Abstract
This paper presents the results of a two-year experiment carried out in mountain areas on soil moisture retrieval from remotely sensed images. In particular, fully polarimetric synthetic aperture radar (SAR) RADARSAT and single polarization ASAR images were used. During acquisitions of SAR over the test area in Alto Adige, Südtirol region, northern Italy, extensive field measurements were carried out to characterize the area in terms of soil moisture variability, vegetation cover and soil properties. The area is equipped with 17 meteorological stations that also provide information on soil moisture on an hourly basis. To deal with the process of soil moisture retrieval from SAR images in this challenging environment, an advanced algorithm based on the support vector regression (SVR) technique and the integration of ancillary data were successfully proposed. The selected method was also combined with an innovative multi-objective strategy for model selection. The results indicate that the use of polarimetric features such as horizontal-horizontal polarization (HH) and horizontal-vertical polarization (HV) channels improves the estimation of SMC in the mountain area investigated, in particular because the HV channel contributes to disentangling the effect of vegetation on the radar signal. The root mean square error (RMSE) indicates accuracy of retrieval for soil moisture values of around 5%, which is well within the requirements for estimating soil moisture from remotely sensed data. This work was carried out with the main aim of demonstrating the feasibility of soil moisture retrieval in mountain areas with high-resolution images in view of the upcoming Sentinel 1 mission.