Soil Moisture Retrieval Using AMSR-E Data By ANN Neural Network For Huaihe River Valley

Soil moisture is the common name of various forms of water in the soil (or state). It is a basic factor for crop drought monitoring. Accurate drought monitoring and prediction are great significance to the sustainable development of agriculture. By drought disaster arising frenquently in the basin, it has brought huge losses of the agricultural production and economy to the Huaihe River Basin as well as China, so it is very necessary that predict soil moisture in Huaihe River Basin. The traditional soil moisture monitoring method is slow and it needs a lot of manpower and resources. Even more importantly, it is difficult that meet the need of real-time and large-scale predict. The traditional AMSR-E soil moisture product which is based on the physical significance and the atmospheric propagation process is a soil moisture inversion model for the global. It is difficult to accurately reflect the soil moisture in a specific area, however, the soil moisture estimation model which is based on ANN artificial neural network and AMSR-E data to establish can get different results in different regions. The model could learn and improve by self continuously in order to achieve the best inversion results. BPNN model is the most popular and widely used among the ANN model, it has a great advantage in the application of soil moisture estimation that can obtain composite nonlinear processing ability by mapping the simple non-linear processing unit complex, so it has a great advantage in inversion of soil moisture. This research that is based on AMSR-E data establish the BPNN artificial neural network and retrieval soil moisture in the Huaihe River Basin and establisht a more accurate prediction of the network model, in order to obtain a better result and provide a more accurate soil moisture product to meet the agricultural drought in Huaihe River Basin finally.In this research, the datas used are observed soil moisture and AMSR-E soil moisture from2006to2010in Huaihe River Basin, then, choosing different bands of the AMSR-E as an input factor. First, we should pretreat the data which is preferred as input factor; Secend, we should construct2-factor,3-factor and4-factor model to invert the soil moisture by the processed data; In the last, we can evaluate and analyze the optimal BP model which is filtered out combine with the rainfall in the researching area, evaporation data and standard precpitation Index (SPI). Through the researching, I get some conclusions as the following:l.At less plains and vegetation areas, the BPNN model which has much preferred factors can get accurater results than small number of preferred factor model; At more hilly and vegetation areas, less preferred number of factors has, better the BPNN model’s effct is. Overall, the effect of2-factors model are poor, they are inferior to the3-factors and4-factors model. It shows that the accuracy of the model is much higher when the preferred input factors are3and4.2. AMSR-E soil moisture product is more sensitive than the retrieval value of the BP-NN model for the extreme change of precipitation. Compared with the observed soil moisture and AMSR-E soil moisture product, the retrieval value of BPNN model has a certain lag when precipitation is little for a long time.3. Not only is there a big error between AMSR-E soil moisture product and the measured,then, AMSR-E soil moisture product can not accurately reflect the size and scope of the observed soil moisture, but also the distributable in time and space is also inconsistent between AMSR-E soil moisture product and the general trend of the the observed soil moisture. However, the retrieval value and distributable trend in time and space of the BP neural network model is very close to the observed soil moisture.4. Compared with AMSR-E soil moisture product, the simulated arid region spatial distributable trend by BPNN model is closer to the reflected by SPI index.Through the research, we learn that the effect of soil moisture retrieval from AMSR-E and BPNN-based model is better than the traditional AMSR-E comes with the equation and the model has a value in use and application in the Huaihe River Basin drought monitoring. All the words, the simulation of the effect which of AMSR-E and BPNN neural network model-based is good, but by the trial,we found that there were some issues to be explored further when we established a model, such as the low spatial resolution of AMSR-E data, it is a major cause of that the simulation results are not so good. If we want to improve the spatial, we need to integrat the multi-source remote sensing data, such as fusing MODIS data and AMSR-E data.