Leveraging satellite observations and machine learning for underwater sound speed estimation.

Abstract

Underwater acoustics plays a vital role in climate science, marine ecosystems, environmental monitoring, mineral exploration, and oceanography. Accurate underwater sound speed data is crucial for acoustic modeling and applications such as sonar systems. However, limited data and computational constraints hinder real-time, high-resolution mapping of three-dimensional sound speed fields. We present an integrated approach that combines remote sensing, machine learning, and underwater acoustics to estimate sound speed across vast ocean regions. By analyzing sea surface temperature and salinity from satellite observations, we use machine learning to rapidly and accurately predict 3D underwater sound speed. Incorporating spatial and temporal variables enables detailed, real-time mapping. Validation against in-situ profiles and Argo float data confirms the model's accuracy across seasons, regions, and timeframes. This approach advances underwater sound speed prediction beyond traditional limits. Acoustic propagation modeling further demonstrates the potential of our model for applications in underwater detection, communication, and noise analysis.