Wednesday, 25 January 2012
Verification of Ensemble River Forecasts At Middle Atlantic River Forecast Center
Hall E (New Orleans Convention Center )
Middle Atlantic River Forecast Center (MARFC) in State College, PA and three other National Weather Service (NWS) River Forecast Centers are producing experimental short-term hydrologic ensemble river forecasts. At MARFC, the operational Ensemble Streamflow Prediction (ESP) system uses ensemble precipitation and temperature forecasts from Short Range Ensemble Forecast (SREF), Global Ensemble Forecast System (GEFS), and North American Ensemble Forecast System (NAEFS) to produce three separate probabilistic river forecast products. MARFC has an archive of four-times daily, 87 hour, SREF ensembles dating back to January 2009, and twice-daily, 168 hour, GEFS and NAEFS ensembles dating back to November 2010. Verification of the hydrologic forecasts were analyzed through the NWS Ensemble Verification System (EVS), using mean error of the ensemble mean forecast, box plot comparisons, Relative Operating Characteristic plots, and Continuous Ranked Probability Scores, among others. These ensemble forecasts account for uncertainty in only the forcing inputs, not hydrologic uncertainties, which reduces the reliability of the forecasts through conditional bias and lack of spread. However, the probability of detection of higher water events far exceeds the probability of false detection, if a low enough probability threshold is used. Mean Error of the ensemble mean forecast shows a wide range of bias caused by differences in hydrologic uncertainty at various forecast points, including conditionally negative bias at some points and over-forecasting bias at others. We conclude by discussing the opportunities for improving the ESP forecasts through ensemble post processing.