Hydrometeorological Forecast Skill of the North American Multimodel Ensemble in the Upper Colorado River Basin

Abstract Water resources management in the Colorado River is supported by the 24-Month Study (24MS), which every month generates operational streamflow forecasts for the next 2 years in the upper basin. The 24MS relies on the ensemble streamflow prediction method that has been shown to overforecast dry conditions as it is based on a wetter historical reference. This study evaluates an alternative streamflow forecasting method where daily climate forecasts from general circulation models (GCMs) of the North American Multimodel Ensemble (NMME) are used to force the Variable Infiltration Capacity (VIC) hydrologic model (NMME + VIC). The skill of precipitation P and temperature T retrospective forecasts from five GCMs, totaling 52 ensemble members, was first assessed in 1982–2011. It was found that P forecasts had a low-to-moderate ability to reproduce spatiotemporal monthly climatologies and forecast 9-month periods, while T forecasts were more accurate. NMME + VIC was then used to issue 9-month streamflow forecasts in the upper basin starting in January and April. Despite errors in climate forecasts, NMME + VIC streamflow forecasts had comparable or better skill than the 24MS. This is due to the use of accurate initial basin wetness conditions, the interaction of different errors and spatial resolutions of the climate and hydrologic models, and the ability of some GCMs to better forecast P and T in high-elevation regions where most runoff is generated. It was also demonstrated that NMME + VIC forecasts could inform operational forecasting by statistically postprocessing the 52 ensemble members, which made it less likely to overforecast dry conditions than the 24MS.