Simulation of microwave backscattering at sea surface with small incidence angle based on back propagation neural network

Abstract This study uses normalized radar cross sections (NRCS) from GPM and wind products from NDBC buoy to simulate microwave backscattering of the sea surface under the incidence angle of less than 18°. First, Spearman correlation coefficient is used to analyse the correlation between NRCS and wind and wave parameters to form different parameter combinations. BPNN is selected as our basic estimation model, and is modified by adjusting the number of network layers and hidden layer neurons, hyper-parameters. Then different combinations of parameters are selected as inputs to the BPNN model. By comparing the different model effect, the optimal BPNN NRCS regression model is selected. The evaluation results show the optimal BPNN model achieve superior NRCS estimation with the root mean square error (RMSE) of 2 dB. Moreover, the model can accurately simulate the relationship between NRCS and wind speed and incidence angle, which solves the problem of the lack of empirical models for backward scattering with small incidence angle.