Medium Range Prediction by a GFDL Global Spectral Model: Results for Three Winter Cases and Sensitivity to Dissipation

Abstract

A preliminary evaluation is made of the medium range predictive capability of a GFDL global spectral model of the atmosphere, based upon three winter blocking cases. Analogous forecasts by a GFDL global grid point model provide a background standard of comparison. The spectral model is rhomboidally truncated at wavenumber 30, has 9 sigma levels, incorporates sub-grid scale physical processes commonly associated with general circulation models and employs semi-implicit time differencing. The grid point model has somewhat finer horizontal resolution and fairly similar sub-grid scale physical processes, and employs explicit time differencing. The spectral model is up to 6 times more economical. The level of forecast skill for the 5 to 15 day range is generally less than practically useful and is more case-dependent than spectral versus grid point model-dependent. In the most successful case, i.e., 16 January 1979, an observed Atlantic blocking ridge is simulated quite well, especially by the spectral model. The predicted Atlantic ridges tend to retard approaching upstream transient disturbances. A zonal bias of the midlatitude circulation, which develops in all three spectral and grid point model predictions is most pronounced in the spectral model forecast from 1 January 1977. Results of a diffusion sensitivity experiment and other evidence suggest that insufficient frictional dissipation may have enhanced the zonal bias of the above forecast. The bias diminishes, consistent with a redistribution of spectral kinetic energy among zonal wavenumbers 0, 1 and 2, if a static stability-dependent pararmeterization of vertical mixing or stronger ?4 horizontal diffusion are used. Also, the predicted-enstrophy spectrum at midlatitudes steepens, given the stronger ?4 horizontal diffusion.