A statistical post-processor for accounting of hydrologic uncertainty in short-range ensemble streamflow prediction

Abstract

Abstract. In addition to the uncertainty in future boundary conditions of precipitation and temperature (i.e. the meteorological uncertainty), parametric and structural uncertainties in the hydrologic models and uncertainty in the model initial conditions (i.e. the hydrologic uncertainties) constitute a major source of error in hydrologic prediction. As such, accurate accounting of both meteorological and hydrologic uncertainties is critical to producing reliable probabilistic hydrologic prediction. In this paper, we describe and evaluate a statistical procedure that accounts for hydrologic uncertainty in short-range (1 to 5 days ahead) ensemble streamflow prediction (ESP). Referred to as the ESP post-processor, the procedure operates on ensemble traces of model-predicted streamflow that reflect only the meteorological uncertainty and produces post-processed ensemble traces that reflect both the meteorological and hydrologic uncertainties. A combination of probability matching and regression, the procedure is simple, parsimonious and robust. For a critical evaluation of the procedure, independent validation is carried out for five basins of the Juniata River in Pennsylvania, USA, under a very stringent setting. The results indicate that the post-processor is fully capable of producing ensemble traces that are unbiased in the mean and in the probabilistic sense. Due primarily to the uncertainties in the cumulative probability distributions (CDF) of observed and simulated flows, however, the unbiasedness may be compromised to a varying degree in real world situations. It is also shown, however, that the uncertainties in the CDF's do not significantly diminish the value of post-processed ensemble traces for decision making, and that probabilistic prediction based on post-processed ensemble traces significantly improves the value of single-value prediction at all ranges of flow.