Seasonal evolution of Antarctic supraglacial lakes in the Sentinel satellite era

Kurzfassung

Supraglacial lakes in Antarctica play a crucial role in ice sheet dynamics, as they influence ice stability by accelerating meltwater-driven processes such as meltwater injection and basal lubrication. The accumulation of meltwater on ice surfaces reduces albedo, causing more solar energy absorption, which in turn accelerates melting in a self-reinforcing cycle. Therefore, it is essential to monitor the seasonal evolution of surface meltwater features and get a better understanding of their dynamics. Earth observation data provide continuous and long-term observations of the vast and remote Antarctic ice sheet, but extracting reliable information on supraglacial lakes requires more than one satellite sensor. During the short period of lake occurrence, optical satellite imagery might miss shortly occurring dynamics in supraglacial lake extent due to heavy cloud cover. In contrast, the extent of supraglacial lakes is very difficult to interpret in cloud and illumination independent SAR data, as slightly overfrozen lakes become easily invisible due to backscatter similarities with the surrounding ice. The Sentinel mission provides the best source to combine advantages from both worlds and monitor the seasonal evolution of supraglacial lakes at best possible accuracy. We developed a deep learning-based classification algorithm to infer the extent of supraglacial lakes based on their spectral and backscattering properties, as well as their spatial appearance. The algorithm is based on a U-Net architecture with an Atrous Spatial Pyramid Pooling module to widen the receptive field and improve the detection of differently sized lakes. For each sensor, we trained the networks separately and reached classification accuracies with an F1-Score of 93% for Sentinel-1 and 91% for Sentinel-2. A final post-processing step combines both classification result to a bi-weekly recurrence map of maximum supraglacial lake extents during the austral summer months (November to March). This presentation addresses advantages and challenges of supraglacial lake mapping with Sentinel-1 and Sentinel-2 data and presents bi-weekly recurrence maps for 19 Antarctic ice shelves and regions with repeated supraglacial lake formation. Statistics on anomalies in maximum lake extents provide a circum-Antarctic picture on the seasonal evolution of supraglacial lakes during the Sentinel era 2014 to 2024. The circum-Antarctic data availability allows to compare the differences in supraglacial lake dynamics between years with high (e.g. 2019/2020) and low (e.g. 2023/2024) circum-Antarctic surface melt occurrence and regionally intense melt years (e.g. Dronning Maud Land 2017/2018). Accurate information on lake dynamics can inform hydrological and surface mass balance models to improve their ability to model surface melt processes.