Behaviour of a Few Popular Verification Scores in Yes/No Forecasting.

摘要

The main objective of verification in meteorology is to assess and quantify the quality of the forecasts. It offers a means to compare results of different models, methods or forecasters. In addition, their performance over different periods or years can be established in order to assess whether there is a general improvement or not. In general three different types of forecasts can be identified; deterministic, probabilistic and categorical ones. A categorical forecast is a forecast in which a statement is made that one and only one of a set of possible events will occur. In probabilistic forecasting probabilities are assigned to the event(s). This report focuses entirely on the issue of verification of categorical forecasts in which the predictand is binary and also the forecasts are issued in binary form. The results can therefore be summarized in a 2x2 contingency table. One can think of yes/no forecasts of, for instance, extreme events like tornadoes, but also of the occurrence of fog, thunderstorms or precipitation.