Combining Conv-LSTM and wind-wave data for enhanced sea wave forecasting in the Mediterranean Sea

This study presents an application of a stateful Convolutional Long Short-Term Memory (Conv-LSTM) model for wave forecasting in the Mediterranean Sea. By leveraging bathymetric data and wind fields, the model predicts key oceanographic variables such as significant wave height ( H s ), peak period ( T p ), and wave direction ( θ ). By incorporating wave buoy measurements into the training data, the Conv-LSTM model effectively captures both spatial and temporal dynamics, particularly in regions characterised by complex wind-wave interactions. While the model shows high accuracy in predicting short-term wave variability, especially in central Mediterranean areas, it exhibits limitations in coastal regions under extreme weather conditions, where local factors and missing variables (e.g., air pressure, air temperature) reduce its accuracy (from 90% to 78%). Validation of measured data confirms the potential of the model to improve operational forecasting, maritime safety, and offshore engineering projects and highlights the need for improving spatial resolution and the inclusion of additional meteorological inputs for future applications. • Conv-LSTM A ppication : Predicts wave height (Hs), peak period (Tp), and direction (θ) in the Mediterranean Sea. • Data & Training: Uses bathymetry, wind data, and buoy measurements to model spatial-temporal wave dynamics. • Performance : Achieves up to 90% accuracy centrally but drops to 78% in coastal areas under extreme weather. • Comparison with CMEMS Data : Captures wave peaks better but may overestimate, improving short-term dynamic accuracy. • Future Improvements : Higher resolution and added meteorological factors like air pressure and temperature for better forecasts.