Getting the Most out of Ensemble Forecasts: A Valuation Model Based on User–Forecast Interactions
Abstract A flexible theoretical model of perceived forecast value is proposed that explicitly includes the effects of user and ensemble characteristics and their interactions. The model can be applied to arbitrary decision problems and is sensitive to a much wider range of factors than traditional forecast valuation models. A simple illustration of its application to the cost–loss decision problem familiar from the forecast valuation literature is discussed. It is shown that perceived value is highly sensitive to perceived model accuracy and that in most cases a high level of perceived accuracy is required for the forecasts to be thought to have any value at all. Decisions with a cost–benefit ratio that is close to the climatological probability of the adverse event are shown to be less sensitive to perceived accuracy. The model shows that it is possible for perceived value to remain unchanged when perceived accuracy increases, thus suggesting an explanation for why forecast uptake often does not increase after improvements in model performance are made. Last, it is argued that attempts to increase forecast uptake should be targeted at those users whose cost–benefit ratios fall in a restricted range that depends on the climatological probability of the event and an objective measure of the ensemble accuracy.