Global Sea Surface Salinity Retrieval Using Heterogeneous Multimodal Spaceborne BDS-3/Galileo Reflected Signals Data Fusion in Complex Marine Environments

Sea Surface Salinity (SSS) is crucial for marine ecosystems. Traditional L-band microwave radiometers can achieve SSS retrieval by correcting sea surface brightness temperature (BT). However, the calibration process is influenced by various complex factors and cannot be quantified to date. The Spaceborne Global Navigation Satellite System-Reflectometry (GNSS-R) can directly reflect changes in sea surface roughness, providing a new method for SSS retrieval. The spaceborne GNSS-R observations of the Cyclone Global Navigation Satellite System (CYGNSS) in conjunction with Soil Moisture and Ocean Salinity (SMOS) have successfully retrieved SSS. However, in addition to the proven factors such as wind speed and wave height, there are still many potential influencing factors for SSS. Therefore, in this study, a Convolutional Neural Networks (CNN)-Transformer fusion model based on BDS-3 and Galileo reflection signal data was proposed to incorporate rainfall and sea surface currents under low to medium wind speeds (<20m/s) into the retrieval. The Tianmu-1 data with global coverage was used to invert global SSS accurately. The results showed that the correlation coefficient between SSS and Soil Moisture Active Passive (SMAP) data retrieved using the CNN-Transformer fusion model based on BDS-3 reflected signals was as high as 0.87. Compared with the Artificial Neural Network (ANN) model, the retrieval accuracy of BDS and Galileo signals in the proposed model has been improved by 33.45% and 34.64%, respectively. The Bais of the BDS-3 signal retrieval ranges from -0.122 psu to 0.049 psu. The Bais of Galileo signal retrieval ranges from -0.568 psu to -0.202 psu. This provides a reference for the future application of spaceborne GNSS-R in SSS retrieval, especially in complex sea conditions and processing high spatiotemporal complexity data, which has significant advantages.