Midterm Projections of Colorado River Streamflow and Water Resources Operations Conditioned on Temperature Projections

Abstract:

Following a severe drought at the onset of the 21st century in the Colorado River Basin (CRB), a variety of tools were developed to improve preparation for future shortages, including modeling efforts, flow forecasting techniques, drought planning, etc. Most of the research in the CRB is focused on seasonal forecasts and century scale flow projections; examples include Ensemble Streamflow Prediction (ESP) and climate model derived flow projections for mid- and end- of century. However, midterm (1 ~ 5 years) flow projections are becoming increasingly important for shortage planning in the CRB reservoir system but have not been the focus like the above techniques, which are generally no better than climatology past the first projection year. We offer a coupled model approach for projection of streamflow and derivative water resource management variables. With the availability of simulated CRB precipitation and temperature from the Community Earth System Model – Large Ensemble (CESM-LE) we examined the relationship between CRB naturalized flow with both observed- and simulated- precipitation and temperature and found strong correlation with temperature at midterm scales, echoing recent studies. We used simulated basin temperature and precipitation as covariates in a K-Nearest Neighbor Block Bootstrap (KNN-BB) method to project ensembles of future 5-year annual flow. The annual flow ensembles are disaggregated to a monthly scale at all gauge locations in the basin. These flows drive a Midterm Operations Model (MTOM) to generate ensembles of water resources decision variables. During a hindcast period of 1928-2013, the climate-conditioned KNN-BB projections of 5-year mean flow were more skillful than climatology slightly less than 50% of the time, based on the rank probability skill score (RPSS) of projected terciles, with a maximum and minimum RPSS of 0.99 and -2.7, respectively. MTOM simulations for 1982-2017 show the KNN-BB method generally outperformed ESP in projecting Lake Powell elevation during forecast years 3-5, reducing median root mean square error (RMSE) by between 11.6% and 13%; although ESP performed best in years 1 and 2. Our results indicate encouraging prospects for use of climate model simulated precipitation and temperature in midterm projections of CRB flow and associated decision-making.