Mapping landslide susceptibility and risk assessment on fragile ecosystem of Himalayan River basins

Landslides pose a significant threat in the Himalayan region due to complex geology, steep terrain, and diverse climatic conditions. This study addresses the need for a multi-dimensional approach by integrating Machine Learning with GIS to map landslide susceptibility across Himalayan River basins. Conditioning variables including topographical, climatological, hydrological, and phenological factors, and surface conditions were analysed using SVM to predict landslide susceptibility. For validation, SHAP, ROC curves, and AUC were used. The model attained 87% accuracy. Risk assessment was performed by intersecting land use/land cover (LULC) data with susceptibility zones to quantify agricultural and Urban and Built-up land exposed to landslides, alongside zonal statistics to estimate population risks. The results indicate that 371.5 thousand hectares are at very high risk of landslides, and 209.2 thousand hectares are at high risk, with the Jhelum River Basin emerging as the most vulnerable in terms of population, agricultural land, and built-up areas. This study demonstrates the dominance of hydrological and vegetation-related variables, such as runoff and forest fires, in driving landslide susceptibility, as revealed by SHAP analysis. Integrating susceptibility models with risk assessment, the study provides insights for regional planning, disaster management, and policy-making, stressing targeted mitigation for vulnerable basins.