In the Himalayan region, permafrost is present at high-altitude areas (> 3800 m above m.s.l) that are vulnerable to degradation due to global warming. Estimating and assessing the permafrost distribution in the Himalayan mountains is required for vulnerability assessment. While current global mountain permafrost distribution models provide a good understanding of permafrost distribution at the national/regional level, they struggle to show its local-scale variability. This work shows finer scale modelling of the permafrost distribution for the Kinnaur district, Himachal Pradesh, India using machine learning algorithm. We utilised six parameters namely slope, aspect, curvature, normalised difference vegetation index (NDVI), normalised difference moisture index (NDMI) and mean annual air temperature (MAAT) along with rock glaciers to prepare the probable permafrost distribution map using support vector machine (SVM). We derived NDVI, NDMI and MAAT using Landsat-8 imagery and obtained slope, aspect, and curvature from ALOS PALSAR DEM. The performance of the model was assessed using five cross-validation techniques, including hold-out validation, k-fold cross-validation, stratified k-fold cross-validation, leave-one-out cross-validation, and repeated random train test split. The model achieved an AUC value of around 91% for all the cases, and the resulting map was validated using field images from crowd-sourced images. This study provides a high-resolution representation of permafrost distribution in the Kinnaur Himalayas. The probable permafrost area was found to be 27.3% of the total study region. This result could be used for better understanding of the impact of climate change on mountain permafrost and the development of effective adaptation strategies.

Permafrost, Rock Glacier, Support Vector Machine, Machine Learning, Northwest Himalaya