Initial down scaling of SMERGE, preliminary results from the MOISST site in Oklahoma

Land surface and satellite-based monitoring of root zone soil moisture (RZSM) can provide critical information supporting a wide range of hydrometeorological applications. Soil MERGE or SMERGE 2.0 is a long-term, daily retrospective RZSM product. It has a 0.125-degree spatial resolution spanning 1979 to 2019 and represents moisture values between the surface and 40 cm depth. This low spatial resolution is not useful for many potential end users because soil moisture has a high variability across both temporal and spatial scales. Advanced machine learning (ML) techniques have been used to downscale soil moisture products like SMAP, SMOS, and ESA-CCI. This study is the first attempt to downscale SMERGE. ML techniques used included eXtreme Gradient Boosting (XGBoost) and Random Forest (RF), as well as a simple linear regression model. Another unique aspect of this work is the use of data from the NASA Airborne Microwave Observatory of Subcanopy and Surface (AirMOSS) initiative for direct validation of SMERGE downscaled results. AirMOSS provided discontinuous P-band (420-440 MHz) radar-based estimates of RZSM at nine North American sites at an approximate 100 m spatial resolution during the warm seasons of 2012 to 2015. This study uses AirMOSS data from the Marena Oklahoma Soil Moisture Active Passive In Situ Testbed (MOISST) site focusing between 36.000 to 36.250 N and 97.125 to 97.500 W. Downscaled SMERGE is examined at 400, 700, 1000, and 1400 m resolutions. The optimal tradeoff between performance and spatial resolution was noted at the 1000 m resolution. The linear regression version of SMERGE exhibited significant improvement in the downscaled version compared with the default resolution. Correlation with AirMOSS increased from 0.49 to 0.62 and unbiased root-mean-square error decreased from 0.050 to 0.043 cubic meters / cubic meters for this downscaled version. This value is close to the standard set for many space-based soil moisture measurement missions. XGBoost and RF did not confer increased performance for downscaled SMERGE. However, these ML approaches were not tuned to optimize performance. In conclusion, the preliminary results support the feasibility of SMERGE downscaling, opening the prospect for the development of a long-term RZSM dataset at a more desirable spatial resolution.