonly used in landslide susceptibility evaluations to build
an appropriate framework in prioritizing the parameterization of it (Vieira et al 2010; Erener and
Düzgün 2010; Conforti et al. 2012; Peethambaran et al. 2020). Despite all of the technological
advancements made by many researchers and scientists on the topic of landslides, most of the
approaches used to evaluate landslide susceptibility and risk have drawbacks. However,
assessing landslide risk and vulnerability is a time-consuming process that requires the
integration of diverse approaches, methodologies, and technologies, as well as the combination
of varied skills (Singh et al. 2020). The major goal of this research is to build a landslide
susceptibility map for the Mao-Maram tehsil of Manipur state by taking into consideration the
availability of data, structural stability provisions, and the geographical characteristics of the
study region. In this study to create maps of landslide susceptibility for the region three models
that are frequency ratio, logistic regression, and analytical hierarchy process models were
utilized with a combination of remote sensing and GIS software. The analytical hierarchy
process, is a semi-quantitative approach also known as the systematic approach, is a multi-
criteria decision analysis method for evaluating landslide susceptibility mapping (Pardeshi et al.
2013; Arabameri et al. 2019). The binary logistic regression and frequency ratio model is
considered a quantitative approach where frequency ratio is among the most effective
quantitative tools for analyzing slope instability whereas the logistic regression model is among
the most dependable mathematical approaches for determining the association between landslide
causative variables and landslide sites (Das and Lepcha 2019).
Recently, several researchers have used multi-criteria decision analysis and statistical model
methods to build landslide susceptibility maps, and some of them have combined such models
with additional methodologies including frequency ratio, analytical hierarchy process and
logistic regression methods. Landslide susceptibility mapping is significant for ensuring human
safety and reducing the damaging effect on a country's regional and national economies. By
creating a landslide susceptibility map, it is vital to analyse the factors that play a larger role in
triggering slope collapse and to limit their socioeconomic repercussions. Statistical approaches
such as frequency ratio, logistic regression, and certainty factor approaches were used for this
aim which is simple to use and provide a well-intentioned outcome (Wubalem 2021). From the
literature review, it is seen that several researchers have used various statistical models such as
frequency ratio and logistic regression approach in the past for landslide susceptibility mapping
(Anbalagan et al. 2015; Das and Lepcha 2019; Shano et al. 2021; Sujatha and Sridhar 2021).
Several studies for landslide susceptibility using multi-criteria analysis like the analytical
hierarchy process was also done in the past (Moragues et al. 2020; Bahrami et al. 2020; Biswas
et al. 2021). Natural hazards like landslides cannot be prevented completely but they can be
minimized with the proper awareness of the scientific method and behaviour of the process
involve and can be a valuable tool for the reduction of natural susceptibility (Gokceoglu and
Sezer 2009; Pourghasemi et al. 2018, 2020). From the literature review it has been observed that
only a few landslides susceptibility study has been done for Manipur such as by Singh et al.
2011; Balamurugan et al. 2016; Okendro and Kushwaha 2018; Chanu and Oinam 2021. From
this review, it is also noticed that there is a lack in the assessment of the feasibility of the terrain
factors and no proper authenticate studies have been carried out so far for the landslide
susceptibility to assess the feasibility of the region. The accuracy obtained from the area under
the curve was also found to be poor and the study carried out for susceptibility study for
landslides was found to only focus along the road side leaving aside other important regions
which are also affected by a landslide resulting in loss of life and damaging properties. The study
for Mao-Maram (Manipur) region is an important issue for the people of the Mao-Maram sub-
district as landslides happen in this region every year. This region with high landslide-prone
areas is very little explored yet in terms of natural disasters and many other fields or
departments. Landslide studies for the terrain like the present study region are very few, and the
few studies found for this terrain are only focusing on the highway, leaving aside some important
areas like human settlements, villages, towns, agricultural land, farmland, etc., which are
affected every year during a landslide incident. The current study covers the entire Mao-Maram
sub-district area and involves the creation of a landslide susceptibility map for the area, which
will be useful in the mitigation, planning, and development processes. In landslide-prone areas,
the local authorities need to know about the landslide hazard area and take appropriate measures
to reduce losses and damage in these areas. Prevention of landslides is very important due to the
huge losses caused by landslides and preventing material costs and human losses should be given
priority.

2. Study Area
The present study region Mao-Maram lies in Manipur and is located in the northmost part of
Manipur, northeastern India bordering Myanmar. Manipur state is also bordered by the Indian
state of Assam to the west, Nagaland to the north, Mizoram to the southwest, and Myanmar
country to the east and south. The study area has an elevation of about 1757 to 1788 meters and
covers latitude 25°24́ʹ-25°14ʹN and longitude 93°45ʹ-93°24ʹE. In the state of Manipur NH-2,
NH-53 and NH-150 connect the rest of the country though, NH-2 is believed to be the sole
lifeline of Manipur since the two other national routes are not sufficiently maintained where
large trucks can't go over these highways. In the tehsil study area, there are two major tribes, the
Mao tribe and the Maram tribe both of which have been settled there for a significant amount of
time. These tribes engage in activities such as Jhum cultivation, terrace cultivation, animal
rearing, and other related activities, and they take advantage of the National Highway 2 (NH-2)
to transport any excess goods they produce to market. The present study area consists of fragile
lithology due to the Disang and Barail groups of rock which belong to a tertiary period of eroded
hilly terrain and fracture geology. These two types of rock have severe extensive weathering,
tectonic collision activities and anthropogenic behaviour. Disang is a dark grey to black shale
interbedded with fine-grained sandstone, splintery and siltstone in a thick pile because of the
junction of bedding with a few notable sets of fracture cleavages (Balamurugan et al. 2016).
Overlying the Disang group, the Barail are characterized by a thick column of arenaceous layers
that are interbedded with shale (Balamurugan et al. 2016). The soil type of Manipur can be
classified into two broad types, firstly alluvium in the valleys and secondly the red ferruginous
soil in the hilly area. The topsoil on the high slopes is very thin as compared to the plain areas
which are considerably thicker, especially in the flood plains. Steep slope soil is considered to be
subjected to high erosion. The present study region is also prone to landslides and every year,
mainly during monsoon, it experiences landslides and blocks the roads for days. Manipur has a
geological setting consisting of the Arakan-Yoma mountains that are characterized by
exceptionally steep, weak slopes, and rocky. Anthropogenic activities in the study region are due
to slope cutting and widening of roads for development which causes frequent landslides in
Manipur state. Triggering factors of landslide incidents in the state of Manipur are heavy rainfall
and seismic activity and most frequently throughout the monsoon season because of heavy
rainfall and also under the major influence of seismic activity due to subduction of plate tectonic
activity, where the Indian plate is subsiding to the Eurasian plate and may anytime cause huge
seismic motion resulting in landslide and earthquake. The present study area Mao-Maram which
is a sub-district in the state of Manipur also falls under Purvanchal Himalaya which is largely
composed of geological sandstone and shale formations.
[Figure: 1.]
3. Landslide inventory of the area
The identification and mapping of historical landslide sites are important to study the connection
between landslide distribution and the various causal factors. In the present study area, a total of
234 landslide location was recorded with the help of field observation, historical information,
and image interpretation. Remote sensing data may be used to identify landslides based on
spectral features, form, and contrast, as well as morphological expression (Kanungo et al. 2006;
Balamurugan et al. 2016; Du et al. 2020). The overall size of the landslides was determined to be
between 25 to 400 meters square but mostly the landslide size was observed to be around 25 to
50 m2. During a field visit to the study area for landslide inventory and field study, it was
discovered that landslides are primarily caused by anthropogenic activity around road
construction. In general, landslides found during field study are rotational type of landslide,
rockfall type of landslide and earthflow type of landslide (Fig 2). One of the major reasons for
finding many landslide locations on these roads is random construction of roads without
scientific knowledge and randomly cutting the slope without being aware of the hazardous
outcome and leaving the scarp face untouched.
Throughout the rainy season, the untreated cut slopes collapse in some spots. Such
breakdowns were reported on both eroded soil, debris, and rocky slopes. Landslides were also
observed more in the area where there is the intersection of two rock formations that is Disang
and Barial group of rock formations. With a comparison between two types of rock formation,
landslides were observed more in the Disang group of rock which is older compared to the Barail
group of rock. During the monsoon season, landslides were also found in the human settlement
area due to poor drainage runoff, improper building construction and site selection for
construction without proper mitigation steps. Many landslide locations were also observed in an
area with steep slopes and along the newly constructed road or old road constructed region where
the cut slope was left untouched after the road construction was done without any scientific
precaution. In some areas, landslides are caused by fragile lithology with the presence of
lineament, which is eroded or weathered for various reasons such as sunlight reaction, weak
material composition, a chemical process or mechanical process, etc.
[Figure: 2.]
4. Data preparation
The identification and creation of a geospatial database of landslide causative factors are
important in the zoning of landslide susceptibility zones. There are no universally accepted rules
for selecting the causative criteria for landslide susceptibility mapping (Balamurugan et al. 2016;
Kumar and Anbalagan 2019). In this study area, for landslide susceptibility analysis 12 causative
factors were selected. Geospatial data used for extraction of geo-environment causative factors
are discussed below in table 1.
[Table: 1.]
All the causative factors were converted into a raster image and were resampled to 25 m × 25 m
resolution according to the required studies to be carried out. Landslide inventory was collected
during the field visit as well as using google earth pro of the past landslide location and history
of landslide incidents. Aside from satellite images, supplementary data such as a topographic
sheet, a soil map, historical landslide data, and a geology map were acquired from several
sources. Ancillary data were transformed into raster format and used for the landslide
susceptibility map based on cell size (25 m × 25 m).

