Model-based polarimetric decomposition can separate to some extent the backscattered radar signals from the vegetation canopy and the underlying ground, hence enabling a strategy for soil moisture retrieval in vegetated agricultural fields. However, the volume scattering models used in previous studies are only applicable to specific cases, making it difficult to completely remove the volume component induced by the vegetation layer. In this paper, three generalized volume scattering models (i.e., generalized volume scattering model (GVSM), simplified adaptive volume scattering model (SAVSM), and simplified Neumann volume scattering model (SNVSM)) are incorporated in the decomposition and evaluated for soil moisture retrieval. Considering the complexity and descriptive ability of the available physical models, a modified two-component model-based decomposition is proposed as the basic decomposition framework. This decomposition is also based on the physical constraint of the dielectric constants included in the model. The employed models combine an X-Bragg surface scattering model with three continuous generalized volume scattering models. The analytic solution of the parameters is obtained, and the minimum residual power criterion is used to determine the optimal solution to fit the model. By using the proposed model-based decomposition framework, the performance of the three models to simulate the canopy scattering and, as a result, to later estimate soil moisture under agricultural vegetation is evaluated. Fully polarimetric RADARSAT-2 C-band images acquired on eight dates in 2013 and 2015 over fields covered by two crops (wheat and soybean) were employed for validation. Results show that the proposed decomposition method, using any of the three volume scattering models, can provide promising inversion results of soil moisture, with RMSEs ranging from 2.89 to 7.43 [vol.%]. Compared with the other two models, the SNVSM simulates the vegetation contribution more accurately in this framework, and it provides a stable soil moisture inversion performance at different crop phenological stages, with an optimal overall accuracy of RMSE = 4.99 [vol.%] and a correlation coefficient of R = 0.78.
Read full abstract