Application of GIS-based data-driven bivariate statistical models for landslide prediction: a case study of highly affected landslide prone areas of Teesta River basin

Predicting landslides has become a critical global challenge for promoting sustainable development in mountainous regions. This study conducts a comparative analysis of landslide susceptibility maps (L.S.M.s) generated using two GIS-based data-driven bivariate statistical models: (a) Frequency Ratio (F.R.) and (b) Evidential Belief Function (E.B.F). These models are applied and evaluated in the high landslide-prone upper and middle Teesta basin of the Darjeeling-Sikkim Himalaya, leveraging geographic information system (GIS) and remote sensing techniques. We compile a comprehensive landslide inventory map containing 2387 regional landslide points. We use approximately 70% of this dataset for model training and reserve the remaining 30% for validation. In the construction of the Landslide Susceptibility maps (LSMs), a comprehensive set of twenty-one landslide-triggering parameters has been considered. These parameters encompass factors such as elevation, distance from drainage, distance from lineament, distance from roads, geology, geomorphology, lithology, land use, and land cover, normalized difference vegetation index, profile curvature, rainfall, relief amplitude, roughness, slope, slope aspect, slope classes, stream power index, sediment transport index, topographic position index, topographic ruggedness index, and topographic wetness index. An examination of multicollinearity statistics reveals no collinearity issues among the twenty-one causative factors utilized in this research. The final L.S.M.s demonstrate that the combined application of the F.R. and E.B.F. models yields the highest training accuracy at 98.10%. The insights derived from this study hold significant promise as valuable tools for assessing environmental hazards and land use planning.