Results from a 3 1/2-yr experimental program of extended-range integrations of the European Centre for Medium-Range Weather Forecasts (ECMWF) numerical weather prediction model are summarized. The topics discussed include
The evolution of extended-range systematic error and skill in forecasting large-scale weather regime transitions;
The dependence of extended-range systematic error and skill on model horizontal resolution;
Monthly mean forecasts of tropical rainfall;
Tropical/extratropical interaction, and the influence of tropical low-frequency variability on extratropical forecast skill;
Ensemble forecasting, including the impact of ensemble averaging on forecast skill, and ensemble dispersion as a measure of forecast reliability; and
Probabilistic forecasting using phase-space cluster analysis.
Our results are broadly consistent with those from other major centers evaluating the feasibility of dynamical extended-range prediction. We believe that operational extended-range forecasting using the ECMWF model may be viable to day 20—and possibly beyond—following further research on techniques for Monte Carlo forecasting, and when model systematic error in the tropics has been reduced significantly.