Explainable <scp>GeoAI</scp> Framework for Landslide Hazard–Vulnerability–Risk Assessment in Rudraprayag, Garhwal Himalaya

ABSTRACT Landslides are one of the most catastrophic of the geological threats in the Himalayas, inflicting loss of life, damage to infrastructure and socio‐economic upheaval. In this study, we introduce an integrated landslide hazard‐vulnerability‐risk assessment (HVRA) for the Rudraprayag district of the central Garhwal Himalaya, Uttarakhand, India, an area characterized by steep slopes, fragile lithology, active tectonics, intense monsoonal rainfall and rapid anthropogenic disturbances. A total of 3383 landslide locations, derived from NRSC, GSI, Google Earth interpretation and extensive field investigations, were used to develop statistical, hybrid and deep learning susceptibility models such as Frequency Ratio‐Shannon Entropy (FR‐SE), UF‐FRDL, ANN, CNN, RNN, DenseNet and ResNet. The modelling framework included 27 topographic, hydrological, geological, environmental, structural and anthropogenic conditioning factors, along with annual post‐monsoon rainfall and Peak Ground Acceleration (PGA). The comparative study showed that ANN had the strongest predictive performance with AUC, accuracy, precision, specificity, F1‐score, and MCC values of 0.975, 0.923, 0.926, 0.928, 0.922, and 0.847, respectively, while FR‐SE and UF‐FRDL also had exceptional predictive performance. Cross‐validation, geographic bootstrap validation, KS‐statistics and field validation demonstrated the adaptability of the developed models with 91% of field observed landslides found in high and very high susceptibility classes. SHAP research showed that the main factors of landslide occurrence were Stream Power Index, slope, terrain roughness, rainfall, and road proximity. The threshold study shows that the vulnerability of landslides is significantly increased for annual rainfall above ~162–200 mm and for PGA values of over 0.20 g, highlighting the combined hydro‐seismic influence on slope instability. The resulting integrated risk map identified 13.01% and 4.78% of the district as high‐risk and very high‐risk zones respectively, predominantly centred along the Mandakini valley and key pilgrimage circuits. The established HVRA framework is an excellent decision support tool for landslide reduction, infrastructure planning and sustainable development of the Himalayas.