Abstract
Experimental 60-day forecasts have been run from observed data using the ECMWF operational spectral model with different resolutions. There still appears to be useful information beyond day 10 in the forecasts in spite of systematic errors. A more realistic sea surface temperature (SST) generally improves the forecasts in the long range. There are occasions when after a period of low predictive skill, the forecast recovers and reaches a level of skill similar to that found for short-range forecasts. The fact that this happens in the cases with the more realistic SST indicates that it is more than mere chance. The experiments, however, involve only a limited number of cases and are therefore not statistically significant. DOI: 10.1111/j.1600-0870.1986.tb00450.x