Abstract

This study presents an advanced spatio-temporal assessment and predictive analysis of glacier dynamics in the Hunza District, Gilgit-Baltistan, Pakistan, over the period 2016-2026. The research integrates Google Earth Engine (GEE), multi-temporal Landsat and Sentinel satellite imagery, and machine learningbased classification to improve glacier delineation accuracy beyond conventional threshold-based approaches. Initial glacier mapping was performed using the Normalized Difference Snow Index (NDSI), while a supervised Random Forest classifier incorporating spectral indices and topographic variables was applied to enhance classification reliability in complex mountainous terrain. The analysis focuses on four major glaciers: Batura, Passu, Gulmit, and Ghulkin. Results indicate a persistent decline in glacier extent over the study period, with Batura Glacier decreasing from 421km² to 406 km², Passu from 61.2km² to 60.4km², Gulmit from 17.4km² to 16.7km², and Ghulkin from 27.1km² to 25.2km². The machine learningbased classification demonstrated improved accuracy compared to NDSI-only extraction, highlighting the effectiveness of AI integration in cryospheric mapping. Climate-glacier interaction analysis using NASA POWER data reveals a strong association between glacier retreat and increasing regional temperatures, which rose from 25.2 °C in 2016 to 31.8 °C in 2026. Precipitation patterns exhibited high interannual variability with an overall declining trend toward the later years of the study period. These climatic shifts have accelerated glacier melt, particularly in low-elevation zones (2800-3500m), where increased solar radiation intensifies ablation processes. Topographic analysis using SRTM-based digital elevation models confirms that elevation, slope, and aspect play a critical role in controlling glacier stability. North-facing slopes exhibit relatively lower retreat rates due to reduced solar radiation exposure, while steeper and lower-altitude regions show higher vulnerability to ice loss. Overall, the study demonstrates that integrating machine learning with Google Earth Engine significantly enhances glacier monitoring accuracy and provides a scalable framework for predictive cryospheric assessment. The findings have important implications for freshwater resource management, hydropower sustainability and climate adaptation strategies in high mountain regions.

Keywords:Glacier retreat; Spatio-temporal analysis; Google earth engine; NDSI; Hunza glacier; Remote sensing; Climate change; Machine learning; Random forest classification; Cryospheric monitoring

Highlights
A. Multi-temporal Landsat and Sentinel satellite data (2016-2026) used to map glacier dynamics
B. All four glaciers show consistent area reduction strongly linked to rising temperature trends
C. Machine learning-based classification improved glacier mapping accuracy over traditional NDSI
D. Google Earth Engine proved highly effective for large-scale cryospheric and climate monitoring