Subhourly ice surface temperature retrieval over the Larsen C Ice Shelf using geostationary satellite data and machine learning

Ice surface temperature (IST) is a key parameter that affects the surface mass balance and energy budget of the Antarctic ice shelves. Existing IST products, primarily derived from polar-orbiting satellites, suffer from insufficient temporal resolution, limiting the understanding of intraday IST variations and impeding in-depth investigations of ice shelf surface physical processes. This study developed an IST retrieval model based on machine learning for the Larsen C Ice Shelf using thermal infrared data from the Geostationary Operational Environmental Satellite−16 (GOES−16) as the primary input, with Moderate Resolution Imaging Spectroradiometer(MODIS) IST products serving as ground truths. High-accuracy radiation emergence angles were calculated pixelwise as critical inputs to correct for angular anisotropy effects. The results demonstrate that the developed IST retrieval model achieved an accuracy comparable to that of the MODIS IST product, with a root mean square error (RMSE) of approximately 1.5 K. More importantly, the temporal resolution of the generated IST in this study increased by approximately 30 times compared with the MODIS IST product, increasing from approximately 4 times per day to 144 times. Multisource data validation revealed RMSE values of 1.234 K compared with IceBridge, 2–3 K compared with automatic weather station broadband thermal infrared measurements under GOES−16 cloud masking, and 1–2 K compared with the Visible Infrared Imaging Radiometer Suite IST product. Based on the trained model, we generated the IST product for the summer days with low cloud cover from December 2019 to March 2022, with a temporal resolution of 10 minutes and a spatial resolution of 2 km. The analysis results show that the intraday IST variation over the Larsen C Ice Shelf has two patterns: a single-peak dominant pattern and a sustained high-IST pattern. The single-peak dominant pattern is severely affected by shortwave downward radiation. The occurrence of the sustained high-IST pattern is relatively rare, mainly caused by intense melting of the ice shelf. The ice melting and freezing processes absorb and release large amounts of latent heat, maintaining the IST near the melting point throughout the day. Notably, the sensible heat caused by the foehn event led to a strong IST increase in the areas near the grounding line, especially when the IST was relatively low, causing significant IST spatial heterogeneity for the Larsen C Ice Shelf.