Pixel-Based Glacial Change Monitoring Across Hindu Kush, Karakoram, & Himalayan Ranges

Le résumé fourni par la source

Glaciers in the high-mountain regions of the Hindu Kush, Karakoram, and Himalayas are critical for sustaining downstream water resources, yet their short-term and localized responses to climate variability remain unclear. To address this gap, we present a machine learning–based framework for high-resolution glacier monitoring using Sentinel-2 imagery from 2016 to 2024. The proposed approach establishes a complete end-to-end pipeline for glacier change analysis, including satellite image preprocessing, supervised SVM-based tri-class classification (snow, partial snow/ice, and no snow), pixel-level change detection, and temporal trend analysis. The tri-class model achieves an overall accuracy of 83%, capturing detailed variations in glacier surface conditions that conventional binary models often overlook. The results indicate a weakening of the Karakoram Anomaly after 2019, with glaciers displaying divergent behaviors and a general trend toward retreat, particularly at lower elevations. While effective, the approach is limited by cloud-prone Sentinel-2 observations, manual and NDSI-based training data, and reduced accuracy in transitional zones such as firn or debris-covered ice. Overall, this framework advances long-term glacier monitoring by improving spatial and temporal resolution, supports water resource management, and provides a foundation for assessing Glacial Lake Outburst Flood risks under changing climatic conditions.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.