Research Publications

The National Water Model (NWM) has undergone several updates since its inception. The performance of its latest release, version 3.0, has yet to be comprehensively evaluated. This study presents the first comprehensive CONUS-wide evaluation of NWMv3.0 retrospective streamflow simulations (1979–2020) and compares them against the preceding NWMv2.1. Using observed streamflow from over 4300 United States Geological Survey (USGS) gages, we jointly evaluated both daily flow and annual peak-flow performance across spatial, seasonal, and categorical dimensions. Results showed that NWMv3.0 generally outperformed NWMv2.1 in both daily flow and peak flow statistics. For daily flow, the mean Kling-Gupta Efficiency (KGE) improved from −0.15 to 0.07, while the mean Percent Bias (PBIAS) dropped from 70.98% to 42.19%. For peak flows, mean Average Annual Peak Flow Bias (AAPFB) decreased from 690% to 234%, and mean Average Annual Peak Timing Error (AAPTE) reduced from 0.62 to 0.27 h from v2.1 to v3. Geographically, the Eastern U.S. generally showed better performance than the Central and Western regions, although the latter showed the greatest relative improvements despite overall lower performance. Seasonally, both versions performed best in spring and worst in winter in terms of mean KGE, with the most substantial statistical improvements observed in winter. Additionally, we also found that the natural gages consistently outperformed human-impacted gages for both versions. Despite overall statistical gains in NWMv3.0, spatial variability in performance still highlights the need for region-specific refinements in future NWM development to better capture diverse hydrologic conditions.