Postprocessing of Precipitation Forecasts For New Configured NCEP Short-range Ensemble Forecasting (SREF) System

Abstract

Since December 2005, the NCEP (National Centers for Environmental Prediction) short-range ensemble forecasting (SREF) system has been updated by constructing new ensemble components (Du et al. 2006). This system contains perturbed initial/boundary conditions, multiple physics, and multimodel. It covers the North American continent and the adjacent maritime zones. It is important to evaluate the performance of precipitation forecasts of the SREF, especially probabilistic quantitative precipitation forecasts (PQPF). Results of precipitation forecasting from this suite of new configured operational SREF system were verified compared to NCEP Stage IV precipitation analyses over the continental United States (CONUS). The reliability curves of PQPF presented light wet biases. It is of interest to conduct calibration for the SREF PQPF and examine how much the postprocessing can benefit an ensemble system with diversified models and physics. One tool used to conduct probabilistic postprocessing is a feedforward artificial neural network (Hsu et al. 1995), which was successfully applied to correct PQPF biases in a high-resolution (12 km) Regional Spectral Model (RSM, Juang and Kanamitsu 1994) ensemble system (Yuan et al. 2005, 2007a). Another tool is a linear regression model, which was applied to calibrate PQPF for heavy precipitation events over the American River Basin during the Hydrometeorolgical Testbed (HMT) program at NOAA/ESRL/GSD (Yuan et al. 2007b). Both tools were implemented to calibrate PQPF from the SREF system over the CONUS. According to hydrological and geographic characteristics, bias correction of PQPF was performed for three major regions – western, central, and eastern. The two seasons are defined as the warm season (April-September) and cool season (October-March), respectively. In this study, preliminary results are present for the two periods: (1) April-September 2006, (2) October-December 2006. Cross-validation was used to compute verification scores and attributes diagrams. The following results are composited from all months during a season in a region.