Research from the Pattern Recognition Lab has been featured by IEEE Spectrum, showcasing a new AI-driven approach for large-scale monitoring of glacier change from satellite imagery.
Accurate tracking of glacier calving fronts is essential for quantifying ice loss and improving projections of sea level rise. However, traditional monitoring relies on manual delineation in satellite images, which is time-consuming and limits global coverage.
The presented method leverages deep learning to automatically segment calving fronts in radar satellite data and significantly improves generalisation to previously unseen glacier regions. By combining minimal region-specific annotations with unlabeled seasonal imagery and geospatial coastline information, the approach achieves substantially improved accuracy compared to baseline models and enables robust transfer to new Arctic environments.
A key demonstration of the method is its application to the Svalbard archipelago, where it enables high-resolution, monthly mapping of calving front positions across all 145 glaciers over multiple years. This provides an unprecedented dataset for analysing glacier dynamics at regional scale.
The research is the result of close collaboration between the Pattern Recognition Lab and the Geography Department at Friedrich-Alexander-Universität Erlangen–Nürnberg, integrating advances in machine learning with domain expertise in glaciology and remote sensing.
Read the full article here: https://spectrum.ieee.org/tracking-glacier-melting-ai
SAR image provided by ESA