An Operational Multifield Analog/Anti-Analog Prediction System for United States Seasonal Temperatures. Part II: Spring, Summer, Fall and Intermediate 3-Month Period Experiments

An expanded version of the multifield analog prediction system developed by Barnett and Preisendorfer (1978) was described and applied to the winter season in Part I of this two-part series (Livezey and Barnston 1988). This second part reviews briefly the major design features detailed in Part I, and then describes the predictive skills in spring, summer, fall and all other intermediate 3-month seasons. In none of the 11 nonwinter seasons is the United States surface temperature predicted with as much skill as in winter. The major winter skill peak (16%) extends partially into the following two seasons (January–February–March and February–March–April), and a secondary maximum in summer (13%) similarly includes the two following seasons (July–August–September and August–September–October). In both skillful periods of the year the skill tends to be greatest over the eastern third of the United States and the immediate Pacific Coast and lowest over the Rockies and Plateau. More predictor variables are used as criteria for analog selection during the skillful times of the year, while fewer are found to contribute to skill levels at other times. The annual cycle of skill of simple persistence forecasts has its primary maximum in August–September–October, when it slightly exceeds the skill of the analog method, and a secondary maximum in winter when it is outperformed by the analog method by a substantial margin. The analog-forecasted temperature patterns are found to be statistically largely independent of the temperature patterns of persistence forecasts at all times of the year. In an exercise aimed at determining which part of the predicted period within the season is most skilfully forecast, it is found that from fall through winter the later month(s) of the season are better predicted than the earlier months, suggesting a potential for useful long-lead forecasting of subseasonal periods in much of the coldest part of the year.