Artificial intelligence for the monitoring of Antarctic supraglacial lake dynamics in 2015-2021 using Sentinel-1 SAR and optical Sentinel-2 data

Earth Observation (EO) provides a wealth of data for the monitoring of the Antarctic continent. In this context, data of the Sentinel-1 Synthetic Aperture Radar (SAR) and optical Sentinel-2 satellite missions of the European Copernicus programme deliver valuable information on key ice sheet parameters including the location of the calving front and grounding line, the ice velocity and elevation as well as the Antarctic surface hydrological network. The monitoring of the latter is crucial for an improved understanding of processes such as hydrofracture triggering ice shelf collapse and ultimately ice flow accelerations and increased ice discharge. To establish a monitoring service for supraglacial lake extent delineation in Sentinel-1 SAR and optical Sentinel-2 imagery, a fully automated processing chain based on machine learning and deep learning was developed and integrated within the internal processing infrastructure of the German Aerospace Center (DLR).

Here, we present first results of the implemented machine learning processing pipeline over six major Antarctic ice shelves. In particular, the full archive of Sentinel-1 and Sentinel-2 was exploited to provide bi-weekly supraglacial lake extent mappings during 2015-2021 at unprecedented 10 m spatial resolution. The results over Antarctic Peninsula ice shelves reveal comparatively low lake coverage in 2015-2018 and high lake coverage during summers 2019-2020 and 2020-2021. Over East Antarctic ice shelves, supraglacial lake extents fluctuated more substantially with comparatively high lake coverage during most of 2016-2019 and low lake coverage throughout melting season 2020-2021. Further, the data reveal a coupling between supraglacial lake formation and the near-surface climate, the local glaciological setting and large-scale atmospheric modes.

The final data products on Antarctic supraglacial lake extent dynamics during 2015-2021 are available via the GeoService of the Earth Observation Center (EOC) at DLR. To establish a near-real-time monitoring service on supraglacial lake dynamics in the future, the full processing pipeline is currently refined and data products will be made readily available for download via the EOC GeoService. In this context, we are building upon the expertise of the Polar Monitor project and IceLines, a processor for automated calving front extraction over the Antarctic coastline.