Soil Moisture Retrieval in Winter Wheat Fields at Different Growth Stages: Integrating a Two-Component Polarimetric SAR Decomposition With CIEM

Polarimetric SAR (PolSAR) offers strong volume penetrability, high resolution, sensitivity to surface dielectric properties, and the ability to acquire abundant ground target information, thus having considerable advantages in soil moisture (SM) inversion at the agricultural field scale. In recent years, numerous PolSAR model-based SM retrieval studies have been published. Nevertheless, there is no consensus within the SAR community when it comes to the identification of those existing physically-based models and algorithms providing the best performances in terms of accuracy, reproducibility, complexity and generalizability (i.e. its application to different crops and growth stages). This paper contributes to the current body of literature on radar-driven SM retrieval approaches by proposing an integrated C-band two-component polarimetric decomposition method applied on winter wheat fields. The proposed technique is built on existing vegetation and soil surface microwave models, therefore, the outcomes of the present paper shed light on the applicability and robustness of such models. Two X-Bragg surface scattering models with broader roughness adaptability and three generalized volume scattering models (namely, the Generalized Volume Scattering Model (GVSM), the Simplified Adaptive Volume Scattering Model (SAVSM) and the Simplified Neumann Volume Scattering Model (SNVSM)) were incorporated into the decomposition framework to mitigate vegetation effects and isolate the surface scattering component. Then, the soil moisture was obtained by combining the calibrated integral equation model (CIEM) with the derived optimal roughness parameter. The performance analysis was conducted by using C-band fully-polarimetric RADARSAT-2 images acquired in 2019 over winter wheat crops in southwestern Ontario, Canada. Experimental results evidence that the proposed methodology achieves a reasonable accuracy in soil moisture inversion. Notably, the X-Bragg model based on a zero-mean normal distribution combined with the SNVSM outperforms all other options with an overall RMSE of 6.17 Vol.% and a correlation coefficient of 0.53.