Soil Moisture Inversion Using AMSR-E Remote Sensing Data: An artificial neural network approach

In this work artificial neural network with a back-propagation learning algorithm (BPNN) is employed to solve soil moisture retrieval for Sichuan Middle Hilly Area in China. Eighteen kinds of BPNN models have been developed using AMSR-E observations to retrieve soil moisture. The results show that the 18.7GHz band has some positive effect on improving soil moisture estimation accuracy while the 36.5GHz may interfere with deriving soil moisture, and vertical brightness temperature has a closer relationship with observed near-surface soil moisture than horizontal TB. The BPNN model driven by vertical and horizontal TB dataset at 6.9GHz and 10.7GHz frequency has the best performance of all the BPNN models withr value of 0.4968 and RMSE 10.2976%. Generally, the BPNN model is more suitable for soil moisture estimation than NASA product for the study area and can provide significant soil moisture information due to its ability of capturing non-linear and complex relationship.