5. Causative factors
In this paper, 12 landslide parameters have been chosen for the analysis namely geology, aspect,
relief, slope, topographic wetness index (TWI), profile curvature, stream power index (SPI),
drainage, road, soil, FFT (fault, fold, and thrust). These data include primary data, secondary
data, and ancillary data.

5.1 Geology
Geology plays a significant causal factor role in influencing instability in hilly terrain. The
geological map of the study region shows two major types of rock Disang and Barial groups in
the area. The oldest forms are the Disang (Eocene Age), which are approximately 58 million
years old. The Disang group of rocks inhabit nearly the whole eastern part of the Manipur state.
The Disang is overlaid by the Barail, which are Oligocene in age (approximately 36 million
years). The Barial group has three types of formations: Renji, Jenam, and Laisong, whereas for
Disang there are two types of formations: Upper Disang and Lower Disang. The Disang group of
rocks is more brittle than the Barial group of rocks. From field investigation, it has been noted
that the rocks are mostly weathered and eroded, with lots of fractures and joints in the bedding.
The most common rock type found in the study region is sandstone and shale with a few
siltstones which are all sedimentary types of rock (table 2). The geological map of the research
region was generated from the published geological map of Manipur (Rajkumar and Klein 2014).
[Table: 2.]
5.2 Land use land cover (LULC)
In hilly terrain, LULC has a crucial role in influencing landslides. Changes in land-use patterns,
such as deforestation, or a rise in agricultural and building activity, change the region and
transform it, making it more prone to landslides. Using remote sensing technology, a Sentinel 2B
satellite image, image processing software, and a geographic information system, the LULC map
was made. The spectral bands of visible light (red) and near-infrared (NIR) were used for
generating the Normalized Difference Vegetation Index (NDVI) using the formula,
NDVI= (NIR-RED)/(NIR+RED) (1)

Five classes were derived from LULC, namely water bodies, barren land or settlements,
agricultural land, scrub forest, and dense forest. The LULC pattern of the terrain has a significant
impact on landslide hazard research. Most landslides were found to be linked to settlements,
barren land, and agricultural land classes.

5.3 Soil cover

A soil map is a type of ancillary data generated from the “food and agriculture organization of
the United Nations (FAO/UNESCO 1974) soil map of the world” by masking out the study area
soil map into a vector and later converting it into a raster with a 25×25 m grid cell. In the study
area, two types of soil are found according to the global soil map and are classified into orthic
acrisols and humic cambisols. Orthic acrisols are light-coloured or thin, have low organic carbon
content which is huge and hard when dry, and have clay-rich subsoils. They are linked to humid,
tropical climates, which frequently support forest regions. Humic cambisol soils spontaneously
develop on medium-to fine-textured parent materials under various climatic, topographic, and
vegetative-cover circumstances. It has been found that orthic acrisols are more affected by
landslides than humic cambisols. This is because more landslides were found in orthic acrisols
than in humic cambisols.

5.4 Slope gradient

The slope gradient defines the rate of elevation change in the orientation of the steepest slope.
The continuous slope angle information was hail from the ALOS PALSAR DEM. Slope angle
was observed to differ in the range of 0–79°, and it was classified into the five relative classes:
very low (0-12°), low (12-21°), medium (21-29°), high (29-39°) and very high (> 39°) due to its
natural effect on the landslide. Slope angle was classified from the continuous range of slope
angle data into categorize relative classes. It is often assumed that places with a steep slope are
more vulnerable to landslides than regions with a low slope gradient (Anbalagan 1992; Kumar
and Anbalagan 2019).
5.5 Aspect
The area aspect was obtained using the ALOS PALSAR DEM and divided into nine categories:
flat, north, northwest, northeast, southeast, south, west, southwest, and east. Landslide
vulnerability is often influenced by topographic factors. The slope aspect affects the
concentration of sunlight on slopes, which is linked to temperature and other environmental
conditions. This research location is located in the Purvanchal Himalayan region, and the
influence of the topographic aspect may be seen in the form of forest on moist and warm south-
facing slopes, while north-facing slopes are mostly dry and chilly. Because of the considerable
precipitation, slopes with a south aspect have a greater tendency to landslides in the Himalayas.

5.6 Relative relief

Relative relief was obtained using ALOS PALSAR DEM and was determined to be in a range of
0-235 m. The differences among the lowest and highest points of elevation between a feature or
region is referred to as relative relief (Kumar and Anbalagan 2019). Relative relief is categorized
into five groups: very low (0-47 m), low (47-97 m), medium (97-141 m), high (141-188 m), and
very high (> 188 m) for landslide susceptibility mapping. Field studies have shown that places
with greater relative relief are more likely to have landslides than places with a small range of
relative relief.

5.7 Fault, Fold, and Thrust (FFT)

The FFT map was derived from GSI (Bhukosh) by downloading the fault and fold of the study
area. Landslides may be generated by surrounding fault activity and use fault planes as a failure
mechanism. In the studied region, geological features such as faults, folds, and thrusts are widely
spread. Fault, fold, and thrust can also be included in linear structural discontinuity and can be
captured in the form of photo lineaments. It is well accepted that landslides are more prone to
occurring along lineaments. The buffer zone of FFT has been created to analyse the influence of
landslides in these regions around the FFT zone, covering 0-100 m, 100-200 m, 200-300 m, 300-
400 m, and > 400 m. It was observed that the density of landslides was observed to be greater in
the vicinity of the FFT zone.

5.8 Distance to drainage

Drainage is an essential component that represents an area's slope development and serves as an
indication of mass wasting and associated erosion issues (Balamurugan et al. 2016). The
drainage buffer map of the study area was generated in digital elevation mode (DEM). The study
region has a high density of drainage and a high stream order of around seven steam orders. The
area with rugged terrain has highly supported and deeply marked ways for drainage systems.
Drainage is often shown to be connected with landslides. Rivers regularly erode their banks and
create steeply plunging slopes in hilly terrain, which are often the location of successive
landslides (Kumar and Anbalagan 2019). The drainage buffer map was classified into five
classes: 0-100 m, 100-200 m, 200-300 m, 300-400 m, and > 400 m.

5.9 Topographic wetness index (TWI)

The topographic wetness index (TWI) is considered to be a secondary attribute that can be
derived from DEM. TWI is related to flow buildup on particular terrain. It is useful in
understanding soil moisture conditions and other associated phenomena. The soil water content
and TWI of a terrain rise as the catchment area expands and the slope gradient decreases. TWI
regulates the flow and collection of water in a landscape. TWI is related to other characteristics
such as evapotranspiration, distribution of water saturation zones, sand or silt content, and
vegetation. TWI can be obtained using the following formula (Kumar and Anbalagan 2019):
CA
TWI = ln (2)
tan slp

where CA denotes the catchment area and slp denotes the area's slope gradient. The TWI
determination is useful in defining the water table conditions and saturation zone. TWI is further
classified into five classes: very low (0.767-3.638), low (3.638-4.485), medium (4.485-5.779),
high (5.79-7.594) and very high (> 7.94).

5.10 Stream power index (SPI)

Stream power index (SPI) is derived from ALOS PALSAR DEM and it is a secondary attribute
that reflects the erosive force of the stream. SPI is related to slope gradient and catchment area.
An increase in the catchment area and slope gradient increases a stream's erosion potential. The
SPI resulted in the determination of the erosive strength of the area's streams, and the following
formula may be used to compute it (Kumar and Anbalagan 2019):

SPI = ln (CA × tan slp) (3)

where CA denotes catchment area and slp denotes the region's slope degree. SPI is classified into
five classes: very low (13.82--5.22), low (-5.22--0.648), medium (-0.648-1.048), high (1.09-
3.49), and very high (> 3.49).

