A Probabilistic Landslide Risk Assessment (LRA) on NH31A and Settlement in Rorachu Watershed, East Sikkim, India by using Bivariate Models and Geospatial Techniques


 The Sikkim Himalaya has been recognized as region enormously susceptible slope instability. The NH 31A road falls with east Sikkim Himalaya which has highly deformed by numerous landslide events. Over the few years the NH 31A road sections and settlement with its surrounding areas are invaded by landslide events. To resolve the problem connected to landslide, landslide susceptibility zones (LSZ) and landslide risk assessment (LRA) is an urgent and safe mitigation measure to helping the strategic planning for local people. The present study is an endeavor to take advantage of bivariate statistical method called frequency ratio (FR), information value (IV) and certainty factor (CF) analysis for LSZ and LRA map and attempt to get out the triggering factors for the LSI and LRA in Rorachu watershed, East Sikkim. The landslide inventory map was made by the more premature reports, aerial photograph, Google Earth image and multiple field visits. A total 153 landslides location were mapped using GIS software and divided into 70 % (107) for training data for the modeling using FR, IV and CF models and remaining 30 % (46) were used for validating the models. The thirteen landslide causative factors Geology, Soil, Elevations, Slope, Curvature, drainage Density (DD), Road Density (RD), Rainfall, Normalize Difference Vegetation Index (NDVI), Land Use Land Cover (LULC), Topographic Position Index (TPI), Stream Power Index (SPI) and Topographic Wetness Index (TWI) were extracted from spatial database for the LSZ mapping using FR, IV and CF models. The landslide susceptibility zonation (LSM) map also tested by the histogram and density plot, this is elicited most of the triggering factors for the landslides in Rorachu watershed. The results have been showing that the slope (35o to 50o), elevations (2,500 – 4,100 m) and rainfall (2000- 2,500 mm and 3,000 – 3,300 mm) is the intensest concentration and density for the landslides. The predictive frequency ratio (FR), information value (IV) and certainty factor (CF) model has been validated by receive operating characteristics (ROC) curve, Success rate curve (SRC) and landslide density (LD) method analysis. The result shows that AUC for success rate curves (SRC) are 0.925 (92.50 %), 0.846 (84.60 %) and 0.868 (86.80 %), respectively for frequency ratio (FR), information value (IV) and certainty factor (CF) models. And the result shows that AUC for prediction rates are 0.828 (82.80 %), 0.750 (75 %) and 0.836 (83.60 %), respectively for the FR, IV and CF models. The element-at-risk (Settlement and Road) is revealed the landslide risk assessment (LRA) have been showing that the most significant risk of settlements areas by the model of FR (9%), IV (38.59%) and CF (20.90%) and the most significant risk of NH 31A road is FR (20.72%), IV (40.91%) and CF (18.78%). These landslide susceptibility maps and landslide risk assessment (LRA) map can be used for the development of land use planning strategies, saves human loss and important for the planners and mitigation purpose. So remarkable attention should be taken into consideration for the highway construction, deforestation and urbanization.


Introduction
Disaster causes by Landslides are one of the most significant geo-environmental disasters of 67 mountain area in the world which has been affected so many human lives around the world and 68 change the evolution of landform. Landslide causes death in the Himalaya approximately more 69 than 200 human live every year (Naithani 1999). According to Centre for Research on the    145 The study area lies in east Sikkim district which is located in Sikkim Himalaya. It is bounded by 146 the latitudes 27 0 17'14.67" N to 27 0 23'48.50" N and the longitudes 88 0 35'51.40" E to 147 88043'11.98" E covering an area of around 69.125 km 2 (Fig. 1). High relative reliefs, steep slope 148 along with immensely rugged surfaces are important physiographic characteristics in this state. 149 The maximum and minimum elevation of this Rorachu watershed is 4100 and 816 m 150 respectively. We have said before that there is a very rugged surface here so this Rorachu   167 Landslide is one of the most complex movements of the earth. It's very difficult to Identification 168 and mapping of suitable landslide factors for the vulnerability modeling and assessment. The scale of the analysis, and data availability were taken into account (Yalcin 2008). In this respect, 175 thirteen factors are considering for Landslide susceptibility Index (LSI) and Landslide risk (LR) 176 modeling and analysis (Fig. 2). It is important to compile a digitized database for execution the 177 landslide vulnerability modeling map using GIS. The spatial database has been design and 178 execute for the landslide vulnerability modeling of this study area as shown in (Table 1)  last all the data were vectorized in ArcGIS 10.3 software. In total 153 major and minor 197 landslides were identified in the Rorachu watershed with total area coverage 0.644 sq. km (Fig. 198 3). All the landslides data has been converted vector to raster format for the model preparation.     266 Drainage density is the total length of all streams and rivers of that grid divided by the total area 267 of that grid (Horton, 1932(Horton, , 1945Strahler, 1952). Drainage density (DD) indicates the measure of 268 how well or how poorly a river watershed is drained by stream channels. Drainage density 269 depends on both physical environment and climatic environments of this particular area.

