Evaluating the U.S. National Water Model Retrospective Evapotranspiration Simulation using Eddy-Covariance Flux Tower Measurements

Contiguous United States (CONUS). The U.S. National Water Model (NWM) provides a multi-decade hindcast of evapotranspiration (ET) across CONUS on a 1 km grid. We conducted the first national-scale performance assessment of NWM ET using monthly comparisons with eddy-covariance flux tower measurements at 72 sites with diverse hydroclimatic conditions. We assessed performance at each site individually, and, to reveal broader spatial and environmental patterns, we grouped site-level results by National Weather Service River Forecast Centers (RFCs), Köppen-Geiger Climate Zones, and land cover classifications. We also evaluated NWM temperature forcings and assessed ET performance under water- and energy-limited conditions Scaled Kling-Gupta Efficiency values ranged from −0.14–0.86 (median: 0.48), and PBIAS from −75 % to + 132 % (median: 11.4 %). Although within-group variability was high, stronger agreement was generally observed in the Ohio and Northeast RFCs, cold climates with warm summers, forested land covers, and colder months. In contrast, lower performance occurred more often in the California-Nevada RFC, temperate dry-summer climates, cultivated croplands, wetlands, and warmer months. Most sites performed better under water-limited regimes than under energy-limited conditions. Temperature forcings showed strong agreement with observations and were not a major source of ET error. These findings highlight opportunities to improve ET simulation in underperforming conditions and support the need for bias correction and expanded ET calibration to enhance model reliability. • We analyzed NWM ET vs AmeriFlux at 72 CONUS sites, finding general overestimation. • Grouping by RFC, climate zone, and land cover revealed high within-group variability. • Humid RFCs, forests, winter months, and water-limited sites showed better accuracy. • Arid RFCs, croplands, wetlands, summer months and energy-limited sites did worse. • Temperature forcing analysis showed it was not a major contributor to ET bias.