5.11 Distance to road

A road map was generated from Google Earth for the study region by digitizing, which is a type
of ancillary data. A considerable number of landslides were seen on the untreated cut-slopes that
happened during the road's construction. The majority of the slopes cut for the construction of
the road are steeply sloping towards the road. During the monsoon season, cut hillsides become
more susceptible to landslides. Road buffers were created for the study area and were classified
into five classes: 0-100 m, 100-200 m, 200-300 m, 300-400 m, and > 400 m.

5.12 Profile curvature

Profile curvature was derived for the study area from ALOS PALSAR DEM by using image
classification software. "Profile curvature is calculated parallel to the greatest slope direction,
with a negative value indicating that the surface is upwardly convex, a positive value indicating
that the surface is upwardly concave, and a value of zero indicating that the surface is linear
(ESRI, 2012)." Profile curvature influences the acceleration or slowdown of flow over the
surface and may be linked to mass movement or erosion processes. Profile curvature has three
classes and has been classified into convex (-28.799--0.1), linear (-0.1-0) and concave (> 0).
[Figure: 3.]
[Table: 3.]

6. Methodology
6.1 Binary logistic regression method (BLR)
The present study used the binary logistic regression (BLR) methodology, which is a
multivariate method, to estimate LSM. The BLR approach was used in this study to identify
LSM. The process begins with training data, which includes recognizing landslide location and
non-landslide location. The BLR technique for LSM uses the landslide dataset as a binary
dependent factor and geo-environmental characteristics as independent factors (variables). A
total of 234 landslide occurrences was recorded in point vector format for the study area, with
155 landslide locations taken for the BLR technique training which is roughly 65% and the
remaining 79 location was saved for validation. The majority of the landslides were found to be
shallow, with diameters roughly similar to the grids (25 m×25 m) employed for this study. A
binary logistic regression dataset was prepared into an equal number of landslides and non-
landslides for analysis. As a result, a geographic data set of 155 landslide occurrences and 155
non-landslide events were labelled with 1 and 0, respectively, which was constructed and
organized using independent components. A significant benefit of the BLR technique for the
LSM over other multivariate statistical approaches is that projected values lie between 0 and 1.
(Ohlmacher and Davis 2003; Vakhshoori et al. 2019; Shano et al. 2021).
To be employed in the BLR approach, independent variables/factor classes (X1, X2,
X3...Xn) might be continuous, categorical, or a mix of both. The following formula may be used
to calculate BLR:

1
P= −Z (4)
1+e
The likelihood of a landslide occurring depending on a substantial independent variable is
indicated by the letter P. Z is a linear combination with a range of -∞ to +∞, with -∞ to 0
indicating a negative effect and 0 to + ∞ indicating a positive impact of independent factors on
the incidence of landslides. The symbol Z may be written as:

n
Z=α + ∑ β i X i (5)
i=1

where α is a constant referring to the method intercept and β i is the coefficient of the
independent variable X i . The BLR technique determines the regression parameters α and β i
based on the involvement of the dependent factor in the independent variable (Kundu et al. 2013;
Riegel et al. 2020; Chowdhuri et al. 2021). Picking the best fit function and, as a result,
computing α and β i are essential components of a BLR technique. The method generates
coefficients (β), which are utilized in cell-by-cell probability estimations for the region in
general.

6.2 Frequency Ratio Method

The landslide frequency ratio is determined by the density of landslides in each factor class. Each
factor class of frequency ratio value represents its importance for landslide occurrence. It follows
the laws of conditional probability, in which a frequency ratio greater than one (>1) indicates a
strong link of factors/classes with landslide occurrences, and a frequency ratio less than one
(<1) indicates a lesser relationship of variables with landslide occurrences. For landslide
susceptibility mapping, normalized values ranging from 0 to 1 were utilized to assign ratings to
the factor classes (Shano et al. 2021). Each component was reclassified based on the rating
values of their respective classes. The frequency ratio can be calculated using the given formula
below,

%l i
frd = (6)
%a i

Where, frd is term as a frequency ratio of the causal factors, % l i is the proportion of landslides
in the causative factor class, and % ai is the area percentage of the causative factor class of the
total map.
6.3 Analytical Hierarchy Process (AHP) Method
The AHP approach was used to calculate the weights of each class factor. Saaty (1980)
developed AHP, a multi-criterion decision-making approach that permits objective as well as
subjective elements to be incorporated in the judgment process (Kumar and Anbalagan 2016; Sur
et al. 2020). Weights are allocated to factors and alternatives on a nine-point ordinal scale based
on the pair-wise comparison. Factors or their classes are organized in the form of a matrix with
an equal number of columns and rows, with scores noted on one side of the diagonal and values
of 1 put in the matrix's diagonal (Saaty 1977; Myronidis et al. 2015; Kumar and Anbalagan
2016; El Jazouli et al. 2019; Sur et al. 2020). Yalcin (2008) underlined the need of considering
both subjective and objective factors in the decision-making process. In this present study area
relative worth of each pair of factors/classes was evaluated using expert information gained
through fieldwork and the prevalence of landslides in those classes. When the variable on the
vertical plane is more significant than the variable on the horizontal plane, this number ranges
between 1 and 9, and when the variable on the horizontal plane is more essential, the value varies
between both the reciprocals 1/2 and 1/9. The eigenvector of a matrix is used to calculate factor
or class weights. The transitivity rule may be violated by the subjective choice rule, resulting in
inconsistency (Feizizadeh and Blaschke 2012; Kumar and Anbalagan 2016; Sur et al. 2020;
Devara et al. 2021). The determination of the consistency ratio (CR) and consistency index (CI)
is an essential part of the AHP approach. Saaty (1980) defined the consistency index as follows:

( λ max−N )
CI = (7)
N−1
Where N is the number of elements in the matrix's row/column and λmax is the maximum
eigenvalue. CR was determined by the ratio CI/RI, where RI stands as a random index. Saaty
(1980) created the random index using several random samples. The highest consistency
threshold of the matrix is 0.1 CR. A CR value greater than 0.1 is considered inconsistent, while a
value of 0 indicates a fully consistent comparison result.
AHP multi-criteria decision-making offers a highly flexible and straightforward decision-
making process that may easily be accommodated in the GIS domain. The Landslide
susceptibility index (LSI) was calculated next. It was calculated using the weighted arithmetic
sum approach, which can be expressed as follows:
 J =n
LSI= ∑ Weight of factor (W J )×Weight of factor classes(W iJ ) (8)
J =1

where WiJ signifies the weight of the ith factor class J.

7. Result and Discussion

In this research paper, GIS software and statistical tools used are ArcGIS (10.7), QGIS (3.18),
and SPSS tools. For the study, 12 selected causative factors influencing landslides are analyzed,
maps are created, and data is extracted using this three software. Remote sensing tools like
satellite images such as Sentinel 2B, ALOS PLASAR, and Google Earth Pro were also used
during the research studies. Field investigation of the study area and collection of ancillaries
from different sources also played an important role in this research paper.
7.1 Binary logistic regression
The statistical analysis in this research was done using SPSS software. It provides multiple
approaches for selecting the best predictors to be included in the procedure in a stepwise manner
(Mathew et al. 2013; Vakhshoori et al. 2019; Saha et al. 2020). The research took into account a
total of 53 independent variables from 12 distinct classifications. The forward step-by-step
method began with 0 variables and ended with 35 variables after ten steps. Variables with a
significance threshold of <0.05 were kept in each phase, while those with a significance
threshold of >0.05 were removed. Equation 9 denotes to Wald chi-square test:

( )
2
2 βi
χ= (9)
SE

From equation 9 SE represents a standard error, which may be written as SE=(s/√n), s is the
standard deviation of the input samples, and n is the sample size in the input data. In categorizing
binary training data, the BLR technique has an accuracy of 85.2 %. Based on the above-
mentioned statistical findings, a logistic regression (equation 10) was developed.
Z = 18.833 – (2.500*NA) – (3.229*NEA) – (2.104*EA) – (2.339*SEA) – (1.671*SA) –
(0.914*SWA) – (1.005*WA) – (0.547*D100M) – (1.558*D200M) + (0.072*D300M) –
(0.201*D400M) – (0.567*GB) + (23.569*LSRS) + (4.253*LSG) + (3.085*LV) + (1.884*RL) +
(3.018*RM) + (1.996*RH) + (1.873*R100M) – (0.768*R200M) – (1.607*R400M) –
(0.016*R800M) – (1.540*SVL) – (3.054*SL) – (3.159*SM) – (1.616*SH) – (0.699*SHC) +
(1.332*SPIVL) + (1.740*SPIL) + (0.981*SPIM) – (0.120*SPIH) – (20.326*TWIVL) –
(18.314*TWIL) – (17.786*TWIM) – (17.940*TWIH) (10)
[Table: 4.]
From the BLR we have also obtained the constant coefficient for the independent variables (table
4). A positive coefficient implies that the independent factor increases the possibility of a
landslide, whereas a negative coefficient suggests that the probability of landslides decreases
(Kumar and Anbalagan 2015; Saha et al. 2020; Riegel et al. 2020). From the eq. 5 and eq. 10
complete research area's landslide probability estimate was generated, with probability values
ranging from 0 to 1. Based on Jenk's Natural Break categorization, the likelihood map was
further classified into the following zones: very low prone, low prone, moderately prone, highly
prone, and very highly prone. The model's ability to identify zones not prone to landslides is
higher (87.7 %) than its capacity to designate areas prone to landslides (82.6 %). Landslide
susceptibility map for binary logistic regression was categorized into five classes namely very
low (-12.432 - -3.24), low (-3.24 – 2.057), medium (2.057 – 10.354), high (10.354 – 18.98) and
very high (18.98 – 44.649).
[Figure: 4.]

