A regression-based method for mapping traffic-related air pollution: application and testing in four contrasting urban environments

Abstract

Accurate, high-resolution maps of traffic-related air pollution are needed both as a basis for assessing exposures as part of epidemiological studies, and to inform urban air-quality policy and traffic management. This paper assesses the use of a GIS-based, regression mapping technique to model spatial patterns of traffic-related air pollution. The model — developed using data from 80 passive sampler sites in Huddersfield, as part of the SAVIAH (Small Area Variations in Air Quality and Health) project — uses data on traffic flows and land cover in the 300-m buffer zone around each site, and altitude of the site, as predictors of NO₂ concentrations. It was tested here by application in four urban areas in the UK: Huddersfield (for the year following that used for initial model development), Sheffield, Northampton, and part of London. In each case, a GIS was built in ArcInfo, integrating relevant data on road traffic, urban land use and topography. Monitoring of NO₂ was undertaken using replicate passive samplers (in London, data were obtained from surveys carried out as part of the London network). In Huddersfield, Sheffield and Northampton, the model was first calibrated by comparing modelled results with monitored NO₂ concentrations at 10 randomly selected sites; the calibrated model was then validated against data from a further 10–28 sites. In London, where data for only 11 sites were available, validation was not undertaken. Results showed that the model performed well in all cases. After local calibration, the model gave estimates of mean annual NO₂ concentrations within a factor of 1.5 of the actual mean (approx. 70–90%) of the time and within a factor of 2 between 70 and 100% of the time. r² values between modelled and observed concentrations are in the range of 0.58–0.76. These results are comparable to those achieved by more sophisticated dispersion models. The model also has several advantages over dispersion modelling. It is able, for example, to provide high-resolution maps across a whole urban area without the need to interpolate between receptor points. It also offers substantially reduced costs and processing times compared to formal dispersion modelling. It is concluded that the model might thus be used as a means of mapping long-term air pollution concentrations either in support of local authority air-quality management strategies, or in epidemiological studies.

Introduction

Growing concern about the effects of traffic emissions on respiratory health, and growing pressures for policy and management action to reduce air pollution levels, have highlighted the need for improved methods of mapping traffic-related pollution in urban areas both for exposure assessment and policy support.

The acute health effects of short-term exposures to traffic-related pollution have been widely demonstrated (for detailed reviews see Committee on Medical Effects of Air Pollutants, 1995, Committee of the Environmental and Occupational Health Assembly of the American Thoracic Society, 1996). Much less, however, is known about the chronic effects of exposure. A number of studies, mainly in the USA, have reported associations with particulates and/or sulfur oxides (Dockery et al., 1989, Schwartz, 1993, Pope et al., 1995), but replication of these effects in European studies has proved difficult. Several studies have found (often relatively weak) associations between chronic morbidity or mortality and traffic-related pollution, measured either in terms of distance from road or traffic volume on the nearest roads (Wjst et al., 1993, Edwards et al., 1994, Weiland et al., 1994), or using estimated exposures to NO₂ as a marker for traffic-related pollution (NILU, 1991, Oosterlee et al., 1996). On the other hand, a number of studies — including several which have attempted to use more specific measures of exposure, such as modelled or measured NO₂ concentrations — have found no detectable effects (Livingstone et al., 1996, Magnus et al., 1998, Wilkinson et al., 1999). The extent to which long-term exposure to relatively low levels of traffic-related air pollution causes or increases susceptibility to respiratory illness thus remains uncertain. Much of this uncertainty relates to the problems of acquiring reliable estimates of exposure to traffic-related pollution, at the individual or small-area level, across large populations and cities. Nevertheless, if such effects do occur, they would have serious public health implications, for though the relative risk may be low, the number of people exposed would be very large, representing a high attributable risk. This would also have significant policy implications, for it would imply the need to reduce background, as well as peak, concentrations of the pollutants concerned.

Maps are needed, equally, to inform management and policy. In the UK, for example, the National Air Quality Strategy (Department of Environment, Transport and the Regions, 1997), which was updated in 1999 (Department of Environment, Transport and the Regions, 1999), introduces air-quality targets to be met by the year 2005 and obliges local authorities to establish Air Quality Management Areas (AQMAs) in zones where these are likely to be exceeded. The Integrated Transport White Paper, published in July 1998, further calls for local traffic management plans aimed at reducing traffic congestion and air pollution, and promises legislation giving greater powers to local authorities to control road traffic in order to improve air quality. Local authorities thus have an urgent need for ways of mapping traffic-related air pollution, to help identify potential AQMAs, to target management action, and to predict and monitor the effects of intervention.

Nevertheless, mapping traffic-related pollution is no trivial task. Levels of road traffic pollution vary substantially, often over distances of metres, and pollution patterns in urban areas are complex (Laxen and Noordally, 1987, Hewitt, 1991). Monitored data on urban air pollution are also sparse. Most urban networks comprise only a few sites, and those which do exist can rarely be taken as representative of the exposures experienced by the population as a whole. Some form of modelling is, therefore, essential if accurate maps of urban air pollution are to be obtained.

A wide range of line-source dispersion models have been developed in recent years, which might ostensibly be used for this purpose. These include: the CALINE models (Benson, 1992); the CAR model (Eerhens et al., 1993); the ADMS model (CERC, 1999); the Operational Street Pollution Model (OPMS) (Berkowicz et al., 1994); and the AERMOD model (USEPA, 1998). In general, however, the performance of line-source models has not always been good (Henriques and Briggs, 1998), and the data demands and intensive processing requirements of the more sophisticated models mean that they are often difficult to apply for pollution mapping across whole cities at the small-area scale. The costs of many of these models may also make them prohibitive for local authority use. For many applications, therefore, the need is for simpler yet more robust methods of pollution mapping which can provide high-resolution, city-wide maps of traffic-related pollution using existing or readily obtainable data.

Regression mapping offers one such approach. It is based on the principles that: (a) environmental conditions for the variable of interest can be estimated from a small number of readily measurable predictor variables; and (b) that the relationship between the target variable and these predictors can be reliably assessed on the basis of a small sample survey or ‘training’ area. Probably the main use of this approach to date has been in the interpretation of remotely sensed data (Fuller et al., 1998, Li et al., 1998), where regression methods are used to determine the relationship between the measured signature (e.g. reflectance) and land cover or other attributes derived from ground truth surveys. Increasingly, however, regression methods have also been used for a wide range of other applications, including: mapping of landscape quality (Briggs and France, 1980); soil conditions (Knotters et al., 1995); salt contamination (Mattson and Godfrey, 1994); and air pollution (Wagner, 1994). The study reported here builds upon one such application — the use of regression mapping to estimate mean annual concentrations of traffic-related pollution as a basis for examining small area variations in air quality and chronic respiratory health (the SAVIAH study).