- Retrieval of soil moisture is crucial for weather and climate change predictions as well as in decision-making for agriculture management practices. Several theoretical, semi-empirical, empirical, and machine learning-based scattering models have been developed for simulating fully polarimetric (FP) synthetic aperture radar (SAR) data. The potential of SAR compact polarimetry (CP) for soil moisture retrieval over bare agriculture fields is still an active area of research. In this study, we develop new CP backscattering models for soil moisture retrieval under bare soil conditions. First, we apply empirical relationships between FP and CP to formulate an empirical CP-advanced integral equation model (AIEM). In addition, we propose two adapted theoretical models CP-AIEM and CP-improved IEM (CP-I2EM) using the direct analytical relationship between FP and CP. We also propose two empirical models CP-Dubois and CP-Oh by recalibrating Dubois and Oh models using CP observations and in situRetrieval of soil moisture is crucial for weather and climate change predictions as well as in decision-making for agriculture management practices. Several theoretical, semi-empirical, empirical, and machine learning-based scattering models have been developed for simulating fully polarimetric (FP) synthetic aperture radar (SAR) data. The potential of SAR compact polarimetry (CP) for soil moisture retrieval over bare agriculture fields is still an active area of research. In this study, we develop new CP backscattering models for soil moisture retrieval under bare soil conditions. First, we apply empirical relationships between FP and CP to formulate an empirical CP-advanced integral equation model (AIEM). In addition, we propose two adapted theoretical models CP-AIEM and CP-improved IEM (CP-I2EM) using the direct analytical relationship between FP and CP. We also propose two empirical models CP-Dubois and CP-Oh by recalibrating Dubois and Oh models using CP observations and in situ data. To compare the soil moisture retrieval performance of developed CP-backscattering models with that of a standard machine learning approach, the random forest (RF) technique is utilized. The six adapted and developed algorithms are tested using the C-band CP data of Canada’s RADARSAT constellation mission (RCM). In situ soil moisture, roughness, and texture measurements were collected from bare agriculture fields of Lennoxville and Montérégie, Quebec, Canada, for calibration and validation of models. Results show that the RF model provides accurate estimates (error: RMSE = 0.04 m3/m3, correlation: r = 0.8 and inversion rate: (IR) = 100%) while requiring extensive training. CP-Oh and CP-Dubois models perform adequately (RMSE = 0.08 m3/m3, r > 0.7 and IR > 60%) having limited applicability range. In the end, CP-AIEM offers the best choice including reasonable accuracy (RMSE = 0.07 m3/m3, r = 0.6, and IR = 64%), wider applicability range and transferability without requiring calibration.…
