The Wgne Assessment of Short-term Quantitative Precipitation Forecasts

APRIL 2003 | C ategorical, or conditional, statistics quantify the skill in the prediction of the occurrence of rain, and are based on the familiar 2 × 2 (yes/no) contingency table. Given a set of matched rain forecasts and observations, the contingency table is a matrix giving the frequencies of predicted and observed rain occurrence and nonoccurrence. “Hits,” H, are correct predictions of rain occurrence. “Misses,” M, indicate rain occurrences that were not predicted, while “false alarms,” F, denote predictions of rain for which no rain occurred. Finally, the frequency of correct forecasts of “no rain” can be denoted by Z (for “zero”). The total number of forecasts is N = Z + F + M + H. Although several categorical statistics can be computed from the entries in the contingency tables for brevity we will concentrate on the two statistics that are used to evaluate model quantified precipitation forecasts (QPFs) at the National Centers for Environmental Prediction (NCEP), namely the bias score and the equitable threat score. The bias score,