Geomorphic Diversity and Landslide Susceptibility in the Balason River Basin, Darjeeling Himalaya

Geomorphic diversity and landslide susceptibility in the Balason River Basin, Darjeeling Himalaya

Subrata Mondal1 and Sujit Mandal2 1 Geography, A.B.N Seal College, Cooch Behar, India 2 Geography, Diamond Harbour Women’s University, Diamond Harbour, India

ABSTRACT

This study attempts to assess the role of basin morphometric parameters in slope instability using a morphometric diversity (MD) model, as well as the role of drainage parameters and relief parameters in slope failure using drainage diversity (DD) and relief diversity (RD) models, respectively. For this, a total of 14 morphometric data layers were considered. The relationship of each data layer to landslide susceptibility was judged using a frequency ratio (FR) approach. Parameters like drainage density (Dd), drainage frequency (Df), relative relief (Rr), drainage texture (Dt), junction frequency (Jf), infiltration number (In), ruggedness index (Ri), dissection index (Di), elevation (E), slope (S), relief ratio (Rra) and hypsometric integral (Hi) were positively related with landslide potentiality while bifurcation ratio (Rb) and drainage intensity (Din) negatively correlated with S failure. The principal component analysis (PCA)-based weight assigned to each data layer in each model was multiplied with unidirectional reclassified data layers for each model using a weighted linear combination (WLC) approach to prepare landslide susceptibility maps. The receiver operating characteristics curve showed that the landslide prediction accuracy of the DD, RD and MD models were 71.4%, 73.9% and 76.3%, respectively. The FR plots of the aforesaid three models suggested that the chance of landslide increases from very low to very high in susceptible zones.

KEYWORDS Relief diversity (RD); drainage diversity (DD); morphometric diversity (MD); weighted linear combination (WIC) approach; landslide susceptibility CONTACT Subrata Mondal [email protected] Received on 20 November 2017

1. Introduction (Rastogi and Sharma, 1976; Nautiyal, 1994; Magesh et al., 2012). Some researchers tried to find out the ground water Landslides are one of the real, normal risks that cause potential zones using morphometric parameters (Gajbhiye roughly 1,000 individual deaths per year worldwide and et al., 2014; Gopal and Saha, 2015). Sub-basin prioritisation around US$4 billion in property damages in the U.S. (Lee (Thakkar and Dhiman, 2007; Amani and Safaviyan, 2015), and Pradhan, 2007). The Himalayan mountain regions, surface water availability (Rai et al., 2014; Khatun and Pal, the Meghalaya plateau and the Western Ghats are the most 2017), and recharge estimation (Avinash et al., 2014; Nidhi significant landslide prone areas in India. The Balason river et al., 2016) were done using morphometric parameters. basin of Darjeeling Himalaya is not an exception to these Several studies were established to show the relationship phenomena, due to its fragile geo-environmental conditions between slope instability or simply landslide occurrences and operating geomorphic process. Every year, a large and drainage morphometric parameters (Jeganathan and amount of people, property and the natural environment Chauniyal, 2000; Ghosh, 2015). In this study, a total of are affected by landslides. To limit the impact of landslide 14 morphometric parameters were calculated and spatial damages, it is important to predict it before it happens. data layers were made accordingly. With the help of these The landslide susceptibility map has turned out to be an parameters, three geomorphic diversity models, i.e. the exceptionally valuable approach to assessing, managing drainage diversity (DD) model, relief diversity (RD) model and mitigating the landslide hazard for the region (Kienholz, and morphometric diversity (MD) model, were applied 1978; Spiker and Gori, 2000; Chau et al., 2004). The to prepare landslide susceptibility maps and attempt to drainage basin morphometry plays an important role in gauge how well each model predicts future landslide quantifying the relief, drainage and hydrological conditions events. Essentially, highly diversified areas are associated which governs the whole landscape of a river basin. The with highly landslide-prone areas and vice versa. The morphometric parameters are intimately associated with other existing landslide susceptibility methods, such as the landslide activities. These parameters portray the surface frequency ratio (FR) technique (Lee and Talib, 2005; Lee chemistry and play vital role in slope instability. Several and Sambath, 2006), logistic regression method (Lee and authors calculated morphometric parameters for terrain Sambath, 2006; Ghosh et al. 2011), information value (Info analysis using remote sensing (RS) and geographic Val) method (Anbalagan, 1992), landslide normal risk factor information system (GIS) in different parts of the world method (Yin and Yin, 1988), discriminant analysis method

(Gupta and Joshi, 1990), artificial neural network (ANN) slope gradient. Most of the area of this basin is covered method (Choi et al., 2011), neuro-fuzzy method (Pradhan, by intensely metamorphosed rocks like gneiss, slates, 2010), support vector machine (Xu et al., 2012), decision phyllite and schists, etc. These rocks are characterised by tree method (Nefeslioglu, 2010), index of entropy (Devkota several fractures, joints and faults. They are also highly et al., 2013), Bayesian probability (Song et al., 2012) and weathered. Such conditions maximise the chances of fractal theory (Majtan et al., 2002), etc., were prepared landslide occurrences. Inceptisol order soils cover almost in combination with several relief, drainage, geological, the entire hilly tract of the Balason river basin, and are anthropological, protective and triggering factors. But in characterised by moderately shallow, well-drained, gravelly this study, the three aforementioned diversity models help loamy soils with a loamy surface and moderate to severe us discover the role of drainage, relief and morphometric erosion, etc., which makes the areas more vulnerable to parameters in landslide susceptibility assessment slope instability. About 85% of the annual rainfall occurs individually. The DD model was calibrated using drainage due to the southwest monsoon. Hilly regions of the basin density (Dd), drainage frequency (Df), bifurcation ratio experience more rainy days (124 days) than plain regions (Rb), relative relief (Rr), drainage intensity (Din), drainage (100 days), often more than 2.5 mm rainfall in a single texture (Dt), junction frequency (Jf), infiltration number day. The hilly region also receives more rainfall than the (In) and ruggedness index (Ri), whereas the RD model plains. The whole basin receives an average of 2,300 mm was constructed using slope (S), Rr, dissection index (Di), (2,000 mm - 5,000 mm) of rainfall (Lama, 2003; Mondal elevation (E), Ri, Rb, relief ratio (Rra) and hypsometric and Mandal, 2017a). The mean temperature of the Balason integral/relative height (Hi). Finally, Dd, Df, Din, Di, Dt, Jf, river basin is not the same in the upper part of the region. In, Rb, Rr, Ri, E, S and Hi were used for the MD model. About 12ºC mean annual temperature can be observed in the hilly region, whereas the recorded mean annual 2. Study area temperature in the plain region is 24ºC. In summer, the mean temperature ranges from 27ºC - 29ºC in the plains The Balason river basin is a section of Bengal’s hilly and from 16ºC - 22ºC in the hills, whereas mean winter region in Darjeeling Himalaya, situated within latitude temperatures range from 12ºC - 14ºC in the plains and 5ºC - 26°40Ń - 27°01Ń and longitude 88°7É - 88°25É and 7ºC in the hills (Lama, 2003).

covering a geographical area of approximately 378.45 sq. km (Figure 1). The major right bank tributary of the Mahananda River, the Balason River originates from Lepchajagat (2,361 m), located on the Ghum-Simana ridge. The main leg of the Balason River flows north to south-east for about 51.92 km and joins the Mahananda River at 26°41´28´N and 88°24´15´E. The total number of streams in the Balason river basin is 3,790 and the total length of all stream segments is 1,476.75 km. The

river basin covers parts of Rangli Rangliot, Naxalbari, Figure 1. Location map of the study area. Matigara, Jorebunglow Sukiapokhri, Mirik and Kurseong Figure 1. Location map of the study area. in the Darjeeling district in West Bengal. The Balason river

basin of Darjeeling Himalaya is characterised by a distinct 3. Materials and methods geographical environment. The evolution of landforms is largely dependent on its geological structure, lithology, In the present study, a total of 14 parameters such as relief configuration, climatic characteristics and bio- Dd, Df, Rb, Rr, Din, Dt, Jf, In, Ri, Di, E, S, Rra and Hi physical processes. The elevation of the region increases were considered for the landslide susceptibility assessment from south to north, 121 m at Matigara but 2,416 m at and prediction, which were prepared from Shuttle Radar Lepchajagat. The network created by different ridges and Topography Mission Digital Elevation Model, Toposheets valleys shows a dendritic drainage pattern. Inclination of and Google Earth imagery (Table 1). Various classification slope is different in various parts of the basin. Some areas methods exist, such as quantile, equal interval, natural in the northern part have an inclination of 30º - 45º, whereas break and standard deviation methods, to classify the other places, especially in the south, have an inclination of aforesaid data layers (Pradhan and Lee, 2010a) . If the data 15º - 30º. The river valley has a steep, ungraded channel, distribution has a positive or negative skew, the quantile or narrow floor, steep valley, etc. in the upper portion of the natural break distribution classifiers can be chosen (Akgun basin. The gradient of the river is controlled by lithology et al. 2012). As suggested by many researchers, the natural and mass movement of the area. The steep slopes and break method was applied to determine different subclasses high altitude, especially in the north, make these areas (Althuwaynee et al., 2014; Shrestha et al., 2017; Mandal vulnerable to slope instability, whereas southern parts of and Mandal, 2017) for this study. the basin are less vulnerable due to the lower altitude and

