Abstract
An idealized model for error growth is used to discuss numerical weather prediction model forecasts of time averages ranging from 0 to 30 days in duration at lags of 0 to 30 days. The idealized model allows a simple assessment of how the initial error, error growth rate and serial correlation, influence a forecast model's prediction of a time average. The simplified nature of the idealized model also allows an easy demonstration of how various filters applied to the raw numerical predictions can help to improve forecast skill.