AN ARTIFICIAL NEURAL-NETWORK FOR INVERSION OF VEGETATION PARAMETERS FROM RADAR BACKSCATTER COEFFICIENTS

An inverse scattering model based on an artificial neural network using the back-error propagation technique is presented in this paper. The vegetation canopy is modelled as a half-space of randomly distributed and orientated circular disks, representing the leaves. A Monte Carlo model is used as the forward scattering model to calculate the radar backscatter coefficients. Using different sets of input parameters such as leaf moisture content, frequency and size of leaf, this forward model generates about 2000 sets of backscatter coefficients. These sets of backscatter coefficients and the corresponding input parameters form a training set for the neural network. The trained neural network constitutes an inverse scatter model. Given a set of backscatter coefficients, our model is able to estimate simultaneously the leaf moisture content, the radius and the thickness of the circular leaf within an accuracy of 10%.