Event-based landslide susceptibility mapping using weights of evidence (WoE) and modified frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh

Shamima Ferdousi Sifa; Tonoy Mahmud; Maria Abdullah Tarin; Dewan Mohammad Enamul Haque

doi:10.1080/24749508.2019.1619222

Event-based landslide susceptibility mapping using weights of evidence (WoE) and modified frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh

Sifa, Shamima Ferdousi; Mahmud, Tonoy; Tarin, Maria Abdullah; Haque, Dewan Mohammad Enamul 2020-07-02 00:00:00 GEOLOGY, ECOLOGY, AND LANDSCAPES 2020, VOL. 4, NO. 3, 222–235 INWASCON https://doi.org/10.1080/24749508.2019.1619222 RESEARCH ARTICLE Event-based landslide susceptibility mapping using weights of evidence (WoE) and modiﬁed frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh Shamima Ferdousi Sifa, Tonoy Mahmud, Maria Abdullah Tarin and Dewan Mohammad Enamul Haque Department of Disaster Science and Management, Faculty of Earth and Environmental Sciences, University of Dhaka, Dhaka, Bangladesh ABSTRACT ARTICLE HISTORY Received 28 December 2018 Landslide hazard of 2017 in Rangamati district had devastating impacts on development, Accepted 11 May 2019 thereby making landslide susceptibility mapping a prerequisite for disaster risk management. This study aims to map the future landslide susceptible areas by overlaying the landslide KEYWORDS inventory of 2017 with causative factor maps using WoE and MFR and compare their results Landslide; susceptibility; to determine that statistical model describes the susceptibility of the landslide occurrence inventory; weights of better than the other. The analysis shows that although both models deﬁne the spatial evidence (WoE); modiﬁed relationship of past landslides with the triggering factors in a same way but in case of frequency ratio (MFR) mapping, MFR had overestimated the high and low susceptible areas and underestimated the moderate susceptible areas than WoE. When validated from success rate curve by plotting the percentage of landslide susceptibility index rank against the percentage of cumulative landslide occurrence, it shows that the WoE model describes the landslides better than the MFR model. About 20% of the high susceptible areas include 85% of the total landslide area in case of the WoE model but the MFR model includes only 20%. On the other hand, the WoE model describes that 30% highly susceptible area covers more than 99% of the total landslide area while MFR deﬁnes only 78%. 1. Introduction appropriate mitigation measures in the landslide suscep- tible areas identiﬁed from susceptibility mapping, which Extreme hydro-meteorological conditions, triggered by will ensure eﬀective and eﬃcient disaster management. climate change thereby, increase the susceptibility of The landslide susceptibility mapping can be done mountainous areas toward landslide hazards (Chen et using heuristic, statistical probabilistic, and determinis- al., 2016), which has become more devastating as the tic models. In heuristic analysis, geo-morphologist frequency and severity of the hazard had increased than identiﬁes the causal factors of landslides and degree of before (Kanungo, Arora, Sarkar, & Gupta, 2009). its inﬂuence by analyzing ﬁeld data, aerial photograph, Anthropogenic activities such as rapid unplanned urba- satellite images, etc. and based on their knowledge of nization and deforestation also combine with geological past events, experience, and expertise, they assign and geomorphological aspects that contribute to the weights to each factor accordingly (Ahmed, 2015; occurrence of the landslide (Horelli, 2005;VanWesten, Intarawichian & Dasananda, 2010; Yilmaz & Yildirim, 2000). 2007, 2008, 2010, 2012, and 2015 are some of the 2006). Deterministic analysis, on the other hand, is a years when this hazard had resulted in a widespread detailed approach, used to quantify landslide hazards in devastating impact on diﬀerent south-eastern regions of individual slope with the help of slope stability model Bangladesh (Sarwar, 2008) but the incidence that took (Cleary, Prakash, & Rothauge, 2010;Zhou, Esaki, place in 2017 has exceeded the past records of death and Mitani, Xie, & Mori, 2003) and calculation of factors economic loss. The disaster claimed the lives of 152 of safety (Van Westen, 2000; van Westen, van Asch, & people in total and had resulted in an economic loss of Soeters, 2005). It considers variables such as normal about USD 223 million (UN RC, 2017), which is thereby stress, the angle of friction, pore water pressure, and a threat to the development opportunities. According to antecedent rainfall, and various hydrological and stabi- Haque et al. (2018), Rangamati, the region (shown in lity models are used together to calculate the probability Figure 1)had suﬀered the worst impacts of 2017 event, of slope failure (Montgomery et al., 1994; Van Westen, has the high future potentiality of slope failure in this 2000; Iverson, 2000; Van Westen et al.,, 2011). The region as the slopes became unstable after 2017 large- landslide susceptibility of the Rangamati district was scale landslide events. This damaging eﬀects of these produced using statistical approach due to high subjec- landslide hazards need to be reduced by taking tivity error in heuristic analysis and complex CONTACT Shamima Ferdousi Sifa sifaferdousi25@gmail.com Department of Disaster Science and Management, Faculty of Earth and Environmental Sciences, University of Dhaka, Dhaka, Bangladesh © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of the International Water, Air & Soil Conservation Society(INWASCON). This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. GEOLOGY, ECOLOGY, AND LANDSCAPES 223 Figure 1. The four Upazilas, Rangamati Sadar, Rajasthali, Kaptai, and Kaukhali of Rangamati district, were chosen as the study area. methodology, huge data requirement in the determinis- involves the calculation of frequencies to map the future tic analysis (Van Westen, 2000). susceptible areas and is well known for producing good There are many statistical models such as informa- results with high precision (Li et al., 2016). The subjec- tion value method, frequency ratio, landslide nominal tivity problem that arises due to the classiﬁcation of the susceptibility factor (LNSF), logistic regression, multiple factor classes is also addressed here and greatly increases regression, discriminant analysis, and artiﬁcial neural the continuity of the frequency ratios. Therefore, this network (ANN). that can be used to produce landslide study aims to map the future landslide susceptible areas susceptible map but most of them are complex and using WoE and MFR and compare their results to diﬃcult to use (Ahmed & Dewan, 2017; Bathrellos, determine which statistical model describes the suscept- Kalivas, & Skilodimou, 2009). In this study, the WoE ibility of the landslide occurrence better. and MFR models will be used to map the future land- slide susceptible areas. Weights of evidence (WoE) is a 2. Materials and method Bayesian probability model that uses a log-linear form of the theorem to map future landslide susceptible areas Landslide susceptibility mapping is carried by over- (Bonham-Carter, 1994). It is a data-driven method laying the landslide inventory with the factor maps where prediction is made from prior and conditional using the WoE and MFR models, a statistical probability when the factor maps are overlaid with past approach that shows how factors contribute to the landslide events (inventory) (Agterberg et al., 1990; landslide occurrence. Therefore, a landslide inventory Bonham-Carter, 1994; Bonham-Carter, Agterberg, & consisting the location of past landslides is produced Wright, 1989). This bivariate method is commonly and combined with triggering factor maps by giving used to deﬁne the statistical association between the diﬀerent weight values derived from each model. Past factors and landslide events in several studies (van landslides were used to train the data because the Westen, Rengers, & Soeters, 2003;Lee &Choi, 2004). “past and present are keys to the future”; therefore, The problems of heterogeneity in ground conditions, the future landslide susceptible areas are estimated lack of detailed maps, and inadequate data are solved by from the past landslides and the preconditioning using this method. It can be used for large areas having factors that are responsible for the slope failure varying data types and information (Neuhäuser, 2014). assuming that the causative factors will also initiate Many researchers consider the method as robust and future landslides under similar conditions. The reliable as it avoids subjectivity and measures uncer- results of the models were later validated from the tainty associated with estimates of probability values gradient of the susceptibility index rank and cumula- (error and relative error values) (Neuhäuser, 2014). On tive landslide occurrence curve. the other hand, modiﬁed frequency ratio is used because The inventory map was produced from the SAR of its simplicity (Mohammady, Pourghasemi, & image, which has a 10-m spatial resolution and the Pradhan, 2012) and easily understandable parameters oﬀset tracking was employed to two SAR single look such as input, calculation, and output procedure, complex (SLC) images sensed on 6 June 2017 and 18 enabling the end users to apply this method with ease June 2017, which gave a 12 days temporal window. To (Lee & Pradhan, 2007; Yilmaz, 2009). This method produce the factor maps, 12.5 m ALOS PALSAR digital 224 S. F. SIFA ET AL. elevation model (DEM) and 30 m Landsat 8 optical along with decorrelated areas. The landslides have a images were used. The rainfall data are collected from speciﬁc coherence threshold. The masking of lower the Bangladesh Meteorological Department. coherence mostly discards the decorrelated areas, thus, eliminating areas of displacements other than landslides. Second, the displacement areas are digi- 2.1. Landslide inventory preparation tized manually. While digitizing, the shape of the displacements was considered. Any displacement The landslide inventory shows the previous occurrence that does not resemble the shapes of landslides was of a landslide that has occurred in the study area, and not included. Lastly, the knowledge of local region is this was prepared using SAR oﬀset tracking technique. also put into consideration. On the other hand, oﬀset Oﬀset tracking provides displacement information par- tracking has a minimum precision of 1/20th of a pixel allel to SAR satellite track (Azimuth oﬀset) and the track (Riveros et al., 2013). This makes the precision of our perpendicular (range oﬀset) (Fialko, Sandwell, Simons, analysis to be 0.5 m that means any ground move- & Rosen, 2005). The window patch intensity is mea- ment larger than 0.5 m will be detected, which is sured to ﬁnd the motion or displacement between two enough for landslide identiﬁcation. Mohammad et images of the same area (Lu, 2016). First, the master al. (2018) in his study mentioned that oﬀset tracking (pre-event) and slave image (post-event) are co-regis- shows better result when compared with subpixel tered using a digital elevation model (DEM). Afterward, correlation of optical images. Afterwards, the displa- to calculate the oﬀset, a ground control point (GCP) cements shown in the oﬀset tracking were digitized to grid is speciﬁed for the master image. The displacement produce the inventory, and some of the landslide is measured by matching frequency peaks of a patch of locations were validated with the location data pro- pixels in master and slave image. If two corresponding vided by Geological Survey of Bangladesh (2001). windows give the same frequency peaks, they are con- The study faced a limitation in inventory produc- sidered stable (Pathier et al., 2006). Each applied to tion as the freely available SAR image utilizes C band oﬀset a correlation value is computed, thus permitting (5–6 cm) wavelength. It is hard for C band wave- calculation of the corresponding matching correlation length to penetrate vegetation cover, resulting in surface (MCS), whose ith and jth elements for the some decorrelation. However, to prevent decorrela- generic pixel of range and azimuth coordinates (x, y) tion, coherence estimation was carried out to mask are the following: (Casu, Manconi, Pepe, & Lanari, out the de-correlated regions. 2011;Sun &Muller, 2016). σMS i;j ðÞ MCS i; j ¼ 2.2. Landslide triggering factor map preparation σMσS i;j Factor maps and inventories were produced from where digital elevation model (DEM), optical and radar sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ X;Y 2 image analysis. Elevation, slope angle, and aspect jj MxðÞ ; y x¼1;y¼1 σ ¼ μ M were derived from DEM, having a spatial resolution XY of 12.5 m and obtained from ALOS PALSAR satellite. sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ On the other hand, landcover classiﬁcation and vege- X;Y 2 jj SxðÞ þ i; y þ j tation index (NDVI) were obtained from Landsat 8 x¼1;y¼1 σS ¼ μ i;j i;j Satellite image, using maximum likelihood algorithm XY based supervised classiﬁcation. Road and waterbodies are the standard deviation values for the master and were digitized and mapped from GeoEye satellite slave amplitude images, respectively, computed on image provided by ESRI, and Euclidean distance the matching window pixels. The normalized cross- was applied to determine the inﬂuence of the follow- correlation (NCC) between the master and slave ing in landslide occurrence (Mohammad et al., 2018). image amplitudes is computed. This gives the quality Data on daily rainfall of about 33 years (from 1977 to of the calculation. 