7.2 Frequency Ratio

Frequency ratio approach analysis was carried out for landslide susceptibility mapping. In this
landslide susceptibility index was generated using ArcGIS software (raster calculator) based on
equation 11 and normalized the obtained frequency ratio values for each subclass (table 5). The
popular technique known as the natural break method was used for the reclassification of
landslide susceptibility index and categorized into five classes very low (3.769 – 5.887), low
(5.887 – 6.637), medium (6.637 – 7.333), high (7.333 – 8.110) and very high (8.110 – 10.602)
figure 5.
n
LSI =∑ ❑ Nfri (11)
i=1

Where N is the number of overall causative factors which is used in the study area.
[Figure: 5.]
[Table: 5.]
7.3 Analytical Hierarchy Process
The AHP approach was used to map the LSM. To weigh factors and their classes, AHP was
utilized. Weight values were applied to raster maps of each factor on a cell-by-cell basis. eq.8
was used to accomplish weighted raster map integration. The integration resulted in a Landslide
susceptibility index (LSI) map including numerical susceptibility information, with higher LSI
values indicating high susceptibility and lower values indicating low susceptibility. The range of
landslide susceptibility index values is 0.090 to 0.375. The Natural Breaks classifier was used to
determine class break scores of the continuous LSI map and the LSI map was categorized into
five groups (figure 6) as a result: very low vulnerability (0.090 – 0.138), low vulnerability (0.138
– 0.167), medium vulnerability (0.167 – 0.197), high vulnerability (0.197 – 0.237) and very high
vulnerability (0.237 – 0.375).

[Figure: 6.]
[Table: 6.]
[Table: 7.]
From table 6, AHP scores were calculated using an AHP calculator for all the 12 factors. For the
AHP score consistency ratio, the max eigen value was also calculated for all the factors.
According to the field visit or ground investigation, a literature review, professional suggestion,
and guide relative score for each factor were given, according to landslide influences. The
consistency ratio (CR) obtained for all the factors was 0.047, and the maximum eigen value was
12.787. Table7 also shows the AHP score calculation for each factor class where the factor
weight of each factor sub-class obtained from the AHP calculation was given to each factor sub-
class accordingly for landslide susceptibility map generation. The CR value and max eigen value
for each factor are as follows: geology (CR 0.0 max eigen 2), aspect (CR 0.087 max eigen
10.008), soil (CR 0.0 max eigen 2), land use/land cover (CR 0.058 max eigen 5.260),
topographic wetness index (CR 0.033 max eigen 5.146), stream power index (CR 0.067 max
eigen 5.300), profile curvature (CR 0.056 max eigen 3.054), road buffer (CR 0.025 max eigen
5.111), relative relief (CR 0.036 max eigen 5.162), slope (CR 0.035 max eigen 5.156), drainage
buffer (CR 0.028 max eigen 5.127) and fault fold and thrust (CR 0.029 max eigen 5.130).

7.4 Field study and landslide influences

The goal of this research was to assess each causative component in landslide susceptibility in
the Mao-Maram, Manipur region, and to compare three models of the spatial likelihood of
landslide occurrence using the FR, BLR, and AHP methodologies. Landslide susceptibility maps
for the three models were finally generated based on the combination of 12 geo-environmental
causing factors and were classified into five zones depending on the probability of landslide
occurrence in those areas. Several villages and small towns were visited during the field
investigation for this region, Mao-Maram. During the field study, landslide incidents were
observed in almost every village or town, but they were mostly observed in areas with 25 to 50
m of landside. Due to tough terrain conditions and poor road conditions, it was difficult to move
around conveniently in the study area, and the landslides observed were mostly along the
roadside. Landslides found near the roads were mainly due to the unscientific construction of
roads, and the roads were randomly cut to make way or pass, leaving the cut slope untouched,
which became very highly prone to landslides. It is also observed in the field that the study area
is completely made up of sedimentary rocks like sandstone, shale, and siltstone. While
considering the hardness of rocks like sandstone and shale, it is generally considered fairly soft.
Sedimentary rock, for example, is frequently the only rock type that includes fossils and is thus
regarded as rather soft. The study area also received a high amount of rainfall and it can rain
continuously for days during the monsoon season, which loosens the earth, especially the area
with fragile lithology, eroded landscape, or weathered due to various reasons and acts as an
external triggering factor, which is the main cause of landslide incidents in the present study
area. Seismic activities in the study area are also considered a highly prone region and have
triggered some landslide incidents in the past, but so far it has not recorded any major landslide
incidents due to seismic activities in study area. Landslides that occurred in the past are likely to
occur again due to similar terrain conditions (Kumar and Anbalagan 2016). During landslide
inventory collection, it was unable to cover the whole area for field base landslide location
collection due to the harsh landscape of the study region, which is considered to be the most
accurate practice for the data-driven method. Remote sensing techniques were used for landslide
inventory collection in a region where it was not possible to reach out or observe landslide
occurrences, making the process much easier, less time-consuming, and less labour-intensive.
In this paper, for three models, 12 causative factors were selected, namely geology, aspect soil,
land use/land cover, topographic wetness index, stream power index, profile curvature, road
buffer, relative relief, slope, drainage buffer, and fault/fold/thrust. After identification and
detailed analysis of the terrain factors influencing landslides using frequency ratio, it is observed
that some causative factors have a high impact on landslides, such as aspect, land use/land cover,
road buffer, slope, drainage, and fault/fold/thrust. In the study area, the highest landslide factor
recorded was in the land use/landcover factor in the sub-class, settlement/barren land and
agricultural land, with a frequency value of 3.851 and 3.309 respectively. This is due to high
landslide incidents that have occurred in the study area due to increasing anthropogenic activity
and unplanned development and planning without proper scientific knowledge and precaution.
For aspect sub-classes like southeast aspect (2.026) and south aspect (2.649), they have a high-
frequency ratio value compared to other sub-classes in aspect factor. This is due to landslides
being more likely to occur in this region, which is more exposed to sunlight and also receives
lots of rainfall. In the high slope (32°-46°) and very high slope (> 46°) slope categories, high
landslide frequency ratios ranging from 1.060 to 1.613 were reported, indicating that the chance
of a landslide increases with increasing slope angle. On steep slopes, the weight of the potential
mobilized material will be greater owing to gravity than on a moderate grade. In the road buffer
classes, a high-frequency value is also seen in the 0-100 m road buffer (2.386), which indicates
that after the road is constructed, the cut slope is left untouched or the road construction without
any scientific precaution in the study area. From this study, it is also observed that the region
with drainage and lineament (fold, fault, and thrust) is more prone to landslide occurrence during
heavy rainfall and seismic activities. From the binary logistic regression analysis, which is a
quantitative approach, it is found that causative factors such as land use and land cover, stream
power index, relative relief, drainage and road factors have high influences on landslides in the
study region. The analytical hierarchy process (AHP) technique is completely based on
subjective decision-making. For the AHP approach, pairwise comparison was given with a
straightforward and appropriate choice rule depending on the field observation, investigation,
and professional guidance. In this AHP analysis for causative factors, high factor weights were
obtained for slope (0.18), road (0.18), relief (0.128), fault/fold/thrust (0.106), drainage (0.082)
and land use/land cover (0.082) compared to other factors for the present study area. From the
three-model analysis for the causative factors, it is observed that factors like road, drainage,
relative relief, land used/ land cover, slope, fault/fold/thrust and stream power index have a high
impact on landslides in the study area. A landslide susceptibility map was generated for three
models with five classes representing their high and low susceptibility zones. This three-
landslide susceptibility map for three models shows slightly different high, moderate, and low
pixel zones depending on the three-method used with the causative factors. The high
susceptibility pixel zone for frequency ratio was observed in the region where there is a
combination of many highly influencing causative factors like settlement, barren land, high slope
angle, slope facing south and southeast, high lineament area, and region closer to road and
drainage. A low susceptibility pixel zone was observed in the region where causative factors
such as flat surface, waterbodies, dense forest, low stream power index, and so on have a very
low influence, whereas a moderate susceptibility pixel zone was observed in the region where
causative factors have a moderate impact. Similarly, for binary logistic regression and the
analytical hierarchy process, each model has acquired higher, lower, and moderate causative
factors influencing landslides. Three models used for generating landslide susceptibility maps
have shown good prediction accuracy using ROC and cumulative percentage curves. When
compared with these three models, binary logistic regression was observed to have slightly better
accuracy than the other two models for this study area.
8. Validation of three model
For binary logistic regression, the landslide inventory map is compared to the continuous
susceptibility maps for the confusion matrix. In this scenario, the LSM was validated using the
cumulative percentage curve and ROC curve. A cumulative percentage curve was created with
79 testing landslide inventory locations with a basic trapezium approach to determine the AUC
and yield an accuracy of 80.32% (table: 8 A). The area under the ROC curve has a maximum
value of one for perfect prediction, while a value approaching 0.5 indicates that the model failed.
In this situation, the ROC curve is determined to be 0.900, which translates to a prediction
accuracy of 90% (table 8).
For frequency ratio cumulative percentage curve approach and ROC curve were used for
validation. A total of 65% of the landslide sites were utilized for the landslide frequency ratio
computation, with the remaining 35% used for validation. For the cumulative percentage curve,
the basic trapezium approach was used to determine the AUC curve, which yielded 77.24%
accuracy (table: 8 B). For ROC approach for AUC has yielded 0.887 which can be converted
into a prediction accuracy of 88.7% (table 8).
 For the analytical hierarchy process cumulative percentage curve approach, which takes
into account previous landslides, was employed in this research to verify the accuracy of the
LSM. The cumulative percentage curve model achieves an AUC value of 0.7322, which
translates to a percentage prediction accuracy of 73.22% (table: 8 C). The ROC curve was also
used in this model for validation. The area under the ROC curve has a yield of 0.774 which can
result in an accuracy of 77.4% (table: 8).
[Figure7]
[Figure8]
[Table 8]
9. Conclusion
The study area of Mao-Maram, Manipur, falls under a highly landslide-prone region in India,
which has been experiencing landslide incidents every year. This region has been highly prone to
landslides. There is no detailed landslide predictor map available for the region where proper
mitigation, planning, and development could be done to minimize property damage and loss of
life. From this research, it is noticed that the study region has a fragile topographic condition and
various causative factors influence the landside in this region. The current study area has a high
influence on triggering factors such as rainfall and seismic activity, as the area receives a lot of
rain during the monsoon season and is close to subduction plate tectonics. Twelve geo-
environmental causal factors were chosen for this study to produce a landslide susceptibility
map. In this research three models namely the analytic hierarchy process, frequency ratio and
binary logistic regression approach were used for generating landslide susceptibility mapping for
the study area. For every three models, five classes were classified for LSM very low, low,
medium, high, and very high. For the frequency ratio approach for LSM, it is observed that the
cumulative percentage curve has yielded 77.24% accuracy, and the ROC curve is 88.7%. The
analytic hierarchy process for the cumulative percentage curve is 73.22% and ROC has obtained
77.4% prediction accuracy. Binary logistic regression for the ROC curve has yielded 90%
prediction accuracy and for the cumulative percentage curve, it has obtained 80.32% of accuracy.
All the three-model prediction accuracy was obtained after validation of all the three models
using the AUC curve. From these models, we can clearly understand that the binary logistic
regression approach has yielded the best prediction accuracy in both the cumulative percentage
curve and ROC curve compared to the other two models with a 90% ROC curve and 80.32%
cumulative percentage curve under the AUC curve and also binary logistic regression
contingency accuracy was 85.2% for the study area. This research emphasizes the need of
producing landslide susceptibility maps in hilly areas while taking tehsil (sub-district)
development and planning into account. It is a must-have tool for developing land management
and mitigation measures. It will also help the city or village planners create sustainable
agricultural techniques. It may assist in finding spots for future expansion in low susceptibility
zones. It may assist policymakers in prevention and mitigation as well as disaster planning and
preparation at the tehsil level.

