ASICast: a novel model for daily antarctic sea ice concentration prediction

Sea ice plays a vital role in shaping the Earth's climate system, it regulates ocean circulation and atmospheric dynamics. Accurate sea ice concentration (SIC) predictions are crucial for improving climate models, supporting maritime navigation, and assessing the effects of climate change. In this study, ASICast, a novel lightweight model for daily Antarctic SIC prediction at a spatial resolution of 6.25 km, is introduced. In the ASICast model, statistical and temporal interpretability are integrated with computational efficiency, providing reliable SIC predictions years in advance from a single initialization. The model achieves stable performance across multiple years of evaluation and consistently outperforms existing data-driven models in medium-range predictions (5th to 15th weeks). Model evaluation reveals that the ASICast model maintains an accuracy greater than 80% throughout most of the year, with the level in more than half of the months exceeding 90% in predicting sea ice margins. Notably, its one-year lead prediction error is comparable to the eight-week lead error of existing models, thus highlighting its robustness in extended prediction. These advancements not only increase the ability to predict sea ice dynamics but also contribute to the refinement of climate models, facilitating improved long-term planning and decision-making in polar research and navigation.