The Study Of Monitoring Soil Moisture Based On Radarsat-2 Remote Sensing Images And BP Artificial Neural Network

Soil moisture related information is been treated an important parameter in domains of metrological analysis,agricultural study,etc,methods on acquiring soil moisture information accurately, abundantly and rapidly are capable of great realistic significance in monitoring crop water scarcity and analyzing hydrological forecast in our country. Results derived from those practical applications of remote sensing in monitoring soil moisture showed that, visible light-infrared related remote sensing is vulnerable to the impact of weather condition, and the accuracy of passive microwave remote sensing monitoring could not meet its requirements. However, the active microwave remote sensing is attached with advantages of round-the-clock tracking, immune to the impacts of cloud and mist, strong penetrating power and so on, serving as compensating factors for those disadvantages of visible light-infrared related remote sensing and passive microwave remote sensing in monitoring soil moisture, plus, providing new approaches in monitoring soil moisture. Currently, it has become the most promising mean in monitoring soil moisture information of surface layer.In the active microwave remote sensing based studies on soil moisture condition monitoring, comparison among different water models was conducted in terms of theoretical basis and application condition. Then the modified integral model (AIEM) was selected which complies with the soil surface roughness of the study area, besides, with the superiority of BP artificial neural network in processing non-linear data been applied. What’s more,AIEM model can better simulate the real scattering characteristics than other inversion models, making the simulation radar of four-wave polarized network model been capable of reveling the relationship between backscattering coefficient and soil dielectric constant.The AIEM model simulated data were selected and been used as training data of BP artificial neural network, so as to obtain the BP artificial neural network processed images related correlation between backscattering coefficient and soil dielectric constant. Combination of ENVI software and given parameters of corresponding radar system was adapted to dispose the data of Radarsat-2 active microwave remote sensing and finally attain the four-wave polarized associated backscattering coefficient values. Subsequently, four polarization modes (HH, HV, VH, VV) related backscattering coefficients were utilized as input values of neural network, training on the neural network was conducted by means of those AIEM model simulated data, along with the establishment of training transfer function, hidden layer nodes quantity and learning methods which aim to accomplish the optimal design of neural network, consequently, with the soil dielectric constant values been exported. The relevance between soil moisture information and soil dielectric constant was illustrated by Topp dielectric model to get the soil gravimetric water content. The experiment plot is situated in Shanba town, Hangjin Rear Banner, Bayannur city, Inner Mongolia, with the application of Radarsat-2 active microwave remote sensing contained C-band data to conduct the inversion analysis on the bare surface in the study area, as well as the correlation coefficients between inversion values and measured values been deployed. Results indicated that the inversion accuracy was satisfying, in 2013 and 2014, the correlation was 0.8503 and 0.8219respectively, could meet the requirements of practical application.