Spatiotemporal Deep Learning for Ocean Wave Forecasting and Vessel Motion Modeling

Abstract

Reliable prediction of ocean wave dynamics and vessel motion is essential for maritime safety and operational efficiency. This paper presents a deep learning framework that integrates spatiotemporal oceanographic data with vessel kinematic signals to jointly model wave evolution and ship response. The proposed model employs a hybrid architecture combining convolutional neural networks for spatial feature extraction and transformer-based temporal modeling to capture long-range dependencies. The dataset includes 10 years of buoy-based wave measurements and onboard sensor data from cargo vessels operating in the North Pacific, totaling over 3.5 TB of multivariate time-series data. Experimental results show that the model achieves a wave height prediction RMSE of 0.21 m and improves vessel roll motion prediction accuracy by 18% compared to physics-based simulation methods. In extreme sea states (significant wave height > 5 m), the model maintains stable performance with less than 12% degradation.