Categorical Forecast Skill: Definitions and Implications

In meteorological forecasts it is usually the events that are not “normal” that generate interest. A model-statistical or numerical essentially models the deterministic component of the signal being forecast. Whereas the deterministic component, by definition, has a tendency to carry the information contained in the more confident region of the distribution of signal modelled as a random variable, the non-deterministic components lie in extreme regions. In order to judge the model’s performance in the extreme regions meteorologists’ resort to categorical forecast skill scores. The definitions of the classes provided in the domain sometimes seem contrary to inherent beliefs and among the researchers too. The contrasting definitions may lead to different set of scores. None can be said to be correct or incorrect. The authors have investigated these possibilities while predicting the UP-East precipitation against the Antarctic Sea ice concentration. The application domain is of immense importance owing to the global warming and other important climate drifts. The results suggest researchers should carefully draft the definitions before validating the models for dichotomous forecast in the extreme area of distribution functions.