Use of the Odds Ratio in Model QPF Verification

Keith Brill



This presentation reviews the treatment of the 2 X 2 contingency table of outcomes obtained in verifying quantitative precipitation forecasts (QPF) against observations with reference to a single threshold value, which is either exceeded or not. The likelihood- base rate factorization of the resulting joint probability distribution between forecasts and observations is used to define the ratio of the odds of detection to the odds of false alarming. This odds ratio is then used to parameterize the relative operating characteristic (ROC) curve and derive an expression for the area under the curve. Using this area to measure forecast skill relative to chance results in an Odds ratio Parameterized ROC Area (OPRA) score for measuring forecast performance. OPRA scores, equitable threat scores, and bias corrected equitable threat scores are compared for measuring Eta and GFS performance.