A convolutional neural network based model to predict nearshore waves and hydrodynamics

This study proposes a convolutional neural network (CNN) based model to predict waves and hydrodynamics in the nearshore zone. Specifically, we address the so-called next-frame prediction of wave and hydrodynamic conditions using the previous-step conditions. A laboratory experiment on unsteady rip currents is simulated using the non-hydrostatic model SWASH. The simulated wave and hydrodynamic results are divided into two groups. The first group of data is used to train the CNN model, and the second group of data is used to evaluate the trained model by comparing the simulated water surface elevation, cross-shore velocity, and longshore velocity by SWASH with the corresponding prediction by the trained CNN model. The result shows that the CNN model accurately predicts the wave propagation over the offshore slope, wave breaking over the sandbar, and bore propagation over the shore including detailed wave crest bending and separation. Time-averaged water level comparison further confirms that the CNN model captures the wave setup induced by wave breaking. The CNN model also correctly predicts the cross-shore velocity and longshore velocity driven by wave breaking over sandbars and wave–current interaction. Finally, the low-frequency nearshore circulation pattern predicted by the CNN model is also evaluated by time-averaging the velocity field. The comparison shows that the CNN model correctly predicts the distribution of circulation cells and the number of cells, as simulated by the SWASH model. This work serves as a paradigm to integrate traditional coastal modeling with machine learning to study coastal processes.