Mapping the spatial distribution and time evolution or snow water equivalent using neural network iterative approach and a snow hydrology model

The estimation of snow parameters by using passive microwave remote sensing data such as SSMR and SSM/I has been studied for many years. In previous parameter retrieval using passive remote sensing, the techniques have been largely limited to using linear regression relationships between measurements and one snow parameter, such as snow-water equivalent. However, the brightness temperatures are not only influenced by snow-water equivalent, but also by snow-grain size, snow temperature, and snow density. Thus, it is important to develop a multi-parametric inversion algorithm using multi-frequency and dual polarization measurements. In this paper, we use a neural network iterative approach for parameter inversion in combination with a snow hydrology model for a priori estimates. The neural network (NN) approach is used to invert the parameters from the measurements for computational efficiency. In the scheme of forward neural network iterative inversion, the initial guesses of snow parameters are given by a physically based snow hydrology model which uses a priori information of weather data as input. The results of applying the algorithm to the Northern Hemisphere are illustrated.