14 HKIE Transactions | Volume 27, Number 1, pp.13–24 locations were used as a traininglocations data were set used and asthe a remainingtraining data 30% set of and locations the remaining were used 30% for of validation locations ofwere used for validation of locations were used as a training data set and2 the remaining 30% of2 locations were used for validation of the model. The average areathe ofmodel. landslide The averagewas 3762.90 area ofm landslide. was 3762.90 m . the model. The average area of landslide was 3762.90 m2. 3.2 Preparation of morphometric3.2 Preparation data layers of morphometric and their role data in slope layers instability and their role in slope instability 3.2 Preparation of morphometric data layers and their role in slope instability A total of 14 parameters Awere total calculated of 14 parameters using a recognised were calculated formula using and a prepared recognised spatial formula and prepared spatial A total of 14 parameters were calculated using a recognised formula and prepared spatial data layers in Arc GIS environmentdata layers in(Table Arc GIS2). The environment relationship (Table between 2). The morphometric relationship parameters between morphometric and parameters and data layers in Arc GIS environment (Table 2). The relationship between morphometric parameters and landslide occurrences werelandslide judged occurrencesusing a frequency were judged ratio (FR) using approach a frequency (Karim ratio et (FR)al., 2011;approach Mondal (Karim et al., 2011; Mondal landslide occurrences were judged using a frequency ratio (FR) approach (Karim et al., 2011; Mondal and Mandal, 2017b), and Mandal, 2017b), and Mandal, 2017b), Table 1. Dataset used in the study. landslide in each class (i), is the total number of FR= . (1) pixels having class (i) in the whole basin,FR= is the . (1) {𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁𝑁𝑁FR=�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁푁100 {𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�.푁𝑁𝑁𝑁𝑁푁푁푁 100 (1) Purpose Data used Data source total number of pixels in each class and is the {∑ 𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁�𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁{𝑁𝑁��푁𝑁𝑁𝑁𝑁푁푁푁𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁𝑁𝑁100 {∑�𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁푁�푁𝑁𝑁𝑁𝑁푁푁100�𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁푁100 Preparation of landslide Google earth historical imageries https://earthexplorer.usgs. total number of pixels in the whole basin. The higher the inventory map (2000 - 2016), Landsat 7 ETM+ gov/, https://glovis.usgs. Where N pix (Si) is the{∑ 𝑁𝑁 number�𝑁𝑁𝑁𝑁𝑁𝑁Where�푁𝑁𝑁𝑁𝑁푁푁 �of 𝑁𝑁N � pixels𝑁𝑁𝑁𝑁𝑁𝑁pix�푁𝑁𝑁𝑁𝑁푁푁푁 (Si) containing 100is the number landslide of pixels in each containing class (i), landslide N pix (Ni in) iseach class (i), N pix (Ni) is imageries (2000 - 2005), Landsat gov, http://bhuvan.nrsc.gov. FR value, the greater the Wherechance Nof pixlandslide (Si) is occurrences. the number of pixels containing landslide in each class (i), N pix (Ni) is 4 - 5 TM imageries (2006 - 2007, in/, Google earth, Arc Map the total number of pixelsthe having total classnumber (i) ofin pixelsthe whole having basin, class ∑N (i )pix in (theSi) wholetotal number basin, ∑Nof pixels pix (Si in) totaleach number of pixels in each 2009 - 2011), LISS III imageries 10.1 world imagery and Field (2008, 2012 - 2013), Landsat 8 investigation using GPS class and ∑N pixthe (totalNi) total number number of pixels of pixels having in theclass whole (i) in basin. the whole The higher basin, the∑N FR pix value, (Si) total the numbergreater theof pixels in each OLI (2014 - 2015), high resolution 3.2.1. Dd class and ∑N pix (Ni) total number of pixels in the whole basin. The higher the FR value, the greater the Sentinel-2 imagery (2016), Arc Map class and ∑N pix (Ni) total number of pixels in the whole basin. The higher the FR value, the greater the high resolution world imagery and chance of landslide occurrences.chance of landslide occurrences. intensive field survey data (2015 - This is an important factor in landslide susceptibility. chance of landslide occurrences. 2016). Dd means the total stream length per unit area. It also Preparation of Rr, Di, E, S, SRTM DEM (30 m spatial earthexplorer.usgs.gov Table 2. Formulae for calculatingTable 2. Formulaeof different for parameters calculating used of different in DD, RDparameters and used in DD, RD and Hi data layers resolution) (2014) indicatesMD the closenessmethod. ofTable spacing 2. Formulae ofMD channel method for per .calculating unit area, of different parameters used in DD, RD and Thematic data layers of Dd, Toposheet (1961 - 1962, 1965 - Survey of India (SOI), which determines the landscape dissection and runoff Df, Rb, Din, Dt, Jf and In 1966, 1969 - 1971) and Google Kolkata. toposheet no. 78A/4, MD method. earth imagery (2015) 78A/8, 78B/1, 78B/5 and potentiality of a region (Strahler, 1964). Dd indirectly 78B/6 and Google earth map Data layers of Ri and Rra SRTM DEM (30 m spatial earthexplorer.usgs.gov, 3.2.1presents Dd the erosional3.2.1 status Dd of the basin and highly resolution) and Toposheet (1961- Survey of India (SOI), correlates with landslide susceptibility. Higher Dd in an 62, 1965-66, 1969-71) and Google Kolkata. toposheet no. 78A/4, 3.2.1 Dd earth imagery (2015) 78A/8, 78B/1, 78B/5 and area indicates a high landslide probability and lower Dd This is an important factorThis in landslide is an important susceptibility. factor in Dd landslide means thesusceptibility. total stream Dd length means per the total stream length per 78B/6 and Google earth map means the opposite (Sarkar and Kanungo, 2004). High This is an important factor in landslide susceptibility. Dd means the total stream length per unit Ddarea. implies It also high indicates runoffunit theand area. closenessconsequently It also ofindicates spacinglow infiltration the of closenesschannel per of unitspacing area, of which channel determines per unit thearea, which determines the 3.1. Preparation of landslide inventory map unit area. It also indicates the closeness of spacing of channel per unit area, which determines the landscaperate, whereas dissection low andDdlandscape impliesrunoff lowpotentiality dissection runoff and ofand consequentlya regionrunoff potentiality(Strahler, 1964). of a regionDd indirectly (Strahler, presents 1964). the Dd indirectly presents the high infiltration.landscape In dissection this study, and Dd runoff varies potentiality from 0 km/ of a region (Strahler, 1964). Dd indirectly presents the A landslide inventory map is the preliminary step in erosionalsq. km status - 15.10 of the km/sq. basinerosional kmand and highlystatus was correlatesof classified the basin with into and landslide 10highly correlatessusceptibility. with Higherlandslide Dd susceptibility. in an area Higher Dd in an area preparing landslide susceptibility map. Several researchers erosional status of the basin and highly correlates with landslide susceptibility. Higher Dd in an area indicatesdifferent a high classes, landslide i.e. indicatesprobability 0 km/sq. a high andkm lowerlandslide - 1.31 Dd km/sq. probabilitymeans km, the oppositeand lower (Nag, Dd means1998; Sarkarthe opposite and Kanungo, (Nag, 1998; Sarkar and Kanungo, incorporated aerial photographs and satellite images to 1.31 km/sq.indicates km - a2.61high km/sq.landslide km, probability 2.61 km/sq. and lowerkm - Dd means the opposite (Nag, 1998; Sarkar and Kanungo, identify different locations of landslide events (Lee and 2004).3.67 High km/sq. Dd implies km, 3.67 high2004). km/sq.runoff, High and kmDd consequentlyimplies- 4.56 km/sq.high runoff,low km, infiltration and consequently rate, whereas low lowinfiltration Dd implies rate, lowwhereas low Dd implies low 2004). High Dd implies high runoff, and consequently low infiltration rate, whereas low Dd implies low Pradhan, 2007; Pradhan, 2010; Choi et al., 2011). In this runoff,4.56 and km/sq. consequently km - 5.39 runoff,high km/sq. infiltration. and consequentlykm, 5.39In this km/sq. study, high infiltration.kmDd varies- from In this 0 km/sq.study, Ddkm varies- 15.10 from km/sq. 0 km/sq. km km - 15.10 km/sq. km study, Google Earth historical imageries (2000 - 2016), 6.27 km/sq.runoff, km, and 6.27 consequently km/sq. km high - 7.28 infiltration. km/sq. Inkm, this study, Dd varies from 0 km/sq. km - 15.10 km/sq. km and was classified into tenand different was classified classes, into i.e. 0ten km/sq. different km classes, - 1.31 km/sq.i.e. 0 km/sq. km, 1.31 km km/sq. - 1.31 kmkm/sq. - 2.61 km, 1.31 km/sq. km - 2.61 Landsat 7 enhanced thematic mapper plus (ETM+) 7.28 km/sq. km - 8.58 km/sq. km, 8.58 km/sq. km - and was classified into ten different classes, i.e. 0 km/sq. km - 1.31 km/sq. km, 1.31 km/sq. km - 2.61 imageries (2000 - 2005), Landsat 4 - 5 thematic mapper km/sq.10.60 km, km/sq. 2.61 km/sq. km and kmkm/sq. 10.60 - 3.67 km/sq.km, km/sq. 2.61 km km/sq.km, - 15.10 3.67 km km/sq. km/sq. - 3.67 km kmkm/sq. - 4.56 km, km/sq. 3.67 km/sq.km, 4.56 km km/sq. - 4.56 km km/sq. - 5.39 km, 4.56 km/sq. km - 5.39 (TM) imageries (2006 - 2007, 2009 - 2011), linear imaging km/sq.(Figure km, 5.392(a)).km/sq. km/sq. km, km 2.61 - 6.27 km/sq. km/sq. km km,- 3.67 6.27 km/sq. km/sq. km, km 3.67 - 7.28 km/sq. km/sq. km km,- 4.56 7.28 km/sq. km/sq. km, km 4.56 - 8.58 km/sq. km - 5.39 self-scanning sensors (LISS) III imageries (2008, 2012 - km/sq. km, 5.39 km/sq. km - 6.27 km/sq. km, 6.27 km/sq. km - 7.28 km/sq. km, 7.28 km/sq. km - 8.58 km/sq. km, 5.39 km/sq. km - 6.27 km/sq. km, 6.27 km/sq. km - 7.28 km/sq. km, 7.28 km/sq. km - 8.58 2013), Landsat 8 operational land imager (OLI) (2014 - km/sq.3.2.2. km, Df 8.58 km/sq. kmkm/sq. - 10.60 km, km/sq. 8.58 km/sq. km and km 10.60 - 10.60 km/sq. km/sq. km -km 15.10 and km/sq.10.60 km/sq. km (Figure km - 315.10(a)). km/sq. km (Figure 3(a)). 2015), high resolution Sentinel-2 imagery (2016), Arc Map km/sq. km, 8.58 km/sq. km - 10.60 km/sq. km and 10.60 km/sq. km - 15.10 km/sq. km (Figure 3(a)). high resolution world imagery and intensive field survey Df is an important geomorphic concept which data (2015 - 2016) were used in the mapping of several 3.2.2measures Df the relative3.2.2 spacing Df of drainage lines. The term large and small landslides. There are three major types of refers to the3.2.2 number Df of stream segments of all orders per landslide identified in the study area - shallow translational unit area (Horton, 1945). It helps for understanding the Df is an important geomorphic concept which measures the relative spacing of drainage debris slides, deep-seated rockslides and shallow drainage texture as well as the textureDf is ofan the important landscape geomorphic concept which measures the relative spacing of drainage Df is an important geomorphic concept which measures the relative spacing of drainage translational rockslides. A total of 295 landslide locations lines.of The an termarea. refers High toDf thelines. indicates number The termfastof stream runoffrefers segments toand the therefore number of all oforders stream per segments unit area of(Horton, all orders 1932 per; Horton,unit area (Horton, 1932; Horton, were mapped (Figure 1) and randomly divided into two landslidelines. susceptibility The term refersis more to thelikely number in a basinof stream with segments of all orders per unit area (Horton, 1932; Horton, 1945). It helps for understanding1945). It the helps drainage for understanding texture as well the asdrainage the texture texture of theas welllandscape as the oftexture an area. of the landscape of an area. parts, i.e. 70% of the total landslide locations were used high stream frequency (Jeganathan and Chauniyal, 1945). It helps for understanding the drainage texture as well as the texture of the landscape of an area. as a training data set and the remaining 30% of locations 2000). In this study, Df was grouped into 10 classes: were used for validation of the model. The average area of 0 streams/sq. km - 4.91 streams/sq. km, 4.91 streams/sq. km - locations were used as a training data2 set and the remaining 30% of locations were used for validation of landslide was 3,762.90 m . 11.05 streams/sq. km, 11.05 streams/sq. km - 2 the model. The average area of landslide was 3762.90 m . 15.96 streams/sq. km, 15.96 streams/sq. km - 3.2. Preparation of morphometric data layers and their locations were used as a training data set and the remaining 30% of locations21.27 were streams/sq. used for validation km, 21.27 of streams/sq. km - role in slope instability 27.00 streams/sq. km, 27.00 streams/sq. km - 3.2 Preparation of morphometric data layers and their 2role in slope instability the model. The average area of landslide was 3762.90 m . 33.54 streams/sq. km, 33.54 streams/sq. km - A total of 14 parameters were calculated using a A total of 14 parameters were calculated using a recognised40.90 formula streams/sq. and prepared km, spatial40.90 streams/sq. km - 3.2 Preparationrecognised of morphometric formula and prepared data layers spatial and data their layers role in in Arc slope instability52.35 streams/sq. km, 52.35 streams/sq. km - data layersGIS in environment.Arc GIS environment The relationship (Table between 2). The morphometricrelationship between 70.35 morphometric streams/sq. parameters km and and70.35 streams/sq. km - landslide parametersoccurrencesA total and ofwere 14landslide parametersjudged occurrences using were a frequency calculated were judged ratio using (FR)using a recognised approach104.30 formula(Karim streams/sq. etand al., preparedkm 2011; (Figure Mondal spatial 2(b)). a FR approach (Karim et al., 2011; Mondal and Mandal, dataand Mandal,layers2017a), in 2017b), Arc GIS environment (Table 2). The relationship between3.2.3. morphometric Rb parameters and landslide occurrences were judged using a frequency ratio (FR) approach (Karim et al., 2011; Mondal FR= . Rb is a dimensionless (1) property of the drainage basin and Mandal, 2017b), (1) {𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁푁100 that may be defined as the ratio between the total number {∑ 𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁�𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁푁100 of stream segments of an order and the next higher order Where is the number of pixels containing FRWhere= N pix (Si) is the number of. pixels containing landslide in each class (1(i)), N pix (Ni) is the total number of pixels{𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁 �having푁𝑁𝑁𝑁𝑁푁푁𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁 class�푁𝑁𝑁𝑁𝑁푁푁푁 (i100) in the whole basin, ∑N pix (Si) total number of pixels in each 15 {∑ 𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁�𝑁𝑁�𝑁𝑁𝑁𝑁𝑁𝑁�푁𝑁𝑁𝑁𝑁푁푁푁100 class and ∑N pixWhere (Ni )N total pix (Si)number is the of number pixels inof theHKIEpixels whole Transactions containing basin. | TheVolume landslide higher 27, Number in the each FR1, pp. classvalue,13–24 (i ),the N greaterpix (Ni )theis thechance total of number landslide of pixelsoccurrences. having class (i) in the whole basin, ∑N pix (Si) total number of pixels in each class and ∑N Tablepix (Ni 2.) totalFormulae number for of calculating pixels in theof wholedifferent basin. parameters The higher used the in FRDD, value, RD andthe greater the chance of landslideMD method occurrences..

Table 2. Formulae for calculating of different parameters used in DD, RD and 3.2.1 Dd MD method.

