Intra-urban variability of air pollution in Windsor, Ontario--measurement and modeling for human exposure assessment

Abstract

There are acknowledged difficulties in epidemiological studies to accurately assign exposure to air pollution for large populations, and large, long-term cohort studies have typically relied upon data from central monitoring stations. This approach has generally been adequate when populations span large areas or diverse cities. However, when the effects of intra-urban differences in exposure are being studied, the use of these existing central sites are likely to be inadequate for representing spatial variability that exists within an urban area. As part of the Border Air Quality Strategy (BAQS), an international agreement between the governments of Canada and the United States, a number of air health effects studies are being undertaken by Health Canada and the US EPA. Health Canada's research largely focuses on the chronic exposure of elementary school children to air pollution. The exposure characterization for this population to a variety of air pollutants has been assessed using land-use regression (LUR) models. This approach has been applied in several cities to nitrogen dioxide (NO2), as an assumed traffic exposure marker. However, the models have largely been developed from limited periods of saturation monitoring data and often only represent one or two seasons. Two key questions from these previous efforts, which are examined in this paper, are: If NO2 is a traffic marker, what other pollutants, potentially traffic related, might it actually represent? How well is the within city spatial variability of NO2, and other traffic-related pollutants, characterized by a single saturation monitoring campaign. Input data for the models developed in this paper were obtained across a network of 54 monitoring sites situated across Windsor, Ontario. The pollutants studied were NO2, sulfur dioxide (SO2) and volatile organic compounds, which were measured in all four seasons by deploying passive samplers for 2-week periods. Correlations among these pollutants were calculated to assess what other pollutants NO2 might represent, and correlations across seasons for a given pollutant were determined to assess how much the within-city spatial pattern varies with time. LUR models were then developed for NO2, SO2, benzene, and toluene. A multiple regression model including proximity to the Ambassador Bridge (the main Canada-US border crossing point), and proximity to highways and major roads, predicted NO2 concentrations with an R2=0.77. The SO2 model predictors included distance to the Ambassador Bridge, dwelling density within 1500m, and Detroit-based SO2 emitters within 3000m resulting in a model with an R2=0.69. Benzene and toluene LUR models included traffic predictors as well as point source emitters resulting in R2=0.73 and 0.46, respectively. Between season pollutant correlations were all significant although actual concentrations for each site varied by season. This suggests that if one season were to be selected to represent the annual concentrations for a specific site this may lead to a potential under or overestimation in exposure, which could be significant for health research. All pollutants had strong inter-pollutant correlations suggesting that NO2 could represent SO2, benzene, and toluene.