2015) were collected from the Bangladesh To prepare a valid inventory, several steps have been Meteorological Department (BMD). The annual aver- taken to identify the past earth movement. First, the age rainfall is then computed from the available data, oﬀset tracking provided the ground displacements P P i y ½ i ðÞ i; j i : i ðÞ i; j i 1 1 2 2 i¼1 j¼1 NCC ¼ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃqﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ P P P P j 2 j 2 i y i y x x i ðÞ i; j i i ðÞ i; j i 1 1 2 2 i¼1 j¼1 i¼1 j¼1 GEOLOGY, ECOLOGY, AND LANDSCAPES 225 and the calculated value is used to estimate the rain- Npix4 fall values in other regions of the study area based on the elevation. Özşahin (2014) shows that for 100 m Slope factor Npix1(nslclass) Npix3 increases in elevation, rainfall increases by 0.54 mm. The geological formation of the study area was digi- tized from the geological map of Bangladesh pro- nclass Landslide duced by the Geological Survey of Bangladesh (2001). Npix2 nslide nmap 2.3. Landslide susceptibility mapping Figure 2. Principle of the WoE model. 2.3.1. Weights of evidence (WoE) model In this method, WoE, the prior probability is calcu- Npix1 ¼ nslclass lated on the basis of past landslides assuming that it will trigger future hazardous event due to the Npix2 ¼ nslide nslclass unstable nature of the slope resulting from the slope failure. When additional information about the fac- Npix3 ¼ nclass nslclass tors are not available, prior probability give a good estimation about the possibility of landslide occur- Npix4 ¼ nmap nslide nclass þ nslclass rence by dividing the number of pixels having land- slides with a total number of pixels in the map The above variables thereby mean (Bonham-Carter, 1994). nslide = Number of pixels with landslides in the map nclass = Number of pixels in the class AreaðÞ Slide nslclass = Number of pixels with landslides in the Pprior ¼ PSfg ¼ AreaðÞ Total class nmap = Total number of pixels in the map However, when information such as presence or extent Both positive and negative weighted values of each of causal factors of landslides is available, the prior variable are then estimated to ﬁnd a degree of correla- probability is further modiﬁed to obtain a conditional tion in the presence or absence of the factor using the probability. This is done by producing a binary map (B) formula described by Bonham-Carter, 1994; Bonham- for that particular factor depending on the presence and Carter et al. (1989): absence of the variable in the map. A relationship is PBfg jS then established between the binary maps with the W ¼ ln landslide inventory, to calculate the conditional prob- fg BjS ability for a certain condition. According to Bonham- Carter (1994), the factors are conditionally independent W ¼ lnððNpix1 of each other, and the conditional probability of occur- ðÞ Npix3 þ Npix4 =ðÞ ðÞ Npix1 þ Npix2 Npix3 ring a landslide given that a particular factor unit is present there is expressed as follows: Pfg BjS W ¼ ln fg BjL PSfg \ B PSfg jB ¼ PBfg W ¼ lnððNpix2 ðÞ Npix3 þ Npix4 =ðÞ ðÞ Npix1 þ Npix2 Npix4 Npixfg S \ B In the presence of factor (B) in landslide area (S) Npixfg B gives a positive weighted value (W ), deﬁning the correlation between them. On the other hand, a In the equation landslide and the factor, variable is negative weight (W ) indicates the absence of the denoted by S and B, respectively. Pixel area in each factor. Then, a weighted contrast factor (C) is map is used in the analysis to ﬁnd about the four obtained to see how much the conditioning factor is possible combinations of probability described in spatially associated with the landslides. A ﬁnal sus- Figure 2, which are when landslides occur in the ceptibility map (LSI) hence, is produced by combin- presence of a potential conditioning factor (Npix1) ing the weighted map of each factor through an or the absence of it (Npix2), when there is no land- overlay operation. slide in the map area but the factor is present (Npix3) and the absence of both landslide and that particular C ¼ W W factor (NPix4). This is obtained by crossing the inventory of landslides with each factor map to Wmap ¼ W W þ W calculate: 226 S. F. SIFA ET AL. the start of the modeling and the remaining 75% are LSI ¼ Wmap used to train the model. The test data are then com- bined with the ﬁnal weighted map that is reclassiﬁed 2.3.2. Modiﬁed frequency ratio (MFR) models into 32 classes to compute percentage The values are Another method used in the study is modiﬁed fre- then plotted in a graph of percentage of cumulative quency ratio, which uses the same assumption as the landslide occurrence against the percentage of suscept- WoE model to ﬁnd a correlation between the land- ibility index rank to obtain a success rate curve (Jebur, slide points and preconditioning factors. Each factor Pradhan, Shafri, Yusoﬀ, & Tehrany, 2015). (F) in this method is classiﬁed into n number of classes after combining the factors with landslide 3. Results and discussion events to ﬁnd out the respective frequencies (PL and PF ) and Frequency ratioðÞ FR for the ith class of 3.1. Landslide inventory and factor maps factor by the formula given below. The studies (Lee About 420 landslide events had been identiﬁed in the & Talib, 2005; Lee & Pradhan, 2007; Deng, Li, & Tan, four Upazilas of Rangamati districts. In case of weights 2017) describe the procedure by the equation: of evidence (WoE) modeling the landslide, locations are PL FR ¼ deﬁned by the polygon attributes whereas point attri- PF butes had been used for modiﬁed frequency ratio (MFR) modeling. The conditioning and triggering fac- the frequency of landslides in the F area ¼ tors that contribute to landslide occurrence in the study the frequency of the F area area had been produced from DEM, optical images, and data from secondary sources. Factors such as elevation, the area of landslides in the F area slope, aspect, landcover, normalized diﬀerence vegeta- =the area of landslides in the study area tion index (NDVI), rain, distance from waterbody, and the area of the F area road have been taken into account and are shown in =the area of the study area Figure 3. The magnitude of FR values indicates the degree of correlation. A value greater than 1 means the strong 3.2. Landslide susceptibility map using WoE and correlation between the factors and landslides; whereas, MFR modeling a value less than 1 describes weak correlation (Pradhan, 2010). Landslide susceptibility map produced by two diﬀerent Modiﬁed frequency ratio is the same as the con- statistical models to show the degree of inﬂuence of ventional frequency ratio up to this point. The diﬀer- each causal factors with past landslide occurrence is ences arise as ratios obtained from the above given below (Figure 4). The future landslide-prone procedure is normalized to a range of 0–1, to use areas are quantiﬁed into high, moderate, low, and very this variable in precision setting and frequency statis- low areas as shown by diﬀerent color representation in tics procedures (Li et al., 2016). The relative fre- the map. To establish a clear relationship, each factor quency ðÞ RF for the ith class, thereby is calculated was divided into several classes to see which class has from the following equation suggested by the highest inﬂuence on landslide occurrence. The rela- Althuwaynee et al. (2016) and summed up together tive frequency value and weight value for diﬀerent to get the ﬁnal susceptibility map known as landslide classes of each factor map for modiﬁed frequency susceptibility indexðÞ LSI . ratio (MFR) and weights of evidence (WoE) model are also provided in Appendix (Tables A1 and A2). As two FR RF ¼ diﬀerent models are used, the parameters and resultant FR outputs are also diﬀerent from each other but relative susceptibility of each class is approximately similar for the frequeny ratio of ith class the methods. In case of the WoE method, W ,W ,and the frquency ratio of the class C factor describe the correlation and spatial association of the landslides with the factors. The positive contrast LSI ¼ RF map factor indicates positive association, that is, more occur- rences fall in the domain than the expected possibility and vice versa for negative contrast factor. The higher 2.4. Validation of landslide susceptibility models weighted value indicates higher degree of inﬂuence on The models are veriﬁed using success rate curve to see landslide occurrence. On the other hand, for MFR, the how well each model describes the susceptibility of the spatial relationship between the factors and landslide is landslide occurrence (Chung & Fabbri, 1999). To vali- deﬁned by the relative frequency value. The closer the date, 25% of the landslide distributions are separated at ratio value is to 1, the stronger is the relationship GEOLOGY, ECOLOGY, AND LANDSCAPES 227 Inventory Elevation Slope Angle Aspect Geology Landuse Pattern NDVI Annual Average Rainfall Distances from Road Distances from Waterbody Figure 3. Landslide inventory and landslide trigerring factor maps produced to understand the contributions of each factors in landslide occurrence. The inventory was produced from radar image analysis by oﬀset tracking technique. The factor maps are produced from DEM and optical image analysis. indication higher percentage of landslide area in the into 20–30, 30–40–40 degrees above has received the class than the percentage of the class in the total area. positive and higher weightage than other classes, but the The weight values derived from each model describe weightage value estimated from the MFR model shows the spatial relationships of the causative factors in con- that slope angle of 40 and above degrees has high value. tributing the landslide occurrences. The association This estimated value shows that if the slope angle is described by both the models is more or less the same. greater than 20 degrees than there is possibility of land- In this study, it has been found that slope angle classiﬁed slideoccurrenceishigh, butiftheslopeangle increasesto 228 S. F. SIFA ET AL. (a) (b) Figure 4. The landslide susceptibility map produced using (a) WoE and (b) MFR model. Landslide causative factor maps were combined with the landslide inventory produced from SAR oﬀset tracking. In the map, the possibility of landslide occurrence in a region is deﬁned by diﬀerent color representation and classifying them into high to very low susceptible classes. 40 degrees the susceptibility of landslide occurrence and water bodies show little correlation with landslide increases greatly as it has received the highest weightage. occurrence as few landslides occurred closer to the road Moreover, the weight values are derived from the past or water bodies; rather, the events were more distributed landslides where the pre-conditioning factors for which (Figure 3). This is because major roads and streams are the slope failed are used to train the dataset and carry out taken into consideration that has aﬀected the results. the modeling. It has also been seen that elevation greater Therefore, there is a further scope of study to consider than 89 m, hill facing in the direction of South-East, East, these factors where minor roads and streams will be taken West direction, the geological formation having an altera- to assess their association with the landslide occurrence. tion of sandstone and shale (Bhuban and Bokabil), the presence of bare and built up area has a higher contribu- 3.3. Comparison of the WoE and MFR models tion to landslide occurrence than other classes of each triggering factor. Vegetation index (NDVI), both factors The research involves quantiﬁcation of future land- show that areas with higher vegetation value have a slide susceptible areas of Rangamati Sadar, Kaukhali, greater probability of landslide occurrence. Upper class Kaptai, and Rajasthali Upazila, by overlaying the con- of vegetation index got higher weight value indicating ditioning factors of landslides with inventory showing that the inﬂuence of heavy rainfall plays an important role landslide events of 2017. Analyzing the two diﬀerent in slope failure as it makes the vegetated area more methods shows that MFR had overestimated the high unstable. Another reason can be indiscriminate logging and low susceptible areas and underestimated the practice. Rainfall was observed to be the main causal moderate susceptible areas than WoE. This is because factor as it received the highest weight value among the not all landslide points (from inventory) that are used other conditioning factors. The distance from road and to identify the susceptible areas of each class fall in waterbody was calculated to see the inﬂuence of the the high or moderate susceptible areas. This is due to position of road and waterbody in case of landslide the complex interaction of the causal factors. occurrence. Road is an important factor that contributes Although both of the models described the classes to landslide occurrence because construction of roads in of each factor in a similar way, the magnitude of hilly areas to ensure adequate communication with other the value assigned by each method was diﬀerent. In regions results in slope instability, since it involves mod- the WoE model, the polygon has been used as train- iﬁcation of slope by means of various activities. On the ing sets to produce the susceptibility map; whereas, other hand, water moving through the stream or rapid points are used in the case of the MFR model. excessive runoﬀ after rainfall are responsible for under- Polygons cover higher areas than points and have cutting and erosion of the slope, thereby increasing sus- higher inﬂuence over the classes and landslide points. ceptibility to sliding. Inﬁltration of the water into the soil To understand the diﬀerence between the two models also reduces the shear strength of the soil by increasing in terms of classifying the study area into diﬀerent the pore water pressure that in turn reduces the friction in classes, the results are presented in the bar diagram in the soil. However, in this study, distances from the road Figure 5. GEOLOGY, ECOLOGY, AND LANDSCAPES 229 Taking the whole area under consideration based to the weight of evidence method. However, there is a on the weight of evidence (WoE) model and modiﬁed signiﬁcant diﬀerence on identifying a moderate sus- frequency ratio (MFR) model, diﬀerent values were ceptible area where the weight of evidence (WoE) found according to the four categories namely very indicates that 62% areas are highly susceptible to low, low, moderate, high, and very high. It is evident landslide hazard and modiﬁed frequency ratio from the chart (Figure 5) that according to the mod- (MFR) model shows 37% areas that is half of the iﬁed frequency ratio, 29% areas are under very high value of the WoE model. Again, 6% areas are identi- landslide susceptible classes while 21% are according ﬁed as low and 12% areas are very low susceptible areas based on WoE whereas MFR shows a diﬀerent result that is, respectively, 20% and 14% areas. The bar diagram (Figure 6(a–d)) for each Upazila in the study area is provided below to show how much of the high, moderate, low, and very low land- slide susceptible areas are present in each Upazila, described by the two models. 