Abbreviations
LSM: Landslide Susceptibility Map; SPI: Stream Power Index; TWI: Topographic Wetness
Index; FFT: Fault, Fold, and Thrust; LULC: Land Use Land Cover; BLR: Binary Logistic
Method; AHP: Analytical Hierarchy Process; FR: Frequency Ratio; NA: North Aspect; NEA:
Northeast Aspect; EA: East Aspect; SEA: Southeast Aspect; SA: South Aspect; SWA:
Southwest Aspect; WA: West Aspect; D100M: Drainage 0-100M; D200M: Drainage 100-200M;
D300M: Drainage 200-300M; D400M: Drainage 300-400M; GB: Geology Barial; LSRS: LULC
sand rock and settlement; LSG: LULC shrub and grassland; LV: LULC vegetation; RL: Relief
low; RM: relief medium; RH: Relief high; R100M: Road 0-100M; R200M: Road 100-200M;
R300M: Road 200-300M; R400M: Road 300-400M; SVL: Slope very low; SL: Slope low; SM:
Slope medium; SH: Slope high; SPIVL: Stream Power Index very low; SPIL: Stream Power
Index low; SPIM: Stream Power Index medium; SPIH: Stream Power Index high; TWIVL:
Topographic Wetness Index very low; TWIL: Topographic Wetness Index low; TWIM:
Topographic Wetness Index medium; TWIH: Topographic Wetness Index high.

Acknowledgements
The authors would like to express their heartfelt thanks to Lovely Professional University in
Punjab for providing laboratory and software facilities to carry out this study. The first author
thanked A.Elow, senior Geologist (Geological Survey of India), for assisting in the collection of
ancillary data during the research. The authors also express gratitude to the Mao Council for their
kindness and hospitality throughout the field research.
Author statement
Both authors have contributed to this paper. The first author has done the fieldwork, data
extraction, analysis, and preparation of the paper on the other hand the second author has guided
the first author throughout the research work.

