Ensemble Streamflow Prediction: Climate signal weighting methods vs. Climate Forecast System Reanalysis

SummaryEnsemble Streamflow Prediction (ESP) provides the means for statistical post-processing of forecasts and estimating the inherent uncertainties. In addition, large scale climate variables provide valuable information for hydrologic predictions. In this study we develop methods to assign weights to ESP ensemble members according to climate signals which are selected based on the spearman’s rank correlation coefficients. Analysis was performed over the snow dominated East River basin to improve the spring streamflow volumetric forecast. Principle Component Analysis (PCA) was found to increase the accuracy of the weighting scheme considerably. We compare five parametric and nonparametric weighting methods including Fuzzy C-Means clustering, Formal Likelihood, Informal Likelihood and two variants of K-Nearest Neighbors approaches. The methods are found to be simple and efficient while the results seem promising. The predictions, based on simple average or the median of the ensemble members, combined with the weighted ensemble forecasts provide improved estimates of probable streamflow ranges and the uncertainty bounds. Improvement in the weighting approach was obtained by selecting the climate signals, choosing the right number of principle components and considering several weighting approaches. As an alternative approach to ESP, an additional climate dataset, the Climate Forecast System Reanalysis (CFSR) provided via the National Centers for Environmental Prediction (NCEP) in its most recent reanalysis project was tested.

► Ensemble members in ESP can be weighted according to climate signals.
► Implementing the PCA on climate signals improves the weighting scheme.
► Delta bias correction of CFSR performed best among other bias correction methods.
► CFSR did not improve the seasonal forecasting as compared with the weighted ESP.