Ensemble‐based ozone forecasts: Skill and economic value

Abstract

In the summer of 2004, seven air quality models provided forecasts of surface ozone concentrations over the eastern United States and southern Canada. Accuracy of these forecasts can be assessed against hourly ozone measurements at over 350 locations. The ensemble of the air quality models is used to issue deterministic and probabilistic forecasts of maximum daily 8‐hour and 1‐hour averaged ozone concentrations. For completeness, a short summary on performance of deterministic forecasts for this ensemble of models, obtained alternatively by averaging model concentrations or by using dynamic linear regression as described by Pagowski et al. (2006), is given on the basis of this work. In parallel, the skill of probabilistic forecasts is discussed. To remove the bias, the probabilistic forecasts are calibrated. The economic value of forecasts, which is calculated using Richardson's cost‐loss decision model, is evaluated for both deterministic and probabilistic cases. It is shown that deterministic forecasts obtained with the ensemble of models provide a greater benefit to decision makers than forecasts issued with individual models. Probabilistic forecasts demonstrate similar advantages over the deterministic forecasts.