The cost of uncertainty in weather prediction: modelling quality-value relationships for yes/no forecasts

Weather forecasts are produced to help people make decisions in weather sensitive situations, so an important aspect of forecast quality is the economic value of the forecasts to decision-makers in real applications. Assessing the economic value of weather forecasts in real-life operations can, however, be complex. The forecasts are usually only one of a number of factors influencing decisions, and it may be difficult to disentangle their effect from other factors. The relevant benefits from correct forecasts or losses from incorrect forecasts may not be known exactly, or may be ‘non-monetary’, for example the value of life or suffering. Information necessary for optimising forecast value in the presence of uncertainty might be unavailable, so that inappropriate decisions are made even when the forecasts are quite accurate. In the case of public weather forecasts, which are used by many different decision-makers, the difficulty is multiplied. Recent studies on the economic evaluation of forecasts include Leigh (1995), Katz and Murphy (1997), Anaman et al. (1998), Zhu et al. (2002) and Richardson (2003). It is plausible that more accurate forecasts will lead to better outcomes in a commercial sense, so verification has often focussed on accuracy, the association between forecasts and the relevant observations, rather than value. It has been found that some measures of accuracy may not be reliable indicators of value, so that forecasts which appear more accurate may not be more valuable to all users (Chen et al. 1987; Murphy and Ehrendorfer 1987). Also, some Aust. Met. Mag. 53 (2004) 111-122