Soil Moisture Retrieval Over Crop Regions Based On Generalized Polarimetric SAR Target Decomposition

Accurate,timely,and large-scale acquisition of soil moisture is critical for irrigation status,disaster assessment and yield estimation in agricultural fields.Polarimetric synthetic aperture radar(Pol SAR),as an active microwave remote sensing technology,has become an important approach for soil moisture retrieval because of its large-scale,high-resolution imaging capability,strong penetration ability,and its sensitivity to target structure and dielectric properties.Over vegetation covered fields,the key to accurately estimate soil moisture is to effectively remove the contribution of vegetation backscattering and retain only the backscattered signal which is associated with the soil surface scattering.The model-based polarimetric SAR target decomposition technique can separate to some extent the backscattered radar signals from the vegetation canopy and the underlying ground by modeling different scattering mechanisms in the scene,hence enabling an effective strategy for soil moisture retrieval in vegetated agricultural fields.However,the simplified scattering models in traditional polarimetric decomposition methods are not adequate to simulate the complex scattering process of the ground and vegetation layers,and there are underdetermined problems in the model parameter solution process,leading to the difficulties in accurately decoupling the observed signals,which consequently affects the inversion performance of soil moisture.In view of this,this thesis carried out the research on the soil moisture retrieval over crop regions based on generalized polarimetric SAR target decomposition.Firstly,the decomposition framework was constructed according to the characteristics of scattering scenes.Then,the generalized scattering models of different scattering mechanisms were established separately,and the optimized model parameter solution strategy was proposed.Finally,soil moisture information of different crop growth stages was obtained using fully polarimetric RADARSAT-2 C-band observation data.The major research content and scientific innovations of this study are summarized as follows:(1)A generalized two-component model-based decomposition method based on the physical constraint of dielectric constant is proposed,which effectively improves the removal effect of vegetation scattering influence and the inversion accuracy of soil moisture.With the aim of the problems that the surface scattering model of traditional polarimetric decomposition is only suitable for low roughness condition,and the discrete volume scattering model cannot accurately describe the complex vegetation structure,the X-Bragg surface scattering model based on zero-mean normal distribution and three generalized volume scattering models are incorporated into the decomposition framework.The analytic expressions of the parameters are derived directly and easily from the observed values,and the dielectric constant physical constraint and the minimum residual power criterion are employed to optimize the model for determining the optimal solution.The performance of three generalized volume scattering models in describing different vegetation types are evaluated from the perspective of theoretical modeling and scattering power,and their application potential in soil moisture retrieval is compared.The experimental results show that any of three generalized volume scattering models in the generalized two-component decomposition framework proposed in this study can provide promising inversion results of soil moisture,with root mean square errors(RMSE)ranging from 2.89% to 7.43%.Compared with the other two models,the simplified Neumann volume scattering model(SNVSM)simulates the vegetation scattering contribution more accurately,and it shows a stable performance in soil moisture inversion at different crop phenological stages,with an overall accuracy of RMSE=4.99% and a correlation coefficient of R=0.78.(2)A soil moisture retrieval method that considers the effect of rough surfacestalk scattering is proposed,which efficiently improves the accuracy of the scattering signal decoupling and the robustness of the soil moisture inversion algorithm.To address the problem that the two-component decomposition model ignores the effect of dihedral scattering caused by crop stalks,a generalized three-component decomposition method is constructed,in which an improved double Fresnel dihedral scattering model considering the surface depolarization effects is introduced,and the seasonal structural changes of vegetation are adaptively modeled using the SNVSM.In response to the underdetermined solution problem of parameters resulted from insufficient observation information,the surface depolarization angle is approximated using the circular polarization correlation coefficient,and the volume scattering power is determined based on the idea of generalized eigenvalue decomposition.Making full use of the polarimetric SAR coherency matrix information,a nonlinear fitting solution strategy is proposed to obtain the surface dielectric constant,and then the surface soil moisture is estimated by the dielectric mixing model.The experimental results show that the proposed method can finely describe different scattering mechanisms in the target scenes at different growth stages of crops and has promising soil moisture inversion performance.The RMSEs of soil moisture inversion results on corn and wheat fields reach 5.74% and 5.70%,respectively,which provides an accurate description of the time-series dynamics of soil moisture under crop cover.