Reference list
Anbalagan R 1992 Landslide hazard evaluation and zonation mapping in mountainous
terrain; Eng. Geol. 32(4) 269-277.
Anbalagan R, Kumar R, Lakshmanan K, Parida S and Neethu S 2015 Landslide hazard zonation
mapping using frequency ratio and fuzzy logic approach, a case study of Lachung Valley,
Sikkim; Geoenvironmental Disasters 2(1).
Arabameri A, Pradhan B, Rezaei K and Lee C W 2019 Assessment of Landslide Susceptibility
Using Statistical- and Artificial Intelligence-based FR–RF Integrated Model and
Multiresolution DEMs; Remote Sens. 11(9) 999.
Balamurugan G, Ramesh V and Touthang M 2016 Landslide susceptibility zonation mapping
using frequency ratio and fuzzy gamma operator models in part of NH-39, Manipur,
India; Nat. Hazards 84(1) 465-488.
Bahrami Y, Hassani H and Maghsoudi A 2020 Landslide susceptibility mapping using AHP and
fuzzy methods in the Gilan province, Iran; GeoJournal 86(4) 1797-1816.
Biswas B, K.S V and Ranjan R 2021 Landslide susceptibility mapping using integrated approach
of multi-criteria and geospatial techniques at Nilgiris district of India; Arabian J. Geosci.
14(11).
Chen Z and Wang J 2007 Landslide hazard mapping using logistic regression model in
Mackenzie Valley, Canada; Nat. Hazards 42(1) 75-89.
Chowdhuri I, Pal S, Chakrabortty R, Malik S, Das B, Roy P and Sen K 2021 Spatial prediction
of landslide susceptibility using projected storm rainfall and land use in Himalayan
region; Bull. Eng. Geol. Environ. 80(7) 5237-5258.
Conforti M, Robustelli G, Muto F and Critelli S 2011 Application and validation of bivariate
GIS-based landslide susceptibility assessment for the Vitravo river catchment (Calabria,
south Italy); Nat. Hazards 61(1) 127-141.
Chanu M L and Oinam B 2021 Impact study for landslide contributing factors using a multi-
criterion approach for landslide susceptibility; Arab. J. Geosci.14 1873
Das G and Lepcha K 2019 Application of logistic regression (LR) and frequency ratio (FR)
models for landslide susceptibility mapping in Relli Khola river basin of Darjeeling
Himalaya, India; SN Appl. Sci. 1(11).
Du H, Song D, Chen Z, Shu H and Guo Z 2020 Prediction model oriented for landslide
displacement with step-like curve by applying ensemble empirical mode decomposition
and the PSO-ELM method; J. Cleaner Prod. 270 122248.
Devara M, Tiwari A and Dwivedi R 2021 Landslide susceptibility mapping using MT-InSAR
and AHP enabled GIS-based multi-criteria decision analysis; Geomatics, Nat. Hazards
Risk 12(1) 675-693.
El Jazouli A, Barakat A and Khellouk R 2019 GIS-multicriteria evaluation using AHP for
landslide susceptibility mapping in Oum Er Rbia high basin (Morocco); Geoenvironmental
Disasters 6(1).
Erener A and Düzgün H S B 2010 Improvement of statistical landslide susceptibility mapping by
using spatial and global regression methods in the case of More and Romsdal (Norway);
Landslides 7(1) 55-68.
ESRI FAQ 2012 What is the Jenks optimization method?
http://support.esri.com/en/knowledgebase/techarticles/detail/26442.
Feizizadeh B and Blaschke T 2012 GIS-multicriteria decision analysis for landslide
susceptibility mapping: comparing three methods for the Urmia lake basin, Iran; Nat.
Hazards 65(3) 2105-2128.
FAO-UNESCO Soil Map of the world 1974. Paris: Unesco (7 Place de Fontenoy, 75700 Paris).
Galve J, Cevasco A, Brandolini P, Piacentini D, Azañón J, Notti D and Soldati M 2016 Cost-
based analysis of mitigation measures for shallow-landslide risk reduction strategies; Eng.
Geol. 213 142-157.
Gokceoglu C and Sezer E 2009 A statistical assessment on international landslide literature
(1945–2008); Landslides 6(4) 345-351.
Kundu S, Saha A, Sharma D and Pant C 2013 Remote Sensing and GIS Based Landslide
Susceptibility Assessment using Binary Logistic Regression Model: A Case Study in the
Ganeshganga Watershed, Himalayas; J. Indian Soc. Remote Sens. 41(3) 697-709.
Kumar R and Anbalagan R 2015 Landslide susceptibility zonation in part of Tehri reservoir
region using frequency ratio, fuzzy logic and GIS; J. Earth Syst. Sci. 124(2) 431-448.
Kumar R and Anbalagan R 2016 Landslide susceptibility mapping using analytical hierarchy
process (AHP) in Tehri reservoir rim region, Uttarakhand; J. Geol. Soc. India 87(3) 271-
286.
Kumar R and Anbalagan R 2019 Landslide susceptibility mapping of the Tehri reservoir rim area
using the weights of evidence method; J. Earth Syst. Sci. 128(6).
Kanungo D, Arora M, Sarkar S and Gupta R 2006 A comparative study of conventional, ANN
black box, fuzzy and combined neural and fuzzy weighting procedures for landslide
susceptibility zonation in Darjeeling Himalayas; Eng. Geol. 85(3-4) 347-366.
Khuaidem K S, Rajkumar H S and Soibam I 2015 Attribute of trace fossils of Laisong flysch
sediments, Manipur, India; J. Earth Syst. Sci. 124(5) 1085-1113.
Myronidis D, Papageorgiou C and Theophanous S 2015 Landslide susceptibility mapping based
on landslide history and analytic hierarchy process (AHP); Nat. Hazards 81(1) 245-263.
Mathew J, Babu D G, Kundu S, Kumar K V and Pant C C 2013 Integrating intensity duration-
based rainfall threshold and antecedent rainfall-based probability estimate towards
generating early warning for rainfall-induced landslides in parts of the Garhwal Himalaya,
India; Landslides 11(4) 575–588.
Moragues S, Lenzano M, Lanfri M, Moreiras S, Lannutti E and Lenzano L 2020 Analytic
hierarchy process applied to landslide susceptibility mapping of the North Branch of
Argentino Lake, Argentina; Nat. Hazards 105(1) 915-941.
National Disaster Management Authority (no date) Home | NDMA, GoI. Available at:
https://ndma.gov.in/.
Ohlmacher G and Davis J 2003 Using multiple logistic regression and GIS technology to predict
landslide hazard in northeast Kansas, USA; Eng. Geol. 69(3-4) 331-343.
Okendro M and Kushwaha S 2018 Landslide hazard zonation in and around litan village along
NH-202, Ukhrul district, Manipur, India; National Geographical Journal of India 68 (1-2).
Rajkumar H and Klein H 2014 First perissodactyl footprints from Flysch deposits of the Barail
Group (Lower Oligocene) of Manipur, India; J. Earth Syst. Sci. 123(2) 413-420.
Rajkumar H, Mustoe G, Khaidem K and Soibam I 2015 Crocodylian Tracks from Lower
Oligocene Flysch deposits of the Barail Group, Manipur, India; Ichnos 22(2) 122-131.
Riegel R, Alves D, Schmidt B, de Oliveira G, Haetinger C, Osório D, Rodrigues M and de
Quevedo D 2020 Assessment of susceptibility to landslides through geographic
information systems and the logistic regression model; Nat. Hazards 103(1) 497-511.
Peethambaran B, Anbalagan R, Kanungo D, Goswami A and Shihabudheen K 2020 A
comparative evaluation of supervised machine learning algorithms for township level
landslide susceptibility zonation in parts of Indian Himalayas; Catena 195 104751.
Pradhan B 2013 A comparative study on the predictive ability of the decision tree, support vector
machine and neuro-fuzzy models in landslide susceptibility mapping using
GIS; Comput. Geosci. 51 350-365.
Pardeshi S, Autade S and Pardeshi S 2013 Landslide hazard assessment: recent trends and
techniques; SpringerPlus 2(1).
Pourghasemi H, Teimoori Y Z, Panagos P and Pradhan B 2018 Analysis and evaluation of
landslide susceptibility: a review on articles published during 2005–2016 (periods of 2005–
2012 and 2013–2016); Arabian J. Geosci. 11(9).
Pourghasemi H, Kornejady A, Kerle N and Shabani F 2020 Investigating the effects of different
landslide positioning techniques, landslide partitioning approaches, and presence-absence
balances on landslide susceptibility mapping; Catena 187 104364.
Shano L, Raghuvanshi T and Meten M 2020 Landslide susceptibility evaluation and hazard
zonation techniques – a review; Geoenvironmental Disasters 7(1).
Shano L, Raghuvanshi T and Meten M 2021 Landslide susceptibility mapping using frequency
ratio model: the case of Gamo highland, South Ethiopia; Arabian J. Geosci. 14(7).
Saaty T 1977 A scaling method for priorities in hierarchical structures;
J. Math. Psychol. 15(3) 234-281.
Saaty T 1980 The Analytic Hierarchy Process: Planning, Priority Setting, Resources Allocation,
McGraw-Hill; New York 287
Sur U, Singh P and Meena S 2020 Landslide susceptibility assessment in a lesser Himalayan
road corridor (India) applying fuzzy AHP technique and earth-observation data; Geomat.
Nat. Hazards Risk 11(1) 2176-2209.
Saha A, Mandal S and Saha S 2020 Geo-spatial approach-based landslide susceptibility mapping
using analytical hierarchical process, frequency ratio, logistic regression and their
ensemble methods; SN Appl. Sci. 2(10).
Singh A, Pal S and Kanungo D 2020 An integrated approach for landslide susceptibility–
vulnerability–risk assessment of building infrastructures in hilly regions of India; Environ.
Dev. Sustain. 23(4) 5058-5095.
Sujatha E and Sridhar V 2021 Landslide Susceptibility Analysis: A logistic regression model
case study in Coonoor, India; Hydrology 8(1) 41.
Singh C D, Behera K K and Rocky W S 2011 Landslide susceptibility along NH-39 between
Karong and Mao, Senapati district, Manipur; J. Geol. Soc. India 78 559-570.
Vieira B, Fernandes N and Filho O 2010 Shallow landslide prediction in the Serra do Mar, São
Paulo, Brazil; Nat. Hazards Earth Syst. Sci. 10(9) 1829-1837.
Varnes D 1978; Slope Movement Types and Processes onlinepubs.trb.org.
Varnes D 1984 Landslide hazard zonation: a review of principles and practice Paris: Unesco.
Vakhshoori V, Pourghasemi H, Zare M and Blaschke T 2019 Landslide Susceptibility Mapping
Using GIS-Based Data Mining Algorithms; Water 11(11) 2292.
Wubalem A 2021 Landslide susceptibility mapping using statistical methods in Uatzau
catchment area, northwestern Ethiopia; Geoenvironmental Disasters 8(1).
Yalcin A 2008 GIS-based landslide susceptibility mapping using analytical hierarchy process
and bivariate statistics in Ardesen (Turkey): Comparisons of results and
confirmations; Catena 72(1) 1-12.

