Implications of model selection: A comparison of publicly available, conterminous US-extent hydrologic component estimates

Abstract

Spatiotemporally continuous estimates of the hydrologic cycle are often generated through hydrologic modeling, reanalysis, or remote sensing (RS) methods and are commonly applied as a supplement to, or a substitute for, in situ measurements when observational data are sparse or unavailable. This study compares estimates of precipitation (P), actual evapotranspiration (ET), runoff (R), snow water equivalent (SWE), and soil moisture (SM) from 87 unique data sets generated by 47 hydrologic models, reanalysis data sets, and remote sensing products across the conterminous United States (CONUS). Uncertainty between hydrologic component estimates was shown to be high in the western CONUS, with median uncertainty (measured as the coefficient of variation) ranging from 11% to 21% for P, 14% to 26% for ET, 28% to 82% for R, 76% to 84% for SWE, and 36%to 96%for SM. Uncertainty between estimates was lower in the eastern CONUS, with medians ranging from 5% to 14% for P, 13% to 22% for ET, 28% to 82% for R, 53% to 63% for SWE, and 42% to 83% for SM. Interannual trends in estimates from 1982 to 2010 show common disagreement in R, SWE, and SM. Correlating fluxes and stores against remote-sensing-derived products show poor overall correlation in the western CONUS for ET and SM estimates. Study results show that disagreement between estimates can be substantial, sometimes exceeding the magnitude of the measurements themselves. The authors conclude that multimodel ensembles are not only useful but are in fact a necessity for accurately representing uncertainty in research results. Spatial biases of model disagreement values in the western United States show that targeted research efforts in arid and semiarid water-limited regions are warranted, with the greatest emphasis on storage and runoff components, to better describe complexities of the terrestrial hydrologic system and reconcile model disagreement.