Coastal Sea Surface Temperature Inversion from Microwave Radiometer using Radial Basis Function Neural Network

A radial basis function neural network (RBFNN) method is proposed to retrieve sea surface temperature (SST) in global coastal waters by passive microwave (PMW) radiometer measurements. Based on the RBFNN method, an optimal feature combination for the model input layer is obtained. The best parameters setting of the model is determined by grid-search method. Finally, a prediction model of RBFNN which is suitable for SST inversion of coastal waters is constructed. The inversion accuracy of the model was evaluated using the new data. Results show that SSTs retrieved by this model has a mean bias of
$\mathbf{0.071}^{\mathbf{o}}\mathbf{C}$
and root-mean-square error (RMSE) of
$1.18^{\mathrm{o}}\mathrm{C}$
in the global coastal waters. The accuracy of this model is not bad and can be used as an effective supplement to the SST inversion in coastal waters.