ON THE COMPARISON OF ONE OR MORE SETS OF PROBABILITY FORECASTS

Abstract

General expressions for the expected score for accuracy, score for skill and for operational value of the forecasts are developed and discussed. The expressions are then applied to the special case of two-class predictors and predictand, and an example is given to illustate how one set of probability forecasts can meet the operational requirements better than another set, even though both sets of forecasts are equally accurate and equally skillful.