3.4. Validation of the models The models need to be validated to see how well the susceptible classes are deﬁned by the models used. This is done by ﬁnding out the success rate for each model. Moreover, 75% of the polygons and points are used in the analysis to training the dataset to produce the susceptibility map. The percentage of landslide susceptibility index rank is plotted against Figure 5. Percentage of each susceptible classes in the study the percentage of cumulative landslide occurrence area deﬁned by the WoE and MFR models. to produce the success rate curve (Figure 7). About (a) (b) (c) (d) Figure 6. Percentage of susceptible classes deﬁned by the WoE and MFR models in (a) Kaptai, (b) Rajasthali, (c) Rangamati Sadar, and (d) Kaukhali Upazila of the study area. 230 S. F. SIFA ET AL. 20% of the high susceptible areas include 85% of (WoE) and modiﬁed frequency ratio (MFR) models thetotal landslidearea in caseof theWoE model to ﬁnd out which model explains the future suscept- but the MFR model includes only 20%. The WoE ibility better. An event-based (2017) landslide inven- model describes that 30% highly susceptible area tory was produced by processing radar image covers more than 99% of the total landslide area through SAR oﬀset tracking. The creation of inven- inthecaseofsuccessrate, whileMFR describes tory involves a knowledge-driven approach, which only 78% that is again lower than the WoE model. was later used in combination with the causal factors According to the success rate, the WoE model to map the future susceptible areas to see the spatial describes the landslide points better than the MFR relationship of each factor with the landslide occur- model. rence that is the factors that were used in the study As WoE describes the landslide susceptible areas was combined with previous landslides to determine better than MFR, so according to the WoE model, the condition of each factor in past landslide occur- 44% of the areas of Rangamati have been detected as rence. Later, these values were used to train the the highest and Kaptai as the lowest landslide suscep- dataset and derive the weight values to see the asso- tible areas (shown in the Figure 8). Rajasthali is the ciation of the factors and possibility of landslide second-most landslide hazard prone area as 25% and occurrence. The factors were then combined to see 33% of its area lie in the high and moderately land- how these factors that were responsible for past land- slide susceptible areas. On the other hand, total 51% slide occurrence also increase the susceptibility for of the areas of Kaukhali lie in the high and moderate future landslide occurrence, assuming that “past and susceptible areas. present are keys to the future.” The study area was then classiﬁed into high, moderate, low, and very low susceptibility classes, and comparison was made 4. Conclusion between the WoE and MFR models, describing how well each model deﬁne the classes. The results of the Rangamati district in Bangladesh got severely aﬀected Landslide susceptibility map of each method were by the landslide event of 2017. This large number of validated from the success rate curves. The percen- events had made the slope unstable, thereby increas- tage of landslide susceptibility index rank is plotted ing the possibility of slope failure in the area greater against the percentage of cumulative landslide occur- than before. The aim of the study was to produce the rence to produce the success rate curve. About 20% of landslide susceptibility map of the area based on the high susceptible areas include 85% of the total landslide inventory of 2017, using weight of evidence Figure 7. Success rate curve for weights of evidence (WoE) and modiﬁed frequency ratio (MFR) model. (a) (b) Figure 8. Percentage of (a) high and (b) moderate landslide susceptible areas of each Upazila in the study area estimated from the WoE model. GEOLOGY, ECOLOGY, AND LANDSCAPES 231 landslide area in case of the WoE model but the MFR Casu,F.,Manconi,A.,Pepe, A.,&Lanari, R. (2011). Deformation time-series generation in areas charac- model includes only 20%. The WoE model describes terized by large displacement dynamics: The SAR that 30% highly susceptible area covers more than amplitude pixel-oﬀset SBAS technique. IEEE 99% of the total landslide area in the case of success Transactions on Geoscience and Remote Sensing, 49 rate, while MFR describes only 78% that is again (7), 2752–2763. lower than the WoE model. As WoE explains the Chen, L., van Westen, C. J., Hussin, H., Ciurean, R. L., Turkington, T., Chavarro-Rincon, D., & Shrestha, D. P. results better than MFR, the model was later used to (2016). Integrating expert opinion with modelling for ﬁnd out which upazila of the study area is at high quantitative multi-hazard risk assessment in the Eastern landslide susceptible areas compared to the other. Italian Alps. Geomorphology, 273(October), 150–167. Rangamati Sadar upazila has been identiﬁed to have Chung, C.-J. F., & Fabbri, A. G. (1999). Probabilistic prediction high susceptibility toward landslide hazard as 46% models for Landslide Hazard mapping. Photogrammetric and 33% of its area fall under high and moderate Engineering & Remote Sensing, 65(12), 1389–1399. Cleary, P. W., Prakash, M., & Rothauge, K. (2010). susceptible classes, respectively. Kaukhali, Rajasthali, Combining digital terrain and surface textures with and Kaptai upazila have 23%, 25%, 8% high, and 28%, large-scale particle-based computational models to pre- 33%, 10% moderate susceptible areas, respectively. dict dam collapse and landslide events. International The results obtained from the research, thereby, Journal of Image and Data Fusion, 1(4), 337–357. might help the policymakers to take appropriate miti- Deng, X., Li, L., & Tan, Y. (2017). Validation of spatial prediction models for landslide susceptibility mapping gation measures to prevent the severity of the land- by considering structural similarity. ISPRS International slide hazard consequences in the study area. Journal of Geo-Information, 6(4), 103. Fialko, Y., Sandwell, D., Simons, M., & Rosen, P. (2005). Three-dimensional deformation caused by the Bam, Disclosure statement Iran, earthquake and the origin of shallow slip deﬁcit. Nature, 435(7040), 295–299. No potential conﬂict of interest was reported by the Geological Survey of Bangladesh. (2001). Description of authors. map units. Geological Survey of Bangladesh (GSB) Haque, D. M. E., Sifa, S. F., Mahmud, T. & Tarin, M. A. (2018). Landslide susceptibility assessment based on ORCID satellite image processing and bi-variate statistical mod- eling for Rangamati district, Bangladesh. Journal of Dewan Mohammad Enamul Haque http://orcid.org/ Physics, 93–106. doi:10.13140/RG.2.2.16092.03203 0000-0001-5439-5528 Horelli, J. A. (2005). Landslides in Hong Kong. University of Helsinki. Intarawichian, N., & Dasananda, S. (2010). Analytical hierarchy References process for landslide susceptibility mapping in lower Mae Chaem Watershed, Northern Thailand. Suranaree Journal of Agterberg, F. P., Bonham-Carter, G. F., & Wright, D. F. Science and Technology, 7(3). Retrieved from http://www. (1990). Statistical pattern integration for mineral thaiscience.info/journals/Article/SJST/10890515.pdf exploration. In D. F. Gaal & G. Merriam (Eds.), Iverson, R. M. (2000). Landslide triggering by rain inﬁltration. Computer application in resource estimation: Prediction Water Resources Research, 36(7), 1897–1910. and assessment for metals and petroleum (pp. 1–21). Jebur,M.N.,Pradhan,B.,Shafri,H.Z.M.,Yusoﬀ,Z.M., & Pergamon Press: Oxford. Tehrany, M. S. (2015). An integrated user-friendly ArcMAP Ahmed, B. (2015). Landslide susceptibility mapping using tool for bivariate statistical modelling in geoscience applica- multi-criteria evaluation techniques in Chittagong tions. GeoscientiﬁcModel Development, 8(3), 881–891. Metropolitan. Journal of the International Consortium Kanungo,D. P.,Arora, M.K.,Sarkar,S., &Gupta,R. P.(2009). on Landslides, Springer Link, 12(2015), 1077–1095. Landslide susceptibility zonation (LSZ) mapping - a review. Ahmed, B., & Dewan, A. (2017). Application of bivariate Journal of South Asia Disaster Studies, 2(1), 81–106. and multivariate statistical techniques in landslide sus- Lee,S.,& Choi,J.(2004). Landslide susceptibility mapping ceptibility modeling in Chittagong city corporation, using GIS and the weight-of-evidence model. International Bangladesh, Remote Sensing, 9, 304, (May 2014). Journal of Geographical Information Science, 18(8), 789–814. Althuwaynee, O. F., Pradhan, B., & Lee, S. (2016). A novel Lee, S., & Pradhan, B. (2007). Landslide hazard mapping at integrated model for assessing landslide susceptibility map- Selangor, Malaysia using frequency ratio and logistic ping using CHAID and AHP pair-wise comparison. regression models. Landslides, 4,33–41. International Journal of Remote Sensing, 37(5), 1190–1209. Lee, S., & Talib, J. A. (2005). Probabilistic landslide suscept- Bathrellos, G. D., Kalivas, D. P., & Skilodimou, H. D. ibility and factor eﬀect analysis. Environmental Geology, 47 (2009). GIS-based landslide susceptibility mapping mod- (7), 982–990. els applied to natural and urban planning in Trikala, Li, L., Lan, H., Guo, C., Zhang, Y., Li, Q., & Wu, Y. (2016). Central Greece. Estudios Geológicos, 65(1), 49–65. A modiﬁed frequency ratio method for landslide sus- Bonham-Carter, G. F. (1994). Geographic information systems ceptibility assessment. Landslides, 14(2), 727–741. for geoscientists, modeling with GIS, 302–334. Oxford: Lu,J.(2016, August). Sentinel-1 Toolbox oﬀset tracking tutorial. Pergamon Press. ESA Bonham-Carter, G. F., Agterberg, F. P., & Wright, D. F. Mohammady, M., Pourghasemi, H. R., & Pradhan, B. (2012). (1989). Weights of evidence modelling: A new approach Landslide susceptibility mapping at Golestan Province, Iran: to mapping mineral potential. Statistical Applications in A comparison between frequency ratio, Dempster-Shafer, Earth Science, 89(9), 171–183. 232 S. F. SIFA ET AL. and weights-of-evidence models. Journal of Asian Earth Van Westen, C. J. (2000). The modelling of landslide hazards Sciences, 61,221–236. using gis (PDF download available). Surveys in Geophysics, Montgomery,D.R.,&Dietrich,W.E.(1994). A physically 21, 241–255. Retrieved from https://www.researchgate.net/ based model for the topographic control on shallow land- publication/233815072_The_Modelling_Of_Landslide_ sliding. Water Resources Research, 30(4), 1153–1171. Hazards_Using_Gis Neuhäuser, B. (2014). Landslide susceptibility and climate vanWesten, C. J.,Ghosh,S., Jaiswal, P.,Martha, T. R.,& change scenarios in ﬂysch areas of the Eastern alps. Kuriakose, S. L. (2011). From landslide inventories to land- Geomorphology, 86(1-2), 12–24 slide risk assessment; an attempt to support methodological Özşahin, E. (2014). Using geographic information system development in India. In the Second World Landslide Forum, and analytical hierarchy process in landslide susceptibil- 1,3–20. ity analysis in Tekirdağ province. The Central and vanWesten, C. J.,Rengers,N., &Soeters,R.(2003). Use Eastern European Online Library, 137–144. doi:10.7366/ of geomorphological information in indirect landslide 1509499516305 susceptibility assessment. Natural Hazards, 30,399– Pathier,E.,Fielding,E. J.,Wright,T. J.,Walker,R.,Parsons,B. 419. Kluwer Academic Publishers. Retrieved from. E., & Hensley, S. (2006). Displacement ﬁeld and slip distri- https://research.utwente.nl/en/publications/use-of-geo bution of the 2005 Kashmir earthquake from SAR imagery. morphological-information-in-indirect-landslide- Geophysical Research Letters, 33(20), 1–5. suscept Pradhan, B. (2010). Landslide susceptibility mapping of van Westen, C. J., van Asch, T. W. J., & Soeters, R. (2005). a catchment area using frequency ratio, fuzzy logic Landslide hazard and risk zonation - Why is it still so and multivariate logistic regression approaches. diﬃcult? Bulletin of Engineering Geology and the Journal of the Indian Society of Remote Sensing, 38 Environment, 65(2), 167–184. (2), 301–320. Yilmaz, I. (2009). Landslide susceptibility mapping using Sarwar, M. G. M. (2008). Landslide tragedy of Bangladesh. In frequency ratio, logistic regression, artiﬁcial neural net- The ﬁrst world landslide forum. Tokyo. Retrieved from works and their comparison: A case study from Kat http://www.academia.edu/3588839/Landslide_Tragedy_of_ landslides (Tokat-Turkey). Computers and Geosciences, Bangladesh 35(6), 1125–1138. Sun, L., & Muller, J. P. (2016). Evaluation of the use of sub- Yilmaz, I., & Yildirim, M. (2006). Structural and geomor- pixel oﬀset tracking techniques to monitor landslides in phological aspects of the Kat landslides (Tokat-Turkey) densely vegetated steeply sloped areas. Remote Sensing, 8 and susceptibility mapping by means of GIS. (8), 659. Environmental Geology, 50(4), 461–472. UN RC. (2017). Bangladesh: HCTT response plan (June- Zhou,G.,Esaki,T.,Mitani,Y.,Xie,M.,&Mori.,J.