An Ocean Emissivity Model Trained from Polarization BRDF Matrix through Multilayer Perceptron Neural Network

The ocean surface emissivity model plays a pivotal role in satellite data assimilation and the retrieval of ocean physical parameters. In our previous research, we developed a physical emissivity model featuring a polarized Bidirectional Reflectance Distribution Function (pBRDF-E). This model effectively ensures the consistency between surface emission and reflection parameters. However, it suffers from low computational efficiency. In this study, we introduce a fast ocean emissivity model, OceanEM. Leveraging the emissivity data output from the pBRDF-E model, OceanEM is developed by using a multilayer perceptron neural network. It can compute the polarization emissivity vector across a wide range of conditions: incidence angles from 0° to 80°, wind speeds from 2 to 50 m s⁻¹, sea surface temperatures from −2° to 30°C, sea surface salinities from 0 to 40 psu, and frequencies from 1.4 to 410 GHz. Alongside the FAST Microwave Emissivity Model (FASTEM6) and SURface Fast Emissivity Model for Ocean (SURFEM-ocean), OceanEM is integrated into the Advanced Radiative Transfer Modeling System (ARMS) as a user-selectable option. To validate the accuracy of OceanEM, we compare it with FASTEM6 and SURFEM-ocean using data from WindSAT, a polarimetric radiometer onboard the Coriolis satellite. The results show that the three models generally yield consistent simulations of WindSAT brightness temperatures. Specifically, for channels at 6.8 GHz, 10.7 GHz (both horizontal and vertical polarization), and 18.7 GHz (vertical polarization), OceanEM demonstrates higher accuracy than FASTEM6 but lower than SURFEM-ocean. Conversely, for channels of 18.7 GHz (horizontal polarization), 23.8 GHz, and 37.0 GHz (both horizontal and vertical polarization), OceanEM outperforms both FASTEM6 and SURFEM-ocean.