Tables
Table:1. The spatial data set used in the study area
Data type Sensor Resolution/scale Derived data Source of data
Multispectral Sentinel 2 B 10 m LULC Copernicus open
data access hub
Landsat series 15 m to 15 cm Landslide Google earth pro
inventory
DEM ALOS PALSAR 12.5 m Slope Alaska satellite
Aspect facility
Relief
Profile curvature
TWI
 Drainage
SPI
Ancillary data Published 1:250000 Geology Khaidem et al.
geology map of 2015
Manipur
Published soil 1:5000000 Soil FAO/UNESCO
map of the world soil map of the
world 1974.
Published Fault, 1:250000 Fault, Fold and GSI Bhukosh
fold and thrust Thrust map
(GSI)
Toposheet 1:250000 Digitized base Survey of India
83G/14 N, map
83G/15 N,
83K/2 N, 83K/3
N
Digitize the road 1:250000 Road Google earth
map of the study
area

Table: 2. Rock formation description of the study area (Rajkumar et al. 2015)
Group Formation Age Lithology
Barial Renji Late Oligocene to Sandstone that is massive and
Late Eocene heavily bedded with flysch
Jenam Sandstone ranges from massive and
heavily bedded, including
carbonaceous shale layers and Flysch
deposits.
Laisong Intercalation of layered sandstone
with siltstone and shale that is
 somewhat thin.
Disang Upper Disang Late Eocene to Shale intercalations with somewhat
Late Cretaceous thin siltstone and sandstone
displaying rhythmites
Lower Disang Splintery, dark grey to black shale
with Flysch deposits

Table 3: Factors classes and range of 12 causative factors

Factor Classes Range/scale
Aspect Flat -1- -0.00001
North -0.00001- 22.5
Northeast 22.5- 67.5
East 67.5- 112.5
Southeast 112.5- 157.5
South 157.5- 202.5
Southwest 202.5- 247.5
West 247.5- 292.5
Northwest 292.5- 337.5
North 337.5- 360
Geology Disang group Not applicable
Barial group
LULC type Water bodies -0.585- 0.05
Settlement/barren land 0.05- 0.2
Agricultural land 0.2- 0.4
Scrub forest 0.4- 0.6
Dense forest 0.6- 0.896
Soil type Orthic acrisols Not applicable
Humic cambisols
Topographic wetness index Very low 0.767 – 3.638
Low 3.638 – 4.485
(TWI)
Medium 4.485 – 5.79
High 5.79 – 7.94
Very high >7.94
Stream power index (SPI) Very low -13.82 - -5.22
Low -5.22 - -0.648
Medium -0.648 – 1.09
High 1.09 – 3.49
Very high >3.49
Profile Curvature Convex -28.80- -0.1
Linear -0.1- 0
Concave 0- 32.139
Road Buffer 0-100 m 0-100 m
100-200 m 100-200 m
200-400 m 200-400 m
400-800 m 400-800 m
>800 m >800 m
Relative relief Very low 0- 47 m
Low 47- 97 m
Medium 97- 141 m
High 141- 188 m
Very high >188 m
Slope category Very low 0-12°
Low 12-21°
Medium 21-29°
High 29-39°
Very high >39°
Drainage buffer 0 – 100 m 0 – 100 m
100 – 200 m 100 – 200 m
200 – 300 m 200 – 300 m
 300 – 400 m 300 – 400 m
>400 m >400 m
Fault, fold and thrust (FFT) 0 – 100 m 0 – 100 m
100 – 200 m 100 – 200 m
200 – 300 m 200 – 300 m
300 – 400 m 300 – 400 m
>400 m >400 m

Table: 4. Coefficients and significant independence variables obtained using BLR method
Causative factor β S.E. Wald df Sig. Exp(β)
(variable)
North Aspect -2.500 1.572 2.531 1 .112 .082
Northeast Aspect -3.229 1.865 2.996 1 .083 .040
East Aspect -2.104 1.551 1.839 1 .175 .122
Southeast Aspect -2.339 1.636 2.043 1 .153 .096
South Aspect -1.671 1.587 1.109 1 .292 .188
Southwest Aspect -.914 1.441 .402 1 .526 .401
West Aspect -1.005 1.118 .808 1 .369 .366
Drainage 0-100 M -.547 .700 .611 1 .435 .579
Drainage 100-200 M -1.558 .595 6.853 1 .009 .210
Drainage 200-300 M .072 .561 .017 1 .897 1.075
Drainage 300-400 M -.201 .616 .107 1 .744 .818
Geology Barial -.567 .457 1.542 1 .214 .567
LULC settlement/barren 23.569 10913.732 .000 1 .998 17215907164.108
land
LULC agricultural land 4.253 .618 47.324 1 .000 70.322
LULC scrub forest 3.085 .535 33.257 1 .000 21.867
Relief low 1.884 1.578 1.425 1 .233 6.577
Relief medium 3.018 1.383 4.761 1 .029 20.457
Relief high 1.99601 1.201 2.763 1 .096 7.360
 6
Road 0-100 M 1.873 .971 3.725 1 .054 6.510
Road 100-200 M -.768 1.096 .491 1 .483 .464
Road 200-400 M -1.607 .893 3.234 1 .072 .201
Road 400-800 M -.016 .608 .001 1 .979 .984
Slope very low -1.540 1.635 .888 1 .346 .214
Slope low -3.054 1.262 5.851 1 .016 .047
Slope medium -3.159 1.121 7.940 1 .005 .042
Slope high -1.616 .963 2.817 1 .093 .199
Soil Humic Cambisols -.699 .555 1.591 1 .207 .497
SPI very low 1.332 1.913 .485 1 .486 3.788
SPI low 1.740 1.895 .843 1 .359 5.699
SPI medium .981 1.789 .301 1 .584 2.667
SPI high -.120 1.675 .005 1 .943 .887
TWI very low -20.326 40194.202 .000 1 1.000 .000
TWI low -18.314 40194.202 .000 1 1.000 .000
TWI medium -17.786 4- .000 1 1.000 .000
0194.202
TWI high -17.940 40194.202 .000 1 1.000 .000
Constant 18.833 40194.202 .000 1 1.000 151046244.047

Table: 5. Frequency ratio calculation for causative factors

Factors & attributes Percentage of No. of landslides Percentage of Frequency
the domain (ai ) landslides (li ) Ratio (fi )
Geology
Disang group 34.80 142 60.68 1.744
Barial group 65.20 92 39.32 0.603
Aspect
Flat 0.09 0 0 0
North 11.22 0 0 0
Northeast 9.98 5 2.14 0.214
East 12.46 27 11.54 0.926
Southeast 14.13 67 28.63 2.026
South 12.10 75 32.05 2.649
Southwest 12.99 43 18.38 1.415
West 14.13 14 5.98 0.423
Northwest 12.90 3 1.28 0.099
Soil type
Orthic acrisols 53.84 198 84.62 1.572
Humic cambisols 46.16 36 15.38 0.333
LULC type
Water bodies 0.12 0 0 0
Settlement/barren land 1.44 13 5.56 3.861
Agricultural land 15.24 118 50.43 3.309
Scrub forest 26.21 88 37.61 1.435
Dense forest 56.99 15 6.41 0.112
TWI
Very low (0.767 - 3.638) 35.54 93 39.74 1.118
Low (3.638 - 4.485) 41.36 112 47.86 1.157
Medium (4.485 - 5.79) 16.49 22 9.40 0.570
High (5.79 - 7.94) 4.94 5 2.14 0.433
Very high (>7.94) 1.68 2 0.85 0.506
SPI
Very low (-13.82 - -5.22) 13.67 43 18.38 1.345
Low (-5.22 - -0.648) 18.87 43 18.38 0.974
Medium (-0.648 - 1.09) 39.33 102 43.59 1.108
High (1.09 - 3.49) 23.90 42 17.95 0.751
Very high (>3.49) 4.23 4 1.71 0.404
Profile Curvature
Convex 40.95 111 47.44 1.158
Linear 12.27 36 15.38 1.253
Concave 46.78 87 37.18 0.795
Road Buffer
0-100 m 3.76 21 8.97 2.386
100-200 m 3.47 9 3.85 1.11
200-400 m 6.51 12 5.13 0.788
400-800 m 11.97 28 11.97 1
>800 m 74.30 164 70.09 0.943
Relative relief
Very low 0.09 0 0 0
Low 8.19 39 16.67 2.035
Medium 24.91 98 41.88 1.681
High 27.20 63 26.92 0.99
Very high 39.61 34 14.53 0.367
Slope category
Very low 15.78 31 13.25 0.84
Low 29.10 63 26.92 0.925
Medium 29.81 70 29.91 1.003
High 19.75 49 20.94 1.060
Very high 5.56 21 8.97 1.613
Drainage buffer
0 – 100 m 10.42 30 12.82 1.230
100 – 200 m 9.77 26 11.11 1.137
200 – 300 m 9.22 32 13.68 1.484
300 – 400 m 8.51 33 14.10 1.657
>400 m 62.08 113 48.29 0.681
Fault, fold and thrust
0 – 100 m 2.90 7 2.99 1.031
100 – 200 m 2.94 3 1.28 0.435
200 – 300 m 2.90 8 3.42 1.179
300 – 400 m 2.89 11 4.70 1.626
>400 m 88.38 205 87.61 0.991
Table: 6. AHP scores of classes/factors, CR and Max eigen value
Geolog Aspec LUL Soil TWI SPI Curvatur Road Relie Slope Drainag FFT Factor
Facto y t C e f e Pairwis
r e
Weight
Nos 1 2 3 4 5 6 7 8 9 10 11 12
1 Geology 1 2 1/2 2 2 2 3 1/3 1/2 1/3 1/2 1/2 0.060
2 Aspect 1/2 1 1/3 1/2 1/2 1/2 2 1/5 1/4 1/5 1/3 1/3 0.027
3 LULC 2 3 1 3 2 3 4 1/3 1/2 1/3 1 1/2 0.082
4 Soil 1/2 2 1/3 1 1/3 1/2 2 1/5 1/4 1/5 1/3 1/4 0.031
5 TWI 1/2 2 1/2 3 1 2 3 1/3 1/3 1/3 1/2 1/2 0.043
6 SPI 1/2 2 1/3 2 1/2 1 4 1/4 1/3 1/4 1/3 1/3 0.049
7 Curvatur 1/3 1/2 1/4 1/2 1/3 1/4 1 1/6 1/5 1/6 1/4 1/4 0.03
e
8 Road 3 5 3 5 3 4 6 1 2 1 3 2 0.18
9 Relief 2 4 2 4 3 3 5 1/2 1 1/2 2 2 0.128
10 Slope 3 5 3 5 3 4 6 1 2 1 3 2 0.18
11 Drainage 2 3 1 3 2 3 4 1/3 1/2 1/3 1 1/2 0.082
12 FFT 2 3 2 4 2 3 4 1/2 1/2 1/2 2 1 0.106
CR: 0.047 Max eigen value= 12.787
Table: 7. Factor classes comparison for AHP

