642 Verification of GEFS Precipitation across the Eastern United States

Wednesday, 13 January 2016
Nicholas Hayden Balderas, NSF, Katy, TX; and A. Mejia and R. Siddique

Flood forecasting greatly relies on accurate precipitation forecasts. Precipitation forecast models have improved due to an increased understanding of atmospheric physics, improved forecasting techniques, and improvements in numerical model techniques. In order to continue improving forecasting models, data verification is required to determine the quality of forecasts from models. In this study, precipitation forecasts from the Global Ensemble Forecast System Reforecast version 2 (GEFSRv2) are verified against multi-sensor precipitation estimates (MPEs). The GEFSRv2 precipitation forecasts are verified for various basin sizes of 1x1 and 5x5 degrees for the Southeast River Forecast Center (SERFC), Northeast River Forecast Center (NERFC), and Middle Atlantic River Forecast Center (MARFC). The quality of the GEFSRv2 forecasts are verified using different metrics, including the correlation coefficient, Brier skill score, relative mean error, reliability diagram, and mean continuous ranked probability score. Overall, the verification results indicate that GEFSRv2 tends to under forecast moderate to heavy precipitation across the Eastern United States (US). Nonetheless, the precipitation forecasts from the GEFSRv2 show some skill up to approximately 8 days relative to climatology. The NERFC shows slightly more skill than the other RFCs for the shorter lead times (~less than 2.5 days). For all the RFCs, there is a seasonal trend of less skillful forecasts in the summer months. The precipitation forecasts are, in general, reliable only for short period lead times. The variations in forecast quality across the Eastern US are expected to affect flood forecasts. The significance of these variations needs to be assessed alongside the uncertainty due to hydrologic sources (e.g., model structure and parameters).
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner