Phase-resolved wave prediction from radar observations based on a multi-scale Fourier neural operator

Phase-resolved ocean wave prediction is a vitally important but challenging task in ocean engineering. Any algorithm used for real-time reconstruction of the wave field must be both efficient and accurate. With this aim in mind, we propose a novel pipeline for phase-resolved wave prediction based on radar observations. The core architecture is based on a multi-scale version of the Fourier Neural Operator (FNO), which enhances the modeling of high-frequency components. Numerical simulations of nonlinear wave fields and their corresponding radar images are then used to train and test the proposed models. It is demonstrated that the proposed method achieves high accuracy and efficiency in wave prediction, with the multi-scale FNO model reducing prediction errors by approximately 14% compared to the standard FNO model. The method also exhibits robustness to noise and can be generalized to unseen sea conditions. Overall, the proposed technique is promising for real-time, phase-resolved wave prediction for use in ocean engineering practice.