A neural network approach for tree height estimation using IFSAR data

There is a recent interest in obtaining tree height/bald earth height estimates from high frequency (X-band) interferometric SAR (IFSAR) measurements. Conventionally, these estimates are obtained through EM modeling of the volume scattering of the vegetation canopy. In this paper, a multilayer perceptron (MLP) neural network based technique for tree height/bald earth height estimation using IFSAR data is presented. Without going through the complex EM modeling process, the neural network based technique uses a bald earth elevation ground truth data set as the target output to perform a training process. The backscatter magnitudes, the correlation coefficients, and the interferometric heights (digital terrain elevation or DTE) are applied to the network as inputs. Training and testing results have shown that the MLP networks are able to approximate the functional relationship between the IFSAR measurements and the tree heights on the ground. The neural network based technique may provide a potential solution for the estimation of bald earth elevation using high frequency IFSAR data