270
Drainage density helps to determine the degree of reducing the shear strength of this slope which 271 has affective for slope instability. In this study, drainage density was assessed by this formula (eq  283 In alliance with all the anthropogenic activities are responsible for slope instability, construction    Where, as is the specific catchment area (SCA) and β is the local slope gradient measured in 331 degrees, respectively. In this Rorachu watershed SPI values was represented in between 0 and 332 145.37 and classified into five classes ( Fig. 7. i).   Table 4. Fig.7. l).   (Table 5).  403 The information value (IV) is a bivariate statistical method that was develops from information (LSI) can be represented by the summation of total information value (IV) in a grid cell j is:  (Table 5).

455
Calculations of certainty factor (CF) in each landslide factors are representing in (Table 5).  460 Multicollinearity is a statistical testing phenomenon in which one predictor variable in a multiple 461 regression model can be linearly predicted from the others with a substantial degree of accuracy.  Table 6. 484 Table 6. Multicollinearity analysis of FR, IV and CF approach.

518
In the present study, we have used three models for landslide susceptibility analysis in Rorachu   (14). The result LSI map is 529 depicted in (Fig. 8).   578 The landslide inventory map was overlaid with the landslide causative factors to dispose the 579 significance of each factor class for landslide occurrence. Using the information value (IV) 580 model, we computed landslide susceptibility map of Rorachu watershed by the equation (15).

581
The final landslide susceptibility map was performed by the IV model is shown (Fig. 9). All the 582 thirteen landslide variables were discerned for the landslide modelling (Table 5).  (Table 5). Then, thirteen landslide 616 conditioning factors were ascertained using Eq. 16. In order to calculate a LSI map (Fig. 10 It can be observed in (table)  The impression of other factors were also been analyzing for the landslide susceptibility analysis.

646
The road densities also important for landslide susceptibility, in this study the moderate road    659 Landslide density is the ratio between the observed landslides in that area and the area of each  susceptible classes (Fig.12). The comparison of the all three statistical model for landslide 671 susceptibility mapping discloses that the maps produced from two statistical models are  703 An appropriate validation is required to prepare a certain landslide susceptibility map of the 704 study area. In the current study, the validations of the landslide susceptibility map was restrained  we could say that 83.60 % prediction accuracy shows the landslide modelling in the Rorachu 724 watershed (Fig 14) with the standard error (FR = 0.016, IV = 0.019 and CF = 0.015) in Table 8.

725
To compare the result of all three bivariate models as quantitatively, the areas under curve 726 (AUC) were recalculated acceptance the total area as 1, which means the perfect probabilistic 727 and deterministic prediction accuracy. In this study, a comparison between FR, IV and CF model  (Fig. 16). The 756 triggering factors has major role for the landslide susceptibility in this area. In this study area, the 757 NH 31A road is more vulnerable to landslide and landslide risk (LR) zonation map also indicates 758 the highest vulnerability. The population is located in west side is safer than the east side.    Figure 1 Location map of the study area Geological map of the study area    Google earth map showing the very high (VH) and high (H) landslide susceptibility of various models (FR, IV and CF).

Figure 12
The landslide density (LD) has been showing the increasing trend to the highest vulnerable areas.

Figure 13
Success rate curve (SRC) for the three models (FR, IV and CF) in the Rorachu watershed Figure 14 Receive operating characteristics (ROC) curve for the FR, IV and CF models. The comparative Bar graph revealing the areal distribution of numerous models (FR, IV and CF) a.
observed landslide area situated in various landslide susceptibility zones b. Areal distribution of landslide susceptibility zones c. Areal distribution of Roads in various landslide risk zones (LRZ) d. Areal distribution of Settlement in various landslide risk zones (LRZ).

Figure 17
The probability of landslide vulnerability in various ranges (Elevation, Slope and Rainfall).