Rapid 3D Mapping of Underwater Sound Speed Using Sea Surface Data-based Machine Learning Model

Accurate underwater sound speed data is crucial for acoustic propagation modeling and applications such as sonar systems. However, conventional methods face challenges in providing real-time, high-resolution mapping of three-dimensional (3D) sound speed fields due to limited data availability and computational constraints. This study presents a machine learning model that leverages readily available sea surface temperature and salinity data from satellite observations to rapidly and accurately estimate 3D underwater sound speed. The model is trained to capture the relationships between surface data and subsurface sound speed, integrating both spatial and temporal variables. Validation against in-situ profiling and Argo float measurements demonstrate the model’s ability to deliver efficient, high-resolution 3D sound speed maps with reasonable accuracy. This approach offers a significant advancement in real-time underwater sound speed prediction, overcoming the limitations of traditional methods. The results of acoustic propagation modeling further suggest the model’s applicability for various underwater operations involving low- to mid-frequency acoustic sources, including detection, communication, and noise propagation.