A GIS-based method for modelling air pollution exposures across Europe

https://doi.org/10.1016/j.scitotenv.2009.09.048Get rights and content

Abstract

A GIS-based moving window approach was developed as a means for generating high resolution air pollution maps over large geographic areas. The approach is demonstrated by modelling annual mean NO2 pollution for the EU-15 (excluding Sweden) at the 1 km level on the basis of emissions and meteorological data. Models were developed using monitoring data from 714 background NO2 sites for 2001 and validated by comparing predicted with observed NO2 concentrations for a reserved set of 228 background sites. First the emission map (NOx) was derived by disaggregating national emissions estimates, categorised by source, to a 1 km grid, using proxies including population and road density, traffic statistics and land cover. A set of annuli was then constructed, of varying radii, and these passed over the emissions grid to derive a calibration between measured annual average concentrations at each monitoring site and distance-weighted emissions in the surrounding area, using a focalsum function. The resulting model was then used to predict concentrations at the reserved set of validation sites, and measures of performance (R2, RMSE and fractional bias) obtained. Validation gave R2 = 0.61, RMSE = 6.59 and FB =  0.01, and indicated performance equivalent to universal kriging and better than ordinary kriging and land use regression.

Introduction

Studies of air pollution epidemiology face a serious dilemma. On the one hand, most studies show small excess risks: for increases of about 10 µg/m3, in the order of 5–20% for atmospheric particles (Stieb et al., 2002, Pope and Dockery, 2006) and 3–14% for NO2 (Filleul et al., 2005, Gehring et al., 2006, Beelen et al., 2008). On the other hand, the large numbers of people potentially exposed means that these may translate into large public health problems. In Austria, Switzerland and France, combined for example, Künzli et al. (2000) estimate that ambient air pollution accounts for 6% of total mortality amongst adults aged 30+ years, or 40 000 deaths annually. Across Europe, some 13 000 deaths annually of infants (ages 0–4) have also been attributed to outdoor air pollution, making it second in terms of the environmental burden of disease only to physical injuries (Valent et al., 2003). To analyse these effects, air pollution epidemiology typically requires large population studies, with sufficient power to detect the small increases in risk against the background of varying base disease rates. Effective risk management likewise needs to be supported by accurate mapping of exposures, both to identify areas and populations at risk, and to give reliable estimation of the overall health impact of present or future policies.

Herein lie the major challenges. Direct measurement of exposures (e.g. using personal exposure metres or biomarkers) is clearly not feasible for such large study populations. In many epidemiological studies, also, the relevant exposures have already occurred (sometimes many years previously), while in many assessments of risk and health impacts the concern is about potential future exposures. Some form of modelling is therefore essential. The traditional approach has been to extrapolate data from neighbouring (usually routine) monitoring sites (Dockery et al., 1993, Pope et al., 1995), though inevitably this leads to substantial exposure misclassification, and tends to attenuate variations in exposure (and dilute any measured associations with health outcome) by falsely assigning the same concentrations to large numbers of people. In an attempt to reduce this problem, several more sophisticated methods of spatial interpolation have been developed, including geostatistical techniques such as kriging in various forms (Liu and Rossini, 1996, Finkelstein et al., 2003, Jerrett et al., 2005b, Henderson et al., 2007), and Bayesian hierarchical modelling (Elliott et al., 2007, Yanosky et al., 2008). Nevertheless, all of these methods are highly dependent on the spatial coverage and representivity of the available monitoring sites, which are generally too sparse to give accurate characterisation of small-area variations in air pollution.

Alternative methods of modelling are therefore needed, which better reflect the factors (e.g. source distribution and propagation processes) that influence spatial variations in air pollution. Source–receptor (or dispersion) models offer one approach, and a number of long-range atmospheric transport models have been developed for use at the continental scale, including CHIMERE (Bessagnet et al., 2004, Konovalov et al., 2005), EURAD (Hass et al., 1993) and EMEP (Simpson et al., 2003). These, however, typically have a spatial resolution in the order of tens of kilometres, and as such are unsuitable for assessing exposures at the individual or small-area scale. Gaussian dispersion models (Hall et al., 2000a, Hall et al., 2000b, Holmes and Morawska, 2006), in contrast, have been widely proven at the local scale (Nyberg et al., 2000, Bellander et al., 2001, Levy et al., 2002), but their use across large study areas is limited by their hungry data demands and heavy processing requirements. In recent years, a range of GIS-based methods have therefore been developed (Briggs, 2005, Jerrett et al., 2005a) in an attempt to provide a more simple and flexible means of modelling source–receptor relationships. Amongst these, land use regression (LUR) has gained particular attention (Briggs et al., 1997, Briggs et al., 2000, Briggs, 2005, Brauer et al., 2003, Jerrett et al., 2005a, Kanaroglou et al., 2005, Ross et al., 2006, Hoek et al., 2008). Although this has been successfully applied for high resolution mapping across large (continental scale) areas (Beelen et al., 2009), however, its main application has been for local (urban-scale) modelling. In this context, somewhat different models have tended to be developed in different study areas, reflecting differences not only in the modelling strategy by the researchers concerned, but also in data availability and quality and, perhaps, the underlying influences on air pollution concentrations (Briggs, 2007, Hoek et al., 2008). For analysis across large and diverse areas, therefore, methods are needed which bring together the effect of land use and other factors in a more consistent form. One way of doing this explicitly is to translate them into measures of emission, and then use the functionality of GIS to model their contributions to pollutant concentrations at any location, on the basis of their spatial proximity.