This is an important factor in landslide susceptibility. Dd means the total stream length per 3.2.1 Dd unit area. It also indicates the closeness of spacing of channel per unit area, which determines the landscape dissectionThis is anand important runoff potentialityfactor in landslide of a region susceptibility. (Strahler, Dd 1964). means Dd the indirectlytotal stream presents length perthe uniterosional area. statusIt also of indicates the basin the and closeness highly correlates of spacing with of landslidechannel persusceptibility. unit area, whichHigher determinesDd in an areathe landscapeindicates a dissectionhigh landslide and probabilityrunoff potentiality and lower of Dd a meansregion the(Strahler, opposite 1964). (Nag, Dd1998; indirectly Sarkar and presents Kanungo, the erosional2004). High status Dd ofimplies the basin high runoff,and highly and consequentlycorrelates with low landslide infiltration susceptibility. rate, whereas Higher low DdDd implies in an arealow indicatesrunoff, and a high consequently landslide probabilityhigh infiltration. and lower In this Dd study,means Dd the varies opposite from (Nag, 0 km/sq. 1998; km Sarkar - 15.10 and km/sq.Kanungo, km 2004).and was High classified Dd implies into tenhigh different runoff, andclasses, consequently i.e. 0 km/sq. low km infiltration - 1.31 km/sq.rate, whereas km, 1.31 low km/sq. Dd implies km - 2.61low runoff,km/sq. andkm, consequently2.61 km/sq. km high - 3.67infiltration. km/sq. km,In this 3.67 study, km/sq. Dd kmvaries - 4.56 from km/sq. 0 km/sq. km, km4.56 - km/sq.15.10 km/sq. km - 5.39 km andkm/sq. was km, classified 5.39 km/sq. into ten km different - 6.27 km/sq. classes, km, i.e. 6.27 0 km/sq. km/sq. km km -- 1.317.28 km/sq.km/sq. km,km, 1.317.28 km/sq.km/sq. kmkm -- 2.618.58 km/sq.km/sq. km, 2.618.58 km/sq. kmkm -- 10.603.67 km/sq.km/sq. km,km and3.67 10.60 km/sq. km/sq. km - km4.56 - 15.10km/sq. km/sq. km, 4.56 km (Figurekm/sq. km3(a) -). 5.39 km/sq. km, 5.39 km/sq. km - 6.27 km/sq. km, 6.27 km/sq. km - 7.28 km/sq. km, 7.28 km/sq. km - 8.58 km/sq.3.2.2 Df km, 8.58 km/sq. km - 10.60 km/sq. km and 10.60 km/sq. km - 15.10 km/sq. km (Figure 3(a)).

3.2.2 Df Df is an important geomorphic concept which measures the relative spacing of drainage lines. The term refers to the number of stream segments of all orders per unit area (Horton, 1932; Horton, 1945). It helpsDf for is understanding an important geomorphicthe drainage concepttexture aswhich well measuresas the texture the relativeof the landscape spacing of of drainage an area. lines. The term refers to the number of stream segments of all orders per unit area (Horton, 1932; Horton, 1945). It helps for understanding the drainage texture as well as the texture of the landscape of an area. S MONDAL AND S MANDAL

(Horton, 1945). It is strongly controlled by Dd, lithological 5.32 streams/km - 6.75 streams/km, 6.75 streams/km - characteristics, basin shape and area, and stream entrance 8.38 streams/km, 8.38 streams/km - 10.23 streams/km, angle, etc. A higher Rb means structurally disturbed 10.23 streams/km - 13.10 streams/km, 13.10 streams/km - watersheds with high distortion in drainage pattern indicate 17.59 streams/km and 17.59 streams/km - 26.07 streams/km a highly landslide prone area (Ghosh, 2015). The Rb of (Figure 2(f)). the Balason river basin ranges from 1.50 - 5.66 and it was reclassified into eight classes, i.e. 1.50 - 2.56, 2.56 - 3.25, 3.2.7. Jf 3.25 - 3.71, 3.71 - 4.01, 4.01 - 4.22, 4.22 - 4.52, 4.52 - 4.98 and 4.98 - 5.66 (Figure 2(c)). This is the number of stream nodes per unit area (Ghosh and Saha, 2015). It reflects the hydrological 3.2.4. Rr condition of a basin. The higher the Jf value, the greater chance of water availability. The river erosional Rr is another important parameter in the preparation process leads to the slope becoming more vulnerable. of a landslide susceptibility zonation map (Mandal and Simply put, it enhances the chance of landslide Mandal, 2016). It can be defined as the difference between occurrences (Jeganathan and Chauniyal, 2000). highest altitude and lowest altitude. The Rr is positively The Jf value ranges from 0 stream nodes/sq.km - associated with landslide susceptibility. High Rr enhances 67.49 stream nodes/sq.km and is grouped into 10 classes, the chance of slope failure and vice versa (Jeganathan i.e. - 0 stream nodes/sq.km - 2.38 stream nodes/sq.km, and Chauniyal, 2000). For the present study, the Rr 2.38 stream nodes/sq.km - 5.82 stream nodes/sq.km, value ranges from 5.19 m/sq. km - 662.00 m/sq. km and 5.82 stream nodes/sq.km - 9.00 stream nodes/sq.km, is divided into 10 different classes, i.e. 5.19 m/sq. km - 9.00 stream nodes/sq.km - 12.45 stream nodes/sq.km, 51.70 m/sq. km, 51.70 m/sq. km - 127.00 m/sq. km, 12.45 stream nodes/sq.km - 16.42 stream nodes/sq.km, 127.00 m/sq. km - 215.00 m/sq. km, 215.00 m/sq. km - 16.42 stream nodes/sq.km - 20.92 stream nodes/sq.km, 292.00 m/sq. km, 292.00 m/sq. km - 346.00 m/sq. km, 20.92 stream nodes/sq.km - 26.74 stream nodes/sq.km, 346.00 m/sq. km - 393.00 m/sq. km, 393.00 m/sq. km - 26.74 stream nodes/sq.km - 34.95 stream nodes/sq.km, 434.00 m/sq. km, 434.00 m/sq. km - 475.00 m/sq. km, 34.95 stream nodes/sq.km - 46.33 stream nodes/sq.km 475.00 m/sq. km - 532.00 m/sq. km and 532.00 m/sq. km - and 46.33 stream nodes/sq.km - 67.49 stream nodes/sq.km 662.00 m/sq. km (Figure 2(d)). (Figure 2(g)).

3.2.5. Din 3.2.8. In

Din may be defined as the ratio of Df to Dd (Faniran, The In of a drainage basin is the product of Dd and 1968). A high Din value indicates that Dd and stream stream frequency (Faniran, 1968). The higher the In, frequency have significant influence on the degree to which the lower the infiltration and the higher the runoff will the land surface of the watershed has been lowered by be. Therefore, a high In is closely associated with slope various denudation processes. Simply put, high Din areas instability (Ghosh, 2015). In values were reclassified into are characterised by more slope failure events. The Din of 10 classes, i.e. 0 In/sq.km - 30.89 In/sq.km, 30.89 In/ the study area ranges from 0 - 48.17 and was reclassified sq.km - 80.32 In/sq.km, 80.32 In/sq.km - 148.28 In/sq.km, into 10 categories, i.e. 0 - 1.51, 1.51 - 3.02, 3.02 - 4.16, 148.28 In/sq.km - 234.78 In/sq.km, 234.78 In/sq.km - 4.06 - 5.67, 5.67 - 8.12, 8.12 - 12.10, 12.10 - 17.96, 17.96 - 339.81 In/sq.km, 339.81 In/sq.km - 481.92 In/sq.km, 25.51, 25.51 - 34.58 and 34.58 - 48.17 (Figure 2(e)). 481.92 In/sq.km - 667.27 In/sq.km, 667.27 In/sq.km - 883.51 In/sq.km, 883.51 In/sq.km - 1,130.66 In/sq.km and 3.2.6. Dt 1,130.66 In/sq.km - 1,575.50 In/sq.km (Figure 2(h)).

Dt denotes the total number of stream segments of all 3.2.9. Ri orders per perimeter of that area (Horton, 1945). It indicates the relative spacing of drainage lines and the cumulative Topographic ruggedness exhibits significant control effect of the infiltration capacity, the amount and type of over the spatial occurrence of landslides and erosion prone vegetation, infiltration capacity and relief aspect of the areas, and portrays the intricate relationship of lithology, terrain, which influences the rate of surface runoff and affects structure, relief and climate. It is the product of Dd and the Dt of a basin. A high textural area indicates greater basin relief divided by 1,000 (Miller et al., 1990). The susceptibility of landslide occurrences (Jeganathan and low ruggedness value of the basin implies that the area is Chauniyal, 2000). In this study, the value of the Dt ranges less prone to soil erosion and landslide, and has intrinsic from 0 streams/km - 26.07 streams/km and is categorised structural complexity in association with relief and Dd. In the into 10 classes, i.e. 0 streams/km - 1.23 streams/km, study, this figure ranges from 0 Ri/sq. km - 6.55 Ri/sq. km, 1.23 streams/km - 2.76 streams/km, 2.76 streams/km - which is partitioned into 10 groups, i.e. 0 Ri/sq. km - 3.99 streams/km, 3.99 streams/km - 5.32 streams/km, .34 Ri/sq. km, .34 Ri/sq. km - .93 Ri/sq. km, .93 Ri/sq. km -

16 HKIE Transactions | Volume 27, Number 1, pp.13–24

1.42 Ri/sq. km, 1.42 Ri/sq. km - 1.83 Ri/sq. km, 1.83 Ri/sq. km - 2.22 Ri/sq. km, 2.22 Ri/sq. km - 2.63 Ri/sq. km, 2.63 Ri/sq. km - 3.12 Ri/sq. km, 3.12 Ri/sq. km - 3.74 Ri/sq. km, 3 .74 Ri/sq. km - 4.48 Ri/sq. km and 4.48 Ri/sq. km - 6.55 Ri/sq. km (Figure 2(i)).

3.2.10. Di

The Di depicts the degree of dissection or vertical erosion, and the stage of landform development in a basin watershed. It may be defined as the ratio between the relative relief and absolute relief of the basin (Nir, 1957) and the value always ranges between 0.0 (complete absence of dissection and thus the dominance of flat topography) and 1 for infrequent cases such as vertical cliff topography at the sea shore or vertical escarpment of hill slope (Farhan et al., 2015). A high Di enhances the chance of S failure and vice versa (Jeganathan and Chauniyal, 2000; Ghosh, 2015). In this study, 10 classes of dissection index were made: .06 Di/sq. km - .13 Di/sq. km, 13 Di/sq. km - .17 Di/sq. km, .17 Di/sq. km - .22 Di/sq. km, .22 Di/sq. km - .26 Di/sq. km, .26 Di/sq. km - .31 Di/sq. km, .31 Di/sq. km - .37 Di/sq. km, .37 Di/sq. km - .42 Di/sq. km, .42 Di/sq. km - .47 Di/sq. km, .47 Di/sq. km - .52 Di/sq. km and .52 Di/sq. km - .59 Di/sq. km (Figure 2(j)).

3.2.11. E

E is the primary causative factor which affects the slope material through the weathering process. It has a direct influence on all others topographic aspects, slope, soil layer arrangement, etc. As the elevation increases, the slope angle also increases, which directly makes the land surface more landslide susceptible (Devkota et al., 2013). Figure 2. Spatial data layers of (a) Dd; (b) Df; (c) Rb; (d) In this study, the E value ranges from 106 m - 2,600 m and Rr; (e) Din; (f) Dt; (g) Jf; (h) In; (i) Ri; (j) Di; (k) E; and (l) S. is divided into 10 classes, i.e. 106 m - 298 m, 298 m - 539 m, 539 m - 792 m, 792 m - 1,030 m, 1,030 m - 1,240 m, 3.2.13. Rra 1,240 m - 1,440 m, 1,440 m - 1,650 m, 1,650 m - 1,880 m, 1,880 m - 2,130 m and 2,130 m - 2,600 m (Figure 2(k)). Rra is an index of relief characteristics and steepness in a drainage basin. It refers to the difference in height between 3.2.12. S the highest and lowest points in the basin and the horizontal distance along the longest dimension of the basin parallel to S is a key factor which strongly influences the landslide the main stream line (Schumm, 1956). The ratio is positively susceptibility of a region. A positive correlation exists correlated with the rate of erosion in the basin and the value between landslides and S which indicates that areas with of the Rra normally increases with decreasing drainage area steep S are more prone to landslide occurrences (Mandal and size of a given catchment. In this study, the Rra varies and Mandal, 2018). As the S angle increases, the tendency from .02 - .30 and was classified into 10 different classes, i.e. of movement increases and shear stress in soil and other un- .02, .02 - .11, .11 - .14, .14 - .18, .18 - .20, .20 - .21, .21 - .22, consolidated material generally increases. Moderate S is .22 - .23, .23 - .27 and .27 - .30 (Figure 3(a)). expected to have a low frequency of landslide (Ayalew et al., 2005). The S of the basin ranges from 0° - 71.10° and 3.2.14. Hi was reclassified into 10 classes, i.e. 0°- 5.03°, 5.03°- 11.30°, 11.30°- 16.80°, 16.80°- 21.90°, 21.90°- 26.60°, 26.60°- Hypsometrics are related to the measurement and 31.10°, 31.10°- 35.50°, 35.50°- 40.60°, 40.60°- 47.00° and analysis of relationships between altitude and basin area. It 47.00°- 71.10° (Figure 2(l)). helps for understanding the degree of dissection and stage of erosion of a basin (Strahler, 1952). The Hi indicates the

