Accurate ocean wave modeling is indispensable for maritime safety and climate research. However, traditional third-generation numerical models (e.g., WAVEWATCH III) are constrained by heavy computational costs and empirical parameterization uncertainties. The central scientific challenge is whether Artificial Intelligence (AI) can evolve beyond simplistic curve-fitting to become a robust surrogate for complex ocean dynamics.

This research represents a typical interdisciplinary synergy between physical oceanography and frontier AI technologies. We leverage deep learning architectures integrated with data assimilation and physical constraints to reconstruct and forecast global wave states. Our research encompasses three key aspects:
First, we established a high-precision physics-guided AI model (published in Science Advances) capable of characterizing global wave height under complex sea conditions with unprecedented efficiency. This work achieved an accuracy that surpasses state-of-the-art numerical models and aligns with the high-fidelity ERA5 reanalysis. Second, to address error growth in long-term rolling forecasts, we developed a data-assimilation-enhanced AI framework (published in Geoscientific Model Development), which significantly suppresses error diffusion and enhances predictive stability. Finally, we developed an AI wave spectrum model that directly predicts the 2D directional wave energy spectrum. By transcending the traditional focus on integrated parameters, this model implements the first comprehensive AI wave model in its full sense.

We hope these advancements demonstrate that the integration of AI into oceanography is not merely an auxiliary tool, but a transformative technology for next-generation global ocean state estimation and real-time forecasting.