Opitimzed WAVEWATCH III for significant wave height computation using machine learning

This paper addresses the improvement of the skill of a wave generation model in calculating the significant wave height of sea states by proposing a wave correction model based on machine learning techniques. The model corrects the simulated significant wave heights(Hs) Hs ) by fitting the discrepancies between numerical simulation results and buoy measurement data. We consider the effects of boundaries and the simplified computation of source terms in the wave model, employing a wave correction model to optimize wave results. The research results demonstrate that the integration of the correction model with the physical calculation model can significantly enhance both accuracy and efficiency in wave computations. The MAPE for Hs decreased from 48.38% to 7.14% and from 20.68% to 5.09%. Simultaneously, computational efficiency improved by factors of 1 and 133.2, respectively. Furthermore, the model demonstrated good generalization capabilities, enabling high- precision wave simulations for other locations within the sea area.