17 HKIE Transactions | Volume 27, Number 1, pp.13–24 Figure 4 Spatial data layers of (a) Rra and (b) Hi.

FigureFigureFigure 4 Spatial 4 FigureSpatial Figure4 Spatial data 44data layersSpatialSpatial data layers layersof data data (a) of layersRraof(a)layers (a) Rraand Rraofof and(b) (a)(a)and Hi. (b) RraRra (b) Hi.andand Hi. (b)(b) Hi.Hi. FigureFigure 4 4SpatialSpatial data data layers layers of of (a) (a) Rra Rraandand (b) (b) Hi. Hi. Figure 4 Spatial data layers of (a) Rra and (b) Hi. 3.3 Uniform scaling and assigning principal component analysis (PCA)-based weight for constructing 3.3 Uniform3.33.3 Uniform Uniform3.3 3.3scaling UniformUniformscaling scaling and and assigning scalingscaling and assigning assigning andand principal assigningassigning principal principal component principal principalcomponent component analysis componentcomponent analysis analysis (PCA)-based analysis(PCA)-basedanalysis (PCA)-based (PCA)-based (PCA)-basedweight weight weight for constructingfor weightweight for constructing constructing forfor constructing constructing 3.33.3 Uniform Uniform scaling scaling and and assigning assigning principal principal component component analysis analysis (PCA)-based (PCA)-basedDD, RD weight and weight MD for for constructing modelsconstructing and landslide susceptibility zonation mapping DD,DD, RDDD, RDand RD DD,and DD,MD and MD RDmodelsRD MD andmodelsand models and MDMD and landslidemodels models and landslide landslide andand susceptibility landslide landslidesusceptibility susceptibility susceptibilitysusceptibility zonation zonation zonation mapping zonationzonationmapping mapping mapping mapping 3.3 UniformDD,DD, scaling RD RD and and and MD MD assigning models models and principal and landslide landslide component susceptibility susceptibility analysis zonation zonation (PCA)-based mapping mapping weight for constructing DD, RD and MD models and landslide susceptibility zonation mapping AfterAfter preparingAfter preparing preparingAfterAfter the preparingpreparing themorphometric the morphometric Aftermorphometric thethe morphometric morphometricpreparing data data layers data layers layers thein datadata GIS inmorphometric layerslayersinGIS platform, GIS platform, inin platform, GISGIS DD, platform, platform,DD, RD dataDD, RDand RD layers DD, andDD,MD and MDRD RDmodels inMD andmodelsand GIS models MDMD platform, modelsmodels DD, RD and MD models AfterAfter preparing preparing the the morphometric morphometric data data layers layers in in GIS GIS platform, platform, DD, DD, RD RD and and MD MD models models werewere usedwere used forwereusedwere forlandslide for usedusedlandslide landslide for forsusceptibility landslide landslidesusceptibility susceptibility susceptibilitysusceptibility zonation zonation zonation mapping zonationzonationmapping mapping (Table mappingmapping(Table (Table 3). 3).Some (Table (Table3). Some Someparameters 3).3). parameters SomeSomeparameters areparametersparameters arecommon are common common are are commoncommon werewereAfter used used preparing for for landslide landslide the morphometric susceptibility susceptibility datazonation zonation layers mapping inmapping GIS platform, (Table (Table 3). DD,3).were Some SomeRD used andparameters parameters MD for models landslideare are common common susceptibility zonation mapping (Table 3). Some parameters are common becausebecausebecause these thesebecausebecause parametersthese parameters parametersthesethese directly parametersparameters directly directly or indirectly or directlydirectly orindirectly indirectly ororrepresent indirectlyindirectly represent represent both represent representboth the both therelief the bothreliefboth andrelief the the anddrainage andreliefrelief drainage drainage and andcharacter, drainagedrainagecharacter, character, such character,character, such as such as as suchsuch asas were usedbecause becausefor landslide these these parameters parameterssusceptibility directly directly zonation or or indirectly indirectly mapping represent represent (Table both 3).both Somethe the relief relief parametersbecause and and drainage drainage theseare commoncharacter, character,parameters such such asdirectly as or indirectly represent both the relief and drainage character, such as Rb RbwhichRb which whichindicatesRbRb indicates whichwhichindicates the indicates indicatesthedrainage the drainage drainage the thecharacteristics drainagedrainagecharacteristics characteristics characteristicscharacteristics and andbreak and break breakof andand Sof characteristics. breakofbreakS characteristics.S characteristics.ofof SS characteristics.characteristics. In theIn Inthesame the same In Inway,same the theway, Risameway,same Ri Riway,way, RiRi because theseRbRb which parameterswhich indicates indicates directly the the ordrainage drainage indirectly characteristics characteristics represent both and and the break break relief of ofand S S characteristics. drainagecharacteristics. character, In In the suchthe same same as way, way, Ri Ri impliesimpliesimplies the arrangementtheimpliesimplies the arrangement arrangement the theRb arrangementofarrangement a whichofdrainage ofa drainage a drainage indicates of ofnetwork aa drainagenetworkdrainage network and the andnetworkruggednessnetwork and drainageruggedness ruggedness andand characteristics ruggednessruggedness characteristicscharacteristics characteristics characteristicscharacteristics of relief. of ofrelief.and relief. of ofbreak relief. relief. of S characteristics. In the same way, Ri Rb whichimplies indicatesimplies theS theMONDAL thearrangement arrangement drainage AND S MANDAL ofcharacteristics of a adrainagedrainage network networkand break and and ruggednessof ruggedness S characteristics. characteristics characteristics In the of of relief.same relief. way, Ri implies the arrangement of a drainage network and ruggedness characteristics of impliesrelief. the arrangement of a drainage network and ruggedness characteristics of relief. TableTable 3Table. Spatial 3. TableSpatial3Table. Spatial data 33 .data .layersSpatialSpatial data layers layersused datadata used in layersusedlayers RD in ,RDin DD usedused RD, DDand ,in inDD RDandMDRDand,,MDDDforDDMD landslideandforandfor landslideMDMD landslide forsusceptibilityfor landslide landslidesusceptibility susceptibility susceptibilitysusceptibility mapping mapping mapping. . mappingmapping. .. TableTable 3 .3Spatial. Spatial data data layers layers used used in in RD RD, DD, DDandandMDMDforfor landslide landslide susceptibility susceptibility mapping mapping. . Table 3. Spatialpercentage data layers area used under in RD the, DDdimensionlessand MD for curve landslide as a goodsusceptibility mappingToTable perform .3. Spatial the aforesaid data layers three usedmodel in analysis, RD, DD and MD for landslide susceptibility mapping. indicator of the basin erosional status or the surface decay, principal compoent analysis (PCA)-based weights were To perform the aforesaid three model analysis, PCA-based weights were assigned to each whereas the erosional integral is the area proportionate to ToTo performassigned performToTo theperformperform to the aforesaideach aforesaid the thedata aforesaid aforesaidthree layer three model of model threethree each analysis, model model analysis,mode. analysis,analysis,PCA-based PCA-based PCA-based PCA-basedPCA-based weights weights were weightsweights were assigned assigned werewere assignedtoassigned toeach each toto eacheach ToTo perform perform the the aforesaid aforesaid three three model model analysis, analysis, PCA-based PCA-based weights weights were were assigned assigned to to each each the area above the curve. The high value dataof thedata layer dataHi layer denotes oflayerdata dataeach of of eachlayer layermode. each weight mode. ofof mode. PCA-based eacheach PCA-basedwas mode.mode.PCA-based also weight PCA-based PCA-basedused weight weight bywas Khatun was also weightweight was also used andalso waswasused by Palused also alsoKhatun by(2017). byKhatun usedused Khatun and Itbyby was and PalKhatunKhatun and Pal(2017). Pal (2017). andand (2017). It PalPal was It (2017).(2017). was Itcalculated was calculated ItItcalculated waswas calculatedcalculated datadata Tolayer layer performthe of activeof each each the erosionalmode. mode.aforesaid PCA-based PCA-based youthful three model stageweight weight analysis,of wasthe was basin,also also PCA-based used whichused by by isKhatun weights Khatuncalculated and andwere Pal Pal assigned using(2017). (2017). Equation Itto It waseach was (2),calculated calculated using Equationusing (2), Equation (2), more slide prone (Ghosh, 2015). The Hi mapusing ofusing the Equation studyEquationusing Equation (2 ),(2), (2), To perform the aforesaid three model analysis, PCA-based weights were assigned to each data layer ofusing each Equation mode. PCA-based (2), weight was also used by Khatun and Pal (2017). It was calculated using Equationarea was (2 divided), into 10 classes, i.e. .18 - .31, .31 - .37, . . . .. (2) (2) (2) (2) (2) (2) using Equation (2), data layer∑𝑎𝑎𝑎𝑎𝑎𝑎 of each mode. PCA-based weight was also used by Khatun and Pal (2017). It was calculated .37 - .41, .41 - .44, .44 - .48, .48 -. .51, . .51 - .54, .54 - .57, ∑𝑎𝑎𝑎𝑎𝑎𝑎 ∑𝑎𝑎𝑎𝑎𝑎𝑎 (2) (2)∑𝑎𝑎𝑎𝑎𝑎𝑎 ∑𝑎𝑎𝑎𝑎𝑎𝑎 Figure 3. Spatial data layers of (a) Dd; (b) Df; (c) Rb; (d) Rr; (e) Din; (f) textured; (g) Jf; (h) In; (i) Ri; (j) Di; (k) E; and (l) 𝑊𝑊𝑊𝑊 = ∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎 S. .57 - .62 and .62 - .75 (Figure. ∑𝑎𝑎𝑎𝑎𝑎𝑎 ∑𝑎𝑎𝑎𝑎𝑎𝑎 3(b)). using 𝑊𝑊𝑊𝑊 𝑊𝑊𝑊𝑊 Equation = (2)∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎= 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎= =(2∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎), Where WmWhere is the weightWm isof the the weight mth parameter, of the mth parameter,amr is the rowamr wise is thesummation row wise of summation of ∑𝑎𝑎𝑎𝑎𝑎𝑎𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊==∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎 WhereWhere WhereWhere WmWm is theis Wm isthe weightthe weightis weight the of weight ofthe of the themth ofmth mparameter, thethparameter, parameter, mth parameter, amrisamr is theisamr the row rowis wisethe wise row summation summation wise summation of of of WhereWhere WmWm is is the the weight weight of of the the mth mthparameter,parameter, theamramr row is is wisethe the row summation row wise wise summation summationof correlation of of coefficient of the mth 𝑊𝑊𝑊𝑊 = ∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎 correlationcorrelationcorrelation coefficientcorrelationcorrelation coefficient coefficient𝑊𝑊𝑊𝑊 of coefficientcoefficient the of of mththe the mth parameter of mthof parameter the the parameter mth mth (avoiding parameter parameter (avoiding (avoiding the (avoiding(avoiding minusthe the minus minussymbol)𝑎𝑎 the𝑎𝑎the symbol)𝑎𝑎 minusminussymbol) .and symbol)and symbol) and andandis the is maximumtheis the maximum ismaximum is the the maximum maximum (2) Where Wm is the weight of the mth parameter, amr is theparameter row𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 wise (avoiding summation𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 the minusof symbol) and 𝑎𝑎𝑎𝑎𝑎𝑎 𝑎𝑎is𝑎𝑎 the𝑎𝑎 𝑎𝑎 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 correlationcorrelation coefficient coefficient of of the the mth mth parameter parameter (avoiding (avoiding the the minus minus symbol) symbol) and and isis the the maximum maximum ∑𝑎𝑎𝑎𝑎𝑎𝑎 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 correlationcorrelationcorrelation coefficientcorrelationcorrelation coefficient coefficient𝑎𝑎𝑎𝑎𝑎𝑎 𝑎𝑎maximumwithin 𝑎𝑎 coefficientcoefficient within withinthe correlation totalthe withinwithinthe total parameter. total parameter. thethecoefficient parameter. totaltotal parameter. parameter. within the total parameter. 𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎 correlation coefficient of the mth parameter (avoiding the minus symbol) and is the maximum 𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 correlationcorrelation coefficient𝑊𝑊𝑊𝑊 coefficient within within the the total total parameter. parameter. 