Development of Remote Sensing-based Dynamic Decision Support System (DSS) for Avalanche Susceptibility Mapping using AI/ML Techniques for NW Himalaya

The North-Western Himalayan region (Jammu and Kashmir) regularly experiences high-impact snow avalanches causing loss of life and disruption to strategic roads, border infrastructure, and settlements. However, current hazard assessment methods struggle due to extreme topography, sparse in-situ observations, and limited real-time monitoring. In this paper, a remote sensing-based dynamic Decision Support System (DSS) that uses multi-sensor Earth observation (EO) satellite data’s to to generate high-resolution avalanche susceptibility maps by analysing terrain parameters, snow cover dynamics, and meteorological drivers. The DSS integrates MODIS (daily snow cover), Sentinel-2 (10 m optical), AMSR-2 (passive microwave snow properties), and SRTM (30 m DEM) to extract terrain, snow, and weather-related indicators for identifying avalanche prone regions. It incorporates two independent yet complementary modelling components. The first employs a knowledge-based Analytic Hierarchy Process (AHP) to establish a transparent susceptibility baseline guided by expert knowledge. The second applies supervised machine learning using five classifiers i.e., Support Vector Machine (SVM), Naïve Bayes, Random Forest, Gradient Boosting, and LightGBM to delineate avalanche-prone areas. Model training uses multi-year historical in situ avalanche records combined with Sentinel-2–detected avalanche events, creating a robust inventory exceeding several hundred mapped occurrences and improving detection in remote high-altitude zones. Among all classifiers, SVM achieved the best performance with a ROC-AUC of ~0.855, demonstrating strong generalization on independent test data. The DSS produces classified susceptibility maps (very low to very high risk) and location-specific risk reports that can be exported as tabular outputs for settlement and road-segment level assessment. The system remains operationally relevant through continuous EO data ingestion and automated updates. This EO-based DSS provides a scalable, data-efficient, and operational framework for avalanche risk assessment in data-scarce mountainous regions, supporting early warning, disaster preparedness, infrastructure planning, and climate-change-driven snow hazard adaptation.