Quantifying Uncertainties in Passive Microwave Remote Sensing of Soil Moisture via a Bayesian Probabilistic Inversion Method

Abstract

Improving the accuracy of remotely sensed soil moisture (SM) is a challenging and popular topic. Quantifying the uncertainty in the SM inversion process and enhancing the confidence of SM retrieval are promising ways to address this challenge but have received little attention. We present a Bayesian probabilistic inversion algorithm that can simultaneously retrieve SM, surface roughness, and vegetation optical depth data and quantify the uncertainty in the inversion. The proposed algorithm is evaluated using airborne polarimetric L-band multibeam radiometer (PLMR) observations. We use three combinations, 3-angular observations at V-polarization (3CV), 3-angular observations at H-polarization (3CH) and 6-channel observations (6CA), to identify the optimal configuration for SM retrieval by taking advantage of the PLMR’s dual-polarization and multiple angles. Uncertainties are quantified by introducing multiple uncertainty quantification metrics into Bayesian posterior distributions of SM retrievals. The estimates are validated against multiscale ground-based measurements, including manual measurements and wireless sensor network (WSN) measurements, and the spatial representativeness of the ground-based reference regarding the validation of pixel-scale SM retrievals is discussed. The 6CA attempt yields the best SM estimates (correlation coefficient (R) ≥ 0.864, root mean square error (RMSE) ≤0.04 m3/m3, and unbiased RMSE (unRMSE) ≤ 0.035 m3/m3), while the 3CH attempt yields the lowest uncertainty. Additionally, dense manual measurements are more representative than sparsely distributed WSN measurements. Overall, combining dual-polarized observations yields the best SM estimates but introduces additional uncertainty. This study highlights uncertainties quantification in SM inversion and thus provides confidence in SM inversion, facilitating improved SM retrieval algorithms.