J12.6 BIAS NORMALIZED PRECIPITATION SCORES Fedor Mesinger* and Keith Brill^ * NCEP/EMC and UCAR, ^ NCEP/CPC, Camp Springs, Maryland Abstract The purpose of the equitable threat or various related scores is to provide information on model’s accuracy in placing precipitation above a given threshold. Yet, they do not quite manage to achieve this because of their dependence on bias, so that a subjective assessment of how and how much these scores might have been affected by the model bias is customary. Conversely and as an opportunity for improving scores in a manner that can be considered as ethically questionable, common wisdom has it that a bias somewhat greater than one is profitable. It is shown that a more satisfactory state of affairs can be arrived at. A simple assumption of the increase of hits per unit increase in bias being proportional to the yet unhit area enables calculation of the number of hits normalized to a perfect bias. Thus, normalization of the equitable threat and related scores to perfect bias is possible. Assumption of the odds ratio being independent of bias can be used to the same end. Examples of the resulting bias normalized equitable threat scores of several operational NCEP models are presented. 1. Introduction The purpose of the threat scores, standard or equitable, is to assess the skill of a model in placing its forecasts of an event, say precipitation above a given threshold. There are quite a few other statistical quantities aiming for roughly the same objective, but equitable threat may well be the most popular. The problem of the skill assessment in a situation where there is a forecast of an event, that can either occur or not occur, is of course quite general, common to many fields. Note Murphy (1996) for an entertaining account of the early weather prediction efforts of more than a century ago, with various quantities reintroduced and renamed later, some more than once. The reason why threat and equitable threat score (or, Gilbert score, Schaefer 1990) are in meteorology more popular than some of the other measures is that threat and equitable threat emphasize skill in forecasting the occurrence of the event more than they do the skill in ____________________________________ * Corresponding author address: Fedor Mesinger, NCEP Environmental Modeling Center, 5200 Auth Road, Room 207, Camp Springs, MD 20746-4304; e-mail: fedor.mesinger@noaa.gov