Multiseason Verification of the MM5. Part II: Evaluation of High-Resolution Precipitation Forecasts over the Northeastern United States

Abstract

This paper evaluates the fifth-generation Pennsylvania State University?National Center for Atmospheric Research Mesoscale Model (MM5) precipitation forecasts over the northeastern United States to show the effects of increasing resolution, the spatial variations in model skill, and the impact of convective parameterizations on the MM5 precipitation forecasts. The MM5 is verified during the cool seasons (November?March) of 1999?2001 and the warm season (May?September) of 2000 using approximately 500 cooperative observer and National Weather Service precipitation sites. During the cool season, the 12-km MM5 produces excessive precipitation immediately downwind of the Great Lakes and along the windward slopes of the Appalachians and too little precipitation in the lee of the barrier. The 36-km MM5 has slightly more skill than at 12-km grid spacing for the light to moderate thresholds, while the 12-km precipitation forecasts are slightly better on average for the heavy precipitation events. During the 2000/01 cool season, two separate MM5 runs were completed twice daily using the National Centers for Environmental Prediction Eta (Eta?MM5) and Aviation (AVN?MM5) Models for initial and boundary conditions. The Eta?MM5 had slightly lower (better) rms errors than the AVN?MM5 for the weak to moderate thresholds (2.54?50.8 mm in 24 h), while for the heavier thresholds the AVN?MM5 had significantly lower rms errors than the Eta?MM5. As a result, the 36-km AVN?MM5 was as skillful as the 12-km Eta?MM5 for these higher thresholds. During the warm season, both the 36- and 12-km grid spacings overpredict precipitation just inland of the coast and significantly underpredict farther inland over the Appalachians. This coastal overprediction originated from an overactive Kain?Fritsch (KF) convective parameterization, while the inland underprediction is associated with a low-level dry bias during the warm season. A representative case study shows that both the Betts?Miller and Grell parameterizations produce less precipitation near the coast than the overactive KF scheme. A new and alternate version of KF (KF2) in the MM5 may also help to reduce this coastal overprediction. The 4-km MM5 explicit precipitation during the summer is sensitive to which convective parameterization is applied in the outer domains. Using KF or KF2 in the 36- and 12-km domains suppresses the explicit precipitation in the 4-km nest, especially for weak to moderate events over western sections of the 4-km domain. For a representative event, the Betts?Miller and Grell convective schemes allowed for a more realistic 4-km precipitation distribution, while a simulation using no convective parameterization in the 36- and 12-km domains produced excessive rain rates in the 4-km forecasts.