GWSM4C: A global wave surrogate model for climate simulation based on a convolutional architecture

Abstract

Accurate and rapid ocean surface wave simulations are essential for navigation safety, marine activity, and climate change research. However, traditional numerical wave models incur substantial computational costs, especially in long-term simulations for climatology research. Deep-learning methods have shown promise in enabling faster simulations with fewer computational resources. However, the cumulative error in current deep-learning wave models limits their use as surrogate component models in climate modeling. In this work, a deep-learning model, named the Global Wave Surrogate Model for Climate simulation (GWSM4C), was developed based on a convolutional architecture, which accounts for the physical processes (wind waves and swells) of waves. Because GWSM4C can automatically generate initial conditions for each simulation moment depending only on historical wind fields, the accumulation of errors from previous simulated wave states can be avoided. The experimental results demonstrated that simulated global significant wave height produced by GWSM4C exhibited a correlation coefficient of 0.94 and a root-mean-square error of 0.21 m compared with the traditional numerical wave model. Furthermore, the generalization test of the GWSM4C indicates that it is feasible to use GWSM4C for long-term wave simulation in climatology-related studies or to integrate it into a coupled climate model as a surrogate wave component model.