A Multiscale Postprocessor for Ensemble Streamflow Prediction

Reducing hydrologic uncertainties in streamflow prediction is of great importance for operational water forecasting and management. Hydrologic uncertainties arise from initial condition uncertainty, model parametric and structural uncertainties, and human control and alternation of movement of water. Post-processing is an approach to reduce these uncertainties by statistical means. The Ensemble Post-Processor, or EnsPost, is such a model developed by the National Weather Service (NWS) to account for hydrologic uncertainties in short-range ensemble streamflow prediction. Because EnsPost models the joint probability distribution of observed and simulated flow, it requires a long period of record for distribution modeling of the all-important upper tail. As such, its performance is susceptible to sampling uncertainties and nonstationarities due to urbanization and climate change. In this presentation, we describe a prototype algorithm for multi-scale ensemble post processing, or MS-EnsPost, which avoids variable transformation and minimizes distribution modeling to reduce data requirement, and utilizes all available skill present in simulated and observed flows via very parsimonious statistical modeling at multiple temporal scales. To validate MS-EnsPost, we use observed and model-simulated streamflow time series for a large number of basins in several NWS River Forecast Centers’ service areas. The streamflow predictions of MS-EnsPost are compared with those of EnsPost. In this presentation, we describe the technique, share the preliminary ensemble verification results, and identify issues and challenges.