Real-time in situ prediction of ocean currents

Abstract The prediction of ocean currents is essential for the path planning and control Autonomous Underwater Vehicles. Regional physics-based forecast models provide valid predictions but are too computationally expensive real-time necessary AUV navigation. While vehicle sensors can measure spatial evolution currents, temporal remains an open problem as existing data-driven with capabilities have only been shown to work at locations where data used develop model. We propose in this paper two predictive tools using deep learning techniques, a Long Short-Term Memory (LSTM) Recurrent Neural Network Transformer, perform in-situ any location. A set from National Oceanic Atmospheric Administration split distinct sets train test models. show that LSTM Transformer averaged Normalized Root Mean Squared Error respectively 0.10 0.11 over all sites standard deviation 0.024 0.031. Comparisons Harmonic Method various territorial sea United States both state-of-the-art accuracy without having trained these sites.