(2003). December 2017) - Bangladesh | ReliefWeb. Retrieved Spatial probabilistic modeling of slope failure using an inte- from https://reliefweb.int/report/bangladesh/bangla grated GIS Monte Carlo simulation approach. Engineering desh-hctt-response-plan-june-december-2017 Geology, 68(3–4), 373–386. GEOLOGY, ECOLOGY, AND LANDSCAPES 233 Appendix Table A1. Relative frequency ratio values for the causative factors using MFR. No. of points % of points cls_area % of cls_area Ratio(+) RF Integer NDVI −0.236,372 0 0 888,624 12.65,263 0 0 0 0.087784–0.285,225 2031.25 4.153,355 403,133 5.739,989 0.723,582 0.237,745 24 0.285,225–0.396,459 13,125 26.83,706 1,675,393 23.855 1.125,008 0.36,964 37 0.396,459–0.56,053 33,750 69.00958 4,056,047 57.75,183 1.194,933 0.392,616 49 AVERAGE ANNUAL RAINFALL (mm) 2559–2560 mm 48,906.25 100 7,023,268 100.0005 0.999,995 0.336,122 34 ASPECT ﬂat 156.25 0.319,489 866,476 12.33,728 0.025896 0.002664 0 north 1562.5 3.194,888 388,229 5.52,778 0.57,797 0.059461 6 north-east 2968.75 6.070288 703,459 10.01617 0.606,049 0.06235 7 east 5937.5 12.14,058 751,505 10.70,027 1.134,605 0.116,728 12 south-east 9843.75 20.1278 781,553 11.1281 1.808,735 0.186,082 19 south 8281.25 16.93,291 795,619 11.32,838 1.494,733 0.153,778 15 south-west 7031.25 14.377 879,217 12.51,869 1.148,443 0.118,152 11 west 5156.25 10.54,313 838,870 11.94,421 0.882,698 0.090812 12 north-west 6406.25 13.09904 732,429 10.42,865 1.256,062 0.129,223 13 north 1562.5 3.194,888 285,879 4.070474 0.784,893 0.08075 8 LANDUSE PATTERN Buit-up area 2343.75 4.792,332 633,618 9.021739 0.531,198 0.190,589 20 Bare Soil 4218.75 8.626,198 612,288 8.718,033 0.989,466 0.355,011 36 Water 0 0 976,037 13.89,725 0 0 0 Vegetation 42,343.75 86.58,147 4,801,360 68.36,393 1.266,479 0.454,401 45 DISTANCES from WATERBODIES (meter) 0–10 m 0 0 1,069,783 15.23,205 0 0 0 10–20 m 0 0 161,065 2.293,316 0 0 0 above 20 m 48,906.25 100 5,792,420 82.47,509 1.212,487 0.269,455 27 GEOLOGY Tipam 7343.75 15.01597 1,103,472 15.71,173 0.955,717 0.140,395 14 Bokabil 22,343.75 45.6869 1,949,704 27.76,076 1.645,736 0.241,759 24 Bhuban 5937.5 12.14,058 484,111 6.892,991 1.761,293 0.258,734 26 Girujan Clay 937.5 1.916,933 246,595 3.511,131 0.545,959 0.080201 8 Dupi Tila 5625 11.5016 979,936 13.95,277 0.824,324 0.121,093 12 Dihing and Dupi Tila 5000 10.22,364 792,991 11.29,096 0.905,471 0.133,014 13 lake 1718.75 3.514,377 1,461,794 20.81,368 0.168,849 0.024804 3 ava 0 0 4633 0.065967 0 0 0 SLOPE ANGLE (degrees) 0–10 degrees 13,750 28.11,502 4,098,000 58.34,917 0.481,841 0.042613 4 10–20 degrees 24,062.5 49.20,128 2,308,687 32.87,213 1.496,748 0.132,368 13 20–30 degrees 10,156.25 20.76,677 535,298 7.621,814 2.72,465 0.24,096 24 30–40 degrees 468.75 0.958,466 70,821 1.008381 0.9505 0.084059 9 40 and above 468.75 0.958,466 10,430 0.148,507 5.653,738 0.499,999 50 DISTANCES from ROADS (meter) 0–10 m 0 0 16,274 0.231,717 0 0 0 10–20 m 0 0 32,162 0.457,937 0 0 0 above 20 m 48,906.25 100 6,974,832 99.3108 1.00694 0.228,959 23 ELEVATION (meter) −80–0 m 3750 7.667,732 3,539,030 50.3903 0.152,167 0.013808 1 0–15 m 10,000 20.44,728 1,051,301 14.9689 1.365,985 0.123,955 13 15–45 m 18,437.5 37.69,968 1,372,905 19.54,804 1.928,566 0.175,006 18 45–89 m 8750 17.89,137 575,766 8.198,016 2.182,403 0.19,804 20 89–151 m 5781.25 11.82,109 255,508 3.638,038 3.249,303 0.294,855 29 151–237 m 2031.25 4.153,355 161,241 2.295,822 1.809,093 0.164,164 16 237–428 m 156.25 0.319,489 67,485 0.960,882 0.332,495 0.030172 3 234 S. F. SIFA ET AL. Table A2. Weighted values for the causative factors using the WoE model. SLIDEN NPIX NMAP NPIXACT NSLIDE NCLS NSCLS NPIX1 NPIX2 NPIX3 NPIX4 WPLUS WNEG C WMP NDVI 0.285,225–0.396,459 unknown 1,673,116 7,023,197 0 9490 1,675,393 2277 2277 7213 1,673,116 5,340,591 0.0058 −0.0018 0.00762 −0.61 0.396,459–0.56,053 unknown 4,049,040 7,023,197 0 9490 4,056,047 7007 7007 2483 4,049,040 2,964,667 0.24,606 −0.4797 0.72,573 0.10,813 0.087784–0.285,225 unknown 402,928 7,023,197 0 9490 403,133 205 205 9285 402,928 6,610,779 −0.9781 0.03733 −1.0154 −1.633 −0.236,372 unknown 888,623 7,023,197 0 9490 888,624 1 1 9489 888,623 6,125,084 −7.092 0.13,537 −7.2274 −7.845 0.396,459–0.56,053 active 7007 7,023,197 7007 9490 4,056,047 7007 7007 2483 4,049,040 2,964,667 0.24,606 −0.4797 0.72,573 0.10,813 0.285,225–0.396,459 active 2277 7,023,197 2277 9490 1,675,393 2277 2277 7213 1,673,116 5,340,591 0.0058 −0.0018 0.00762 −0.61 0.087784–0.285,225 active 205 7,023,197 205 9490 403,133 205 205 9285 402,928 6,610,779 −0.9781 0.03733 −1.0154 −1.633 −0.236,372 active 1 7,023,197 1 9490 888,624 1 1 9489 888,623 6,125,084 −7.092 0.13,537 −7.2274 −7.845 AVERAGE ANNUAL RAINFALL (mm) 2559.25–2560.27 unknown 7,013,778 7,023,268 0 9490 7,023,268 9490 9490 0.000001 7,013,778 0.000001 0 6.60,539 −6.6054 6.6054 2559.25–2560.27 active 9490 7,023,268 9490 9490 7,023,268 9490 9490 0.000001 7,013,778 0.000001 0 6.60,539 −6.6054 6.6054 ASPECT north unknown 387,967 7,023,236 0 9490 388,229 262 262 9228 387,967 6,625,779 −0.6949 0.02891 −0.7239 −0.9244 North unknown 285,588 7,023,236 0 9490 285,879 291 291 9199 285,588 6,728,158 −0.2836 0.01043 −0.294 −0.4945 North-east unknown 702,814 7,023,236 0 9490 703,459 645 645 8845 702,814 6,310,932 −0.3882 0.0352 −0.4234 −0.6239 East unknown 750,141 7,023,236 0 9490 751,505 1364 1364 8126 750,141 6,263,605 0.29,555 −0.0421 0.3376 0.13,708 South-east unknown 779,349 7,023,236 0 9490 781,553 2204 2204 7286 779,349 6,234,397 0.7372 −0.1465 0.8837 0.68,318 South-west unknown 878,162 7,023,236 0 9490 879,217 1055 1055 8435 878,162 6,135,584 −0.1189 0.01592 −0.1348 −0.3353 South unknown 793,379 7,023,236 0 9490 795,619 2240 2240 7250 793,379 6,220,367 0.73,556 −0.1492 0.88,476 0.68,424 West unknown 838,267 7,023,236 0 9490 419,435 603 603 8887 418,832 6,594,914 0.06208 −0.0041 0.06616 −0.1344 North-west unknown 731,617 7,023,236 0 9490 732,429 812 812 8678 731,617 6,282,129 −0.1981 0.02072 −0.2188 −0.4194 ﬂat unknown 866,462 7,023,236 0 9490 866,476 14 14 9476 866,462 6,147,284 −4.4277 0.13,039 −4.5581 −4.7586 South-east active 2204 7,023,236 2204 9490 781,553 2204 2204 7286 779,349 6,234,397 0.7372 −0.1465 0.8837 0.68,318 East active 1364 7,023,236 1364 9490 751,505 1364 1364 8126 750,141 6,263,605 0.29,555 −0.0421 0.3376 0.13,708 North-west active 812 7,023,236 812 9490 732,429 812 812 8678 731,617 6,282,129 −0.1981 0.02072 −0.2188 −0.4194 West active 603 7,023,236 603 9490 419,435 603 603 8887 418,832 6,594,914 0.06208 −0.0041 0.06616 −0.1344 North active 291 7,023,236 291 9490 285,879 291 291 9199 285,588 6,728,158 −0.2836 0.01043 −0.294 −0.4945 South active 2240 7,023,236 2240 9490 795,619 2240 2240 7250 793,379 6,220,367 0.73,556 −0.1492 0.88,476 0.68,424 North-east active 645 7,023,236 645 9490 703,459 645 645 8845 702,814 6,310,932 −0.3882 0.0352 −0.4234 −0.6239 South-west active 1055 7,023,236 1055 9490 879,217 1055 1055 8435 878,162 6,135,584 −0.1189 0.01592 −0.1348 −0.3353 north active 262 7,023,236 262 9490 388,229 262 262 9228 387,967 6,625,779 −0.6949 0.02891 −0.7239 −0.9244 ﬂat active 14 7,023,236 14 9490 866,476 14 14 9476 866,462 6,147,284 −4.4277 0.13,039 −4.5581 −4.7586 LANDUSE PATTERN Vegetation unknown 4,793,182 7,023,303 0 9490 4,801,360 8178 8178 1312 4,793,182 2,220,631 0.2319 −0.8286 1.06049 −0.2138 Built Up Area unknown 633,238 7,023,303 0 9490 633,618 380 380 9110 633,238 6,380,575 −0.813 0.05376 −0.8668 −2.1411 Bare Soil unknown 611,357 7,023,303 0 9490 612,288 931 931 8559 611,357 6,402,456 0.11,822 −0.0121 0.13,028 −1.144 Water unknown 976,036 7,023,303 0 9490 976,037 1 1 9489 976,036 6,037,777 −7.1859 0.14,974 −7.3356 −8.6099 Vegetation active 8178 7,023,303 8178 9490 4,801,360 8178 8178 1312 4,793,182 2,220,631 0.2319 −0.8286 1.06049 −0.2138 Bare Soil active 931 7,023,303 931 9490 612,288 931 931 8559 611,357 6,402,456 0.11,822 −0.0121 0.13,028 −1.144 Built Up Area active 380 7,023,303 380 9490 633,618 380 380 9110 633,238 6,380,575 −0.813 0.05376 −0.8668 −2.1411 Water active 1 7,023,303 1 9490 976,037 1 1 9489 976,036 6,037,777 −7.1859 0.14,974 −7.3356 −8.6099 DISTANCES from WATERBODIES (meter) above 20 unknown 5,783,030 7,023,268 0 9490 5,792,418 9388 9388 102 5,783,030 1,230,748 0.18,214 −2.7928 2.97,491 −2.2548 10–20 m unknown 160,896 7,023,268 0 9490 160,959 63 63 9427 160,896 6,852,882 −1.24 0.01655 −1.2565 −6.4862 0–10 m unknown 1,069,852 7,023,268 0 9490 1,069,891 39 39 9451 1,069,852 5,943,926 −3.6141 0.16,139 −3.7755 −9.0051 above 20 active 9388 7,023,268 9388 9490 5,792,418 9388 9388 102 5,783,030 1,230,748 0.18,214 −2.7928 2.97,491 −2.2548 10–20 m active 63 7,023,268 63 9490 160,959 63 63 9427 160,896 6,852,882 −1.24 0.01655 −1.2565 −6.4862 0–10 m active 39 7,023,268 39 9490 1,069,891 39 39 9451 1,069,852 5,943,926 −3.6141 0.16,139 −3.7755 −9.0051 (Continued) GEOLOGY, ECOLOGY, AND LANDSCAPES 235 Table A2. (Continued). SLIDEN NPIX NMAP NPIXACT NSLIDE NCLS NSCLS NPIX1 NPIX2 NPIX3 NPIX4 WPLUS WNEG C WMP GEOLOGY Bhuban unknown 483,380 7,023,236 0 9490 484,111 731 731 8759 483,380 6,530,366 0.11,124 −0.0087 0.11,999 −0.1195 Bokabil unknown 1,944,924 7,023,236 0 9490 1,949,704 4780 4780 4710 1,944,924 5,068,822 0.59,685 −0.3758 0.97,264 0.73,313 Tipam unknown 1,102,360 7,023,236 0 9490 1,103,472 1112 1112 8378 1,102,360 5,911,386 −0.2937 0.04636 −0.34 −0.5795 Predominantly Water unknown 1,461,443 7,023,236 0 9490 1,461,794 351 351 9139 1,461,443 5,552,303 −1.7288 0.19,597 −1.9247 −2.1642 Dihing and Dupi Tila unknown 792,211 7,023,236 0 9490 792,991 780 780 8710 792,211 6,221,535 −0.3179 0.03409 −0.352 −0.5915 Dihing and Dupi Tila active 780 7,023,236 780 9490 792,991 780 780 8710 792,211 6,221,535 −0.3179 0.03409 −0.352 −0.5915 Dupi Tila unknown 978,394 7,023,236 0 9490 979,936 1542 1542 7948 978,394 6,035,352 0.15,256 −0.0271 0.17,964 −0.0599 Bokabil active 4780 7,023,236 4780 9490 1,949,704 4780 4780 4710 1,944,924 5,068,822 0.59,685 −0.3758 0.97,264 0.73,313 Predominantly Water active 351 7,023,236 351 9490 1,461,794 351 351 9139 1,461,443 5,552,303 −1.7288 0.19,597 −1.9247 −2.1642 Tipam active 1112 7,023,236 1112 9490 1,103,472 1112 1112 8378 1,102,360 5,911,386 −0.2937 0.04636 −0.34 −0.5795 Dupi Tila active 1542 7,023,236 1542 9490 979,936 1542 1542 7948 978,394 6,035,352 0.15,256 −0.0271 0.17,964 −0.0599 Girujan Clay unknown 246,401 7,023,236 0 9490 246,595 194 194 9296 246,401 6,767,345 −0.5415 0.01511 −0.5566 −0.7961 ava unknown 4633 7,023,236 0 9490 4633 0 0.001 9490 4633 7,009,113 −8.7433 0.00066 −8.744 −8.9835 Bhuban active 731 7,023,236 731 9490 484,111 731 731 8759 483,380 6,530,366 0.11,124 −0.0087 0.11,999 −0.1195 Girujan Clay active 194 7,023,236 194 9490 246,595 194 194 9296 246,401 6,767,345 −0.5415 0.01511 −0.5566 −0.7961 SLOPE ANGLE (degrees) 0–10 degrees unknown 4,095,289 7,023,236 0 9490 4,098,000 2711 2711 6779 4,095,289 2,918,457 −0.7149 0.54,041 −1.2553 −1.0027 10–20 degrees unknown 2,304,231 7,023,236 0 9490 2,308,687 4456 4456 5034 2,304,231 4,709,515 0.35,714 −0.2357 0.59,288 0.84,546 20–30 degrees unknown 533,386 7,023,236 0 9490 535,298 1912 1912 7578 533,386 6,480,360 0.97,429 −0.1459 1.12,019 1.37,277 30–40 degrees unknown 70,461 7,023,236 0 9490 70,821 360 360 9130 70,461 6,943,285 1.32,868 −0.0286 1.35,725 1.60,984 40 and above unknown 10,379 7,023,236 0 9490 10,430 51 51 9439 10,379 7,003,367 1.28,967 −0.0039 1.29,358 1.54,617 0–10 degrees active 2711 7,023,236 2711 9490 4,098,000 2711 2711 6779 4,095,289 2,918,457 −0.7149 0.54,041 −1.2553 −1.0027 10–20 degrees active 4456 7,023,236 4456 9490 2,308,687 4456 4456 5034 2,304,231 4,709,515 0.35,714 −0.2357 0.59,288 0.84,546 20–30 degrees active 1912 7,023,236 1912 9490 535,298 1912 1912 7578 533,386 6,480,360 0.97,429 −0.1459 1.12,019 1.37,277 30–40 degrees active 360 7,023,236 360 9490 70,821 360 360 9130 70,461 6,943,285 1.32,868 −0.0286 1.35,725 1.60,984 40 and above active 51 7,023,236 51 9490 10,430 51 51 9439 10,379 7,003,367 1.28,967 −0.0039 1.29,358 1.54,617 DISTANCES from ROADs (meter) above 20 m unknown 6,965,328 7,023,236 0 9490 6,974,775 9447 9447 43 6,965,328 48,418 0.00239 −0.421 0.42,342 −0.4139 above 20 m active 9447 7,023,236 9447 9490 6,974,775 9447 9447 43 6,965,328 48,418 0.00239 −0.421 0.42,342 −0.4139 10–20 m unknown 32,151 7,023,236 0 9490 32,180 29 29 9461 32,151 6,981,595 −0.4055 0.00153 −0.407 −1.2444 0–10 m unknown 16,267 7,023,236 0 9490 16,281 14 14 9476 16,267 6,997,479 −0.4524 0.00085 −0.4533 −1.2906 0–10 m active 14 7,023,236 14 9490 16,281 14 14 9476 16,267 6,997,479 −0.4524 0.00085 −0.4533 −1.2906 10–20 m active 29 7,023,236 29 9490 32,180 29 29 9461 32,151 6,981,595 −0.4055 0.00153 −0.407 −1.2444 ELEVATION (meter) 0–15 m unknown 1,049,074 7,023,236 0 9490 1,051,301 2227 2227 7263 1,049,074 5,964,672 0.45,038 −0.1054 0.55,581 0.77,643 15–45 m unknown 1,370,013 7,023,236 0 9490 1,372,905 2892 2892 6598 1,370,013 5,643,733 0.44,476 −0.1461 0.59,091 0.81,153 −80–0m unknown 3,537,566 7,023,236 0 9490 3,539,030 1464 1464 8026 3,537,566 3,476,180 −1.1846 0.53,439 −1.719 −1.4984 45–89 m unknown 574,158 7,023,236 0 9490 575,766 1608 1608 7882 574,158 6,439,588 0.72,748 −0.1003 0.82,773 1.04834 89–151 m unknown 254,639 7,023,236 0 9490 255,508 869 869 8621 254,639 6,759,107 0.92,513 −0.0591 0.98,419 1.20,481 −80–0m active 1464 7,023,236 1464 9490 3,539,030 1464 1464 8026 3,537,566 3,476,180 −1.1846 0.53,439 −1.719 −1.4984 0–15 m active 2227 7,023,236 2227 9490 1,051,301 2227 2227 7263 1,049,074 5,964,672 0.45,038 −0.1054 0.55,581 0.77,643 237–428 m unknown 67,410 7,023,236 0 9490 67,485 75 75 9415 67,410 6,946,336 −0.1957 0.00172 −0.1974 0.02322 151–237 m unknown 160,886 7,023,236 0 9490 161,241 355 355 9135 160,886 6,852,860 0.48,906 −0.0149 0.50,398 0.72,459 15–45 m active 2892 7,023,236 2892 9490 1,372,905 2892 2892 6598 1,370,013 5,643,733 0.44,476 −0.1461 0.59,091 0.81,153 89–151 m active 869 7,023,236 869 9490 255,508 869 869 8621 254,639 6,759,107 0.92,513 −0.0591 0.98,419 1.20,481 m active 1608 7,023,236 1608 9490 575,766 1608 1608 7882 574,158 6,439,588 0.72,748 −0.1003 0.82,773 1.04834 45–89 151–237 m active 355 7,023,236 355 9490 161,241 355 355 9135 160,886 6,852,860 0.48,906 −0.0149 0.50,398 0.72,459 237–428 m active 75 7,023,236 75 9490 67,485 75 75 9415 67,410 6,946,336 −0.1957 0.00172 −0.1974 0.02322 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Geology Ecology and Landscapes Taylor & Francis http://www.deepdyve.com/lp/taylor-francis/event-based-landslide-susceptibility-mapping-using-weights-of-evidence-BAnTuTWWDW

