Monitor-to-monitor temporal correlation of air pollution in the contiguous US

Abstract

Numerous studies have reported short-term associations between ambient air pollution concentrations and mortality and morbidity. Particulate matter (PM) was often implicated as the most significant predictor of the health outcomes among the various air pollutants. However, a question remains as to the potential role played by the relative error of exposure estimation associated with each pollutant in defining their relative strengths of association. While most of the recent studies on PM exposure measurements have focused on the temporal correlation between personal exposures and the concentrations observed at ambient air quality monitors (within a few miles from the subjects), there have been few studies that systematically evaluated spatial uniformity of temporal correlation of air pollution within the scale of a city (several tens of miles) for which mortality or morbidity outcomes are aggregated in time-series studies. In this study, spatial uniformity of temporal correlation was examined by computing monitor-to-monitor correlation using available multiple monitors for PM₁₀ and gaseous criteria pollutants (NO₂, SO₂, CO, and O₃) in the nationwide data between 1988 and 1997. For each monitor, the median of temporal correlation with other monitors within the Air Quality Control Region (AQCR) was computed. The resulting median monitor-to-monitor correlation was modeled as a function of qualitative site characteristics (i.e., land-use, location-setting, and monitoring-objective) and quantitative information (median separation distance, longitude/latitude or regional indicators) for each pollutant. Generalized additive models (GAM) were used to fit the smooth function of the separation distance and regional variation. The intercepts of the models across pollutants showed the overall rankings in monitor-to-monitor correlation on the average to be: O₃, NO₂, and PM₁₀, (r∼0.6 to 0.8)>CO (r<0.6)>SO₂ (r<0.5). Both the separation distance and regional variation were important predictors of the correlation. For PM₁₀, for example, the correlation for the monitors along the East Coast was higher by ∼0.2 than western regions. The qualitative monitor characteristics were often significant predictors of the variation in correlation, but their impacts were not substantial in magnitude for most categories. These results suggest that the apparent regional heterogeneity in PM effect estimates, as well as the differences in the significance of health outcome associations across pollutants, may in part be contributed to by the differences in monitor-to-monitor correlations by region and across pollutants.