Long-term wave height forecasting using VMD-informer

Abstract Accurate oceanic weather forecasting plays a crucial role in various marine applications, from wave energy resource assessment to the establishment of operational safety limits for maritime activities. Among the key oceanic parameters, significant wave height is of particular importance due to its direct impact on marine operations. Traditional numerical simulations, while effective, require precise boundary conditions and substantial computational resources, often leading to long processing times. In contrast, deep learning approaches, leveraging powerful neural networks, have gained increasing attention for their ability to generalize and model complex, nonlinear relationships in data. However, current deep learning-based predictive models still face challenges regarding prediction accuracy and generalizability, particularly over extended forecast periods. To address these challenges, we propose an innovative predictive framework, VMD-Informer, which combines deep learning techniques with signal processing methods to improve the accuracy of significant wave height predictions over long forecasting horizons. The framework utilizes the Variational Mode Decomposition (VMD) method to decompose wave signal data during the preprocessing stage, enhancing both processing efficiency and prediction accuracy. The model construction incorporates the Informer model, which is specifically designed to ensure high accuracy across multi-step long-term time series predictions. Using data from NOAA's global buoy station 46,078, covering the years 2018–2019, our experiments demonstrate that the VMD-Informer model outperforms traditional machine learning models, particularly in predicting significant wave height for longer forecast intervals. These results highlight the potential of the VMD-Informer approach for advancing the accuracy of long-term oceanic weather predictions, providing valuable insights for marine forecasting systems.