A possibilistic approach to soil moisture retrieval from ERS synthetic aperture radar backscattering under soil roughness uncertainty

Radar remote sensing of bare soil surfaces has been shown to be very useful for retrieving soil moisture. However, the error on the retrieved value depends on the accuracy of the roughness parameters (RMS height and correlation length). Several studies have demonstrated that these parameters show a high variability within a field, and therefore a lot of soil roughness profiles need to be measured to obtain accurate estimates. However, in an operational mode, soil roughness measurements are not available and therefore, for different types of tillage, roughness parameters are ill known. Possibility theory offers a way of handling this type of uncertainty, by modeling roughness parameters by means of possibility distributions. Inverting the integral equation model then leads to a possibility distribution for soil moisture. After transforming these possibilities into probabilities, mean soil moisture values and the uncertainty thereupon (given by the standard deviation) are obtained. It is found that the uncertainty depends on the wetness state of the soil. An application of our possibilistic retrieval algorithm to field observations at two sites in Belgium and one site in Italy resulted in accurate soil moisture observations (RMS error less than 6 vol %).