A Statistical Evaluation of the Predictive Abilities of Climatic Averages

Abstract

The predictive abilities of NOAA normals and running means of 2?30 years length are tested statistically. Heating degree-day (HDD) data from six northern United States sites are tested using root-mean-square error of prediction (RMSE) tests, mean absolute error (MAE) tests, and a ?best versus worst? predictor methodology. Monte Carlo tests using biased and unbiased numbers are presented for the RMSE and ?best versus worst? analyses. Results are consistent with past research in showing that running means 10?30 years in length perform better than shorter averaging periods for predictive purposes. The MAE values are generally found to be lowest for running mean lengths shorter than that for the RMSE statistic at the six sites. For the 30 years studied, NOAA HDD normals performed well along the east coast, indicating a possible regional difference that requires more detailed investigation. Limitations of the ?best versus worst? predictor method are discussed, and it is suggested that such a procedure should not be solely relied on in determining the optimum length of prediction.