Artificial Intelligence for Atmospheric and Oceanic Forecasting and Research: A Fudan University Perspective

Rapid advances in artificial intelligence (AI), together with the rapid growth of observational and reanalysis datasets, are reshaping atmospheric and oceanic sciences. This paper presents a Fudan University perspective on recent AI applications in this field, with particular emphasis on representative studies from the Department of Atmospheric and Oceanic Sciences and the Artificial Intelligence Innovation and Incubation Institute at Fudan University. We highlight the development of AI models for atmosphere, ocean and cloud systems, represented respectively by FuXi, Xihe and DaYu, together with related advances in AI-based data assimilation, forecast utilization, hybrid modeling and end-to-end forecasting systems. Beyond forecasting, we further discuss Fudan-related work that uses AI as a scientific tool, including studies on interpretability, predictability analysis, causality diagnosis and emerging multimodal foundation models for Earth system science. Current challenges facing AI applications in atmospheric and oceanic sciences, including physical consistency, extreme event prediction, generalization and long horizon stability, data limitations and computational cost are discussed. Finally, we outline future directions for integrating AI with physics-based understanding in atmospheric and oceanic sciences. Overall, this perspective suggests that AI should be viewed not only as a tool for improving prediction, but also as a framework for advancing scientific understanding of atmospheric and oceanic processes.