Developing Site-Specific Rainfall Thresholds for Landslide Prediction in the Himalayas: A Comparative Assessment between Northwestern and Northeastern Himalayas

Landslides, predominantly triggered by intense and prolonged rainfall, pose a critical hazard in the Himalayan region, with Indian Himalayas contributing approximately 15% of global rainfall-triggered landslides. Despite advances in landslide prediction, existing thresholds often fail to account for the diverse climatic and geophysical conditions across the Himalayas. To address these gaps, this study establishes both at-site and regional rainfall thresholds for landslide prediction by integrating advanced statistical techniques and environmental analyses. Seasonal rainfall thresholds were established to define rainy days, revealing higher winter thresholds in the Northwestern Himalayas (NWH) due to snowmelt contributions and elevated monsoon thresholds in the Northeastern Himalayas (NEH), driven by prolonged rainfall and antecedent moisture saturation. Building on this, we derived empirical event-duration (E-D) thresholds using a novel non-crossing quantile regression approach to ensure robustness against lower quantile crossing issues. The derived regional thresholds for NEH (E = -11.10 + 0.62D) and NWH (E = -12.00 + 0.63D) fits within global bounds . Land use/land cover (LULC) analysis and probabilistic mutual information ─ based analysis further identified critical environmental controls shaping these thresholds. In the NWH, built-up areas, elevation, and vegetation emerged as key factors playing significant roles in shaping rainfall thresholds to trigger landslides, while elevation, rangeland, and the Standardized Precipitation Index (SPI) were significant in the NEH. These insights underscore the need for region-specific E-D thresholds for landslide prediction and disaster management in the Himalayan region. By integrating environmental controls into a 'physics-based statistical learning' framework, this study overcomes limitations of conventional empirical rainfall threshold for landslide prediction models, delivering region-specific thresholds, thereby enhancing disaster preparedness, a step towards developing a climate-resilient landslide early warning system in the Himalayas.