The Sentinel-1 Synthetic Aperture Radar (SAR) constellation can map soil moisture at high spatial resolution, but in arid and semi-arid regions, radar pulses penetrate dry soil and scatter off subsurface rocks and stones. This subsurface scattering undermines the core assumption of a positive relationship between backscatter and soil moisture, causing widespread false wetting signals in current retrieval products. We address this by exploiting the fine-scale spatial heterogeneity of subsurface scattering: pixels with no sensitivity to soil moisture are masked at 20m resolution before aggregating backscatter to 1km. This sidesteps the need to model subsurface scattering by exploiting the fine spatial resolution of the SAR data. Additionally, we fit a zeroth-order radiative transfer model that ingests satellite-derived Leaf Area Index to correct for vegetation bias in the backscatter signal. Over a study region covering the Iberian Peninsula and northwestern Africa, masking improved the correlation with passive microwave reference soil moisture by 0.1 on average, with the strongest gains over sclerophyllous vegetation associated with arid and semi-arid regions (0.19). Vegetation modelling yielded a further improvement of 0.14. Validation against in-situ stations confirmed that the combined approach outperforms the current Copernicus 1km soil moisture operational product (mean correlation 0.6 versus 0.5). Both methods are operationally feasible and would be directly applicable to the Copernicus pipeline. They are also relevant for upcoming L-band missions such as NISAR and ROSE-L, where deeper soil penetration may amplify subsurface scattering effects.
Raml et al. (Thu,) studied this question.