Loading next page...

References

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher: Taylor & Francis
Copyright: © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of the International Water, Air & Soil Conservation Society(INWASCON).
ISSN: 2474-9508
DOI: 10.1080/24749508.2019.1619222
Publisher site: See Article on Publisher Site

Abstract

GEOLOGY, ECOLOGY, AND LANDSCAPES 2020, VOL. 4, NO. 3, 222–235 INWASCON https://doi.org/10.1080/24749508.2019.1619222 RESEARCH ARTICLE Event-based landslide susceptibility mapping using weights of evidence (WoE) and modiﬁed frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh Shamima Ferdousi Sifa, Tonoy Mahmud, Maria Abdullah Tarin and Dewan Mohammad Enamul Haque Department of Disaster Science and Management, Faculty of Earth and Environmental Sciences, University of Dhaka, Dhaka, Bangladesh ABSTRACT ARTICLE HISTORY Received 28 December 2018 Landslide hazard of 2017 in Rangamati district had devastating impacts on development, Accepted 11 May 2019 thereby making landslide susceptibility mapping a prerequisite for disaster risk management. This study aims to map the future landslide susceptible areas by overlaying the landslide KEYWORDS inventory of 2017 with causative factor maps using WoE and MFR and compare their results Landslide; susceptibility; to determine that statistical model describes the susceptibility of the landslide occurrence inventory; weights of better than the other. The analysis shows that although both models deﬁne the spatial evidence (WoE); modiﬁed relationship of past landslides with the triggering factors in a same way but in case of frequency ratio (MFR) mapping, MFR had overestimated the high and low susceptible areas and underestimated the moderate susceptible areas than WoE. When validated from success rate curve by plotting the percentage of landslide susceptibility index rank against the percentage of cumulative landslide occurrence, it shows that the WoE model describes the landslides better than the MFR model. About 20% of the high susceptible areas include 85% of the total landslide area in case of the WoE model but the MFR model includes only 20%. On the other hand, the WoE model describes that 30% highly susceptible area covers more than 99% of the total landslide area while MFR deﬁnes only 78%. 1. Introduction appropriate mitigation measures in the landslide suscep- tible areas identiﬁed from susceptibility mapping, which Extreme hydro-meteorological conditions, triggered by will ensure eﬀective and eﬃcient disaster management. climate change thereby, increase the susceptibility of The landslide susceptibility mapping can be done mountainous areas toward landslide hazards (Chen et using heuristic, statistical probabilistic, and determinis- al., 2016), which has become more devastating as the tic models. In heuristic analysis, geo-morphologist frequency and severity of the hazard had increased than identiﬁes the causal factors of landslides and degree of before (Kanungo, Arora, Sarkar, & Gupta, 2009). its inﬂuence by analyzing ﬁeld data, aerial photograph, Anthropogenic activities such as rapid unplanned urba- satellite images, etc. and based on their knowledge of nization and deforestation also combine with geological past events, experience, and expertise, they assign and geomorphological aspects that contribute to the weights to each factor accordingly (Ahmed, 2015; occurrence of the landslide (Horelli, 2005;VanWesten, Intarawichian & Dasananda, 2010; Yilmaz & Yildirim, 2000). 2007, 2008, 2010, 2012, and 2015 are some of the 2006). Deterministic analysis, on the other hand, is a years when this hazard had resulted in a widespread detailed approach, used to quantify landslide hazards in devastating impact on diﬀerent south-eastern regions of individual slope with the help of slope stability model Bangladesh (Sarwar, 2008) but the incidence that took (Cleary, Prakash, & Rothauge, 2010;Zhou, Esaki, place in 2017 has exceeded the past records of death and Mitani, Xie, & Mori, 2003) and calculation of factors economic loss. The disaster claimed the lives of 152 of safety (Van Westen, 2000; van Westen, van Asch, & people in total and had resulted in an economic loss of Soeters, 2005). It considers variables such as normal about USD 223 million (UN RC, 2017), which is thereby stress, the angle of friction, pore water pressure, and a threat to the development opportunities. According to antecedent rainfall, and various hydrological and stabi- Haque et al. (2018), Rangamati, the region (shown in lity models are used together to calculate the probability Figure 1)had suﬀered the worst impacts of 2017 event, of slope failure (Montgomery et al., 1994; Van Westen, has the high future potentiality of slope failure in this 2000; Iverson, 2000; Van Westen et al.,, 2011). The region as the slopes became unstable after 2017 large- landslide susceptibility of the Rangamati district was scale landslide events. This damaging eﬀects of these produced using statistical approach due to high subjec- landslide hazards need to be reduced by taking tivity error in heuristic analysis and complex CONTACT Shamima Ferdousi Sifa sifaferdousi25@gmail.com Department of Disaster Science and Management, Faculty of Earth and Environmental Sciences, University of Dhaka, Dhaka, Bangladesh © 2019 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group on behalf of the International Water, Air & Soil Conservation Society(INWASCON). This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. GEOLOGY, ECOLOGY, AND LANDSCAPES 223 Figure 1. The four Upazilas, Rangamati Sadar, Rajasthali, Kaptai, and Kaukhali of Rangamati district, were chosen as the study area. methodology, huge data requirement in the determinis- involves the calculation of frequencies to map the future tic analysis (Van Westen, 2000). susceptible areas and is well known for producing good There are many statistical models such as informa- results with high precision (Li et al., 2016). The subjec- tion value method, frequency ratio, landslide nominal tivity problem that arises due to the classiﬁcation of the susceptibility factor (LNSF), logistic regression, multiple factor classes is also addressed here and greatly increases regression, discriminant analysis, and artiﬁcial neural the continuity of the frequency ratios. Therefore, this network (ANN). that can be used to produce landslide study aims to map the future landslide susceptible areas susceptible map but most of them are complex and using WoE and MFR and compare their results to diﬃcult to use (Ahmed & Dewan, 2017; Bathrellos, determine which statistical model describes the suscept- Kalivas, & Skilodimou, 2009). In this study, the WoE ibility of the landslide occurrence better. and MFR models will be used to map the future land- slide susceptible areas. Weights of evidence (WoE) is a 2. Materials and method Bayesian probability model that uses a log-linear form of the theorem to map future landslide susceptible areas Landslide susceptibility mapping is carried by over- (Bonham-Carter, 1994). It is a data-driven method laying the landslide inventory with the factor maps where prediction is made from prior and conditional using the WoE and MFR models, a statistical probability when the factor maps are overlaid with past approach that shows how factors contribute to the landslide events (inventory) (Agterberg et al., 1990; landslide occurrence. Therefore, a landslide inventory Bonham-Carter, 1994; Bonham-Carter, Agterberg, & consisting the location of past landslides is produced Wright, 1989). This bivariate method is commonly and combined with triggering factor maps by giving used to deﬁne the statistical association between the diﬀerent weight values derived from each model. Past factors and landslide events in several studies (van landslides were used to train the data because the Westen, Rengers, & Soeters, 2003;Lee &Choi, 2004). “past and present are keys to the future”; therefore, The problems of heterogeneity in ground conditions, the future landslide susceptible areas are estimated lack of detailed maps, and inadequate data are solved by from the past landslides and the preconditioning using this method. It can be used for large areas having factors that are responsible for the slope failure varying data types and information (Neuhäuser, 2014). assuming that the causative factors will also initiate Many researchers consider the method as robust and future landslides under similar conditions. The reliable as it avoids subjectivity and measures uncer- results of the models were later validated from the tainty associated with estimates of probability values gradient of the susceptibility index rank and cumula- (error and relative error values) (Neuhäuser, 2014). On tive landslide occurrence curve. the other hand, modiﬁed frequency ratio is used because The inventory map was produced from the SAR of its simplicity (Mohammady, Pourghasemi, & image, which has a 10-m spatial resolution and the Pradhan, 2012) and easily understandable parameters oﬀset tracking was employed to two SAR single look such as input, calculation, and output procedure, complex (SLC) images sensed on 6 June 2017 and 18 enabling the end users to apply this method with ease June 2017, which gave a 12 days temporal window. To (Lee & Pradhan, 2007; Yilmaz, 2009). This method produce the factor maps, 12.5 m ALOS PALSAR digital 224 S. F. SIFA ET AL. elevation model (DEM) and 30 m Landsat 8 optical along with decorrelated areas. The landslides have a images were used. The rainfall data are collected from speciﬁc coherence threshold. The masking of lower the Bangladesh Meteorological Department. coherence mostly discards the decorrelated areas, thus, eliminating areas of displacements other than landslides. Second, the displacement areas are digi- 2.1. Landslide inventory preparation tized manually. While digitizing, the shape of the displacements was considered. Any displacement The landslide inventory shows the previous occurrence that does not resemble the shapes of landslides was of a landslide that has occurred in the study area, and not included. Lastly, the knowledge of local region is this was prepared using SAR oﬀset tracking technique. also put into consideration. On the other hand, oﬀset Oﬀset tracking provides displacement information par- tracking has a minimum precision of 1/20th of a pixel allel to SAR satellite track (Azimuth oﬀset) and the track (Riveros et al., 2013). This makes the precision of our perpendicular (range oﬀset) (Fialko, Sandwell, Simons, analysis to be 0.5 m that means any ground move- & Rosen, 2005). The window patch intensity is mea- ment larger than 0.5 m will be detected, which is sured to ﬁnd the motion or displacement between two enough for landslide identiﬁcation. Mohammad et images of the same area (Lu, 2016). First, the master al. (2018) in his study mentioned that oﬀset tracking (pre-event) and slave image (post-event) are co-regis- shows better result when compared with subpixel tered using a digital elevation model (DEM). Afterward, correlation of optical images. Afterwards, the displa- to calculate the oﬀset, a ground control point (GCP) cements shown in the oﬀset tracking were digitized to grid is speciﬁed for the master image. The displacement produce the inventory, and some of the landslide is measured by matching frequency peaks of a patch of locations were validated with the location data pro- pixels in master and slave image. If two corresponding vided by Geological Survey of Bangladesh (2001). windows give the same frequency peaks, they are con- The study faced a limitation in inventory produc- sidered stable (Pathier et al., 2006). Each applied to tion as the freely available SAR image utilizes C band oﬀset a correlation value is computed, thus permitting (5–6 cm) wavelength. It is hard for C band wave- calculation of the corresponding matching correlation length to penetrate vegetation cover, resulting in surface (MCS), whose ith and jth elements for the some decorrelation. However, to prevent decorrela- generic pixel of range and azimuth coordinates (x, y) tion, coherence estimation was carried out to mask are the following: (Casu, Manconi, Pepe, & Lanari, out the de-correlated regions. 2011;Sun &Muller, 2016). σMS i;j ðÞ MCS i; j ¼ 2.2. Landslide triggering factor map preparation σMσS i;j Factor maps and inventories were produced from where digital elevation model (DEM), optical and radar sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ X;Y 2 image analysis. Elevation, slope angle, and aspect jj MxðÞ ; y x¼1;y¼1 σ ¼ μ M were derived from DEM, having a spatial resolution XY of 12.5 m and obtained from ALOS PALSAR satellite. sﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ On the other hand, landcover classiﬁcation and vege- X;Y 2 jj SxðÞ þ i; y þ j tation index (NDVI) were obtained from Landsat 8 x¼1;y¼1 σS ¼ μ i;j i;j Satellite image, using maximum likelihood algorithm XY based supervised classiﬁcation. Road and waterbodies are the standard deviation values for the master and were digitized and mapped from GeoEye satellite slave amplitude images, respectively, computed on image provided by ESRI, and Euclidean distance the matching window pixels. The normalized cross- was applied to determine the inﬂuence of the follow- correlation (NCC) between the master and slave ing in landslide occurrence (Mohammad et al., 2018). image amplitudes is computed. This gives the quality Data on daily rainfall of about 33 years (from 1977 to of the calculation. 2015) were collected from the Bangladesh To prepare a valid inventory, several steps have been Meteorological Department (BMD). The annual aver- taken to identify the past earth movement. First, the age rainfall is then computed from the available data, oﬀset tracking provided the ground displacements P P i y ½ i ðÞ i; j i : i ðÞ i; j i 1 1 2 2 i¼1 j¼1 NCC ¼ qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃqﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ P P P P j 2 j 2 i y i y x x i ðÞ i; j i i ðÞ i; j i 1 1 2 2 i¼1 j¼1 i¼1 j¼1 GEOLOGY, ECOLOGY, AND LANDSCAPES 225 and the calculated value is used to estimate the rain- Npix4 fall values in other regions of the study area based on the elevation. Özşahin (2014) shows that for 100 m Slope factor Npix1(nslclass) Npix3 increases in elevation, rainfall increases by 0.54 mm. The geological formation of the study area was digi- tized from the geological map of Bangladesh pro- nclass Landslide duced by the Geological Survey of Bangladesh (2001). Npix2 nslide nmap 2.3. Landslide susceptibility mapping Figure 2. Principle of the WoE model. 2.3.1. Weights of evidence (WoE) model In this method, WoE, the prior probability is calcu- Npix1 ¼ nslclass lated on the basis of past landslides assuming that it will trigger future hazardous event due to the Npix2 ¼ nslide nslclass unstable nature of the slope resulting from the slope failure. When additional information about the fac- Npix3 ¼ nclass nslclass tors are not available, prior probability give a good estimation about the possibility of landslide occur- Npix4 ¼ nmap nslide nclass þ nslclass rence by dividing the number of pixels having land- slides with a total number of pixels in the map The above variables thereby mean (Bonham-Carter, 1994). nslide = Number of pixels with landslides in the map nclass = Number of pixels in the class AreaðÞ Slide nslclass = Number of pixels with landslides in the Pprior ¼ PSfg ¼ AreaðÞ Total class nmap = Total number of pixels in the map However, when information such as presence or extent Both positive and negative weighted values of each of causal factors of landslides is available, the prior variable are then estimated to ﬁnd a degree of correla- probability is further modiﬁed to obtain a conditional tion in the presence or absence of the factor using the probability. This is done by producing a binary map (B) formula described by Bonham-Carter, 1994; Bonham- for that particular factor depending on the presence and Carter et al. (1989): absence of the variable in the map. A relationship is PBfg jS then established between the binary maps with the W ¼ ln landslide inventory, to calculate the conditional prob- fg BjS ability for a certain condition. According to Bonham- Carter (1994), the factors are conditionally independent W ¼ lnððNpix1 of each other, and the conditional probability of occur- ðÞ Npix3 þ Npix4 =ðÞ ðÞ Npix1 þ Npix2 Npix3 ring a landslide given that a particular factor unit is present there is expressed as follows: Pfg BjS W ¼ ln fg BjL PSfg \ B PSfg jB ¼ PBfg W ¼ lnððNpix2 ðÞ Npix3 þ Npix4 =ðÞ ðÞ Npix1 þ Npix2 Npix4 Npixfg S \ B In the presence of factor (B) in landslide area (S) Npixfg B gives a positive weighted value (W ), deﬁning the correlation between them. On the other hand, a In the equation landslide and the factor, variable is negative weight (W ) indicates the absence of the denoted by S and B, respectively. Pixel area in each factor. Then, a weighted contrast factor (C) is map is used in the analysis to ﬁnd about the four obtained to see how much the conditioning factor is possible combinations of probability described in spatially associated with the landslides. A ﬁnal sus- Figure 2, which are when landslides occur in the ceptibility map (LSI) hence, is produced by combin- presence of a potential conditioning factor (Npix1) ing the weighted map of each factor through an or the absence of it (Npix2), when there is no land- overlay operation. slide in the map area but the factor is present (Npix3) and the absence of both landslide and that particular C ¼ W W factor (NPix4). This is obtained by crossing the inventory of landslides with each factor map to Wmap ¼ W W þ W calculate: 226 S. F. SIFA ET AL. the start of the modeling and the remaining 75% are LSI ¼ Wmap used to train the model. The test data are then com- bined with the ﬁnal weighted map that is reclassiﬁed 2.3.2. Modiﬁed frequency ratio (MFR) models into 32 classes to compute percentage The values are Another method used in the study is modiﬁed fre- then plotted in a graph of percentage of cumulative quency ratio, which uses the same assumption as the landslide occurrence against the percentage of suscept- WoE model to ﬁnd a correlation between the land- ibility index rank to obtain a success rate curve (Jebur, slide points and preconditioning factors. Each factor Pradhan, Shafri, Yusoﬀ, & Tehrany, 2015). (F) in this method is classiﬁed into n number of classes after combining the factors with landslide 3. Results and discussion events to ﬁnd out the respective frequencies (PL and PF ) and Frequency ratioðÞ FR for the ith class of 3.1. Landslide