Abstract
This study proposes a novel deep learning model, the graph convolutional gated recurrent unit (GC-GRU), to address the critical challenge of accurate forecasting of ocean wave heights due to the complex nonlinear spatiotemporal variability of wave dynamics. The proposed model, which integrates the strengths of graph convolutional networks (GCNs) for spatial feature extraction and gated recurrent units (GRUs) for temporal feature extraction, allows for effective capture of complex spatiotemporal patterns in wave height data and is evaluated on a dataset of 666 observation stations in the Gulf of Mexico, forecasting wave heights up to 36 h in advance. Comparative experiments with traditional CNN and GRU models demonstrate the superior predictive performance of the GC-GRU approach. Additionally, we introduce the shapley additive explanation (SHAP) values to provide physical insights into the key physical variables and historical patterns driving the model's predictions. The results show that wind speed and mean wave period are the most influential factors related to wave height variations. It is expected that this work presents a significant advancement in wave height forecasting by introducing the innovative GC-GRU architecture and leveraging SHAP analysis to interpret the model's inner workings. The findings are expected to have important implications for enhancing coastal and maritime operations as well as informing our understanding of complex ocean wave dynamics.