High-Temporal Antarctic Glacier Terminus and Ice Shelf Front Mapping from Sentinel-1 – A Deep Learning Approach

Kurzfassung

Antarctic glacier termini and ice shelf fronts are sensitive indicators of glaciological and environmental change. Mapping Antarctic calving front change in a high-temporal and spatial resolution has been difficult due to the lack of suitable data and the time-consuming manual delineation of fronts. Since the launch of Sentinel-1 year-round SAR imagery over the Antarctic coastline exists with at least weekly revisit times. To exploit the abundance of data it is necessary to implement an automated extraction algorithm for glacier and ice shelf fronts. Novel improvements in deep learning offer great opportunities for scene classification in remote sensing data even when facing complex structures. Our developed approach uses a modified U-Net for semantic segmentation classifying Sentinel-1 scenes for glacier ice and ocean. Accurate front positions can be obtained also for glacier termini enclosed by icebergs and mélange. Nevertheless, surface melt can be challenging in some regions. To demonstrate the model’s performance, we present high-temporal time-series of calving front positions for fast moving glaciers (e.g. David Glacier). The frequent mapping of glacier termini reveals changes in front fluctuations in unprecedented detail and could be used as input data for ice dynamic modelling.