inventory and factor maps factor by the formula given below. The studies (Lee About 420 landslide events had been identiﬁed in the & Talib, 2005; Lee & Pradhan, 2007; Deng, Li, & Tan, four Upazilas of Rangamati districts. In case of weights 2017) describe the procedure by the equation: of evidence (WoE) modeling the landslide, locations are PL FR ¼ deﬁned by the polygon attributes whereas point attri- PF butes had been used for modiﬁed frequency ratio (MFR) modeling. The conditioning and triggering fac- the frequency of landslides in the F area ¼ tors that contribute to landslide occurrence in the study the frequency of the F area area had been produced from DEM, optical images, and data from secondary sources. Factors such as elevation, the area of landslides in the F area slope, aspect, landcover, normalized diﬀerence vegeta- =the area of landslides in the study area tion index (NDVI), rain, distance from waterbody, and the area of the F area road have been taken into account and are shown in =the area of the study area Figure 3. The magnitude of FR values indicates the degree of correlation. A value greater than 1 means the strong 3.2. Landslide susceptibility map using WoE and correlation between the factors and landslides; whereas, MFR modeling a value less than 1 describes weak correlation (Pradhan, 2010). Landslide susceptibility map produced by two diﬀerent Modiﬁed frequency ratio is the same as the con- statistical models to show the degree of inﬂuence of ventional frequency ratio up to this point. The diﬀer- each causal factors with past landslide occurrence is ences arise as ratios obtained from the above given below (Figure 4). The future landslide-prone procedure is normalized to a range of 0–1, to use areas are quantiﬁed into high, moderate, low, and very this variable in precision setting and frequency statis- low areas as shown by diﬀerent color representation in tics procedures (Li et al., 2016). The relative fre- the map. To establish a clear relationship, each factor quency ðÞ RF for the ith class, thereby is calculated was divided into several classes to see which class has from the following equation suggested by the highest inﬂuence on landslide occurrence. The rela- Althuwaynee et al. (2016) and summed up together tive frequency value and weight value for diﬀerent to get the ﬁnal susceptibility map known as landslide classes of each factor map for modiﬁed frequency susceptibility indexðÞ LSI . ratio (MFR) and weights of evidence (WoE) model are also provided in Appendix (Tables A1 and A2). As two FR RF ¼ diﬀerent models are used, the parameters and resultant FR outputs are also diﬀerent from each other but relative susceptibility of each class is approximately similar for the frequeny ratio of ith class the methods. In case of the WoE method, W ,W ,and the frquency ratio of the class C factor describe the correlation and spatial association of the landslides with the factors. The positive contrast LSI ¼ RF map factor indicates positive association, that is, more occur- rences fall in the domain than the expected possibility and vice versa for negative contrast factor. The higher 2.4. Validation of landslide susceptibility models weighted value indicates higher degree of inﬂuence on The models are veriﬁed using success rate curve to see landslide occurrence. On the other hand, for MFR, the how well each model describes the susceptibility of the spatial relationship between the factors and landslide is landslide occurrence (Chung & Fabbri, 1999). To vali- deﬁned by the relative frequency value. The closer the date, 25% of the landslide distributions are separated at ratio value is to 1, the stronger is the relationship GEOLOGY, ECOLOGY, AND LANDSCAPES 227 Inventory Elevation Slope Angle Aspect Geology Landuse Pattern NDVI Annual Average Rainfall Distances from Road Distances from Waterbody Figure 3. Landslide inventory and landslide trigerring factor maps produced to understand the contributions of each factors in landslide occurrence. The inventory was produced from radar image analysis by oﬀset tracking technique. The factor maps are produced from DEM and optical image analysis. indication higher percentage of landslide area in the into 20–30, 30–40–40 degrees above has received the class than the percentage of the class in the total area. positive and higher weightage than other classes, but the The weight values derived from each model describe weightage value estimated from the MFR model shows the spatial relationships of the causative factors in con- that slope angle of 40 and above degrees has high value. tributing the landslide occurrences. The association This estimated value shows that if the slope angle is described by both the models is more or less the same. greater than 20 degrees than there is possibility of land- In this study, it has been found that slope angle classiﬁed slideoccurrenceishigh, butiftheslopeangle increasesto 228 S. F. SIFA ET AL. (a) (b) Figure 4. The landslide susceptibility map produced using (a) WoE and (b) MFR model. Landslide causative factor maps were combined with the landslide inventory produced from SAR oﬀset tracking. In the map, the possibility of landslide occurrence in a region is deﬁned by diﬀerent color representation and classifying them into high to very low susceptible classes. 40 degrees the susceptibility of landslide occurrence and water bodies show little correlation with landslide increases greatly as it has received the highest weightage. occurrence as few landslides occurred closer to the road Moreover, the weight values are derived from the past or water bodies; rather, the events were more distributed landslides where the pre-conditioning factors for which (Figure 3). This is because major roads and streams are the slope failed are used to train the dataset and carry out taken into consideration that has aﬀected the results. the modeling. It has also been seen that elevation greater Therefore, there is a further scope of study to consider than 89 m, hill facing in the direction of South-East, East, these factors where minor roads and streams will be taken West direction, the geological formation having an altera- to assess their association with the landslide occurrence. tion of sandstone and shale (Bhuban and Bokabil), the presence of bare and built up area has a higher contribu- 3.3. Comparison of the WoE and MFR models tion to landslide occurrence than other classes of each triggering factor. Vegetation index (NDVI), both factors The research involves quantiﬁcation of future land- show that areas with higher vegetation value have a slide susceptible areas of Rangamati Sadar, Kaukhali, greater probability of landslide occurrence. Upper class Kaptai, and Rajasthali Upazila, by overlaying the con- of vegetation index got higher weight value indicating ditioning factors of landslides with inventory showing that the inﬂuence of heavy rainfall plays an important role landslide events of 2017. Analyzing the two diﬀerent in slope failure as it makes the vegetated area more methods shows that MFR had overestimated the high unstable. Another reason can be indiscriminate logging and low susceptible areas and underestimated the practice. Rainfall was observed to be the main causal moderate susceptible areas than WoE. This is because factor as it received the highest weight value among the not all landslide points (from inventory) that are used other conditioning factors. The distance from road and to identify the susceptible areas of each class fall in waterbody was calculated to see the inﬂuence of the the high or moderate susceptible areas. This is due to position of road and waterbody in case of landslide the complex interaction of the causal factors. occurrence. Road is an important factor that contributes Although both of the models described the classes to landslide occurrence because construction of roads in of each factor in a similar way, the magnitude of hilly areas to ensure adequate communication with other the value assigned by each method was diﬀerent. In regions results in slope instability, since it involves mod- the WoE model, the polygon has been used as train- iﬁcation of slope by means of various activities. On the ing sets to produce the susceptibility map; whereas, other hand, water moving through the stream or rapid points are used in the case of the MFR model. excessive runoﬀ after rainfall are responsible for under- Polygons cover higher areas than points and have cutting and erosion of the slope, thereby increasing sus- higher inﬂuence over the classes and landslide points. ceptibility to sliding. Inﬁltration of the water into the soil To understand the diﬀerence between the two models also reduces the shear strength of the soil by increasing in terms of classifying the study area into diﬀerent the pore water pressure that in turn reduces the friction in classes, the results are presented in the bar diagram in the soil. However, in this study, distances from the road Figure 5. GEOLOGY, ECOLOGY, AND LANDSCAPES 229 Taking the whole area under consideration based to the weight of evidence method. However, there is a on the weight of evidence (WoE) model and modiﬁed signiﬁcant diﬀerence on identifying a moderate sus- frequency ratio (MFR) model, diﬀerent values were ceptible area where the weight of evidence (WoE) found according to the four categories namely very indicates that 62% areas are highly susceptible to low, low, moderate, high, and very high. It is evident landslide hazard and modiﬁed frequency ratio from the chart (Figure 5) that according to the mod- (MFR) model shows 37% areas that is half of the iﬁed frequency ratio, 29% areas are under very high value of the WoE model. Again, 6% areas are identi- landslide susceptible classes while 21% are according ﬁed as low and 12% areas are very low susceptible areas based on WoE whereas MFR shows a diﬀerent result that is, respectively, 20% and 14% areas. The bar diagram (Figure 6(a–d)) for each Upazila in the study area is provided below to show how much of the high, moderate, low, and very low land- slide susceptible areas are present in each Upazila, described by the two models. 3.4. Validation of the models The models need to be validated to see how well the susceptible classes are deﬁned by the models used. This is done by ﬁnding out the success rate for each model. Moreover, 75% of the polygons and points are used in the analysis to training the dataset to produce the susceptibility map. The percentage of landslide susceptibility index rank is plotted against Figure 5. Percentage of each susceptible classes in the study the percentage of cumulative landslide occurrence area deﬁned by the WoE and MFR models. to produce the success rate curve (Figure 7). About (a) (b) (c) (d) Figure 6. Percentage of susceptible classes deﬁned by the WoE and MFR models in (a) Kaptai, (b) Rajasthali, (c) Rangamati Sadar, and (d) Kaukhali Upazila of the study area. 230 S. F. SIFA ET AL. 20% of the high susceptible areas include 85% of (WoE) and modiﬁed frequency ratio (MFR) models thetotal landslidearea in caseof theWoE model to ﬁnd out which model explains the future suscept- but the MFR model includes only 20%. The WoE ibility better. An event-based (2017) landslide inven- model describes that 30% highly susceptible area tory was produced by processing radar image covers more than 99% of the total landslide area through SAR oﬀset tracking. The creation of inven- inthecaseofsuccessrate, whileMFR describes tory involves a knowledge-driven approach, which only 78% that is again lower than the WoE model. was later used in combination with the causal factors According to the success rate, the WoE model to map the future susceptible areas to see the spatial describes the landslide points better than the MFR relationship of each factor with the landslide occur- model. rence that is the factors that were used in the study As WoE describes the landslide susceptible areas was combined with previous landslides to determine better than MFR, so according to the WoE model, the condition of each factor in past landslide occur- 44% of the areas of Rangamati have been detected as rence. Later, these values were used to train the the highest and Kaptai as the lowest landslide suscep- dataset and derive the weight values to see the asso- tible areas (shown in the Figure 8). Rajasthali is the ciation of the factors and possibility of landslide second-most landslide hazard prone area as 25% and occurrence. The factors were then combined to see 33% of its area lie in the high and moderately land- how these factors that were responsible for past land- slide susceptible areas. On the other hand, total 51% slide occurrence also increase the susceptibility for of the areas of Kaukhali lie in the high and moderate future landslide occurrence, assuming that “past and susceptible areas. present are keys to the future.” The study area was then classiﬁed into high, moderate, low, and very low susceptibility classes, and comparison was made 4. Conclusion between the WoE and MFR models, describing how well each model deﬁne the classes. The results of the Rangamati district in Bangladesh got severely aﬀected Landslide susceptibility map of each method were by the landslide event of 2017. This large number of validated from the success rate curves. The percen- events had made the slope unstable, thereby increas- tage of landslide susceptibility index rank is plotted ing the possibility of slope failure in the area greater against the percentage of cumulative landslide occur- than before. The aim of the study was to produce the rence to produce the success rate curve. About 20% of landslide susceptibility map of the area based on the high susceptible areas include 85% of the total landslide inventory of 2017, using weight of evidence Figure 7. Success rate curve for weights of evidence (WoE) and modiﬁed frequency ratio (MFR) model. (a) (b) Figure 8. Percentage of (a) high and (b) moderate landslide susceptible areas of each Upazila in the study area estimated from the WoE model. GEOLOGY, ECOLOGY, AND LANDSCAPES 231 landslide area in case of the WoE model but the MFR Casu,F.,Manconi,A.,Pepe, A.,&Lanari, R. (2011). Deformation time-series generation in areas charac- model includes only 20%. The WoE model describes terized by large displacement dynamics: The SAR that 30% highly susceptible area covers more than amplitude pixel-oﬀset SBAS technique. IEEE 99% of the total landslide area in the case of success Transactions on Geoscience and Remote Sensing, 49 rate, while MFR describes only 78% that is again (7), 2752–2763. lower than the WoE model. As WoE explains the Chen, L., van Westen, C. J., Hussin, H., Ciurean, R. L., Turkington, T., Chavarro-Rincon, D., & Shrestha, D. P. results better than MFR, the model was later used to (2016). Integrating expert opinion with modelling for ﬁnd out which upazila of the study area is at high quantitative multi-hazard risk assessment in the Eastern landslide susceptible areas compared to the other. Italian Alps. Geomorphology, 273(October), 150–167. Rangamati Sadar upazila has been identiﬁed to have Chung, C.-J. F., & Fabbri, A. G. (1999). Probabilistic prediction high susceptibility toward landslide hazard as 46% models for Landslide Hazard mapping. Photogrammetric and 33% of its area fall under high and moderate Engineering & Remote Sensing, 65(12), 1389–1399. Cleary, P. W., Prakash, M., & Rothauge, K. (2010). susceptible classes, respectively. Kaukhali, Rajasthali, Combining digital terrain and surface textures with and Kaptai upazila have 23%, 25%, 8% high, and 28%, large-scale particle-based computational models to pre- 33%, 10% moderate susceptible areas, respectively. dict dam collapse and landslide events. International The results obtained from the research, thereby, Journal of Image and Data Fusion, 1(4), 337–357. might help the policymakers to take appropriate miti- Deng, X., Li, L., & Tan, Y. (2017). Validation of spatial prediction models for landslide susceptibility mapping gation measures to prevent the severity of the land- by considering structural similarity. ISPRS International slide hazard consequences in the study area. Journal of Geo-Information, 6(4), 103. Fialko, Y., Sandwell, D., Simons, M., & Rosen, P. (2005). Three-dimensional deformation caused by the Bam, Disclosure statement Iran, earthquake and the origin of shallow slip deﬁcit. Nature, 435(7040), 295–299. No potential conﬂict of interest was reported by the Geological Survey of Bangladesh. (2001). Description of authors. map units. Geological Survey of Bangladesh (GSB) Haque, D. M. E., Sifa, S. F., Mahmud, T. & Tarin, M. A. (2018). Landslide susceptibility assessment based on ORCID satellite image processing and bi-variate statistical mod- eling for Rangamati district, Bangladesh. Journal of Dewan Mohammad Enamul Haque http://orcid.org/ Physics, 93–106. doi:10.13140/RG.2.2.16092.03203 0000-0001-5439-5528 Horelli, J. A. (2005). Landslides in Hong Kong. University of Helsinki. Intarawichian, N., & Dasananda, S. (2010). Analytical hierarchy References process for landslide susceptibility mapping in lower Mae Chaem Watershed, Northern Thailand. Suranaree Journal of Agterberg, F. P., Bonham-Carter, G. F., & Wright, D. F. Science and Technology, 7(3). Retrieved from http://www. (1990). Statistical pattern integration for mineral thaiscience.info/journals/Article/SJST/10890515.pdf exploration. In D. F. Gaal & G. Merriam (Eds.), Iverson, R. M. (2000). Landslide triggering by rain inﬁltration. Computer application in resource estimation: Prediction Water Resources Research, 36(7), 1897–1910. and assessment for metals and petroleum (pp. 1–21). Jebur,M.N.,Pradhan,B.,Shafri,H.Z.M.,Yusoﬀ,Z.M., & Pergamon Press: Oxford. Tehrany, M. S. (2015). An integrated user-friendly ArcMAP Ahmed, B. (2015). Landslide susceptibility mapping using tool for bivariate statistical modelling in geoscience applica- multi-criteria evaluation techniques in Chittagong tions. GeoscientiﬁcModel Development, 8(3), 881–891. Metropolitan. Journal of the International Consortium Kanungo,D. P.,Arora, M.K.,Sarkar,S., &Gupta,R. P.(2009). on Landslides, Springer Link, 12(2015), 1077–1095. Landslide susceptibility zonation (LSZ) mapping - a review. Ahmed, B., & Dewan, A. (2017). Application of bivariate Journal of South Asia Disaster Studies, 2(1), 81–106. and multivariate statistical techniques in landslide sus- Lee,S.,& Choi,J.