Local real-time forecasting of ozone exposure using temperature data

Rigorous and rapid assessment of ambient ozone exposure is important for informing the public about ozone levels that may lead to adverse health effects. In this paper, we use hierarchical modeling to enable real-time forecasting of 8-hr average ozone exposure. This contrasts with customary retrospective analysis of exposure data. Specifically, our contribution is to show how incorporating temperature data in addition to the observed ozone can significantly improve forecast accuracy, as measured by predictive performance and empirical coverage. We adopt two-stage autoregressive models, also introducing periodicity and heterogeneity while still maintaining our objective of forecasting in real time. The entire effort is illustrated through modeling data collected at the Village Green monitoring station in Durham, North Carolina.

Duke Scholars

Author Alan E. Gelfand Statistical Science