Implications of Model Selection: Inter-Comparison of Publicly-Available CONUS Extent Hydrologic Component Estimates

Components of the hydrologic cycle are estimated through a variety of methods, including hydrologic modeling, reanalysis algorithms, and remote sensing-derived products. However, a comprehensive inter-comparison of these products (e.g. estimates of snow water equivalent or actual evapotranspiration) is lacking in the scientific literature in the context of the complete hydrologic system. A primary goal of this work is to enhance discussion of the impact of model selection in both operational and research settings. We compare a total of 82 distinct estimates of precipitation (P) (n = 15), actual evapotranspiration (AET) (n = 18), runoff (R) (n = 14), snow water equivalent (SWE) (n = 17), and rootzone soil moisture (RZSM) (n = 18) generated from 42 continental or global extent hydrologic models, reanalysis, or remote sensing products aggregated at monthly and annual time steps. Datasets were spatially aggregated by mean to the extent of the continental United States (CONUS), as well as ten large CONUS-wide ecoregions. Comparison results show that datasets can have wide ranges and be in complete disagreement in terms of inter-annual magnitudes, long-term trends, and/or seasonality. Aggregated mean P and AET magnitudes vary by 10-35%, and R and SWE by 50-200%, depending on the ecological regime. Long-term inter-annual trends in P, AET, R, and RZSM generally show higher agreement in the arid regions of the western CONUS. However, long-term SWE estimates show contradictory positive and negative trends in western ecoregions, including the Marine West Coast Forest and Northwestern Forested Mountains. Seasonality, in terms of annual peak month, varies widely in the arid regions of the CONUS, yielding standard deviations of 1-4 months. Spearman's rho correlation of monthly values against remote sensing products shows poorest agreement in the arid and semi-arid regimes of the western CONUS. The eastern CONUS, dominated by the Eastern Temperate Forests ecoregion that constitutes 30% of the CONUS, generally yields the lowest overall disagreement between datasets. Ensembles of estimates from multiple sources can be used to constrain ranges and likely provide improved confidence intervals for research results and model calibration and validation.