Abstract Glaciers play a critical role in our society, impacting everything from sea‐level rise and access to clean water to the tourism industry. Their accelerated melt represents a key indicator of the changing climate, highlighting the need for efficient monitoring techniques. The traditional way of assessing glacier area change is by rebuilding glacier inventories. This often relies on manual correction of semi‐automated outputs from satellite imagery, which is time‐consuming and susceptible to human biases. However, recent advancements in Deep Learning have enabled significant progress toward fully automatic glacier mapping. In this work, we introduce DL4GAM: a multi‐modal Deep Learning‐based framework for Glacier Area Monitoring, available open‐source. It includes uncertainty quantification through ensemble learning and a procedure to identify the imagery with the best mapping conditions independently for each glacier. DL4GAM is trained and evaluated on the European Alps, a region for which experts estimated an annual change rate of around −1.3% over 2003–2015. We use DL4GAM to investigate the glacier evolution from 2015 to 2023 using Sentinel‐2 imagery and elevation (change) maps. By employing geographic cross‐validation, our models, based on U‐Net ensembles, demonstrate strong generalization capabilities. We then apply the models on 2023 data and estimate the area change at both the glacier and regional levels. Regionally, we estimate an area change rate of −1.90 1.26% per year. We provide quality‐controlled individual estimates over 2015–2023 for about 900 glaciers, covering around 70% of the region. Debris‐covered regions remain the most uncertain.
Diaconu et al. (Mon,) studied this question.