A Bayesian Spatial Temporal Model of Ozone Submitted in Fulfillment of Integrative Exercise 2008-2009

We would like to thank Professor Katie St. Clair, our advisor for this project, for her guidance and thoughtful comments. We would also like to thank the Kari Palmer and Gre-gory Pratt of the Minnesota Pollution Control agency for providing ozone data and relevant information. Additionally, we are grateful to Wei-Hsin Fu for help with ArcGIS. We model and attempt to forcast low level atmospheric ozone concentration across Min-nesota in a Bayesian setting. Our data include 2007 ozone measurements from the Min-nesota Pollution Control Agency as well as meteorological data. Our model, based on McMillan et al., is hierarchical and incorporates nearest neighbor spatial and time series parameters. We compute the posterior distributions using Gibbs sampling and slice sam-pling, implemented with WinBUGS. We provide an overview of Bayesian hierarchical mod-eling and our data, and discuss the rationale for the model. Our results indicate that ozone concentration is primarily dependent on the previous day’s ozone concentration in the same location and the concentration to the west, with temperature and pressure also significant factors. Our predictions are, at least for two days into the future, fairly accurate but mis