STATISTICAL AND DYNAMICAL LONG-RANGE ATMOSPHERIC FORECASTS - EXPERIMENTAL COMPARISON AND HYBRIDIZATION

We perform a direct comparison between a statistical forecast model us ing space-time principal components as predictors and a series of expe rimental long-range (up to 44 days) dynamical forecasts performed at M eteo-France using a simplified-physics version of the former french op erational forecast model ''Emeraude''. The comparison is made possible by forecasting the same upper air quantity, the monthly averaged 50 k Pa geopotential height, at the same forecast dates and lead times, usi ng the same skill measure. A disappointing result is that most of the skill differences are nonsignificant, due to the small number of forec ast cases used (40). Moreover, the skill comparisons are obscured by i nherent biases due to the combination of trends, interdecadal variabil ity and systematic errors. We use a very conservative significance tes ting procedure taking into account these problems. A careful examinati on of the skill leads to the conclusion that the statistical model per forms better in the long run than the dynamical one. Of particular rel evance is the question whether the valuable information contained in t he empirical and the dynamical forecasts differ. If such is the case a n appropriate combination of bath forecasts, and/or forecasting algori thms could lead to an improvement of the skill. This is our second pur pose: we propose here two hybrid procedures which combine objectively the dynamical and statistical predictive information contents. The ski ll of these hybrid models is compared to that of the two original mode ls. One of the two hybrid schemes is shown to be significantly more sk illful, at long lead times, than the dynamical model.