Development of Hierarchical Ensemble Model and Estimates of Soil Water Retention With Global Coverage

Correct quantification of mass and energy exchange processes between land surface and atmosphere requires an accurate description of unsaturated soil hydraulic properties. Soil pedotransfer functions (PTFs) have been widely used to predict soil hydraulic parameters. Here, 13 PTFs were grouped according to input data requirements and evaluated against a well‐documented database (National Cooperative Soil Survey Characterization [NCSS]) covering the continental United States (87.7% of data) and other regions of the globe (12.3% of data). Weighted ensembles were shown to have improved performance over individual PTFs in terms of evaluation criteria. Validation of moisture content estimated from the ensemble models against observations showed promising results. Global maps of soil water retention data from the ensemble models as well as their uncertainty were provided. Our full 13‐member ensemble model provides more accurate estimates than PTFs that are currently being used in Earth system models, which may, therefore, provide improved water fluxes and reduce uncertainty of the estimations. The availability of soil water retention data is essential for quantifying mass and energy exchange processes at the interface between land surface and atmosphere. In large‐scale applications, soil water retention characteristics usually are estimated with empirical models that, unfortunately, use nonuniform predictors and were developed on subsets of the global distribution of soils. Their reliability for global estimates is often unknown. Using a global database, we developed an ensemble of up to 13 previously published models allowing estimates of soil water retention data under data‐poor to data‐rich conditions. High‐resolution global maps of key points in soil water retention characteristics (and their uncertainties) were produced. These maps suggest that middle and high latitudes in the Northern Hemisphere have larger variability of the estimates. The new model provides more accurate estimates than models currently being used in Earth system models. The performance of 13 popular models for estimating soil water retention was quantified using a data set with global coverageRelative to individual models, weighted multimodel ensembles had improved performance with the best obtained with the full ensembleHigh‐resolution global maps of soil water retention properties suitable for Earth system modeling were produced