Abstract
The global T42 version of the French numerical weather prediction model has been used to produce monthly mean forecasts. A study based on 21 cases of 44-day forecasts (for winter months from 1983 to 1990) is presented. Nine forecasts in this database may be directly compared with ECMWF 30-day forecasts. Some skill of 15-day running means exist for both models beyond day 15, and it is better with the ECMWF model. Beyond day 30, the predictive skill does not completely vanish: after systematic error correction, the 50-kPa height anomaly correlation over the Northern Hemisphere is 0.27 for day 15–44 avenges; 3 out of 21 values are negative, and 4 values exceed 0.50. The amplitude of the forecast anomaly explains a small part of this case-to-case skill variability. Similar results are found for the other atmospheric field. However, such a marginal skill could be useful only in association with other predictors in a statistical postprocessing.
