A Bayesian Spatio-Temporal framework to improve exposure measurements combining observed and numerical model output

The high resolution Air Dispersion Modelling System (ADSM)-Urban represents an advanced model to simulate the local traffic and non traffic related contribution of PM10. The aim of our study is to provide a Bayesian framework to improve exposure estimates of PM10 combining observed data from monitoring sites with ADMS-Urban numerical model output. To illustrate our approach we use PM10 daily averaged values for 46 monitoring sites in London, over the period 2002-2003 and output from ADMS-Urban. Different spatio-temporal structures are investigated and compared in performance. We demonstrate that adding covariates on environmental characteristics of sites and meteorological changes over time improve the precision and accuracy of the concentration estimates.