𝑎𝑎𝑎𝑎𝑎𝑎 The preparation of a landslide𝑊𝑊𝑊𝑊 = susceptibility∑𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎 map 𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎Where Wm is the weight of the mth parameter, amr is the row wise summation of correlation coefficient within the total parameter. TheThe preparationThe𝑎𝑎𝑎𝑎𝑎𝑎푎 requirespreparation preparation𝑎𝑎TheThe𝑎𝑎𝑎𝑎 of thatpreparationpreparation aof thelandslide ofa landslidefactorsa landslide ofof susceptibility a aconsideredlandslide landslidesusceptibility susceptibility susceptibility susceptibility bemap combinedmap requires map requires requires map mapwiththat requiresthatrequires the that thefactorsthe factorsthatthat factors consideredthethe considered factorsfactors considered be consideredconsidered be be bebe TheThe preparation preparation of of a alandslidelandslide susceptibility susceptibility map map requires requiresrespect that that to thetheirthe factors factorsrelative considered consideredimportance be bein landslide occurrence combinedcombinedcombined withcombinedcombined with respect with respect respect withcorrelationtowith their to respectrespect totheir relative their relative to to coefficientrelative their theirimportance importance relativerelative importance of in importanceimportance landslide thein inlandslide mth landslide occurrenceininparameter landslide landslideoccurrence occurrence ( occurrenceSarkaroccurrence (avoiding (Sarkar (Sarkar and and Kanungo,((Sarkar Sarkar andthe Kanungo, Kanungo,minus andand 2004 Kanungo,Kanungo, 2004 ;symbol) 2004; ; 20042004 and;; is the maximum The preparation of a landslide susceptibility map requires that the(Sarkar factors and considered Kanungo, be 2004). 𝑊𝑊𝑊𝑊 The intra-class ranking 𝑎𝑎𝑎𝑎𝑎𝑎 combinedcombined with with respect respect to to their their relative relative importance importance in in landslide landslide occurrence occurrence ( Sarkar(Sarkar and and Kanungo, Kanungo, 2004 2004; ; KanungoKanungoKanungo et al.Kanungo Kanungoet, al.et2006 ,al.2006)., of et 2006etThe 14al.).al. The ,).,intraparameters2006 2006The intra-class ).).intra The-Theclass ranking-was class intraintra ranking done -ranking-classclass of based14 of rankingranking parametersof14 on14parameters a ofparametersof 10-point 1414 wasparametersparameters was donescale was done baseddoneto was was based basedon donedone aon10 based onbaseda-point 10a -10 pointonon-scalepoint aa10scale10 to--scalepointpoint to toscalescale toto combined with respect to their relative importance in landslide occurrence (Sarkarcorrelation and Kanungo, coefficient 2004; within the total parameter. KanungoKanungo et et al. al., 2006, 2006). ).The The intra intra-class-class ranking ranking of of 14 14parametersparameters was was makedone done thebased based parameters on on a a1010- pointunidirectional-pointscalescale to to in terms of landslide 𝑎𝑎𝑎𝑎𝑎𝑎푎𝑎𝑎𝑎𝑎𝑎𝑎 makemake themake theparameters makethe makeparameters parameters thethe unidirectional parametersparameters unidirectional unidirectional unidirectionalunidirectional in terms in interms ofterms landslideinofin termsoftermslandslide landslide ofsusceptibilityof landslide landslidesusceptibility susceptibility susceptibilitysusceptibility, with, with ,10with points10 , ,10pointswithwith pointsassigned 1010assigned pointspointsassigned to the assignedassignedto tothe the toto thethe Kanungo makeetmake al., the2006 the parameters ).parameters The intra unidirectional -unidirectionalclass ranking in ofin terms 14termsparameters of of landslide landslide was susceptibility susceptibilitydone basedsusceptibility, ,on,withwith a 10 10 10-point points withpointsscale 10assignedassigned points to toassigned to the the to the class which classclass whichclass which whichenhancesclassclass enhances whichwhichenhances theenhances enhances enhanceschancethe the chance thechance of thethechance landslide of chancechance oflandslideThe oflandslide landslide occurrencesofpreparationof landslide landslideoccurrences occurrences occurrences occurrencesofoccurrences each ofof ofeacha dataof eachlandslide each data oflayerof data eacheach datalayer (Table layer datadatasusceptibility (Table (Tables layerlayer4, s54 (Tableands(Table, 54 , and65). sand s Exceptmap464,)., 56 5Except ). and and Exceptrequires 66). ). ExceptExcept that the factors considered be make the classparametersclass which which unidirectionalenhances enhances the the chance inchance terms of of oflandslide landslide landslide occurrences occurrences susceptibility of of each ,eachwith data data 10 layer pointslayer (Table (Tableassigneds s4,4 5 , to 5and andthe 6 ).6 ).Except Except the Rbthethe dataRb Rb data layer,the thedata layer, RbRb alllayer, datadata thealllayer. layer,alldatalayer,the the Exceptdata layersall alldata layers thethe thelayerswere datadata Rbwere divided layers werelayersdata divided dividedlayer, were wereinto into divided10alldivided into classesthe10 10 classes data intointo classes to layers 10 10simplify to classesclasses simplifyto weresimplifythe toto simplifyworkthesimplifythe work flow. workthe theflow. After flow. workwork After giving After flow.flow. giving giving AfterAfter giving giving class which enhances the chance of landslide occurrences of each data layer (Tabledividedcombineds 4, into5 and 10with 6 ).classes Except respect to simplify to their the relativeworkflow. importance After in landslide occurrence (Sarkar and Kanungo, 2004; thethe Rb Rb data data layer, layer,allall the the data data layers layers were were divided dividedeach into into its 10 relative 10 classes classes importance to to simplify simplify onthe the work 10work-point flow. flow. scale After After(Table giving givings 7, 8 and 9), each unidirectional and reclassified Figure 4. Spatial data layers of (a) Rra and (b) Hi. eacheach its itsrelativeeacheach relative itsits importance relativerelative importance importanceimportance on onthe the 10 on-on10point the-thepoint 10scale10 -scale-pointpoint(Table (Table scalescales 7,(Tables(Table 87, and 8 andss 9 7,7,) , 988each ) and,and each unidirectional 99) ),unidirectional, eacheach unidirectionalunidirectional and and reclassified reclassified andand reclassifiedreclassified Figure 3. Spatial data layers of (a) Rra and (b) Hi. giving each its relative importance on the 10-point scale the Rb dataeacheach layer, its its relativeall relative the data importance importance layers were on on the dividedthe 10 10-point-point into scale 10scale classes(Table(Table tos s7, simplify 7, 8 8and and 9 the)9, )each, Kanungoworkeach unidirectional unidirectionalflow. Afteret al. giving,and and2006 reclassified reclassified ). The intra -class ranking of 14 parameters was done based on a 10-point scale to datadata layerdata layer was layerdatadata was multiplied was layerlayer multiplied multiplied(Table waswas with multipliedmultiplied 3), with the eachwith individualthe unidirectionalthe individual withwith individual thethe PCA individualindividual PCA -andbased PCA -reclassifiedbased- parbased PCAPCA ameterpar- -basedparbasedameter dataameter weight par parlayer weightameterameter weight andwas andfinally weightweight and finally finallysummed andandsummed finallyfinallysummed up by summedupsummed the upby bythe upthe up byby thethe each its relativedata layerimportance was multiplied on the 10 with-point the scale individual(Table sPCA 7, 8- basedand 9) ,par eachameter unidirectional weightmultiplied and andwith finally reclassified the summedindividual up PCA-based by the parameter weight data layer3.3. was Uniform multiplied scaling with and the assigningindividual principal PCAweighted-based component linearparameterweighted combination weight linearmake and methodcombination the finally parameters (WLCM)summed method, up (WLCM)unidirectional by the , in terms of landslide susceptibility, with 10 points assigned to the weightedweightedweighted linear linear combination andlinearcombination finally combination method summed method (WLCM) method up(WLCM) by the (WLCM), ,weighted, linear combination data layerweighted wasweighted multipliedanalysis linear linear combinationwith (PCA)-basedcombination the individual method methodweight PCA (WLCM) (WLCM)for-based constructing, par, ameter DD, weight RD and finally summed up by the method (WLCM), weighted linear combinationand MD models method and (WLCM) landslide, susceptibility zonation class which enhances the chance of landslide occurrences of each data layer (Tables 4, 5 and 6). Except mapping the Rb data layer, all the. . data . . layers . . were divided (3) (3) into (3) (3) 10 classes (3)(3) to simplify the work flow. After giving 𝑛𝑛 𝑛𝑛 After preparing the morphometric data .layers . in GIS (3) (3) 𝑗𝑗푗푗 𝑛𝑛 𝑛𝑛 𝑛𝑛 each WLCM its relative = ∑ 𝑎𝑎𝑎𝑎𝑗𝑗푗푗importance𝑗𝑗푗푗× 𝑤𝑤𝑤𝑤 𝑗𝑗푗푗𝑗𝑗푗푗 on the 10-point scale (Tables 7, 8 and 9), each unidirectional and reclassified platform, DD, RD and MD models𝑛𝑛 𝑛𝑛 . were used for landslide Where Where aj is(3)WhereWhere Where the aj WLCM unidirectional is WLCM theaj ajis =unidirectionalis isWLCM theWLCM the∑ =the ∑unidirectional unidirectionalunidirectional 𝑎𝑎𝑎𝑎and = =𝑎𝑎𝑎𝑎×∑ ∑reclassified𝑤𝑤× and𝑤𝑤𝑤𝑤𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑤𝑤 reclassified ×and × andand𝑤𝑤𝑤𝑤 𝑤𝑤reclassified data𝑤𝑤 reclassifiedreclassified layer data of layerdata datadatathe of layerjthlayer the attribute, ofofjth thethe attribute, jthjth wj attribute,attribute, is thewj is thewjwj isis thethe 𝑗𝑗푗푗𝑗𝑗푗푗 Where aj is the unidirectional and reclassified data layer of the jth attribute, wj is the susceptibilityWhereWhere zonation ajaj is WLCM is theWLCM 𝑛𝑛mappingthe unidirectional unidirectional = =∑ (Table∑ 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 2).× and× 𝑤𝑤andSome𝑤𝑤𝑤𝑤 reclassified𝑤𝑤 reclassified parameters data data layer layerlayer of of the thethe jththjth attribute,attribute, attribute, wjiswj theis is the PCA-based the weight of the 𝑗𝑗푗푗 PCA-basedPCA-basedPCA-based weightPCA-basedPCA-based weight weight of 𝑎𝑎𝑎𝑎the ofdata weight weightofthejth the layer attribute.jth ofof jthattribute. thethe wasattribute. jthjth multiplied attribute. attribute. with the individual PCA𝑗𝑗 -based 𝑤𝑤par𝑤𝑤 ameter weight and finally summed up by the Whereare commonaj is WLCM the because unidirectional = ∑ these𝑎𝑎𝑎𝑎 parameters× and𝑤𝑤𝑤𝑤 reclassified directly ordata indirectly layer of the thjth𝑎𝑎𝑎𝑎 attribute. attribute,𝑎𝑎𝑎𝑎 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 wj is the 𝑗𝑗 𝑗𝑗 𝑗𝑗𝑗𝑗 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤 PCA-basedPCA-based weight weight of of the the jthjth attribute. attribute. represent both𝑎𝑎𝑎𝑎 the𝑎𝑎𝑎𝑎 relief and drainage character, such as Rb 𝑗𝑗 𝑗𝑗 𝑤𝑤𝑤𝑤𝑤𝑤 𝑤𝑤 weighted𝑗𝑗 𝑗𝑗 𝑗𝑗 linear𝑗𝑗𝑗𝑗 combination method (WLCM), PCA-based weight of 𝑎𝑎𝑎𝑎the jth attribute. 𝑗𝑗 𝑤𝑤𝑤𝑤 which indicates the𝑗𝑗 𝑗𝑗drainage characteristics and break of S 3.4. Validation methods characteristics.𝑗𝑗 In the same way, Ri implies the arrangement of a drainage network and ruggedness characteristics of relief. Two well recognised validation methods were employed in this study, i.e. the receiver operating . (3) Table 2. Spatial data layers and composite equations used in characteristics (ROC) curve and FR. Several researchers RD, DD and MD methods. 𝑛𝑛 used the ROC curve for the accuracy assessment𝑗𝑗푗푗 of the landslide susceptibilityWhere map (Buiaj et is WLCMal.,2012; the unidirectional Ozdemir = ∑ and𝑎𝑎𝑎𝑎 × and𝑤𝑤𝑤𝑤 reclassified data layer of the jth attribute, wj is the Method Number of Name of Composite equation Altural, 2013). The ROC curve is the graphical presentation spatial data spatial data ofPCA-based the sensitivity weight on the ofy axisthe andjth 1-specificity attribute. of the layers layers 𝑎𝑎𝑎𝑎 𝑗𝑗 𝑤𝑤𝑤𝑤 DD 9 Dd, Df, Rb, DD Model = (Reclass of Dd × 1.000) + (Reclass of Df × 1.000) + model on the x axis (Mohammady et al., 2012). The area Rr, Din, Dt, (Reclass of Rb × 0.707) + (Reclass of Rr × 0.769) + (Reclass of Din × under curve (AUC) value of the ROC curve varies from Jf, In and Ri 0.638) + (Reclass of Dt × 1.000) + (Reclass of Jf ×.0.971) + (Reclass of 𝑗𝑗 In × 0.891) + (Reclass of Ri × 0.5 - 1.0, where values below 0.5 indicate a random fit 0.967) and values close to 1 mean excellent prediction accuracy. RD 8 S, Rr, Di, E, RD Model = (Reclass of S × 0.811) + (Reclass of Rr × 1.000) + (Reclass Ri, Rb, Rra of Di × 0.588) + (Reclass of Altitude × 0.851) + (Reclass of Ri × 0.890) The range value of AUC was sorted into five classes, and Hi + (Reclass of Rb × 0.831) + (Reclass of Rra × 0.878) + (Reclass of Hi × demonstrating the predictive accuracy of the map: 0.9 - 1.0 0.602) means excellent, 0.8 - 0.9 means very good, 0.7 - 0.8 means MD 13 Dd, Df, Din, MD Model = (Reclass of Dd × 1.000) + (Reclass of Df × 0.981) + Di, Dt, Jf, (Reclass of Din × 0.671) + (Reclass of Di × 0.452) + (Reclass of Dt × good, 0.6 - 0.7 means average and 0.5 - 0.6 means poor In, Rb, Rr, 0.981) + (Reclass of Jf × 0.935) + (Reclass of In × 0.836) + (Reclass of (Yesilnacar, 2005). On the other hand, for the consideration Ri, E, S and Rb × 0.765) + (Reclass of Rr × 0.895) + (Reclass of Ri × 1.000) + Hi (Reclass of Altitude × 0.848) + (Reclass of S × 0.720) + (Reclass of Hyi of the ideal landslide suspectiblity map, the FR value must × 0.527) have a greater value in very high susceptibility zones and it gradually decreased to a very low susceptibility zone (Tsangaratos and Ilia, 2016; Mondal and Mandal, 2017a, Mondal and Mandal, 2017b). 18 HKIE Transactions | Volume 27, Number 1, pp.13–24 TableTable 3 .3. Factors Factors weight, weight, sub class sub ranking class and ranking FR value. and FR value.

