A novel technique for automated mapping of Antarctic supraglacial lakes in Sentinel-1 SAR imagery using deep learning

Supraglacial meltwater accumulation on ice sheets and ice shelves can have considerable impact on ice discharge, mass balance and global sea-level-rise. With further increasing surface air temperatures, surface melting and resulting processes including hydrofracturing, meltwater penetration to the glacier bed as well as surface runoff will cumulate and most likely trigger unprecedented ice mass loss from the Greenland and Antarctic ice sheets. To date, the Antarctic surface hydrological network remains understudied calling for increased monitoring efforts and circum-Antarctic mapping strategies. This is particularly important given that Antarctica’s future contribution to global sea-level-rise is the largest uncertainty in current projections.

In this study, we present a novel methodology for Antarctic supraglacial lake extent mapping in Sentinel-1 Synthetic Aperture Radar imagery using state-of-the-art deep learning techniques. The method was implemented with the aim of complementing a previously developed supraglacial lake detection algorithm applying Machine Learning on optical Sentinel-2 data in order to deliver a more complete picture of Antarctic meltwater ponding compared to single-sensor mapping products. The deep learning model was trained on 21,200 Sentinel-1 image patches using a modified ResUNet for semantic segmentation of supraglacial lakes and evaluated by means of ten spatially or temporally independent Sentinel-1 test acquisitions distributed across the Antarctic continent. Besides, George VI Ice Shelf is analyzed for intra-annual lake dynamics throughout austral summer 2019/2020 and decision-level fused Sentinel-1 and Sentinel-2 maximum lake extent mapping products are presented for selected time periods. Future work involves the integration of more training data as well as the generation of circum-Antarctic mapping products using both, Sentinel-2 and Sentinel-1 derived lake extent mappings. These will be crucial for intra-annual analyses on supraglacial lake occurrence across the whole continent and associated drivers and impacts.