Content area
Full text
1. Introduction
Forecast skill of quantitative precipitations forecasts (QPFs) in the National Weather Service (NWS) is low, especially during the warm season (Fritsch and Carbone 2004). While gradual improvements in QPF skill have been made over the years, the rate of improvement has lagged corresponding improvements in large-scale circulation forecasts by operational numerical weather prediction (NWP) models. The skill disparity is widely believed to be due to the relatively coarse spatial resolution in operational NWP models: intense precipitation is associated with mesoscale systems, whereas these NWP models primarily resolve synoptic-scale circulations. This implies that a key to improving QPF resides in downscaling predictive information from current NWP models or using finer-resolution models.
The performance of precipitation forecasts from operational NWP models reflects the large-scale nature of the models. Light precipitation is overpredicted, while intense precipitation is underpredicted (Jensenius 1990), which the authors have found in archived National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS; Kanamitsu 1989; Kanamitsu et al. 1991; Iredell and Caplan 1997) precipitation forecasts from 2001 to the present. We have also seen a similar bias trend in the NCEP Nonhydrostatic Mesoscale Model (Rodgers et al. 2005), though to a somewhat lesser degree. (Documentation of these findings is planned for a forthcoming article.) Such model bias suggests inadequacy of the parameterization of convective precipitation in these coarse-mesh operational models, as recent experimentation with fine-grid nonhydrostatic (“convection allowing”) numerical models to improve precipitation forecasts have yielded encouraging results (e.g., Clark et al. 2009).
Model output statistics (MOS) postprocessing of NWP output has been used by the NWS to produce unbiased QPFs and provide uncertainty (probability) estimates for over three decades (Antolik 2000; Ruth et al. 2008). However, conventional MOS QPF applications are hampered by short samples from stable NWP models (Carter et al. 1989; Antolik and Baker 2009), the necessity of using broad precipitation categories as predictands, poor sharpness of forecast probabilities (Wilks 2006) for rare heavy-precipitation categories, and the inability to predict very heavy precipitation events. Another limitation of the operational MOS QPFs is that the MOS station distribution is quite irregular over the forecast domain (Gilbert et al. 2009).
The recent implementation of the gridded National Digital Forecast Database (NDFD; Glahn and Ruth 2003) by the NWS has created a...