Tests of a Fine-Mesh Model over Europe and the United States

Abstract

Thirty?two 24 h forecasts have been run over western Europe and the eastern United States using a six-layer, 60 km mesh primitive equation model. The forecasts show considerable skill in forecasting cyclogenesis over the Mediterranean and the United States in spite of the inadvertent neglect of surface friction over half of the domain. The average 24 h S 1score for sea level pressure is 39.1 compared to an average of 45.9 for Fleet Numerical Weather Central's operational model and 73.4 for persistence. The initialization scheme is based on an objective analysis of the horizontal wind field. Following the wind analysis, we infer geopotential and temperature from the rotational part of the wind with a nonlinear form of the balance equation. We present detailed results from one initial analysis and error statistics from 30 analyses occurring from December 1977 through April 1978. Typical root-mean-square (rms) differences between first-guess and balanced analyses of geopotential and temperature are ?20 m and 2°C, while rms vector differences between analyzed and balanced winds are ?5 m s?1 Three forecasts are discussed in detail. The first is a case of cyclogenesis in the Gulf of Genoa that was forecast well by the model. The second is a forecast of the intense Ohio blizzard of 26 January 1978, which was also reasonably successful. The structure of the model planetary boundary layer (PBL) in the second forecast is studied with the aid of a one-dimensional, second-order closure PBL model. The third forecast greatly overpredicted the intensity of a cyclone along the southern coast of the United States. Latent heating and the parameterization of cumulus convection were dominant factors in producing this fictitious intensification. In the latter two cases the sensitivity of the model's 24 h forecast of cyclogenesis to surface friction and parameterization of cumulus convection is established. A major conclusion from this study is that significant improvements in 24 h sea level pressure forecasts were obtained from a model with high horizontal resolution, even though the vertical resolution was coarse and the physics in the model was simple. It appears likely that further increases in forecast accuracy are possible by refining the vertical resolution and improving the physics.