Improved Ensemble Streamflow Prediction through New ESP Weighting Schemes and Bias Correction

Statistical Water Supply Forecast and Ensemble Streamflow Prediction (ESP) are the two major approaches for operational streamflow forecast in U.S. ESP generates an ensemble of future streamflow trajectories based on the current hydrologic states, and the previous meteorological records. It provides the means for statistical post-processing of the results and estimating the inherent uncertainties. It has been found that large scale climate variables provide valuable information for hydrologic predictions. In this study we assign weights to ESP ensemble members based on the large scale climate signals. We compare previous nonparametric approaches and those that limit the ensemble member generation to years of similar ENSO and PDO with the new weighting techniques presented in this study. These methods rely on fuzzy clustering and Bayesian Model Averaging techniques. The effect of bias correction using different methods are also investigated and compared for various short-middle and long term periods. The results indicate that ensemble member weighting outperforms the traditional no weight approach while the new methods show that further increased skill can be obtained from the streamflow ensemble.