Geology
Nos Classes Disang Barial Classes weight
1 Disang group 1 3 0.75
2 Barial group 1/3 1 0.25
CR= 0.0 Max eigen value= 2

Aspect
Nos Classes Flat North Northeast East Southeast South Southwest West

1 Flat 1 1/2 1/3 1/4 1/6 1/7 1/5 1/4

2 North 2 1 1/3 1/4 1/6 1/7 1/5 1/4
3 Northeast 3 3 1 1/2 1/5 1/7 1/4 1/3
4 East 4 4 2 1 1/4 1/6 4 2
5 Southeast 6 6 5 4 1 1/3 3 5
6 South 7 7 7 6 3 1 4 6
7 Southwest 5 5 4 1/4 1/3 1/4 1 4
8 West 4 4 3 1/2 1/5 1/6 1/4 1
9 Northwest 2 2 2 1/2 1/6 1/7 1/3 1/2
CR= 0.087 Max eigen value = 10.008

Soil
Nos Orthic acrisols Humic cambisols Classes weight
1 Orthic acrisols 1 3 0.75
2 Humic cambisols 1/3 1 0.25
CR= 0.0 Max eigen value = 2

LULC
Nos Water Sand, rock or Shrub and Vegetation High
bodies settlement grassland vegetation
1 Water bodies 1 1/7 1/5 1/3 1/2
2 Settlement/barren 7 1 3 5 6
 land
3 Agricultural land 5 1/3 1 4 5
4 Scrub forest 3 1/5 1/4 1 3
5 Dense forest 2 1/6 1/5 1/3 1
CR= 0.058 Max eigen value = 5.260

TWI
Nos Very low Low Medium High Very high

1 Very low 1 1/2 1/3 1/3 2

2 Low 2 1 1/2 1/3 2
3 Medium 3 2 1 1/2 3
4 High 3 3 2 1 3
5 Very high 1/2 1/2 1/3 1/3 1
CR= 0.033 Max eigen value = 5.146

SPI
Nos Very low Low Medium High Very high

1 Very low 1 2 3 2 2
2 Low 1/2 1 1/2 2 2
3 Medium 1/3 2 1 2 3
4 High 1/2 1/2 1/2 1 2
5 Very high 1/2 1/2 1/3 1/2 1
CR=0.067 Max eigen value = 5.300

Profile Curvature
Nos Convex Linear Concave
1 Convex 1 1/2 3
2 Linear 2 1 3
3 Concave 1/3 1/3 1
CR=0.056 Max eigen value= 3.054
Road Buffer
Nos 0-100 m 100-200 m 200-400 m 400-800 m >800 m
1 0-100 m 1 2 3 3 4
2 100-200 m 1/2 1 2 2 3
3 200-400 m 1/3 1/2 1 2 3
4 400-800 m 1/3 1/2 1/2 1 2
5 >800 m 1/4 1/3 1/3 1/2 1
CR= 0.025 Max eigen value = 5.111

Relative relief
Nos Very low Low Medium High Very high

1 Very low 1 1/5 1/4 1/2 1/2

2 Low 5 1 3 4 5
3 Medium 4 1/3 1 3 3
4 High 2 1/4 1/3 1 2
5 Very high 2 1/5 1/3 1/2 1
CR= 0.036 Max eigen value = 5.162

Slope category
Nos Very low Low Medium High Very high

1 Very low 1 1/2 1/2 1/2 1/4

2 Low 2 1 1/2 1/2 1/3
3 Medium 2 2 1 1/2 1/3
4 High 2 2 2 1 1/3
5 Very high 4 3 3 3 1
CR= 0.035 Max eigen value = 5.156

Drainage buffer
Nos 0 – 100 m 100 – 200 m 200 – 300 m 300 – 400 m >400 m
 1 0 – 100 m 1 2 3 3 4
2 100 – 200 m 1/2 1 2 3 3
3 200 – 300 m 1/3 1/2 1 2 3
4 300 – 400 m 1/3 1/3 1/2 1 2
5 >400 m 1/4 1/3 1/3 1/2 1
CR= 0.028 Max eigen value = 5.127

Fault, fold and thrust

Nos 0 – 100 m 100 – 200 m 200 – 300 m 300 – 400 m >400 m

1 0 – 100 m 1 2 2 3 3
2 100 – 200 m 1/2 1 2 2 3
3 200 – 300 m 1/2 1/2 1 2 3
4 300 – 400 m 1/3 1/2 1/2 1 2
5 >400 m 1/3 1/3 1/3 1/2 1
CR= 0.029 Max eigen value = 5.130
Table: 8. The area under the ROC curve for three models.
Landslide susceptibility model Area under the curve value
Binary logistic regression 0.900
Frequency ratio 0.887
Analytical hierarchy process 0.774

Figure with captions

Fig: 1. Study area location map of Mao-Maram region extracted using Sentinel 2B
A B C

D E F

H I
G

Fig: 2. Field photographs taken during fields study: A) Rotational type of landslide B)
Rotational type of landslide C) Rotational type of landslide D) Rockfall type of landslide E)
Rotational type of landslide F) Earthflow type of landslide G) Rockfall type of landslide H)
Rotational type of landslide I) Earthflow type of landslide.
A B

C D

E F
G H

I J

K L

Figure: 3. Landslide causative factors of the study area A) Geology B) Soil C) Relief D) Slope E)
LULC F) Profile curvature G) Aspect H) TWI I) SPI J) Drainage K) Road buffer L) Fault, fold
and thrust.
Fig: 4. Landslide susceptibility map of the study area using Binary logistic
regression

Fig: 5. Landslide susceptibility map of the study area using frequency ratio
 Fig: 6. Landslide susceptibility map of the study area using AHP.

A B
C

Fig: 7. ROC curve for 3 models A) Binary logistic regression B) Frequency ratio C) Analytical
hierarchy process

A 100 B 100
Cumulative percentage of

Cumulative percentage of

80 80
landslide

60 60
landslide

40 40
20 20
0 0
0 20 40 60 80 100 0 20 40 60 80 100
Cumulative percentage of LSI
Cumulative percentage of LSI (Decresing)

C 100
Cumulative perentage of

80

60
landslide

40

20

0
0 10 20 30 40 50 60 70 80 90 100

Cumulative percentage of LSI (Decresing)

Fig: 8. Cumulative percentage curve of 3 models A) Binary logistic regression B) Frequency

ratio C) Analytical hierarchy process (AHP)