Soil Moisture Retrieval Using AMSR-E Data By BP Neural Network For Sichuan Middle Hilly Area

Soil moisture content is a crucial parameter in research fields of Meteorology, Hydrology, Ecology and Science of Agriculture, and one of the most important environmental process factor of drought. Exact description of its dynamic change and spatial distribution is the key to learn the situation and degree of drought, and is also of significance to soil moisture conditions monitoring and study of spatial pattern. The common soil moisture forecast models include Soil Water Dynamic Research method, Water Balance method. Empirical Formula method, Time Series Model method etc. however, all of these methods has shown certain problems. For example, many factors are to be measured, and the required data are tremendous. Microwave Remote Sensing, such as Advanced Microwave Scanning Radiometer-EOS, AMSR-E, has the advantages of strong penetration, insusceptible to atmosphere, cloud and fog. Compared with traditional soil moisture forecast model. Artificial Neural Network, like Radial Basis Function Network, is able to deal with multi-factor and non-linear problem. Given this, this article aims to build a model to invert soil moisture for research area based on AMSR-E remote sensing data and RBF neural network.After a large number of experiments, the author of current research combined data adopted from four AMSR-E wave bands-Tb6.9V, Tb6.9H、Tb10.7V、Tb10.7H-with vector diagram of hilly area in Sichuan and site actual observed value, built a RBF Neural Network Model by applying GIS technology and Matlab software inverted soil moisture of the target area. This research also conducted a correlation analysis of the inversion result and site actual observed soil moisture data, assessed inversion accuracy and carried out a space-time validation by combining data of rainfall of target area and soil moisture carried by AMSR-E data (AMSR-E soil moisture product value). This research has reached conclusions as following:1.Viewing from time series, the soil moisture value of RBF model and actual observed data has certain delay on the time drought occurred compared with accumulative rainfall; AMSR-E soil moistures products is more sensitive to accumulative rainfall than simulation value of RBF model. However, the inverting soil moisture value is about25percent less than both the actual observed data and simulation value of RBF model. While, viewing from the overall trend of time series, the simulation value of RBF model is more close to the actually observed, but its time series curve is more smooth than that of the later and smoothes the larger or smaller value of the pixel.2.From the perspective of space, the numerical size of simulation value of RBF model is consistent with observed value on space distribution, and is more close to the later. On the other hand, AMSR-E soil moisture product value is smaller than both observed value and simulation value of RBF model. On space distribution, the trend of the numerical size of AMSR-E soil moisture inversion value and that of observed value are unstable. Compared with AMSR-E soil moisture inversion value, the simulation value of RBF model in the northwest region of Sichuan hilly area conforms more to the actual observed value, but different in their overall trends. On the whole, simulation value of2009is much better than that of2010, and much closer to actual observed value.In the south region of Sichuan hilly area, there is a relatively bigger difference between the simulation value of RBF model and actual observed value due to the negative influence the complex terrain and climate had on RBF simulation. Both the simulated value of soil moisture by RBF model and the observed soil moisture value cannot reflect the influence of some factor change on soil moisture. For example, the decrease of soil moisture value under both circumstances will show some delay on time affected by long-term less rainfall, but AMSR-E can better reflect the change of soil moisture content caused by factors like raining. This may be because AMSR-E soil moisture product mainly monitors the moisture of1-2cm soil layer, while observed value and simulation value of RBF model focus on that of10cm soil layer.To sum up, after comparing accumulative rainfall and AMSR-E soil moisture product and inverting soil moisture of Sichuan hilly area by applying AMSR-E and RBF Neural Network Model, the author reached a conclusion that the inversion outcome of RBF Neural Network Model is closer to actual observed value than AMSR-E soil moisture product both on time series and special distribution.