True global error maps for SMAP, SMOS, and ASCAT soil moisture data based on machine learning and triple collocation analysis

Authors: KIM, H. - US Department Of Agriculture (USDA); Crow, Wade; LI, X. - University Of Bordeaux; WAGNER, W. - Vienna University Of Technology; HAHN, S. - Vienna University Of Technology; LAKSHMI, V. - University Of Virginia

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/17/2023
Publication Date: 9/5/2023
Citation: Kim, H., Crow, W.T., Li, X., Wagner, W., Hahn, S., Lakshmi, V. 2023. True global error maps for SMAP, SMOS, and ASCAT soil moisture data based on machine learning and triple collocation analysis. Remote Sensing of Environment. 298: Article e113776. https://doi.org/10.1016/j.rse.2023.113776.
DOI: https://doi.org/10.1016/j.rse.2023.113776

Interpretive Summary: Remotely sensed soil moisture products contribute significantly to the global monitoring of agricultural drought. However, users of these products still face considerable uncertainty when attempting to assess their quality. This work employs machine learning approaches to quantify how uncertainty in satellite-based soil moisture products varies continuously across space and, hence, how truly useful these products are for agricultural and hydrological applications. Specifically, this is the first study to provide spatially continuous accuracy information for satellite-based soil moisture retrievals on a truly global scale. In the future, stakeholders will use these results to improve their use of remotely sensed soil moisture data to monitor the extent and severity of agricultural drought.

Technical Abstract: Quantifying the accuracy of the satellite-based soil moisture (SM) data is important for a number of key applications, such as: combining satellite-based SM products for long-term SM analyses, assimilating SM data into land surface models, and providing quality flags to mask bad quality SM data. Many statistical methods have been proposed to calculate errors in large-scale SM data sources including the: instrumental variable (IV) method, triple collocation analysis (TCA), and quadruple collocation analysis (QCA). Nonetheless, TCA-based methods still cannot provide truly global error maps for satellite SM products due to the limited number of independent SM products and difficulties with baseline TCA assumptions. Moreover, temporal sampling requirements for TCA are often impractical because of low SM retrieval skill in forested and arid areas – as well as regions prone to radio frequency interference. Here, we seek to fill significant spatial gaps in TCA results using machine learning (ML) and therefore provide spatially complete error maps for the satellite-based SM data products derived from the Soil Moisture Active Passive (SMAP), Soil Moisture and Ocean Salinity (SMOS), and Advanced Scatterometer (ASCAT) systems. Globally, and across all three products, 72.0% of missing error information in a TCA-based analysis, due to either the lack of valid data or the inability of TCA to provide reliable results, can be reconstructed from the ensemble prediction mean of the ML models. For the SMAP SM product, we find that in 17.5% of global land areas between 60i S to 60i N) , current data quality-control practices mask viable SM retrievals (combination of a.m. and p.m. data) and are therefore associated with no significant improvement in data quality. In addition, over 13.3% (combination of a.m. and p.m. data) of the Earth’s surface, SM dynamics cannot be reliably estimated from SMAP.