AI-driven 6-hour ahead nowcasting of sea-surface currents using HF Radar

This study introduces novel AI-driven models, Bidirectional Encoding-Forecasting (BiEF) and Variational Bidirectional Encoding-Forecasting (VBiEF), for nowcasting sea-surface currents using High Frequency (HF) Radar data. These models leverage advanced deep learning techniques to predict the dynamics of sea currents with accuracy and temporal resolution. Our research demonstrates that these AI models significantly outperform traditional persistence-based methods, providing skillful forecasts up to six hours ahead. While the VBiEF model, in particular, showcases good skill in capturing both the spatial and temporal complexities of sea currents, as well as in reconstructing intricate oceanographic features such as vorticity, divergence fields, and the rate of deformation tensor, several challenges remain to be addressed to further increase predictability levels. Furthermore, the good performance of these models in areas beyond their training domain suggests their adaptability and scalability for global ocean studies, opening new avenues for future research and application, highlighting the potential of AI in marine science. • AI models outperform simpler strategies by capturing sea currents’ complexities. • VBiEF excel in sea-surface nowcasting showing the highest accuracy. • AI demonstrates remarkable skill in accurately reconstructing sea features such as vorticity. • Proposed AI methods prove adaptable, showing strong performance in new areas. • Field-based forecasting accurately captures dynamic ocean structures.