Forecast Intercomparisons from Three Numerical Weather Prediction Models

Abstract

A forecast intercomparison study using six different initial states was carried out with forecasts made by the 2.5°, six-layer, second-generation National Center for Atmospheric Research (NCAR) general circulation model (GCM); the 4° ? 5°, nine-layer Goddard Institute for Space Studies (GISS) GCM; and the 4° (at 50°N), six-layer National Meteorological Center (NMC) operational model. The initial conditions for geopotential and velocity were obtained from the NMC operational analyses during December 1972 and January 1973 for the Northern Hemisphere. The operational NMC analyses were used for verification of the forecast 1000 and 500 mb heights. Forecast projections to either three or five days were examined using conventional skill scores, wavenumber analysis of longitude-time plots constructed from the forecasts, and maps of the actual difference (forecast minus verification). The forecast errors were defined for each model and the common errors in all three forecasts were identified. The GISS and NMC models proved to be nearly equal in skill with the GISS forecasts being slightly better beyond two days. The present version of the NCAR model appears to be significantly less skillful, apparently due to initialization sensitivity, systematic bias and damping of the small scales. All three models tend to produce similar errors in the phase speeds of all waves. Smaller scale waves not only show systematic phase and amplitude errors, but also fail to develop and decay at the proper times. Phase and amplitude errors in the planetary wave structure are comparable in magnitude to errors made at higher wavenumbers. There was also a distinct tendency for errors in each model to be highly correlated with one another. The NCAR model produces a large mean error and systematically damps the smaller scale waves during the forecasts. The GISS model exhibited the best skill in the planetary-wave structure but the phase speeds of shorter waves were, on the average, slower than those in the other models.