The economic value of a weather forecast is modelled, in a signal detection framework, as a function of skill, climatalogical frequency of the event and, most importantly, decision threshold. An expression has been derived, from first principles, for the optimal decision threshold, which minimises cost with respect to a perfect forecast. This optimal threshold is a function only of the costs of misses and false alarms.

An experiment involving Meteorologists in various offices of the Australian Bureau of Meteorology has been in progress for two years. They have been logging their percentage confidence, at various lead times, that the weather at each airport will be below the airport's alternate minimum. This is the level which causes airlines to take protective action by way of extra fuel. Results are shown of reliability diagrams and relative operating characteristics. It is shown that individual forecasters adopt varying decision thresholds, and that the difference in decision thresholds between individuals is more significant than differences in skill.

It is shown that forecasts at lead times of 3 hours or less generally fail to match persistence. An argument is mounted that providing TAFs as a probability of breaching the instrument approach minimum would provide a cost outcome superior to the current categorical system. Examples of likely savings for particular flights are presented. It will be shown that even the moderately reliable TAFs as currently produced, if given in probabilities, will produce considerable savings for airlines, at no greater forecaster skill, compared to the current categorical product.