(2004). Landslide susceptibility mapping ceptibility modeling in Chittagong city corporation, using GIS and the weight-of-evidence model. International Bangladesh, Remote Sensing, 9, 304, (May 2014). Journal of Geographical Information Science, 18(8), 789–814. Althuwaynee, O. F., Pradhan, B., & Lee, S. (2016). A novel Lee, S., & Pradhan, B. (2007). Landslide hazard mapping at integrated model for assessing landslide susceptibility map- Selangor, Malaysia using frequency ratio and logistic ping using CHAID and AHP pair-wise comparison. regression models. Landslides, 4,33–41. International Journal of Remote Sensing, 37(5), 1190–1209. Lee, S., & Talib, J. A. (2005). Probabilistic landslide suscept- Bathrellos, G. D., Kalivas, D. P., & Skilodimou, H. D. ibility and factor eﬀect analysis. Environmental Geology, 47 (2009). GIS-based landslide susceptibility mapping mod- (7), 982–990. els applied to natural and urban planning in Trikala, Li, L., Lan, H., Guo, C., Zhang, Y., Li, Q., & Wu, Y. (2016). Central Greece. Estudios Geológicos, 65(1), 49–65. A modiﬁed frequency ratio method for landslide sus- Bonham-Carter, G. F. (1994). Geographic information systems ceptibility assessment. Landslides, 14(2), 727–741. for geoscientists, modeling with GIS, 302–334. Oxford: Lu,J.(2016, August). Sentinel-1 Toolbox oﬀset tracking tutorial. Pergamon Press. ESA Bonham-Carter, G. F., Agterberg, F. P., & Wright, D. F. Mohammady, M., Pourghasemi, H. R., & Pradhan, B. (2012). (1989). Weights of evidence modelling: A new approach Landslide susceptibility mapping at Golestan Province, Iran: to mapping mineral potential. Statistical Applications in A comparison between frequency ratio, Dempster-Shafer, Earth Science, 89(9), 171–183. 232 S. F. SIFA ET AL. and weights-of-evidence models. Journal of Asian Earth Van Westen, C. J. (2000). The modelling of landslide hazards Sciences, 61,221–236. using gis (PDF download available). Surveys in Geophysics, Montgomery,D.R.,&Dietrich,W.E.(1994). A physically 21, 241–255. Retrieved from https://www.researchgate.net/ based model for the topographic control on shallow land- publication/233815072_The_Modelling_Of_Landslide_ sliding. Water Resources Research, 30(4), 1153–1171. Hazards_Using_Gis Neuhäuser, B. (2014). Landslide susceptibility and climate vanWesten, C. J.,Ghosh,S., Jaiswal, P.,Martha, T. R.,& change scenarios in ﬂysch areas of the Eastern alps. Kuriakose, S. L. (2011). From landslide inventories to land- Geomorphology, 86(1-2), 12–24 slide risk assessment; an attempt to support methodological Özşahin, E. (2014). Using geographic information system development in India. In the Second World Landslide Forum, and analytical hierarchy process in landslide susceptibil- 1,3–20. ity analysis in Tekirdağ province. The Central and vanWesten, C. J.,Rengers,N., &Soeters,R.(2003). Use Eastern European Online Library, 137–144. doi:10.7366/ of geomorphological information in indirect landslide 1509499516305 susceptibility assessment. Natural Hazards, 30,399– Pathier,E.,Fielding,E. J.,Wright,T. J.,Walker,R.,Parsons,B. 419. Kluwer Academic Publishers. Retrieved from. E., & Hensley, S. (2006). Displacement ﬁeld and slip distri- https://research.utwente.nl/en/publications/use-of-geo bution of the 2005 Kashmir earthquake from SAR imagery. morphological-information-in-indirect-landslide- Geophysical Research Letters, 33(20), 1–5. suscept Pradhan, B. (2010). Landslide susceptibility mapping of van Westen, C. J., van Asch, T. W. J., & Soeters, R. (2005). a catchment area using frequency ratio, fuzzy logic Landslide hazard and risk zonation - Why is it still so and multivariate logistic regression approaches. diﬃcult? Bulletin of Engineering Geology and the Journal of the Indian Society of Remote Sensing, 38 Environment, 65(2), 167–184. (2), 301–320. Yilmaz, I. (2009). Landslide susceptibility mapping using Sarwar, M. G. M. (2008). Landslide tragedy of Bangladesh. In frequency ratio, logistic regression, artiﬁcial neural net- The ﬁrst world landslide forum. Tokyo. Retrieved from works and their comparison: A case study from Kat http://www.academia.edu/3588839/Landslide_Tragedy_of_ landslides (Tokat-Turkey). Computers and Geosciences, Bangladesh 35(6), 1125–1138. Sun, L., & Muller, J. P. (2016). Evaluation of the use of sub- Yilmaz, I., & Yildirim, M. (2006). Structural and geomor- pixel oﬀset tracking techniques to monitor landslides in phological aspects of the Kat landslides (Tokat-Turkey) densely vegetated steeply sloped areas. Remote Sensing, 8 and susceptibility mapping by means of GIS. (8), 659. Environmental Geology, 50(4), 461–472. UN RC. (2017). Bangladesh: HCTT response plan (June- Zhou,G.,Esaki,T.,Mitani,Y.,Xie,M.,&Mori.,J.(2003). December 2017) - Bangladesh | ReliefWeb. Retrieved Spatial probabilistic modeling of slope failure using an inte- from https://reliefweb.int/report/bangladesh/bangla grated GIS Monte Carlo simulation approach. Engineering desh-hctt-response-plan-june-december-2017 Geology, 68(3–4), 373–386. GEOLOGY, ECOLOGY, AND LANDSCAPES 233 Appendix Table A1. Relative frequency ratio values for the causative factors using MFR. No. of points % of points cls_area % of cls_area Ratio(+) RF Integer NDVI −0.236,372 0 0 888,624 12.65,263 0 0 0 0.087784–0.285,225 2031.25 4.153,355 403,133 5.739,989 0.723,582 0.237,745 24 0.285,225–0.396,459 13,125 26.83,706 1,675,393 23.855 1.125,008 0.36,964 37 0.396,459–0.56,053 33,750 69.00958 4,056,047 57.75,183 1.194,933 0.392,616 49 AVERAGE ANNUAL RAINFALL (mm) 2559–2560 mm 48,906.25 100 7,023,268 100.0005 0.999,995 0.336,122 34 ASPECT ﬂat 156.25 0.319,489 866,476 12.33,728 0.025896 0.002664 0 north 1562.5 3.194,888 388,229 5.52,778 0.57,797 0.059461 6 north-east 2968.75 6.070288 703,459 10.01617 0.606,049 0.06235 7 east 5937.5 12.14,058 751,505 10.70,027 1.134,605 0.116,728 12 south-east 9843.75 20.1278 781,553 11.1281 1.808,735 0.186,082 19 south 8281.25 16.93,291 795,619 11.32,838 1.494,733 0.153,778 15 south-west 7031.25 14.377 879,217 12.51,869 1.148,443 0.118,152 11 west 5156.25 10.54,313 838,870 11.94,421 0.882,698 0.090812 12 north-west 6406.25 13.09904 732,429 10.42,865 1.256,062 0.129,223 13 north 1562.5 3.194,888 285,879 4.070474 0.784,893 0.08075 8 LANDUSE PATTERN Buit-up area 2343.75 4.792,332 633,618 9.021739 0.531,198 0.190,589 20 Bare Soil 4218.75 8.626,198 612,288 8.718,033 0.989,466 0.355,011 36 Water 0 0 976,037 13.89,725 0 0 0 Vegetation 42,343.75 86.58,147 4,801,360 68.36,393 1.266,479 0.454,401 45 DISTANCES from WATERBODIES (meter) 0–10 m 0 0 1,069,783 15.23,205 0 0 0 10–20 m 0 0 161,065 2.293,316 0 0 0 above 20 m 48,906.25 100 5,792,420 82.47,509 1.212,487 0.269,455 27 GEOLOGY Tipam 7343.75 15.01597 1,103,472 15.71,173 0.955,717 0.140,395 14 Bokabil 22,343.75 45.6869 1,949,704 27.76,076 1.645,736 0.241,759 24 Bhuban 5937.5 12.14,058 484,111 6.892,991 1.761,293 0.258,734 26 Girujan Clay 937.5 1.916,933 246,595 3.511,131 0.545,959 0.080201 8 Dupi Tila 5625 11.5016 979,936 13.95,277 0.824,324 0.121,093 12 Dihing and Dupi Tila 5000 10.22,364 792,991 11.29,096 0.905,471 0.133,014 13 lake 1718.75 3.514,377 1,461,794 20.81,368 0.168,849 0.024804 3 ava 0 0 4633 0.065967 0 0 0 SLOPE ANGLE (degrees) 0–10 degrees 13,750 28.11,502 4,098,000 58.34,917 0.481,841 0.042613 4 10–20 degrees 24,062.5 49.20,128 2,308,687 32.87,213 1.496,748 0.132,368 13 20–30 degrees 10,156.25 20.76,677 535,298 7.621,814 2.72,465 0.24,096 24 30–40 degrees 468.75 0.958,466 70,821 1.008381 0.9505 0.084059 9 40 and above 468.75 0.958,466 10,430 0.148,507 5.653,738 0.499,999 50 DISTANCES from ROADS (meter) 0–10 m 0 0 16,274 0.231,717 0 0 0 10–20 m 0 0 32,162 0.457,937 0 0 0 above 20 m 48,906.25 100 6,974,832 99.3108 1.00694 0.228,959 23 ELEVATION (meter) −80–0 m 3750 7.667,732 3,539,030 50.3903 0.152,167 0.013808 1 0–15 m 10,000 20.44,728 1,051,301 14.9689 1.365,985 0.123,955 13 15–45 m 18,437.5 37.69,968 1,372,905 19.54,804 1.928,566 0.175,006 18 45–89 m 8750 17.89,137 575,766 8.198,016 2.182,403 0.19,804 20 89–151 m 5781.25 11.82,109 255,508 3.638,038 3.249,303 0.294,855 29 151–237 m 2031.25 4.153,355 161,241 2.295,822 1.809,093 0.164,164 16 237–428 m 156.25 0.319,489 67,485 0.960,882 0.332,495 0.030172 3 234 S. F. SIFA ET AL. Table A2. Weighted values for the causative factors using the WoE model. SLIDEN NPIX NMAP NPIXACT NSLIDE NCLS NSCLS NPIX1 NPIX2 NPIX3 NPIX4 WPLUS WNEG C WMP NDVI 0.285,225–0.396,459 unknown 1,673,116 7,023,197 0 9490 1,675,393 2277 2277 7213 1,673,116 5,340,591 0.0058 −0.0018 0.00762 −0.61 0.396,459–0.56,053 unknown 4,049,040 7,023,197 0 9490 4,056,047 7007 7007 2483 4,049,040 2,964,667 0.24,606 −0.4797 0.72,573 0.10,813 0.087784–0.285,225 unknown 402,928 7,023,197 0 9490 403,133 205 205 9285 402,928 6,610,779 −0.9781 0.03733 −1.0154 −1.633 −0.236,372 unknown 888,623 7,023,197 0 9490 888,624 1 1 9489 888,623 6,125,084 −7.092 0.13,537 −7.2274 −7.845 0.396,459–0.56,053 active 7007 7,023,197 7007 9490 4,056,047 7007 7007 2483 4,049,040 2,964,667 0.24,606 −0.4797 0.72,573 0.10,813 0.285,225–0.396,459 active 2277 7,023,197 2277 9490 1,675,393 2277 2277 7213 1,673,116 5,340,591 0.0058 −0.0018 0.00762 −0.61 0.087784–0.285,225 active 205 7,023,197 205 9490 403,133 205 205 9285 402,928 6,610,779 −0.9781 0.03733 −1.0154 −1.633 −0.236,372 active 1 7,023,197 1 9490 888,624 1 1 9489 888,623 6,125,084 −7.092 0.13,537 −7.2274 −7.845 AVERAGE ANNUAL RAINFALL (mm) 2559.25–2560.27 unknown 7,013,778 7,023,268 0 9490 7,023,268 9490 9490 0.000001 7,013,778 0.000001 0 6.60,539 −6.6054 6.6054 2559.25–2560.27 active 9490 7,023,268 9490 9490 7,023,268 9490 9490 0.000001 7,013,778 0.000001 0 6.60,539 −6.6054 6.6054 ASPECT north unknown 387,967 7,023,236 0 9490 388,229 262 262 9228 387,967 6,625,779 −0.6949 0.02891 −0.7239 −0.9244 North unknown 285,588 7,023,236 0 9490 285,879 291 291 9199 285,588 6,728,158 −0.2836 0.01043 −0.294 −0.4945 North-east unknown 702,814 7,023,236 0 9490 703,459 645 645 8845 702,814 6,310,932 −0.3882 0.0352 −0.4234 −0.6239 East unknown 750,141 7,023,236 0 9490 751,505 1364 1364 8126 750,141 6,263,605 0.29,555 −0.0421 0.3376 0.13,708 South-east unknown 779,349 7,023,236 0 9490 781,553 2204 2204 7286 779,349 6,234,397 0.7372 −0.1465 0.8837 0.68,318 South-west unknown 878,162 7,023,236 0 9490 879,217 1055 1055 8435 878,162 6,135,584 −0.1189 0.01592 −0.1348 −0.3353 South unknown 793,379 7,023,236 0 9490 795,619 2240 2240 7250 793,379 6,220,367 0.73,556 −0.1492 0.88,476 0.68,424 West unknown 838,267 7,023,236 0 9490 419,435 603 603 8887 418,832 6,594,914 0.06208 −0.0041 0.06616 −0.1344 North-west unknown 731,617 7,023,236 0 9490 732,429 812 812 8678 731,617 6,282,129 −0.1981 0.02072 −0.2188 −0.4194 ﬂat unknown 866,462 7,023,236 0 9490 866,476 14 14 9476 866,462 6,147,284 −4.4277 0.13,039 −4.5581 −4.7586 South-east active 2204 7,023,236 2204 9490 781,553 2204 2204 7286 779,349 6,234,397 0.7372 −0.1465 0.8837 0.68,318 East active 1364 7,023,236 1364 9490 751,505 1364 1364 8126 750,141 6,263,605 0.29,555 −0.0421 0.3376 0.13,708 North-west active 812 7,023,236 812 9490 732,429 812 812 8678 731,617 6,282,129 −0.1981 0.02072 −0.2188 −0.4194 West active 603 7,023,236 603 9490 419,435 603 603 8887 418,832 6,594,914 0.06208 −0.0041 0.06616 −0.1344 North active 291 7,023,236 291 9490 285,879 291 291 9199 285,588 6,728,158 −0.2836 0.01043 −0.294 −0.4945 South active 2240 7,023,236 2240 9490 795,619 2240 2240 7250 793,379 6,220,367 0.73,556 −0.1492 0.88,476 0.68,424 North-east active 645 7,023,236 645 9490 703,459 645 645 8845 702,814 6,310,932 −0.3882 0.0352 −0.4234 −0.6239 South-west active 1055 7,023,236 1055 9490 879,217 1055 1055 8435 878,162 6,135,584 −0.1189 0.01592 −0.1348 −0.3353 north active 262 7,023,236 262 9490 388,229 262 262 9228 387,967 6,625,779 −0.6949 0.02891 −0.7239 −0.9244 ﬂat active 14 7,023,236 14 9490 866,476 14 14 9476 866,462 6,147,284 −4.4277 0.13,039 −4.5581 −4.7586 LANDUSE PATTERN Vegetation unknown 4,793,182 7,023,303 0 9490 4,801,360 8178 8178 1312 4,793,182 2,220,631 0.2319 −0.8286 1.06049 −0.2138 Built Up Area unknown 633,238 7,023,303 0 9490 633,618 380 380 9110 633,238 6,380,575 −0.813 0.05376 −0.8668 −2.1411 Bare Soil unknown 611,357 7,023,303 0 9490 612,288 931 931 8559 611,357 6,402,456 0.11,822 −0.0121 0.13,028 −1.144 Water unknown 976,036 7,023,303 0 9490 976,037 1 1 9489 976,036 6,037,777 −7.1859 0.14,974 −7.3356 −8.6099 Vegetation active 8178 7,023,303 8178 9490 4,801,360 8178 8178 1312 4,793,182 2,220,631 0.2319 −0.8286 1.06049 −0.2138 Bare Soil active 931 7,023,303 931 9490 612,288 931 931 8559 611,357 6,402,456 0.11,822 −0.0121 0.13,028 −1.144 Built Up Area active 380 7,023,303 380 9490 633,618 380 380 9110 633,238 6,380,575 −0.813 0.05376 −0.8668 −2.1411 Water active 1 7,023,303 1 9490 976,037 1 1 9489 976,036 6,037,777 −7.1859 0.14,974 −7.3356 −8.6099 DISTANCES from WATERBODIES (meter) above 20 unknown 5,783,030 7,023,268 0 9490 5,792,418 9388 9388 102 5,783,030 1,230,748 0.18,214 −2.7928 2.97,491 −2.2548 10–20 m unknown 160,896 7,023,268 0 9490 160,959 63 63 9427 160,896 6,852,882 −1.24 0.01655 −1.2565 −6.4862 0–10 m unknown 1,069,852 7,023,268 0 9490 1,069,891 39 39 9451 1,069,852 5,943,926 −3.6141 0.16,139 −3.7755 −9.0051 above 20 active 9388 7,023,268 9388 9490 5,792,418 9388 9388 102 5,783,030 1,230,748 0.18,214 −2.7928 2.97,491 −2.2548 10–20 m active 63 7,023,268 63 9490 160,959 63 63 9427 160,896 6,852,882 −1.24 0.01655 −1.2565 −6.4862 0–10 m active 39 7,023,268 39 9490 1,069,891 39 39 9451 1,069,852 5,943,926 −3.6141 0.16,139 −3.7755 −9.0051 (Continued) GEOLOGY, ECOLOGY, AND LANDSCAPES 235 Table A2. (Continued). SLIDEN NPIX NMAP NPIXACT NSLIDE NCLS NSCLS NPIX1 NPIX2 NPIX3 NPIX4 WPLUS WNEG C WMP GEOLOGY Bhuban unknown 483,380 7,023,236 0 9490 484,111 731 731 8759 483,380 6,530,366 0.11,124 −0.0087 0.11,999 −0.1195 Bokabil unknown 1,944,924 7,023,236 0 9490 1,949,704 4780 4780 4710 1,944,924 5,068,822 0.59,685 −0.3758 0.97,264 0.73,313 Tipam unknown 1,102,360 7,023,236 0 9490 1,103,472 1112 1112 8378 1,102,360 5,911,386 −0.2937 0.04636 −0.34 −0.5795 Predominantly Water unknown 1,461,443 7,023,236 0 9490 1,461,794 351 351 9139 1,461,443 5,552,303 −1.7288 0.19,597 −1.9247 −2.1642 Dihing and Dupi Tila unknown 792,211 7,023,236 0 9490 792,991 780 780 8710 792,211 6,221,535 −0.3179 0.03409 −0.352 −0.5915 Dihing and Dupi Tila active 780 7,023,236 780 9490 792,991 780 780 8710 792,211 6,221,535 −0.3179 0.03409 −0.352 −0.5915 Dupi Tila unknown 978,394 7,023,236 0 9490 979,936 1542 1542 7948 978,394 6,035,352 0.15,256 −0.0271 0.17,964 −0.0599 Bokabil active 4780 7,023,236 4780 9490 1,949,704 4780 4780 4710 1,944,924 5,068,822 0.59,685 −0.3758 0.97,264 0.73,313 Predominantly Water active 351 7,023,236 351 9490 1,461,794 351 351 9139 1,461,443 5,552,303 −1.7288 0.19,597 −1.9247 −2.1642 Tipam active 1112 7,023,236 1112 9490 1,103,472 1112 1112 8378 1,102,360 5,911,386 −0.2937 0.04636 −0.34 −0.5795 Dupi Tila active 1542 7,023,236 1542 9490 979,936 1542 1542 7948 978,394 6,035,352 0.15,256 −0.0271 0.17,964 −0.0599 Girujan Clay unknown 246,401 7,023,236 0 9490 246,595 194 194 9296 246,401 6,767,345 −0.5415 0.01511 −0.5566 −0.7961 ava unknown 4633 7,023,236 0 9490 4633 0 0.001 9490 4633 7,009,113 −8.7433 0.00066 −8.744 −8.9835 Bhuban active 731 7,023,236 731 9490 484,111 731 731 8759 483,380 6,530,366 0.11,124 −0.0087 0.11,999 −0.1195 Girujan Clay active 194 7,023,236 194 9490 246,595 194 194 9296 246,401 6,767,345 −0.5415 0.01511 −0.5566 −0.7961 SLOPE ANGLE (degrees) 0–10 degrees unknown 4,095,289 7,023,236 0 9490 4,098,000 2711 2711 6779 4,095,289 2,918,457 −0.7149 0.54,041 −1.2553 −1.0027 10–20 degrees unknown 2,304,231 7,023,236 0 9490 2,308,687 4456 4456 5034 2,304,231 4,709,515 0.35,714 −0.2357 0.59,288 0.84,546 20–30 degrees unknown 533,386 7,023,236 0 9490 535,298 1912 1912 7578 533,386 6,480,360 0.97,429 −0.1459 1.12,019 1.37,277 30–40 degrees unknown 70,461 7,023,236 0 9490 70,821 360 360 9130 70,461 6,943,285 1.32,868 −0.0286 1.35,725 1.60,984 40 and above unknown 10,379 7,023,236 0 9490 10,430 51 51 9439 10,379 7,003,367 1.28,967 −0.0039 1.29,358 1.54,617 0–10 degrees active 2711 7,023,236 2711 9490 4,098,000 2711 2711 6779 4,095,289 2,918,457 −0.7149 0.54,041 −1.2553 −1.0027 10–20 degrees active 4456 7,023,236 4456 9490 2,308,687 4456 4456 5034 2,304,231 4,709,515 0.35,714 −0.2357 0.59,288 0.84,546 20–30 degrees active 1912 7,023,236 1912 9490 535,298 1912 1912 7578 533,386 6,480,360 0.97,429 −0.1459 1.12,019 1.37,277 30–40 degrees active 360 7,023,236 360 9490 70,821 360 360 9130 70,461 6,943,285 1.32,868 −0.0286 1.35,725 1.60,984 40 and above active 51 7,023,236 51 9490 10,430 51 51 9439 10,379 7,003,367 1.28,967 −0.0039 1.29,358 1.54,617 DISTANCES from ROADs (meter) above 20 m unknown 6,965,328 7,023,236 0 9490 6,974,775 9447 9447 43 6,965,328 48,418 0.00239 −0.421 0.42,342 −0.4139 above 20 m active 9447 7,023,236 9447 9490 6,974,775 9447 9447 43 6,965,328 48,418 0.00239 −0.421 0.42,342 −0.4139 10–20 m unknown 32,151 7,023,236 0 9490 32,180 29 29 9461 32,151 6,981,595 −0.4055 0.00153 −0.407 −1.2444 0–10 m unknown 16,267 7,023,236 0 9490 16,281 14 14 9476 16,267 6,997,479 −0.4524 0.00085 −0.4533 −1.2906 0–10 m active 14 7,023,236 14 9490 16,281 14 14 9476 16,267 6,997,479 −0.4524 0.00085 −0.4533 −1.2906 10–20 m active 29 7,023,236 29 9490 32,180 29 29 9461 32,151 6,981,595 −0.4055 0.00153 −0.407 −1.2444 ELEVATION (meter) 0–15 m unknown 1,049,074 7,023,236 0 9490 1,051,301 2227 2227 7263 1,049,074 5,964,672 0.45,038 −0.1054 0.55,581 0.77,643 15–45 m unknown 1,370,013 7,023,236 0 9490 1,372,905 2892 2892 6598 1,370,013 5,643,733 0.44,476 −0.1461 0.59,091 0.81,153 −80–0m unknown 3,537,566 7,023,236 0 9490 3,539,030 1464 1464 8026 3,537,566 3,476,180 −1.1846 0.53,439 −1.719 −1.4984 45–89 m unknown 574,158 7,023,236 0 9490 575,766 1608 1608 7882 574,158 6,439,588 0.72,748 −0.1003 0.82,773 1.04834 89–151 m unknown 254,639 7,023,236 0 9490 255,508 869 869 8621 254,639 6,759,107 0.92,513 −0.0591 0.98,419 1.20,481 −80–0m active 1464 7,023,236 1464 9490 3,539,030 1464 1464 8026 3,537,566 3,476,180 −1.1846 0.53,439 −1.719 −1.4984 0–15 m active 2227 7,023,236 2227 9490 1,051,301 2227 2227 7263 1,049,074 5,964,672 0.45,038 −0.1054 0.55,581 0.77,643 237–428 m unknown 67,410 7,023,236 0 9490 67,485 75 75 9415 67,410 6,946,336 −0.1957 0.00172 −0.1974 0.02322 151–237 m unknown 160,886 7,023,236 0 9490 161,241 355 355 9135 160,886 6,852,860 0.48,906 −0.0149 0.50,398 0.72,459 15–45 m active 2892 7,023,236 2892 9490 1,372,905 2892 2892 6598 1,370,013 5,643,733 0.44,476 −0.1461 0.59,091 0.81,153 89–151 m active 869 7,023,236 869 9490 255,508 869 869 8621 254,639 6,759,107 0.92,513 −0.0591 0.98,419 1.20,481 m active 1608 7,023,236 1608 9490 575,766 1608 1608 7882 574,158 6,439,588 0.72,748 −0.1003 0.82,773 1.04834 45–89 151–237 m active 355 7,023,236 355 9490 161,241 355 355 9135 160,886 6,852,860 0.48,906 −0.0149 0.50,398 0.72,459 237–428 m active 75 7,023,236 75 9490 67,485 75 75 9415 67,410 6,946,336 −0.1957 0.00172 −0.1974 0.02322

