Physics-Based Spatio-Temporal Modeling With Machine Learning for the Prediction of Oceanic Internal Waves

Abstract

Accurately predicting the occurrence of oceanic internal waves in the northeastern South China Sea is of great importance to marine ecosystems, and economy. The traditional physics-based models for monitoring the occurrence of internal waves require complex parameterization, and the partial differential equations (PDEs) are relatively difficult to solve. The emergence of integrating physical knowledge and data-driven models brings light to solving the problem, which improves interpretability and meets the physical consistency. It not only inherits the advantages of machine learning in massive data processing but also makes up for the “black box” characteristics. In this paper, we propose a physics-based spatio-temporal data analysis model based on the widely used LSTM framework to achieve oceanic internal wave prediction. The results show higher prediction accuracy compared with the traditional LSTM model, and the introduction of physical laws can improve data utilization while enhancing interpretability.