Numerical study of a CNN-based model for regional wave prediction

Abstract

The accurate prediction of large-scale waves is of great significance for marine resource assessment and ocean disaster prevention. In recent years, a new generation of wave models represented by the Simulating Waves Nearshore (SWAN) model has been widely applied in the field of large-scale wave prevention. However, high accuracy of long-term wave simulation using the SWAN model is costly in computation resources. Considering the high prediction efficiency of machine learning methods, this paper proposes a CNN-based regional wave prediction (CNN-RWP) model. We apply a convolutional neural network (CNN) to construct the mapping relationship between wind data and wave data, which takes an hourglass configuration. In this paper, wind datasets from the European Center for Medium-Term Weather Forecast (ECMWF) and the validated wave datasets simulated by SWAN were employed in model-training optimization, verification, and generalization analysis. A comparison between the CNN-RWP model and the SWAN model is addressed using a dataset from the Gulf of Mexico. The mean absolute error between the prediction result of the CNN-RWP model and the SWAN output result is less than 10%, while the calculation efficiency is improved by about 1000 times. The proposed CNN-RWP model provides an effective approach for achieving high-precision wave prediction.