Sea Ice Edge Constraint Improves Antarctic Sea Ice Seasonal Prediction in Deep Learning Model

Predicting Antarctic sea ice is of substantially academic and practical significance. However, current prediction models including deep learning (DL)-based models show notable bias in the marginal ice zone (MIZ). In this study we developed a pure data-driven DL model for predicting Antarctic austral summer monthly to seasonal sea ice concentration (SIC) by incorporating a novel hybrid sea ice edge constraint loss function (HybridLoss). The model is referred to as ASICNet. Independent test based on the recent five years (2019–2023) demonstrates that ASICNet with HybridLoss achieves significantly higher skills than ASICNet otherwise, with a reduced mean absolute error (MAE) of 0.021 from 0.022, a reduced integrated ice edge error (IIEE) of 1.714×10⁶ from 1.794×10⁶ km², but an increased pattern correlation coefficient (PCC) of 0.40 from 0.38, although both ASICNet outperform dynamical and statistical models. Furthermore, this study developed enhanced heat maps to interpret the predictability sources of sea ice within DL-based models, and the results suggest that the Antarctic sea ice predictability is attributed to the factors like the Antarctic Dipole (ADP), Amundsen Sea Low (ASL), and Southern Ocean sea surface temperature (SST) as revealed in previous studies. Thus, ASICNet is an efficient tool for austral summer Antarctic SIC prediction.