This study is intended to determine the spatially varying optimal time periods for calculating seasonal climate normals over the entire United States based on temperature data at 344 United States climate divisions during the period of 1931–1993. This is done by verifying the seasonal climate normals as a forecast for the same season next year, The forecast skill is measured by the correlation between the predicted and observed anomalies relative to the 30-yr normal. The optimal time periods are chosen to produce the highest correlation between the forecasts and the observation.
The results indicate that generally (all seasons and all locations) annually updated climate normals averaged over shorter than 30-yr periods are better than the WMO specified 30-yr normal (updated only every 10 years), in terms of the skill in predicting the upcoming year. The spatial pattern of the optimal averaging time periods changes with season. The skill of optimal normals comes from both the annual updating and the shorter averaging time periods of these normals. Using optimal climate normals turns out to be a reasonably successful forecast method. Utility is further enhanced by realizing that the lead time of this forecast is almost one year. Forecasts at leads beyond one year (skipping a year) are also reasonably skillful.
The skill obtained from the dependent verification is lowered to take account of the degradation expected on independent data.
In practice the optimal climate normals with a variable averaging period were found to be somewhat problematic. The problems had to do primarily with the temporal continuity and spatial consistency of the forecasts. For the time being, a constant time period of 10 years is used in the operational seasonal temperature forecasts for all seasons and locations.