Sentinel-1 soil moisture content and its uncertainty over sparsely vegetated fields

• Main Authors: Harm-Jan F. Benninga, Rogier van der Velde, Zhongbo Su
• Format: Article
• Language: English
• Published: Elsevier 2020-12-01
• Series: Journal of Hydrology X
• Subjects: Soil moisture content; Remote sensing; Sentinel-1 satellites; Retrieval uncertainty; Uncertainty sources; Soil surface roughness
• Online Access: http://www.sciencedirect.com/science/article/pii/S2589915520300171

Soil moisture content (SMC) retrievals from synthetic aperture radar (SAR) observations do not exactly match with in situ references due to imperfect retrieval algorithms, and uncertainties in the model parameters, SAR observations and in situ references. Information on the uncertainty of SMC retrievals would contribute to their applicability. This paper presents a methodology for deriving the SMC retrieval uncertainty and decomposing this in its constituents. A Bayesian calibration framework was used for deriving the total uncertainty and the model parameter uncertainty. The methodology was demonstrated with the integral equation method (IEM) surface scattering model, which was employed for reproducing Sentinel-1 backscatter (σ0) observations and the retrieval of SMC over four sparsely vegetated fields in the Netherlands. For two meadows the calibrated surface roughness parameter distributions are remarkably similar between the ascending and the descending Sentinel-1 orbits as well as between the two meadows, and yield consistent SMC retrievals for the calibration and validation periods (RMSDs of 0.076 m3 m−3 to 0.11 m3 m−3). These results are promising for operational retrieval of SMC over meadows. In contrast, the surface roughness parameter distributions of two fallow maize fields differ significantly and the surface roughness conditions changing over time result in less consistent SMC retrievals (calibration RMSDs of 0.096 m3 m−3 and 0.13 m3 m−3 versus validation RMSDs of 0.26 m3 m−3). The SMC retrieval uncertainty derived with the Bayesian calibration successfully reproduces the uncertainty estimated empirically using in situ references. The main uncertainty originates from the in situ references and the Sentinel-1 observations, whereas the contribution from the surface roughness parameters is relatively small. The presented research yields further insights into the surface roughness of agricultural fields and SMC retrieval uncertainties, and these insights can be used to guide SAR-based SMC product developments.