An Integrated Geospatial Modelling Framework for Improved Landslide Hazard Zonation in the Nilgiris District

Landslides represent one of the most critical geomorphic hazards in the Western Ghats of India in which the Nilgiris district consistently identified as a high risk zone. Although numerous landslide studies exist, many suffer from limited predictor variables and single algorithm modelling, constraining their applicability for operational disaster management. This study enhances the scientific rigor of landslide susceptibility assessment by integrating seventeen conditioning factors spanning topographic, geological, climatic, and anthropogenic domains within a GIS based analytical framework. A comprehensive inventory of 292 landslides, supported by 290 non-landslide points, served as the dataset for trained and validating three predictive models: Generalised Linear Regression (GLR), Random Forest (RF) and Gradient Boosted Trees (GBT). Susceptibility outputs were categorised into five hazard classes using Jenks natural breaks, while model performance was quantified through Receiver Operating Characteristic (ROC) and Success Rate Curve (SRC) analyses. The findings demonstrated that RF achieved the highest accuracy (AUC = 0.935) provides the highest predictive skill, followed by GBT (AUC = 0.920), both of which significantly outperform the GLR baseline (AUC = 0.860). These findings underscore the scientific utility of ensemble machine learning algorithms for generating high resolution susceptibility maps that support early warning systems, climate resilience, infrastructure planning and evidence based risk mitigation in the Nilgiris.