Factors Subclasses Subclasses Factor Factor Factor FR Factors Subclasses Subclasses Factor Factor Factor FR rank weight weight weight value rank weight weight weight value of DD of RD of MD of DD of RD of MD method method method method method method Dd 0-1.31 1 1.000 Not 1.000 0.00 Df 0-4.91 1 1.000 Not 0.981 0.00 1.31-2.61 2 used 0.01 4.91-11.05 2 used 0.36 2.61-3.67 3 1.00 11.05-15.96 3 2.17 3.67-4.56 4 1.97 15.96-21.27 4 0.99 4.56-5.39 5 1.38 21.27-27.00 5 0.92 5.39-6.27 6 1.29 27.00-33.54 6 1.87 6.27-7.28 7 1.46 33.54-40.90 7 4.37 7.28-8.58 8 1.85 40.90-52.35 8 0.69 8.58-10.60 9 0.87 52.35-70.35 9 2.76 10.60-15.10 10 2.94 70.35-104.30 10 4.59 Din 0-1.51 1 0.638 Not 0.671 0.00 Di .06-.13 1 Not 0.588 0.452 0.10 1.51-3.02 2 used 0.46 .13-.17 2 used 1.43 3.02-4.16 3 1.74 .17-.22 3 0.22 4.06-5.67 4 1.60 .22-.26 4 0.29 5.67-8.12 5 1.35 .26-.31 5 0.45 8.12-12.10 6 0.58 .31-.37 6 0.40 12.10-17.96 7 0.23 .37-.42 7 1.32 17.96-25.51 8 0.00 .42-.47 8 2.92 25.51-34.58 9 0.00 .47-.52 9 5.29 34.58-48.17 10 0.00 .52-.59 10 0.89 Dt 0-1.23 1 1.000 Not 0.981 0.00 Jf 0-2.38 1 0.971 Not 0.935 0.00 1.23-2.76 2 used 0.36 2.38-5.82 2 used 0.21 2.76-3.99 3 2.17 5.82-9.00 3 2.63 3.99-5.32 4 0.99 9.00-12.45 4 0.80 5.32-6.75 5 0.92 12.45-16.42 5 0.81 6.75-8.38 6 1.87 16.42-20.92 6 3.38 8.38-10.23 7 4.37 20.92-26.74 7 1.84 10.23-13.10 8 0.69 26.74-34.95 8 1.19 13.10-17.59 9 2.76 34.95-46.33 9 3.41 17.59-26.07 10 4.59 46.33-67.49 10 4.52 In 0-30.89 1 0.891 Not 0.836 0.01 Rb 1.50-2.56 1 0.707 0.831 0.765 0.00 30.89-80.32 2 used 1.76 2.56-3.25 2 0.18 80.32-148.28 3 0.92 3.25-3.71 3 1.63 148.28-234.78 4 1.54 3.71-4.01 5 4.14 234.78-339.81 5 3.13 4.01-4.22 6 0.82 339.81-481.92 6 1.63 4.22-4.52 8 0.38 481.92-667.27 7 2.09 4.52-4.98 9 0.42 667.27-883.51 8 3.11 4.98-5.66 10 0.31 883.51-1130.66 9 5.47 1130.66-1575.50 10 3.63 Rr 5.19-51.70 1 0.769 1.000 0.895 0.00 Ri 0-.34 1 0.967 0.890 1.000 0.00 51.70-127.00 2 0.00 .34-.93 2 0.04 127.00-215.00 3 0.08 .93-1.42 3 0.21 215.00-292.00 4 0.25 1.42-1.83 4 0.57 292.00-346.00 5 0.88 1.83-2.22 5 1.78 346.00-393.00 6 0.74 2.22-2.63 6 3.53 393.00-434.00 7 0.51 2.63-3.12 7 1.56 434.00-475.00 8 1.43 3.12-3.74 8 2.85 475.00-532.00 9 1.69 3.74-4.48 9 1.38 532.00-662.00 10 10.33 4.48-6.55 10 0.24 E 106-298 1 Not 0.851 0.848 0.00 S 0-5.03 1 Not 0.811 0.720 0.04 298-539 2 used 0.59 5.03-11.30 2 used 0.37 539-792 3 1.99 11.30-16.80 3 0.73 792-1030 4 2.20 16.80-21.90 4 0.98 1030-1240 5 2.46 21.90-26.60 5 0.89 1240-1440 6 0.92 26.60-31.10 6 1.13 1440-1650 7 0.49 31.10-35.50 7 1.48 1650-1880 8 0.16 35.50-40.60 8 2.22 1880-2130 9 0.47 40.60-47.00 9 4.42 2130-2600 10 2.67 47.00-71.10 10 13.57 Rra .02 1 Not 0.878 Not 0.00 Hi .18-.31 1 Not 0.602 0.527 0.35 .02-.11 2 used used 0.23 .31-.37 2 used 0.39 .11-.14 3 0.15 .37-.41 3 0.41 .14-.18 4 2.37 .41-.44 4 0.55 .18-.20 5 0.08 .44-.48 5 0.87 .20-.21 6 2.01 .48-.51 6 0.54 .21-.22 7 3.09 .51-.54 7 0.97 .22-.23 8 0.01 .54-.57 8 1.47 .23-.27 9 0.10 .57-.62 9 2.98 .27-.30 10 0.13 .62-.75 10 3.59

19 HKIE Transactions | Volume 27, Number 1, pp.13–24 S MONDAL AND S MANDAL

4. Results and discussion landslide occurrences was found in the class of 0 - 1.23 (0.00) (Table 3). For the Jf, the class of 46.33 - 67.49 has 4.1. Relationship between morphometric parameters and the highest FR value at 4.52, indicating a strong association landslide susceptibility with S failure events. Similarly, the class of 0 - 2.38 has the lowest impact on landslide occurrences with a FR value of The relationship between morphometric parameters 0.00 accounting 0.14% of the total landslide area (Table 3). and landslide susceptibility has been done using FR value. In the case of In, the class of 883.51 - 1130.66 acquired highest degree of FR value of 5.47. The classes of 1130.66 4.1.1. Dd, Df, Rb, Rr and their landslide potentiality - 1575.50 (3.63), 234.78 - 339.81 (3.13), 667.27 - 883.51 (3.11) and 481.92 - 667.27 also have the high probability of It can be observed that Dd in the range of 10.60 - landslide occurrences. 0 - 30.89 class has the least degree 15.10 has the highest FR value of 2.94, representing the (0.01) of association with landslides (Table 3). greater probability of movement of materials down the In general, the chance of landslide occurrences reduces slope. Also, the Dd classes of 3.67 - 4.56 (1.97), 7.28 - 8.58 with increasing Din. On the other hand, Dt, Jf and In were (1.85), 6.27 - 7.28 (1.46), 4.56 - 5.39 (1.38) and 5.39 - 6.27 positively linked with S failure. (1.29) have a strong relationship with landslides events. On the other hand, the probability of landslide decreased 4.1.3. Ri, Di, altitude, S and their role in S failure in the Dd class of 0 - 1.31 (0.00) (Table 3). In the Df map, the class of 70.35 - 104.30 had the maximum probability For Ri, the highest FR value of 3.53 was obtained in of S failure events (4.59), followed by 52.35 - 70.35 (2.76), the class of 2.22 - 2.63, which shows greater potentiality 11.05 - 15.96 (2.17), 27.00 - 33.54 (1.87), 15.96 - 21.27 for S failures. The classes of 3.12 - 3.74 (2.85), 1.83 - 2.22 (0.99), 21.27 - 27.00 (0.92), 40.90 - 52.35 (0.69), 4.91 - (1.78), 2.63 - 3.12 (1.56) and 3.74 - 4.48 (1.38) also have a 11.05 (0.36) and 0 - 4.91 (0.00), respectively (Table 3). A stronger relationship with landslides. The least probability Rb of 3.71 - 4.01 has highest FR value of 4.14, representing of landslide occurrences was found in the class of 0 - .34 the maximum chance of landslide occurrences in future (0.00) (Table 3). In the case of Di, the class of .47 - .52 containing 42.10% of landslide pixels to the total pixels acquired the highest FR value of 5.29, covering 33.71% of followed by classes of 3.25 - 3.71 (1.63), 4.01 - 4.22 (0.82), the landslide area to the total landslide area, followed by 4.52 - 4.98 (0.42), etc., respectively. On the other hand, the .42 - .47 (2.92), .13 - .17 (1.43), .37 - .42 (1.32), .52 - .59 lowest probability of landslide occurrences was found in (0.89), .26 - .31 (0.45), .31 - .37 (0.40), .22 - .26 (0.29), .17- the class of 1.50 - 2.56 (0.00), having 0.05% of landslide .22 (0.22) and .06 - .13 (0.10), respectively (Table 3). It can pixels (Table 3). In the case of relative relief, the class of be observed that elevation in the range of 2130 m - 2600 m 532.00 - 662.00 registered a maximum FR value of 10.33, has the highest FR value of 2.67 and represents a greater representing a greater probability of landslide phenomena probability of movement of materials down the S. Also, the over 34.51% of landslide area, followed by 475.00 - 532.00 classes of 1030 m - 1240 m (2.46), 792 m - 1030 m (2.20) (1.69), 434.00 - 475.00 (1.43), etc., respectively. The and 539 m - 792 m (1.99) have a stronger relationship with lowest landslide probability was found in the classes of landslides events. On the other hand, the probability of 5.19 - 51.70 (0.00) and 51.70 - 127.00 (0.00) (Table 3). landslides was lessened in the elevation class of 106 m - The trend line of FR value reveals that Dd, Df and Rr were 298 m (0.00) (Table 3). S angle of 47.00°- 71.10° has the positively correlated with S failure, which means that S highest FR value of 13.57, representing a maximum chance instability increases as this value increases. On the other of landslide occurrences in future, containing 13.44% of hand, there was negative relation between Rb and landslide landslide pixels to the total pixels followed by S ranges occurrences. of 40.60°- 47.00° (4.42), 35.50° - 40.60° (2.22), 22.03° - 25.03° (1.48), etc., respectively. On the other hand, the 4.1.2. Din, Dt, Jf, In and their role in S instability lowest probability of landslide occurrences was found in the class of 0° - 5.03° (0.04) S range (Table 3). In general, The Din class of 3.02 - 4.16 has the larger FR value all four factors were positively associated with S instability. of 1.74, which indicates a maximum S instability condition If the value of the factors increases, the chance of landslide in future, with 57.26% of landslides. Similarly, the classes also increases. of 4.06 - 5.67 (1.60) and 5.67 - 8.12 (1.35) have the high probability of materials sliding. On the other hand, the 4.1.4. Rra, Hi and their role in landslide occurrences classes of 0 - 1.51 (0.00), 17.96 - 25.51 (0.00), 25.51 - 34.58 (0.00) and 34.58 - 48.17 (0.00) were characterised In the Rra layer, the highest FR value (3.09) was by a minimum probability of S failure events (Table 3). observed in the class of .21 - .22, indicating greater The highest probability was found in the Dt class of 17.59 probability of landslide occurrences. The other classes - 26.07, having an FR value of 4.59 followed by 8.38 - of .14 - .18 (2.37) and .20 - .21 (2.01) also have a high 10.23 (4.37), 13.10 - 17.59 (2.76), 2.76 - 3.99 (2.17), 6.75 probability of S failure. On the other hand, the lowest - 8.38 (1.87) etc., respectively. The lowest probability of probability of landslides was observed in the class of .02

20 HKIE Transactions | Volume 27, Number 1, pp.13–24 with FR value of 0.00 (Table 3). The FR value of the Hi showed that the class of .62 - .75 has the most effect on the incidence of landslides with an FR value of 3.59, followed by the classes of .57 - .62 (2.98), .54 - .57 (1.47), .51 - .54 (0.97), .44 - .48 (0.87), .41 - .44 (0.55), .48 - .51 (0.54), .37 - .41 (0.41), .31 - .37 (0.39) and .18 - .31 (0.35), respectively (Table 3). Generally, the chance of landslide occurrences increases with the increasing value of the Rra and Hi.

