Effective Predictions of Snow Scattering Characteristics Using Neural Networks and The Quasi-Crystalline Approximation

Prediction of snow scattering characteristics is critical to active and passive microwave remote sensing of snow. In this paper, we combine the quasi-crystalline approximation (QCA) scattering model and a neural network to effectively predict the scattering parameters of snow. These include the effective permittivity, the scattering coefficient, the absorption coefficient, and the phase matrix. The physics driven neural network is shown to outperform a purely data-driven neural network. The developed framework significantly speeds up the computational efficiency compared to the QCA scattering model.