Multi-parametric inversion in microwave remote sensing with an artificial neural network trained with multiple scattering theory

Summary form only given. The inversion problem in remote sensing consists of multiparametric inversion from multifrequency and polarimetric remote sensing measurements. The authors performed such inversions with artificial neural networks and constrained iterative inversions trained with multiple scattering theory. They performed the inversion of four snow parameters from passive microwave remote sensing measurements with a neural network trained with the dense media radiative transfer theory. The four parameters were snow-water equivalent (or snow depth), mean-grain size of ice particles in snow, snow density, and snow temperature. The inversion was from passive microwave remote sensing measurements of five brightness temperatures: 19 GHz V-polarization, 19 GHz H-polarization, 22 GHz V-polarization, 37 GHz V-polarization, and 37 GHz H-polarization. The neural network worked very well on synthetic testing data. The authors demonstrated mapping of an entire region based on synthetic data and the inversion technique.<>