Permafrost, ground that stays below 0 °C for at least two consecutive years, is essential for ecosystem stability and water balance in mountain environments. Its degradation raises the risk of landslides and other cryospheric hazards. This study used reanalysis data, rock glacier inventories, and Machine Learning algorithms to create high-resolution (30 m) permafrost probability maps for the Kali Gandaki River Basin (KGRB) in the central Himalaya. Three models Gradient Boosting (GB), Random Forest (RF), and Support Vector Machine (SVM) were trained with WorldClim, ERA5, and CMIP6 datasets to analyze changes in permafrost over time and space. According to the data, WorldClim shows a 6.60% (119 km 2 ) increase in permafrost from 1961−1970 to 2010−2018, while ERA5 reports a 26.65% (724 km 2 ) decrease over the same period. Future CMIP6 projections suggest changes ranging from a 0.47% (9 km 2 ) increase under low emission (SSP1-2.6) to a 9.29% (179 km 2 ) decrease under moderate emission (SSP2-4.5), and a 40.94% (822 km 2 ) decrease under high emission (SSP5-8.5) by 2091−2100. Continuous permafrost is highly vulnerable, with findings stressing that strong climate mitigation is vital for maintaining permafrost stability and reducing carbon feedback. The mean annual air temperature has increased by 0.011–0.019 °C per year during 1961–2020 and is expected to rise to 0.053 °C per year by 2100. All datasets consistently show that permafrost is increasingly concentrated at higher elevations (5,000–6,000 m), reflecting the same climatic refuge zones.
Sthapit et al. (Wed,) studied this question.