High-Precision Prediction Model for Sea Surface Salinity

To build a highprecision ocean surface salinity prediction model, the back propagation (BP) neural network method is utilized to establish a sea surface salinity prediction model based on soil moisture and ocean salinity (SMOS) satellite level 1C brightness and temperature data and auxiliary data. The array for realtime geostrophic oceanography (ARGO) buoy observations are used as the measured value of sea surface salinity to test the accuracy of the new model^ʼs prediction results, and the verification set is used to verify the accuracy of the model. The results show that the sea surface salinity predicted by the new model (referred to as SSS0) is more accurate than the three roughness model salinity products of soil moisture ocean salinity (SMOS) satellites (referred to as SSS1, SSS2, and SSS3). The accuracy of the root mean square errors of SSS0, SSS1, SSS2, SSS3 compared to SSS_ARGO are 0.847 3, 2.041 7, 2.028 8 and 2.080 5, and the absolute average errors are 0.755 3, 1.422 6, 1.421 6 and 1.456 6. Both of the root mean square error and absolute average error of SSS0 are significantly smaller than SSS1, SSS2, and SSS3. Therefore, it shows that the sea surface salinity prediction model established in this paper has higher accuracy, and it provides a novel way for generating the sea surface salinity inversion algorithm.