Spatial wave assimilation by integration of artificial neural network and numerical wave model

Ocean wave information is generally limited, particularly at nearshore, and attempts have been made to reconstruct spatial wave information, such as by using numerical wave models. The simulation results, however, often contain errors, and thus, wave assimilation is essential. This study aims to develop a spatial wave assimilation algorithm for significant wave height (Hs) and peaked wave period (Tp) using an artificial neural network (ANN) with data at a specific site. The ANN model is applied to correct the numerical simulation errors. The ANN inputs include the wave attenuation, numerical simulated waves, offshore wave and wind. The ANN predicted errors are then coupled back to numerical simulated results to perform wave assimilation. The case study in an Australia coast indicate that ANN-assimilated model improved the accuracy of Hs and Tp on average, 42% and 16% for RMSE, 30% and 10% for Correlation Coefficient, and 66% and 35% for Scatter Index, respectively, when compared to numerical simulated results. It also shows that the accuracy depends on the distance from the trained site, particularly at a non-linear coastline, but it could be overcome by introducing longshore wave attenuation and wave refraction. The developed spatial ANN wave assimilation model presented here can provide higher accurate wave information for nearshore regions where the numerical model is employed, and the technique developed ensured it can be transferrable to other nearshore regions.