Reconstructing the Three-Dimensional Thermohaline Structure of Mesoscale Eddies in the South China Sea Using In Situ Measurements and Multi-Sensor Satellites

The evolution of the three-dimensional thermohaline structure of mesoscale eddies is crucial for assessing energy and mass transfer during their long-distance propagation in the ocean. However, the understanding and quantitative evaluation of the role that mesoscale eddies play in driving variations of thermohaline in the deep sea remains constrained due to the scarcity of in situ observations, particularly in marginal seas such as the South China Sea (SCS). In this study, we propose an artificial intelligence (AI)–physics-based deep learning model that integrates satellite measurements and Argo data from 2003 to 2021 to reconstruct the three-dimensional thermohaline structure of mesoscale eddies in the SCS. Besides utilizing basic sea surface hydrodynamic parameters obtained from satellite data for model training, an additional branch incorporating eddy physical parameters was introduced to optimize the model. The results demonstrate that the model effectively reconstructs thermohaline properties within mesoscale eddies in the SCS. Compared to Argo observations, the average root mean square error (RMSE) for temperature (salinity) within anticyclonic eddies was 0.34 °C (0.036 PSU), while it was 0.36 °C (0.032 PSU) within cyclonic eddies in the upper 1500 m. Further validation using high-resolution glider observations tracking an anticyclonic eddy originating in the SCS confirms the model’s efficiency, achieving an RMSE of 0.2962 °C (0.0138 PSU) for temperature (salinity). The accuracy of our proposed model significantly outperforms that of HYCOM and GLORYS simulations, with the RMSE reduced by 40% to 60%. The distinctive capabilities provide valuable insights into understanding the fine-scale structures of mesoscale eddies, especially in regions with limited in situ data.