Neural Network-based Analytical Model for Biomass Estimation in Poyang Lake Wetland Using Envisat Asar Data

Poyang Lake is the largest freshwater lake in China with an area of about 3,000 km2. Its wetland ecosystem has a significant impact on China’ s environment change. Traditional way to monitor biomass change on this area is to use linear or nonlinear model from TM/ETM data. In this paper we discuss the neural network algorithms(NNA) to retrieve wetland biomass from multipolarization(HH and VV) backscattering values using Envisat ASAR data. Two field measurements were carried out concomitant to the acquisition of ASAR images in this area through the hydrological cycle from Apr and Nov. Training data of the network are generated by MIMICS model which is often used in the forest. We simplify the model to make it available on the wetland system. The model input parameters are defined according to the real wetland circumstance. NNA retrieval results are validated with experimental data. The inversion results show that the NNA is capable of performing the retrieval with good accuracy. Finally, the trained neural network is used to estimate the overall biomass of the Poyang Lake. The wetland biomass reaches a level of 1.06x109 kg,1.72x108 kg, 1.0x109 kg in April, July and November 2007.