4.2. Landslide susceptibility mapping using DD, RD and Figure 7. Landslide susceptibility zonation mapping using (a) DD; (b) RD; and (c) MD methods. MD models Figure 4. Landslide suspectibility zonation mapping using (a) DD; (b) RD; and (c) MD metway. Landslide susceptibility maps based on the three different models were prepared using PCA-based weight Table 4. Statistics of different susceptibility classes of the and their composite equations were presented in Table 2. prepared landslide susceptibility maps using RD, DD and Table 11. StatisticsMD methods. of different susceptibility classes of the prepared landslide susceptibility maps using RD, DD and MD The DD model was prepared to identify highlymethods. Very low Low Moderate High Very diversified drainage area where geomorphic process is Landslide susceptibility zones high actively engaged in landform modification, whereas the RD Absolute 381311 267158 524092 371199 138240 Total pixels Percentage 22.67 15.88 31.16 22.07 8.22 model was constructed to find out probable areas where Using Landslide Absolute 3 195 1056 3000 1233 DD landslide might strike based on the relief configuration occurrence 0.05 3.56 19.25 54.67 22.47 method Percentage parameters. Finally, the MD model was prepared by using pixels FR value 0.00 0.22 0.62 2.48 2.73 drainage and relief parameters. Total pixels Absolute 294320 161612 371767 560142 294159 In this study, the values of the landslide susceptibility Using Percentage 17.50 9.61 22.10 33.30 17.49 RD Landslide Absolute 1 49 219 2087 3132 method occurrence map using the DD model varied from 7.94 - 88.46, with Percentage 0.02 0.89 3.98 38.03 57.08 pixels the higher value indicating greater possibility of landslide FR value 0.00 0.09 0.18 1.14 3.26 Absolute 379664 254090 537934 376104 134208 occurrences and vice versa. For visual interpretation, the Total pixels Percentage 22.57 15.11 31.98 22.36 7.98 Using continuous landslide susceptibility value was classified Landslide Absolute 3 124 1121 3000 1233 MD occurrence 0.05 2.26 20.43 54.67 22.47 into five different categorical landslides susceptibility method Percentage pixels zones using the natural breaks classification method, i.e. FR)value 0.00 0.15 0.64 2.45 2.82 very low (7.94 - 16.47), low (16.47 - 27.52), moderate (27.52 - 36.68), high (36.68 - 47.73) and very high (47.73 4.3. Validation and accuracy assessment - 88.46) (Figure 4(a)), covering an area of 22.67%, 15.88%, 31.16%, 22.07% and 8.22%, respectively, of the The AUC value of the ROC curve indicated that the total area of the Balason river basin (Table 4). It was seen prediction accuracy of the DD, RD and MD models was that 77.14% of landslide pixels were found in very high 71.4%, 73.9% and 76.3%, respectively (Figure 5(a)). The and high susceptibility zones, while the minimum figure FR plot of the aforementioned three models indicated that lies in the class of very low susceptibility zones (0.05%) the chance of landslide occurrences gradually increases (Table 4). Similarly, both the landslide susceptibility from very low to very high susceptibility class (Figure values of RD and MD models were divided into five 5(b) and Table 4). The two-validation procedures showed different classes using the natural breaks classification that the DD, RR and MD models yield reasonable method (Figures 4(b) and 4(c)). The very low, low, prediction accuracy and can be used for identifying the moderate, high and very high susceptibility zones covered role of drainage, relief and morphometric parameters in an area of 17.50%, 9.61%, 22.10%, 33.30% and 17.49%, landslide susceptibility assessment. It can be mentioned respectively, using the RD model. In case of the MD here that the findings of Mondal and Mandal (2017a) model, these zones covered 22.57%, 15.11%, 31.98%, and Mondal and Mandal (2017b) and their work on 22.36% and 7.98%, respectively (Table 4). The proportion the Balason river basin suggests that the prediction of landslide pixels to total landslide pixels decreased accuracy of the landslide susceptibility maps was higher from very high to very low susceptibility zones using RD using FR and logistic regression (LR) models (94.20% model. About 70% of landslide pixels were confined to and 96.10%, respectively). They incorporated several the very low to moderate susceptibility classes using the geomorphological, lithological, hydrological, protective, MD model (Table 4). anthropogenic and triggering factors in the preparation of landslide susceptibility maps.

21 HKIE Transactions | Volume 27, Number 1, pp.13–24 Note on Contributors

Mr Subrata Mondal is an Assistant Professor of Geography at A.B.N Seal College, Cooch Behar, India from 21.09.2019. He completed his B.A and M.A. degrees from the University of Gour S MONDAL AND S MANDAL Banga, West Bengal, India. He is the author of three research articles published in journals by Springer and Taylor & Francis, and several research articles submitted to reputable international journals. He is quite efficient in Geoinformatics. He completed his M.Phil degree, zones of the Balasonentitled river “Slope basin instability in Darjeeling analysis andHimalaya, delineation of potential landslide (a) (b susceptibility zones of the Balason river basin in Darjeeling Himalaya, 3.5 ) West Bengal” fromWest the Bengal”University from theof UniversityGour Banga, of Gour West Banga, West Bengal, India 3 Bengal, India in 2017.in 2017.At present, At present, he ishe pursuing is pursuing a aPh.D. Ph.D. HeHe participated in several 2.5 national and international seminars and conferences and presented 2 participated in several national and international seminars 1.5 research papers on landslides in Darjeeling Himalaya. FR value 1 and conferences and presented research papers on landslides 0.5 in Darjeeling Himalaya. 0 Very low Low Moderate High Very high Landslide susceptibility zones Prof Sujit Mandal is a Professor FR value of DD model FR value of RD model Prof Sujit Mandal is a Professor of Geography at Diamond Harbour FR value of MD model ofWomen’s Geography University, at DiamondWest Bengal, HarbourIndia since 2005. He completed Figure 8. Validation of the landslide susceptibility maps using landslide locations (a) ROC curve and (b) FR plot. Women’sB.Sc. and University,M.Sc. degrees West from CalcuttaBengal, University, India as well as his Ph.D. thesis, entitled “An analysis of Slope instability in the Shivkhola Figure 5. Validation of the landslide susceptibility maps sinceWatershed 2005. of DarjeelingHe completed Himalaya: B.Sc. A scientific and approach towards the using landslide locations (a) ROC curve and (b) FR plot. M.Sc.management degrees of fromland, waterCalcutta and soil”.University, During his teaching career, he participated and presented papers in 30 national and international asseminars well asand his conferences Ph.D. onthesis, several entitled aspects of landslides. He is the 4.4. Limitations of the prepared landslide susceptibility “Anauthor analysis of 40 ofresearch Slope articles instability published in thein reputable national and Shivkholainternational Watershedjournals, of which of Darjeeling12 have been published by Springer maps and two by Taylor & Francis. He is the first author of the book Semi- Himalaya: A scientificquantitative approach approaches towards for the Landslide management Susceptibility Assessment and Predictionof land,, published water andby Springer soil”. PublishingDuring his House. teaching He is also career, the principal he investigator of one The prepared landslide susceptibility maps do not minor research project sponsored by university grants commission (UGC) and one major research define the chronological characteristics of S failure events projectparticipated sponsored by and Indian presented Council of Socialpapers Science in 30 Research national (ICSSR) and and detailed risk zones. The susceptibility maps should be international seminars and conferences on several aspects of viewed at the initial landslide mapping. This study does landslides. He is the author of 40 research articles published not delineate the areas which become unstable during an Referencesin reputable national and international journals, of which earthquake. In addition, the limitations of the study also 12 have been published by Springer and two by Taylor & include an incomplete landslide inventory map due to the [1]Francis. Akgun A, He Sezer is the EA, first Nefesliogl author HA, of theGokceoglu book Semi-quantitativeC and Pradhan B (2012). An easy-to-use MATLABapproaches program for (MamLand) Landslide for Susceptibility the assessment of Assessment landslide susceptibility and using a Mamdani restriction of air photos and lack of information about past fuzzy algorithm. Comput Geosci, 38, pp. 23-34. landslides, generation of errors in the model due to scale Prediction, published by Springer Publishing House. He is and mapping roughness of the landslide causative factors. also the principal investigator of one minor research project [2]sponsored Althuwaynee by OF, university Pradhan B, grantsPark H andcommission Lee JH (2014). and A one novel major ensemble bivariate statistical 5. Conclusion evidentialresearch belief project function sponsored with knowledge-based by Indian analyticalCouncil hierarchyof Social process and multivariate statisticalScience logistic Research. regression for landslide susceptibility mapping. Catena, 114, pp. 21-36. It can be concluded that each model can be treated as unique in terms of terrain processes and forms, and [3]References Amani M and Safaviyan A (2015). Sub-basins prioritization using morphometric analysis- can be used to identify the role of drainage and relief in S remote sensing technique and GIS-Golestan-Iran. International Letters of Natural Sciences, 38, pp. 56-65. failures individually. The DD model on its own predicts [1] Akgun A, Sezer EA, Nefesliogl HA, Gokceoglu C 71.40% of landslides, whereas the RD model predicts and Pradhan B (2012). An easy-to-use MATLAB 73.9% of landslides. Landslide prediction is slightly better program (MamLand) for the assessment of landslide when DD and RD are examined together. The MD model susceptibility using a Mamdani fuzzy algorithm. (a combination of RD and DD) has an area ratio .763 and Comput Geosci, 38, pp. 23-34. predicts 76.3% of landslides. However, one must be careful [2] Althuwaynee OF, Pradhan B, Park H and Lee while using these geomorphic diversity models for specific JH (2014). A novel ensemble bivariate statistical site development because of the scale of analysis where evidential belief function with knowledge-based other S factors need to be considered. analytical hierarchy process and multivariate statistical Note on Contributors logistic regression for landslide susceptibility Note on Contributors mapping. Catena, 114, pp. 21-36. [3] Amani M and Safaviyan A (2015). Sub-basins Mr Subrata Mondal is an Assistant Professor of Geography at MrA.B.N Subrata Seal MondalCollege, hasCooch been Behar, an Assistant India from 21.09.2019.prioritization He using morphometric analysis- Professorcompleted ofhis Geography B.A and M.A. at A.B.Ndegrees fromSeal the Universityremote of Gour sensing technique and GIS-Golestan-Iran. Banga, West Bengal, India. He is the author of three researchInternational articles Letters of Natural Sciences, 38, pp. 56- College,published Cooch in journals Behar, by IndiaSpringer since and 2019.Taylor & Francis, and several Heresearch completed articles his submitted B.A. and to M.A.reputable degrees international journals.65. He is fromquite the efficient University in Geoinformatics. of Gour Banga, He Westcompleted his[4] M.Phil Anbalagan degree, R (1992). Landslide hazard evaluation entitled “Slope instability analysis and delineation of potentialand landslide zonation mapping in mountainous terrain. Bengal,susceptibility India. zones He ofis the the Balason author river of basinfive in Darjeeling Himalaya, researchWest Bengal” articles from published the University in journals of Gour byBanga, West Bengal,Engineering India Geology, 32, pp. 269-277. Springerin 2017. and At present,Taylor &he Francis,is pursuing and a severalPh.D. He participated[5] Avinash in several K, Deepika B and Jayappa KS (2014). national and international seminars and conferences andBasin presented geomorphology and drainage morphometry research articles submittedresearch to papers reputable on landslides international in Darjeeling journals. Himalaya. He is the author of two books, published by Springer. He parameters used as indicators for groundwater completed his M.Phil. degree, entitled “Slope instability prospect: Insight from geographical information analysis and delineation of potential landslide susceptibility system (GIS) Technique. Journal of Earth Science, 25(6), pp. 1018-1032. Prof Sujit Mandal is a Professor of Geography at Diamond Harbour Women’s University, West Bengal, India since 2005. He completed B.Sc. and M.Sc. degrees from Calcutta University, as well as his Ph.D. 22 thesis, entitled “An analysis of Slope instability in the Shivkhola Watershed of DarjeelingHKIE Himalaya: Transactions A scientific | Volume approach 27, Number towards 1, pp. the13–24 management of land, water and soil”. During his teaching career, he participated and presented papers in 30 national and international seminars and conferences on several aspects of landslides. He is the author of 40 research articles published in reputable national and international journals, of which 12 have been published by Springer and two by Taylor & Francis. He is the first author of the book Semi- quantitative approaches for Landslide Susceptibility Assessment and Prediction, published by Springer Publishing House. He is also the principal investigator of one minor research project sponsored by university grants commission (UGC) and one major research project sponsored by Indian Council of Social Science Research (ICSSR)

References

[1] Akgun A, Sezer EA, Nefesliogl HA, Gokceoglu C and Pradhan B (2012). An easy-to-use MATLAB program (MamLand) for the assessment of landslide susceptibility using a Mamdani fuzzy algorithm. Comput Geosci, 38, pp. 23-34.

