<p>As one of the most important parameters in earth surface, soil moisture plays a crucial role in in many fields, such as agriculture, environment, hydrology, ecology and water management. With the development of earth observation technology, Synthetic Aperture Radar (SAR) provides a powerful method to estimate soil moisture at diverse spatial and temporal scales. However, in agricultural area, soil moisture estimated by SAR often obstructed by vegetation cover. Volume scattering and vegetation attenuation can complex the received SAR backscatter signal when microwave interacts with vegetation canopy. In this study, a model-based polarimetric decomposition and the two-way attenuation parameter in Water Cloud Model (WCM) were adopted to remove the effect of volume scattering and vegetation attenuation respectively. And a deorientation process of SAR data was applied to remove the influence of randomly distributed target angles before polarimetric decomposition. After that, the Dubois model was used to describe the underlying soil backscattering and retrieve soil moisture. Optimal surface roughness was adopted to parameterize the Dubois model due to the difficulty of soil roughness measurement under vegetation cover. This soil moisture estimation method was applied to soybean fields with time-series RADARSAT-2 SAR data. Validation based on in-situ measured soil moisture demonstrates that the proposed method is capable of estimating soil moisture over soybean fields, with Root Mean Square Errors (RMSEs) of 9.2 vol.% and 8.2 vol.% at HH and VV polarization respectively.</p>
Field-Scale Soil Moisture Retrieval Using PALSAR-2 Polarimetric Decomposition and Machine Learning