This paper proposes a new method to retrieve salinity profiles from the sea surface salinity （SSS） observed by the Soil Moisture and Ocean Salinity （SMOS） satellite. The main vertical patterns of the salinity prof...This paper proposes a new method to retrieve salinity profiles from the sea surface salinity （SSS） observed by the Soil Moisture and Ocean Salinity （SMOS） satellite. The main vertical patterns of the salinity profiles are firstly extracted from the salinity profiles measured by Argo using the empirical orthogonal function. To determine the time coefficients for each vertical pattern, two statistical models are developed. In the linear model, a transfer function is proposed to relate the SSS observed by SMOS （SMOS_SSS） with that measured by Argo, and then a linear relationship between the SMOS_SSS and the time coefficient is established. In the nonlinear model, the neural network is utilized to estimate the time coefficients from SMOS_SSS, months and positions of the salinity profiles. The two models are validated by comparing the salinity profiles retrieved from SMOS with those measured by Argo and the climatological salinities. The root-mean-square error （RMSE） of the linear and nonlinear model are 0.08-0.16 and 0.08-0.14 for the upper 400 m, which are 0.01-0.07 and 0.01-0.09 smaller than the RMSE of climatology. The error sources of the method are also discussed.展开更多
基金The Startup Foundation for Introducing Talent of Nanjing University of Information Science and Technology under contract No.S8113078001
文摘This paper proposes a new method to retrieve salinity profiles from the sea surface salinity （SSS） observed by the Soil Moisture and Ocean Salinity （SMOS） satellite. The main vertical patterns of the salinity profiles are firstly extracted from the salinity profiles measured by Argo using the empirical orthogonal function. To determine the time coefficients for each vertical pattern, two statistical models are developed. In the linear model, a transfer function is proposed to relate the SSS observed by SMOS （SMOS_SSS） with that measured by Argo, and then a linear relationship between the SMOS_SSS and the time coefficient is established. In the nonlinear model, the neural network is utilized to estimate the time coefficients from SMOS_SSS, months and positions of the salinity profiles. The two models are validated by comparing the salinity profiles retrieved from SMOS with those measured by Argo and the climatological salinities. The root-mean-square error （RMSE） of the linear and nonlinear model are 0.08-0.16 and 0.08-0.14 for the upper 400 m, which are 0.01-0.07 and 0.01-0.09 smaller than the RMSE of climatology. The error sources of the method are also discussed.
