Deep Learning for Skillful Regional Wave Downscaling: Remote Wind as Boundary Conditions

Accurate modelling of integral wave parameters, significant wave height (SWH), mean wave direction (MWD), and mean wave period (MWP) is critical for ocean engineering, coastal hazard mitigation, and wave climate analysis.Although numerical wave models (NWMs) offer high accuracy grounded in physical principles, their computational demands often limit their utility in real-time forecasting and resource-constrained environments.To address these limitations, we propose a deep learning framework capable of simultaneously predicting relatively high-resolution (4-arcminute) regional SWH, MWD, and MWP.The model is driven by local wind forcing at 0.5° resolution and further incorporates large-scale wind fields at 1.5° resolution to implicitly represent boundary influences, capturing the generation and propagation of swells from remote sources.Evaluation against NWM outputs in the Australian region demonstrates that the model can reproduce fine-scale wave fields with high fidelity and computational efficiency.These results highlight the framework's potential for operational implementation in applications requiring rapid forecasts, such as emergency response and operational wave prediction, as well as for longterm regional wave climate analyses and projections.