A Method for Evaluation of Ensemble Streamflow Predictions

Kristie J. Franz, Holly C. Hartmann, Soroosh Sorooshian, and Roger Bales

Department of Hydrology and Water Resources

The University of Arizona

The National Weather Service (NWS) Ensemble Streamflow Prediction (ESP) forecasting system uses present basin conditions and historical meteorological time-series as hydrologic model inputs to generate numerous streamflow scenarios (traces). A statistical analysis of the ensemble of traces is performed from which a forecaster can make probabilistic statements about the anticipated discharge. To conduct an evaluation of the ESP forecast technique and illustrate an ESP verification approach, simulated historical streamflow predictions for fourteen watersheds in the Colorado River Basin were created and analyzed. The forecast analyses include scalar, probabilistic, and conditional verification methods; the importance of a comprehensive diagnostic approach is highlighted. The simulated ESP forecasts illustrate an average 10-30% improvement over climatology forecasts for prediction of total seasonal discharge, with the highest improvement occurring later in the forecast season. In addition, forecasts with 2-3 month lead times show good discrimination for high and low flows, while forecasts with 3-month lead times show near perfect reliability. The verification technique applied in this study is currently being investigated for incorporation into NWS forecast evaluation procedures. The goal is to assist the development of a consistent evaluation method that allows the generation of quantitative information to support forecast credibility and judge future forecasts.