IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (Jan 2022)
Assessment of NASA SMAP Soil Moisture Products for Agricultural Regions in Central Mexico: An Analysis Based on the THEXMEX Dataset
Abstract
Accurate knowledge of soil moisture (SM) is crucial in hydrological, micrometeorological, and agricultural applications; however, the SM estimation is particularly challenging in agricultural regions due to high spatial variability and dynamic vegetation conditions. The need for information about SM conditions is even more evident in developing countries with limited monitoring infrastructure. Satellite SM products are a useful tool as a proxy for SM conditions on the ground, but they need to be evaluated for specific regions. In this study, we assess the quality of the soil moisture active passive (SMAP) SM retrievals at 36, 9, and 3 km in an agricultural region in Central Mexico using in situ measurements during the Terrestrial Hydrology Experiments in Mexico 2018 and 2019. In addition, we provide insights into soil and vegetation parameters in the retrieval algorithms compared to those observed in the region. It was found that the SM spatial variability at the SMAP pixel grids was well represented by upscaled in situ SM measurements (SM$_{\text{up}}$) from five monitoring stations using the soil-weighted averaging and the Voronoï diagrams. Overall, the SMAP SM retrievals are highly correlated with SM$_{\text{up}}$ at all scales, but they estimated wetter conditions and the average root-mean-square difference (RMSD) $>$ 0.045 m$^{3}$/m$^{3}$. The lowest RMSD was obtained for the SM product at 36 km, while the highest RMSD was found for the SM product at 3 km. In addition, the single-channel algorithm using H-polarization provided the lowest RMSD for the products at 36 and 9 km. The main sources of uncertainty in the region may arise from the higher clay fraction used in the SMAP retrieval algorithm, by 13% compared to that observed, and a nonrepresentative characterization of land cover heterogeneity for vegetation water content estimation. The incorporation of in situ values into an SM retrieval algorithm resulted in differences $< $0.04 m$^{3}$/m$^{3}$ between SM estimates and in situ SM for the complete growing season. Particularly, the use of in situ information helped in improving SM estimation when optimizing V- and dual-polarization brightness temperature observations.
Keywords
