Given the increasing interest in the association between exposure to air pollution and adverse health outcomes, the development of models that provide accurate spatio-temporal predictions of air pollution concentrations at small spatial scales is of great importance when assessing potential health effects of air pollution. The methodology presented here has been developed as part of the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air), a prospective cohort study funded by the US EPA to investigate the relationship between chronic exposure to air pollution and cardiovascular disease. We present a spatio-temporal framework that models and predicts ambient air pollution by combining data from several different monitoring networks with the output from deterministic air pollution model(s). The model can accommodate arbitrarily missing observations and allows for a complex spatio-temporal correlation structure.

We apply the model to predict long-term average concentrations of gaseous oxides of nitrogen (NOx) ─ one of the primary pollutants of interest in the MESA Air study ─ during a ten year period in the Los Angeles area, based on measurements from the EPA Air Quality System and MESA Air monitoring. The measurements are augmented by a spatio-temporal covariate based on the output from a source dispersion model for traffic related air pollution (Caline3QHC) and the model is evaluated using cross-validation. The predictive ability of the model is good with cross-validated R2 of approximately 0.7 at subject sites.

The incorporation of a dispersion model output into the overall prediction model was feasible, but the particular implementation of Caline3QHC used here did not improve predictions in a model that also includes road information. However, excluding the road information the inclusion of model output improves predictions and we find some evidence that the source dispersion model can replace road covariates.

The model presented in this paper has been implemented in an R package, SpatioTemporal, which will be available on CRAN shortly.


Statistical Models