Neural network algorithm and backscattering model for biomass estimation of wetland vegetation in Poyang Lake area using Envisat ASAR data

Abstract

Poyang Lake is the largest freshwater lake in China with an area of about 3000 km2. Its wetland ecosystem has a significant impact on China's environment change. In this paper, we discuss the neural network algorithms (NNA) to retrieve wetland vegetation biomass using the alternating polarization Envisat ASAR data. Two field measurements were carried out coincident with the satellite overpasses at this area through the hydrological cycle from April and November. Training data of the neural network are generated by the Michigan Microwave Canopy Scattering (MIMICS) model which is often used for the tree canopy. We modified the model to make it applicable to herbaceous wetland ecosystems. The model input parameters are defined according to the wetland circumstance. NNA retrieval results are validated with ground measured data. The inversion results show the NNA combined with MIMICS model is capable of performing the retrieval with good accuracy. Finally, the trained neural network is used to estimate the overall biomass of Poyang Lake wetland vegetation.