SoilMERGE (SMERGE) is a 0.125-degree root zone soil moisture (RZSM) product (0 to 40 cm depth) covering the contiguous United States. The study area included most of Oklahoma and Kansas, a region where SMERGE exhibited superior performance. The time frame examined was the warm season from 2008 to 2019. In this study, evaluation of a prototype downscaled (500 m) version of SMERGE was made using (1) Ranked correlation (R2) benchmarking against Normalized Difference Vegetation Index (NDVI) datasets and (2) Ranked correlation (R2) analysis of antecedent RZSM with storm-event streamflow across a range of precipitation intensities (5 to >35 mm/day) at a watershed scale. In the NDVI benchmarking, all three downscaled products outperformed (0.52 to 0.59) default SMERGE (0.44). EXtreme Gradient Boosting (XGB) and Gradient Boost recorded a higher ranked correlation (0.59) than Random Forest (0.52). Within the study area, ranked correlation analysis of antecedent RZSM with storm-event United States Geological Survey streamflow was examined in five watersheds. For the most intense storm events (>35 mm), antecedent XGB downscaled SMERGE (0.64) outperformed antecedent streamflow (0.43) and all other versions of SMERGE (0.52 to 0.56) as a predictor of storm event response. The results of this study demonstrated broad-scale benefits of Machine Learning-assisted downscaling, providing proof of concept for the development of state-based SMERGE products across the US Great Plains.
Soil moisture regulates water and energy exchange between the land surface and the atmosphere, yet most current satellite products have a resolution on the order of tens of kilometers. SoilMERGE (SMERGE) is a 0.125° root zone soil moisture product covering the contiguous United States based on fusion of satellite and land model datasets. Building on earlier tree-based downscaling work, 500 m versions of SMERGE across Kansas, Oklahoma, North Texas, and South Texas were produced. Three comparisons were evaluated: per-state vs. pooled models, land cover as an input variable, and deep learning vs. tree-based models with five algorithms: Yggdrasil Decision Forests random forest (YDFRF) and XGBoost (XGB) as tree-based models. Deep learning models included Gated Linear Units (GLU), TabNet, and a tabular network derived from the Temporal Fusion Transformer (TFT). All downscaled products showed improved ranked correlation with AVHRR/VIIRS NDVI compared to the original SMERGE, with the best per-region model raising the ranked correlation coefficient by 0.04 to 0.09 while preserving fidelity to the default SMERGE (anomaly RMSE 0.0136 to 0.0231 m3/m3). In Kansas and Oklahoma, per-state and pooled models performed similarly (mean matched-pixel ΔR −0.016 to +0.007; equivalent within ±0.02 for every model under a spatial block bootstrap). Adding land cover as an input changed the ranked correlation by 0.0053 on average in absolute value. Evaluated through agreement with vegetation response, these results support further development of a state-based 500 m version of SMERGE in the Southern Great Plains.