Journal

Geology Ecology and Landscapes – Taylor & Francis

Published: Jul 2, 2020

Keywords: Landslide; susceptibility; inventory; weights of evidence (WoE); modified frequency ratio (MFR)

References

Landslide susceptibility mapping using multi-criteria evaluation techniques in Chittagong Metropolitan
Application of bivariate and multivariate statistical techniques in landslide susceptibility modeling in Chittagong city corporation, Bangladesh
A novel integrated model for assessing landslide susceptibility mapping using CHAID and AHP pair-wise comparison
GIS-based landslide susceptibility mapping models applied to natural and urban planning in Trikala, Central Greece
Weights of evidence modelling: A new approach to mapping mineral potential
Deformation time-series generation in areas characterized by large displacement dynamics: The SAR amplitude pixel-offset SBAS technique
Integrating expert opinion with modelling for quantitative multi-hazard risk assessment in the Eastern Italian Alps
Probabilistic prediction models for Landslide Hazard mapping
Combining digital terrain and surface textures with large-scale particle-based computational models to predict dam collapse and landslide events
Validation of spatial prediction models for landslide susceptibility mapping by considering structural similarity
Three-dimensional deformation caused by the Bam, Iran, earthquake and the origin of shallow slip deficit
Analytical hierarchy process for landslide susceptibility mapping in lower Mae Chaem Watershed, Northern Thailand
Landslide triggering by rain infiltration
An integrated user-friendly ArcMAP tool for bivariate statistical modelling in geoscience applications
Landslide susceptibility zonation (LSZ) mapping - a review
Landslide susceptibility mapping using GIS and the weight-of-evidence model
Landslide hazard mapping at Selangor, Malaysia using frequency ratio and logistic regression models
Probabilistic landslide susceptibility and factor effect analysis
A modified frequency ratio method for landslide susceptibility assessment
Landslide susceptibility mapping at Golestan Province, Iran: A comparison between frequency ratio, Dempster-Shafer, and weights-of-evidence models
A physically based model for the topographic control on shallow landsliding
Using geographic information system and analytical hierarchy process in landslide susceptibility analysis in Tekirdağ province
Displacement field and slip distribution of the 2005 Kashmir earthquake from SAR imagery
Landslide susceptibility mapping of a catchment area using frequency ratio, fuzzy logic and multivariate logistic regression approaches
Evaluation of the use of sub-pixel offset tracking techniques to monitor landslides in densely vegetated steeply sloped areas
Bangladesh: HCTT response plan (June-December 2017) - Bangladesh | ReliefWeb
The modelling of landslide hazards using gis (PDF download available)
From landslide inventories to landslide risk assessment; an attempt to support methodological development in India
Use of geomorphological information in indirect landslide susceptibility assessment
Landslide hazard and risk zonation - Why is it still so difficult?
Landslide susceptibility mapping using frequency ratio, logistic regression, artificial neural networks and their comparison: A case study from Kat landslides (Tokat-Turkey)
Structural and geomorphological aspects of the Kat landslides (Tokat-Turkey) and susceptibility mapping by means of GIS
Spatial probabilistic modeling of slope failure using an integrated GIS Monte Carlo simulation approach

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Event-based landslide susceptibility mapping using weights of evidence (WoE) and modified frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh

Event-based landslide susceptibility mapping using weights of evidence (WoE) and modified frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Event-based landslide susceptibility mapping using weights of evidence (WoE) and modified frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh

Event-based landslide susceptibility mapping using weights of evidence (WoE) and modified frequency ratio (MFR) model: a case study of Rangamati district in Bangladesh

References

Abstract

Journal

Recommended Articles

References

Our policy towards the use of cookies