Assessment of a bayesian multivariate interpolation approach for health impact studies

TLDR: Sun and Zidek as mentioned in this paper proposed a Bayesian approach for estimating air pollution at locations where monitoring data are not available, using the concentrations observed at other monitoring stations and possibly at different time periods.

Abstract: Health impact studies of air pollution often require estimates of pollutant concentrations at locations where monitoring data are not available, using the concentrations observed at other monitoring stations and possibly at different time periods. Recently, a Bayesian approach for such a temporal and spatial interpolation problem has been proposed by Le. Sun and Zidek (1997). One special feature of the method is that it does not require all sites to monitor the same set of pollutants. This feature is particularly relevant in environmental health studies where pollution data are often pooled together from several monitoring networks which may or may not monitor the same set of pollutants. The methodology is applied to the data in the Province of Ontario, where monthly average concentrations for summer months ofnitrogen dioxide (NO 2 in μg/m 3 ), ozone (O 3 in ppb), sulphur dioxide (SO in μg/m 3 ) and sulfate ion (SO 4 in μg/m 3 ) are available for the period from January I of 1983 to December 31 of 1988 at 31 ambient monitoring sites. Detailed descriptions of spatial interpolation for air pollutant concentrations at 37 approximate centroids of Public Health Units in Ontario using all available data are presented. The methodology is empirically assessed by a cross-validation study where each of the 31 sites is successively removed and the remaining sites are used to predict its concentration levels. The methodology seems to perform well.