Deep Learning Insights into Glacier Dynamics in the Austrian Alps (2015–2024): Potential and Challenges

Glacier retreat is a critical indicator of climate change, with significant implications for hydrology, biodiversity, and local economies. This study employs an Attention U-Net deep learning (DL) model to analyse glacier retreat in selected regions in the Austrian Alps from 2015 to 2024. By integrating Sentinel-2 imagery and a high-resolution Digital Elevation Model (DEM), we quantify ice loss across four major glacier zones: Großglockner, Großvenediger, Sonnblick, and Wildspitze. The model achieved an Intersection over Union (IoU) of 80.56% and a recall of 89.72%, demonstrating its effectiveness in automated glacier mapping. Our results indicate a total glacier area reduction from 154.90 km² in 2015 to 130.56 km² in 2024, resulting in a 15.72% loss. These findings confirm the accelerated pace of glacier retreat in the Alps. Despite inconsistencies in data for certain years, partly caused by temporal mismatches in satellite imagery, the study underscores the potential of deep learning for accurate and scalable glacier monitoring. This study provides a framework for future climate impact assessments and adaptation strategies in alpine environments.