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

https://doi.org/10.1016/j.envres.2007.09.004Get rights and content

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 1500 m, and Detroit-based SO2 emitters within 3000 m 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.

Introduction

There is an increasing interest in the impact that chronic exposure to ambient air pollution has upon both cardiac and respiratory health. Large cohort studies investigating chronic exposure to air pollution and the potential impacts this has upon mortality and morbidity have typically focused on variations in exposure across very large regions. This has ranged from the Eastern half of the US (Dockery et al., 1993; Laden et al., 2000) to all of the US (Pope et al., 1995, Pope et al., 2002). In more recent years, chronic exposure to air pollution gradients over much smaller regions (i.e. intra-urban) have been associated with both cardiac and respiratory health outcomes (Kuenzli et al., 2005). Similarly, a growing number of studies have emphasized exposure gradients resulting from proximity to roadways (Hoek et al., 2002; Jerrett et al., 2005a) and have shown that there are significant detrimental impacts of locally derived traffic-related pollutants on the respiratory health of children (Van Vliet et al, 1997; Zmirou et al, 2004).

There are acknowledged difficulties in accurately assigning exposure to large populations, and several large, long-term cohort studies have relied upon data collected by central monitoring stations. This approach has generally been adequate when populations span large areas or diverse cities (Dockery et al., 1993; Pope et al., 2002). However, when the effects of intra-urban differences in exposure are being studied, the use of existing central sites that are part of national monitoring networks are likely to be inadequate due to the lack of available monitors. Factors, such as traffic, are considered to be responsible for much of the intra-urban variability, which central monitoring stations typically cannot capture (Briggs et al., 2000; Gilbert et al., 2005; Hoek et al., 2002; Jerrett et al., 2005b). It is important to account for such variability in order to assign more appropriate chronic exposures to each individual.

To better characterize the smaller-scale spatial patterns in ambient air pollutants, a number of different techniques have been used. Interpolation techniques such as Kriging (Schaug et al., 1993; Briggs et al., 1997) and more recently, land-use regression modeling (LUR) (Briggs et al., 1997) attempt to address the intra-urban variability of air pollution. In addition, simple proxies such as distance to roadway have been found to be effective in some health studies (English et al., 1999; Lin et al., 2002; Van Vliet et al., 1997; Zmirou et al., 2004). It has been suggested that a well-calibrated LUR can portray localized variations in air pollution more effectively than Kriging (Jerrett et al., 2005b). These models predict pollutant concentrations at a given location based on surrounding traffic, land use and other air pollution-related indicators (e.g. point source locations). A number of LUR models have been developed in Europe, and also in North America, with varying degrees of success (Hoek et al., 2001; Kanaroglou et al., 2005; Gilbert et al., 2005; Jerrett et al., 2005b). Most of these studies have focussed on nitrogen dioxide (NO2), due to its link to vehicle emissions and other combustion sources, as well as the existence of well-established passive measurement techniques. The exposure patterns portrayed are often dominated by the distribution of traffic and thus, the NO2 exposures are assumed to be a good indicator of traffic pollution in general. Thus, a key question to consider is what other air pollutants, traffic related or not, could NO2 be representing? This question is important because there is currently debate regarding the extent to which NO2 exposure in itself, at current concentrations, has detectable health effects (Brook et al., 2007). Another issue is that most of the previous LUR models have been developed from monitoring data collected for a limited time period, often in only one season. As the overall goal for this study is to estimate chronic or long-term exposure, it is important to develop LUR models that predict actual long-term patterns. Thus, the concentration dataset used to develop the model should be representative of this pattern. The other key question to examine is how well the true within city spatial variability of NO2 and other traffic-related pollutants can be characterized by a single saturation monitoring campaign. The lack of measurements in multiple seasons has been cited as a potential source of error when developing previous LUR models (Jerrett et al., 2005b).

