Changes in permafrost distribution and geo-hazard susceptibility in the central Himalayas: Insights from machine learning and remote sensing techniques

High-altitude Himalayan permafrost is rapidly degrading under sustained climate warming. This degradation generates increasing environmental and infrastructure hazards that require urgent, evidence-based assessment. However, long-term studies linking permafrost dynamics to hazard evaluation using machine learning remain limited. This gap restricts informed decision-making for climate adaptation in vulnerable mountain regions. This study quantifies spatiotemporal changes in permafrost distribution and evaluates associated hazard susceptibility in Nepal's Kali Gandaki River Basin (KGRB) during the last six decades (1961-2020). It provides the first comprehensive temporal assessment for this climatically sensitive watershed. Unlike previous studies, this work integrates permafrost mapping and hazard assessment within a unified framework. It combines in-situ meteorological data, WorldClim reanalysis, and a rock glacier inventory (n = 354). A comparative geospatial Machine Learning approach is applied using Support Vector Machine (SVM) and Generalized Additive Model (GAM). Results show that cumulative warming of 1.6 ℃ lowered the permafrost limit by 655 m. Degradation affected 15%-20% of the basin, primarily at lower elevations. In contrast, 5%-10% of northeastern high-altitude areas showed localized intensification. SVM results indicate a shift in dominant controls from topographic factors to thermal and freeze-thaw processes, highlighting the increasing influence of climate warming. Hazard risk increased markedly, with high-risk zones expanding by 38 km². Approximately 13% of settlements and 11% of hydropower facilities are now classified as highly susceptible. The resulting hazard maps provide critical tools for infrastructure planning, early warning, and climate adaptation in high-mountain environments.