[2] Althuwaynee OF, Pradhan B, Park H and Lee JH (2014). A novel ensemble bivariate statistical evidential belief function with knowledge-based analytical hierarchy process and multivariate statistical logistic regression for landslide susceptibility mapping. Catena, 114, pp. 21-36.

[3] Amani M and Safaviyan A (2015). Sub-basins prioritization using morphometric analysis- remote sensing technique and GIS-Golestan-Iran. International Letters of Natural Sciences, 38, pp. 56-65. [6] Ayalew L, Yamagishi H, Marui H and Kanno T (2005). [18] Horton RE (1945). Erosional development of streams Landslides in Sado island of Japan: part II, GIS-based and their drainage basins, a hydrophysical approach susceptibility mapping with comparisons of results to quantitative morphology. Geological Society of from two methods and verifications. Engineering American Bulletin, 56, pp. 275-370. Geology, 81, pp. 432-445. [19] Jeganathan C and Chauniyal DD (2000). An evidential [7] Bui DT, Pradhan B, Lofman O, Revhaug I and Dick weighted approach for landslide hazard zonation from OB (2012). Spatial prediction of landslide hazards geo-environmental characterization: A case study of in Hoa Binh province (Vietnam): A comparative Kelani area. Current Science, 79(2), pp. 238-243. assessment of the efficacy of evidential belief [20] Karim S, Jalileddin S and Ali MT (2011). Zoning functions and fuzzy logic models. Catena, 96, pp. landslide by use of frequency ratio method (case 28-40. study: Deylaman Region). Middle-East Journal of [8] Chau KT, Sze YL, Fung MK, Wong WY, Fong EL Scientific Research, 9(5), pp. 578-583. and Chan LCP (2004). Landslide hazard analysis [21] Khatun S and Pal S (2017). Categorization of for Hong Kong using landslide inventory and GIS. morphometric surface through morphometric diversity Computers & Geosciences, 30, pp. 429-443. analysis in Kushkarani River Basin of Eastern India. [9] Choi J, Oh HJ, Lee HJ, Lee C and Lee S (2011). Asian Journal of Physical and Chemical Sciences, Combining landslide susceptibility maps obtained 2(1), pp. 1-19. from frequency ratio logistic regression and artificial [22] Lama IL (2003). Study of the environmental neural network models using aster images and GIS. geomorphology in the Balason Basin. Ph.O. North Engineering Geology, 124, pp. 12–23. Bengal University. [10] Devkota KC, Regmi AD, Pourghasemi HR, Yoshida [23] Lee S and Pradhan B (2007). Landslide hazard K, Pradhan B, Ryu IC, Dhital MR and Althuwaynee mapping at Selangor Malaysia using frequency ratio OF (2013). Landslide susceptibility mapping and logistic regression models. Landslides, 4, pp. using certainty factor index of entropy and logistic 33-41. regression models in GIS and their comparison [24] Lee S and Sambath T (2006). Landslide susceptibility at Mugling-Narayanghat road section in Nepal mapping in the Damrei Romel area Cambodia using Himalaya. Natural Hazards, 65, pp. 135-165. frequency ratio and logistic regression models. [11] Faniran A (1968). The index of drainage intensity - A Environmental Geology, 50, pp. 847-855. provisional new drainage factor. Australian Journal of [25] Lee S and Talib JA (2005). Probabilistic landslide Science, 31, pp. 328-330. susceptibility and factor effect analysis. Environmental [12] Farhan Y, Anbar A, Enaba O and Al-Shaikh N (2015). Geology, 47, pp. 982-990. Quantitative analysis of geomorphometric parameters [26] Magesh NS, Jitheshlal K, Chandrasekar N and Jini of Wasi Kerak, Jordan, using remote sensing and GIS. K (2012). GIS based morphometric evaluation of Journal of Water Resource and Protection, 7, pp. 456- Chimini and Mupily watersheds, parts of Western 475. Ghats, Thrissur district, Kerala, India. Earth Science [13] Gajbhiye S, Mishra SK and Pandey A (2014). Informatics, 5, pp. 111-121. Prioritizing erosion-prone area through morphometric [27] Majtan S, Omura H and Morita K (2002). Fractal analysis: An RS and GIS perspective. Applied Water dimension as an indicator of probability for landslides Science, 4(1), pp. 51-61. in North Matsuura Japan. Geograficky Casopis, 54, [14] Ghosh D (2015). Landslide susceptibility analysis pp. 5-19. from morphometric parameter analysis of riyong [28] Mandal B and Mandal S (2016). Assessment of khola basin, west sikkim, india: a geospatial approach. mountain slope instability in the Lish river basin of International Journal of Geology, 5(1), pp. 54-65. Eastern Darjeeling Himalaya using frequency ratio [15] Ghosh S, Carranza EJM, van Westen CJ, Jetten VG model (FRM). Earth Systems and Environment, 2, and Bhattacharya DN (2011). Selecting and weighting p. 121. spatial predictors for empirical modeling of landslide [29] Mandal S and Mandal K (2017). Bivariate statistical susceptibility in the Darjeeling Himalayas (India). index for landslide susceptibility mapping in the Geomorphology, 131(1–2), pp. 35-56. Rorachu river basin of eastern Sikkim Himalaya, [16] Gopal KG and Saha S (2015). Identification of soil India. Spatial Information Research, 26 (1), pp. 59-75. erosion susceptible areas in Hinglo river basin, [30] Mandal S and Mandal K (2018). Modeling and Eastern India based on geostatistics. Universal mapping landslide susceptibility zones using GIS Journal of Environmental Research and Technology, based multivariate binary logistic regression (LR) 5(3), pp. 152-164. model in the Rorachu river basin of eastern Sikkim [17] Gupta RP and Joshi BC (1990). Landslide hazard Himalaya, India. Modeling Earth Systems and zonation using the GIS approach—A case Study from Environment, 4 (1), pp. 69-88. the Ramganga Catchment Himalayas. Engineering Geology, 28, pp. 119–131.

23 HKIE Transactions | Volume 27, Number 1, pp.13–24 S MONDAL AND S MANDAL

[31] Miller JR, Ritter DF and Kochel RC (1990). [43] Pradhan B and Lee S (2010b). Landslide susceptibility Morphometric assessment of lithologic controls on assessment and factor effect analysis: Backpropagation drainage basin evolution in the Crawford Upland, artificial neural networks and their comparison with South-Central Indiana. America Journel of Science, frequency ratio and bivariate logistic regression 290, pp. 569-599. modelling. Environmental Modelling & Software, [32] Miller VC (1953). A quantitative geomorphic study of 25(6), pp. 747-759. drainage basin characteristics in the Clinch mountain [44] Rai PK, Mishra S, Ahmad A and Mohan K (2014). A area. Virginia and Tennessee: Deptt. of Geology, GIS-based approach in drainage morphometric analysis Columbia University. of Kanhar River Basin, India. Applied Water Science, [33] Mohammady M, Pourghasemi HR and Pradhan B 7, pp. 217-232. (2012). Landslide susceptibility mapping at Golestan [45] Rastogi RA and Sharma TC (1976). Quantitative Province, Iran: A comparison between frequency ratio, analysis of drainage basin characteristics. Journal of dempster-shafer, and weights-of-evidence models. Soil and Water Conservation in India, 26 (1&4), pp.18- Journal of Asian Earth Sciences, 61, pp. 221-236. 25. [34] Mondal S and Mandal S (2017a). Application of [46] Sarkar S and Kanungo DP (2004). An integrated frequency ratio (FR) model in spatial prediction of approach for landslide susceptibility mapping using landslides in the Balason river basin, Darjeeling remote sensing and GIS. Photogrammetric Engineering Himalaya. Spatial Information Research, 25, pp. 337- & Remote Sensing, 70(5), pp. 617-625. 350. [47] Schumm SA (1956). Evolution of drainage systems [35] Mondal S and Mandal S (2017b). RS & GIS-based and slopes in badlands at Perth Amboy, New Jersey, landslide susceptibility mapping of the Balason River Geological Society of American Bulletin, 67(5), p. 597. basin, Darjeeling Himalaya, using logistic regression [48] Shrestha S, Kang T and Suwal MS (2017). An (LR) model. Georisk: Assessment and Management of ensemble model for co-seismic landslide susceptibility Risk for Rngineered Systems and Geohazards. 12(1), using GIS and random forest method. International pp. 29-44. Journal of Geo-Information, 6, p. 365. [36] Nautiyal MD (1994). Morphometric analysis of [49] Song Y, Gong J, Gao S, Wang D, Cui T, Li Y and Wei drainage basin, district Dehradun, Uttar Pradesh. B (2012). Susceptibility assessment of earthquake Journal of the Indian Society of Remote Sensing, 22, induced landslides using Bayesian network: A case pp. 252-262. study in Beichuan China. Computers & Geosciences, [37] Nefeslioglu HA, Sezer E, Gokceoglu C, Bozkir AS 42, pp.189-199. and Duman TY (2010). Assessment of landslide [50] Spiker EC and Gori PL (2000). National landslide susceptibility by decision trees in the metropolitan hazards mitigation strategy: A framework for loss area of Istanbul Turkey. Mathematical Problems in reduction. US: Department of the interior, U.S. Engineering, Article ID: 901095. Geological Survey, p. 59. [38] Nidhi K, Chowdary VM, Tiwari KN, Shinde V and [51] Strahler AN (1952). Hypsometric (area-altitude) Dadhwal VK (2016). Assessment of surface water analysis of erosional topography. Geological Society of potential using morphometry and curve number-based American Bulletin, 63, pp. 117-142.[65] approaches. Geocarto International, 32(11), pp. 1206- [52] Strahler AN (1964). Handbook of Applied Hydrology. 1228. United States: McGraw-Hill. pp. 4.39-4.76. [39] Nir D (1957). The ratio of relative and absolute [53] Thakkar AK and Dhiman SD (2007). Morphometric altitude of Mt.Camel. Geographical Review, 47, pp. analysis and prioritization of miniwatersheds in 564-569. Mohr watersheds, Gujarat using remote sensing and [40] Ozdemir A and Altural T (2013). A comparative GIS techniques. Journal of Indian Society of Remote study of frequency ratio, weights of evidence and Sensing, 33(1), pp. 25-38. logistic regression methods for landslide susceptibility [54] Tsangaratos P and Ilia I (2016). Comparison of a mapping: Sultan Mountains, SW Turkey. Journal of logistic regression and native bayes classifier in Asian Earth Sciences, 64, pp. 180-197. landslide susceptibility assessments: The influence of [41] Pradhan B (2010). Use of GIS-based fuzzy logic models complexity and training dataset size. Catena, relations and its cross application to produce landslide 145, pp. 164-179. susceptibility maps in three test areas in Malaysia. [55] Xu C, Dai F, Xu X and Lee YH (2012). GIS-based Environmental Earth Sciences, 63, pp. 329-349. support vector machine modeling of earthquake- [42] Pradhan B and Lee S (2010a). Delineation of triggered landslide susceptibility in the Jianjiang River landslide hazard areas on Penang Island, Malaysia, by watershed China. Geomorphology, 145-146, pp. 70-80. using frequency ratio, logistic regression, and artificial [56] Yesilnacar EK (2005). The application of neural network models. Environmental Earth Science, computational intelligence to landslide susceptibility 60(5), pp. 1037-1054. mapping in Turkey. Ph.D. University of Melbourne. [57] Yin K J and Yin TZ (1988). Statistical prediction model for slope instability of metamorphosed rock. In: Proceedings of 5th International Symposium on Landslides. Lausanne: Balkema 2, pp. 1269-1272.

24 HKIE Transactions | Volume 27, Number 1, pp.13–24