Producing and Assessing Short-Term Temperature Ensembles for Ensemble Streamflow Prediction
Abstract
An ensemble pre-processor has been developed to generate the short-term precipitation and temperature ensemble forecasts needed for the National Weather Service (NWS) Ensemble Streamflow Prediction (ESP) system that produces probabilistic streamflow forecasts. The meteorological ensemble forecasts are constructed to incorporate the skill of the current single-value forecasts and to account for the forecast uncertainty. The statistical pre-processing estimates the conditional distribution for the future events given the current forecast, from historical pairs of forecasts and observations. A distribution mapping process is then used to re-scale the original ensembles according to this conditional distribution. Finally the resulting synthetic ensembles are ingested by ESP in place of the historical data to produce streamflow ensembles that reflect the meteorological uncertainty. In support of the Advanced Hydrologic Prediction Service, the method has been developed in an operational forecasting environment integrating the existing forecasts used at the River Forecast Centers (RFCs) and is being tested in pilot projects at four RFCs. Verification results are presented for temperature ensembles for the Juniata River basins (Pennsylvania) at Middle-Atlantic RFC and the American River basins (California) at California-Nevada RFC where temperature drives the winter snow hydrology operations. Temperature forecasts are the single-value daily maximum and minimum Model Output Statistics (MOS) temperature forecasts from the Global Forecast System for lead times of one to five days with three to five years of data. The corresponding daily maximum and minimum temperature observed values are generated from the mean areal temperature time series using a fixed diurnal cycle. Daily maximum and minimum temperature ensembles are then generated for lead days one to five and merged to produce a 6-hour mean temperature ensembles based on a user-defined diurnal cycle. Retrospective forecast verification procedures have been developed to compute the Nash-Sutcliffe efficiency, the Heidke and Brier skill scores among other statistics.
- Publication:
-
AGU Spring Meeting Abstracts
- Pub Date:
- May 2004
- Bibcode:
- 2004AGUSM.H21E..04D
- Keywords:
-
- 1833 Hydroclimatology;
- 1854 Precipitation (3354);
- 1860 Runoff and streamflow;
- 1869 Stochastic processes;
- 1884 Water supply