On the Use of Analogues to Improve Regression Forecasts

Abstract

Three methods of objectively forecasting maximum temperature using multiple-stepwise screening are described and compared. The first method is the traditional “perfect prog” approach whereby a historical series of upper air predictors are screened against maximum temperature to establish a prediction equation. The second method employs a stratification of the historical data on the basis of mean sea level pressure gradient analogues prior to the regression step. A new prediction equation is required for each forecast. The third method is similar to the second except that random stratification is used instead of analogue stratification. A comparison of the accuracy of the three methods when tested on independent data indicates that the best forecasts are obtained when analogue stratification is used. Application of the analogue stratification method to “model output statistics” forecasts and the statistical correction of prognoses is discussed.