A Sea Surface Salinity error correction model integrating physical models and complex environmental influences

To address the significant error issue of coastal Sea Surface Salinity (SSS) products caused by multiple interference factors, this study proposes an SSS error correction model for Soil Moisture and Ocean Salinity (SMOS) that integrates physical models and complex environmental influences. Based on SMOS satellite Level-2 data and relevant auxiliary data, the study constructs an SSS error correction model by fusing the Klein–Swift dielectric constant model with machine learning methods. This model significantly improves the accuracy of SMOS SSS products. Experimental validation is conducted in the South China Sea (4°E–26°E, 104°N–125°N) as the main study area. Results show that the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of the proposed method are 0.14 psu and 0.11 psu, respectively. Compared with original SMOS SSS data, these errors decrease by 2.17 psu and 1.71 psu. When compared with machine learning and deep learning models such as Transformer and LSTM, the proposed method also delivers significant advantages. Furthermore, through cross-year and cross-season experiments conducted in the South China Sea, as well as cross-regional validation in the Coral Sea, the proposed model is demonstrated to exhibit strong adaptability and generalization capability across diverse environmental conditions. It provides a new method and idea for obtaining high-precision SSS in complex coastal environments.