This paper describes the long-term verification of the Penn State University-NCAR Mesoscale model (MM5) and NCEP Eta model over the eastern two-thirds of the United States and surrounding coastal waters during the cool and warm seasons of 1999-2001. During this period the MM5 was run in real-time down to 4 km resolution around Long Island region (see http://atmos.msrc.sunysb.edu/html/alt_mm5.cgi for details). Verification statistics are calculated by interpolating model forecasts to the observation sites. The horizontal and vertical distribution of model errors will be presented as well as the diurnal trends. Many of the MM5 and Eta errors have slowly-evolving intraseasonal trends. In particular, the cool bias amplifies during the winter and the summer dry bias in the MM5 increases during prolonged wet periods. The sea-level pressure errors are episodic, with clusters of negative and positive mean errors lasting approximately 3-6 weeks, thereby suggesting a dependence on the large-scale flow.

Because of the complex coastal geometry around southern New England there is some improvement in the temperature and wind forecasts with decreased grid spacing from 36- to 12-km; however, this improvement is limited during the cool season by the amplification of the temperature and wind biases over the major metropolitan areas at high resolution as well as MM5 over-prediction of precipitation over the Appalachians. For the warm season sea breeze events, the 12-km MM5 has significantly more wind and temperature skill along the coast than the 36 km, but there is little improvement from 12 to 4 km. The sea breezes in the MM5 are too early on average and are associated with a late afternoon cool bias. The 4-km MM5 explicit precipitation during the summer is sensitive to the convective parameterization applied in the outer domains. Using Kain-Fritsch in the 36-/12-km suppresses the explicit precipitation in the 4-km while the Betts-Miller and Grell convective schemes allows for a more realistic 4-km precipitation distribution.