Predicting long-term average concentrations of traffic-related air pollutants using GIS-based information

Abstract

Global regression models were developed to estimate individual levels of long-term exposure to traffic-related air pollutants. The models are based on data of a one-year measurement programme including geographic data on traffic and population densities. This investigation is part of a cohort study on the impact of traffic-related air pollution on respiratory health, conducted at the westerly end of the Ruhr-area in North-Rhine Westphalia, Germany. Concentrations of NO 2, fine particle mass (PM 2.5) and filter absorbance of PM 2.5 as a marker for soot were measured at 40 sites spread throughout the study region. Fourteen-day samples were taken between March 2002 and March 2003 for each season and site. Annual average concentrations for the sites were determined after adjustment for temporal variation. Information on traffic counts in major roads, building densities and community population figures were collected in a geographical information system (GIS). This information was used to calculate different potential traffic-based predictors: (a) daily traffic flow and maximum traffic intensity of buffers with radii from 50 to 10 000 m and (b) distances to main roads and highways. NO 2 concentration and PM 2.5 absorbance were strongly correlated with the traffic-based variables. Linear regression prediction models, which involved predictors with radii of 50 to 1000 m, were developed for the Wesel region where most of the cohort members lived. They reached a model fit ( R 2) of 0.81 and 0.65 for NO 2 and PM 2.5 absorbance, respectively. Regression models for the whole area required larger spatial scales and reached R 2 = 0.90 and 0.82. Comparison of predicted values with NO 2 measurements at independent public monitoring stations showed a satisfactory association ( r = 0.66 ). PM 2.5 concentration, however, was only slightly correlated and thus poorly predictable by traffic-based variables ( r < 0.3 ). We concluded that NO 2 and soot can be considered truly traffic-related pollutants, and that GIS-based regression models offer a promising approach to assess individual levels of exposure to these pollutants.