Recently, we have applied LUR models to characterize chronic exposure in Windsor, Ontario, Canada (Luginaah et al., 2006). This has been in support of prospective studies on children's respiratory health. An intra-urban monitoring network was established in Windsor and 2-week integrated air pollutant measurements were obtained for NO2, sulfur dioxide (SO2) and volatile organic compounds (VOCs) using passive samplers. This was conducted in each of the four seasons in 2004. In addition to developing LUR models for Windsor, these data were used in this paper to assess the two key issues raised above: the degree to which a two-week saturation monitoring campaign to capture spatial variability from a single season can represent other seasons and potentially the true long-term pattern, and; the relationship between the spatial pattern of NO2 and other pollutants. In addition, the difference in the model fit for other pollutants compared with the commonly studied NO2 was examined (i.e. how well does LUR work for other pollutants?)

Windsor is impacted by both domestic and international sources of air pollution including; local industrial and municipal point sources, commercial and residential area sources, and transportation sources that contribute to both domestic and transboundary air quality issues on both sides of the border. There are border crossings, which tend to ‘focus’ the traffic, at the Ambassador Bridge and the Detroit-Windsor Tunnel, which connect the two cities. These crossings, along with the Bluewater Bridge, which connects Sarnia and Port Huron, are the busiest international crossings between Canada and the United States. They represent nearly 50% of the traffic volume crossing the entire border between Canada and the United States, with over 75,000 vehicles traveling between the two countries each day.

This paper focuses on NO2, SO2, benzene, and toluene, which are expected to provide insight into the intra-urban variability of air pollution from a range of sources known to impact upon Windsor. The predominant source of NO2 is the combustion of fossil fuels such as gasoline, diesel and natural gas. Typically, in urban ambient air, NO2 concentrations are correlated with traffic volumes. SO2 is emitted by the combustion of sulfur-containing material; major sources of SO2 in ambient air include diesel-powered vehicles, coal-fired power plants, petroleum refineries, and metal-processing plants. Benzene and toluene are both constituents of gasoline and are typically emitted by vehicles, as well as through a variety of industrial processes some of which are related to the automotive industry and paint processing, both of which are present in Windsor and Detroit.

Section snippets

Materials and methods

Passive monitoring of NO2, SO2, and VOCs over a 2-week integrated period in each of four seasons was conducted throughout 2004. The methods employed for the passive monitoring include the patented method from Maxxam Analytics (Maxxam PASS, Maxxam Analytics, Calgary, Edmonton) for NO2 and SO2. VOCs were collected using 3M #3500 badges (Guillevan, Montreal). Analysis of the Maxxam samplers was completed by ion-chromatography and the 3M #3500 badges were analyzed by gas chromatography–mass

Results and discussion

The pollutant measurements from a total of 54 monitoring locations within the city of Windsor were included in the analysis and the locations of these sites are shown in Fig. 1. Blank corrected annual concentrations of NO2 ranged from 6.9 to 20.2 ppb, SO2 from 3.8 to 7.7 ppb, benzene from 0.5 to 1.4 μg/m3, and toluene from 1.3 to 6.3 μg/m3 (Table 3), all samples exceeded method detection limits of 3×standard deviation of the field blanks. The highest concentrations for the samples occurred during

Acknowledgments

James Brown of EnWin Utilities Ltd. for permission to place the monitors on lightpoles around Windsor. Students at the University of Windsor and Health Canada staff for conducting the field work. Alberta Environment for loaning rainshelters. Douglas Simpson at Environment Canada, Ontario Region for providing NAPS and meteorological data. Michael Brauer for guidance with developing the LUR model predictors.

References (23)

  • D.W. Dockery et al.

    An association between air pollution and mortality in six US cities

    N. Engl. J. Med.

    (1993)
  • Cited by (162)

    • Analysis of air quality spatial spillover effect caused by transportation infrastructure

      2022, Transportation Research Part D: Transport and Environment
    View all citing articles on Scopus

    Funding sources: The research was supported under the Border Air Quality Strategy. Funding was made available from the Air Health Effects Division, Health Canada.

    View full text