This paper describes, demonstrates and tests an approach of this form. Developed as part of the EU-funded APMoSPHERE project (an acronym list is presented in Appendix A), it uses moving window techniques to derive a distance-weighted model of relationships between area-level emissions and monitored concentrations at a set of training sites, and then applies this to estimate concentrations at unsampled locations across the EU. The methodology is demonstrated using nitrogen oxide (NOx) emissions, and monitored NO2 data from the European Airbase database. The resulting NO2 model is compared to two more conventional methods—kriging and land use regression (Beelen et al., 2009).

Section snippets

Principles of the moving window approach

Moving window techniques and associated focal functions are now widely available in many GIS. As the name implies, these techniques involve passing a window, cell-by-cell, across a grid to derive a new value for the central (target) cell as some function of the other cells covered by the window. Several focal functions are available for this purpose. Focalsum, the function used here, sums the cell values covered by the window. The window configuration is determined by a text file called a

Calibration

Table 3 shows the results for the best-fit models at the training sites, obtained using different types of window, both with and without wind speed and direction.

Overall, model performance is highly consistent, notwithstanding the differences in both the distribution of weights and the shape of the annuli. Optimisation of the weights only marginally improves model performance compared to the starting (1/d2) models and circular annuli are only slight improvements on rectangular windows.

Discussion

Despite the relatively small number of available monitoring sites and the large area over which the predictions were made, the focalsum models developed here appear to provide realistic and reliable predictions of NO2 concentrations across the EU. Exposure estimates based on these models differ substantially from those based on the monitoring sites alone, and suggest that the monitoring network may give a somewhat biased indication of exposure distributions. This has important implications for

Conclusion

The contradictory needs for both wide geographical coverage and high spatial resolution in exposure assessment pose severe challenges for exposure modelling. This research demonstrates the use of focalsum methods to model annual average concentrations of nitrogen dioxide across a large continental area, at 1 km resolution. The method is seen to be capable of providing estimates of pollutant concentrations at least as reliable as the other approaches currently available. While it shares with

Acknowledgements

This work was largely undertaken in the framework of the EU-funded APMoSPHERE project (EVK2-2002-00577) between December 2002 and September 2005. It was also partly supported through the EU 6th Framework Programme INTARESE project (018385-2), and by funding from the Ruimte voor Geo-Informatie (RGI), Netherlands. The authors gratefully acknowledge the financial support given by the funders, and the scientific input and advice of colleagues working on these projects. Particular thanks are due to

References (76)

  • N.S. Holmes et al.

    A review of dispersion modelling and its application to the dispersion of particles: an overview of different dispersion models available

    Atmos Environ

    (2006)
  • M. Jerrett et al.

    Particulate air pollution, social confounders, and mortality in small areas of an industrial city

    Soc Sci Med

    (2005)
  • P.S. Kanaroglou et al.

    Establishing an air pollution monitoring network for intra-urban population exposure assessment: a location–allocation approach

    Atmos Environ

    (2005)
  • N. Künzli et al.

    Public-health impact of outdoor and traffic-related air pollution: a European assessment

    Lancet

    (2000)
  • J.I. Levy et al.

    Using CALPUFF to evaluate the impacts of power plant emissions in Illinois: model sensitivity and implications

    Atmos Environ

    (2002)
  • L.J.S. Liu et al.

    Use of kriging models to predict 12-hour mean ozone concentrations in metropolitan Toronto—a pilot study

    Environ Int

    (1996)
  • W. Loibl et al.

    From national emission totals to regional ambient air quality information for Austria

    Adv Environ Res

    (2001)
  • R.V. Martin

    Satellite remote sensing of surface air quality

    Atmos Environ

    (2008)
  • H. Pleijel et al.

    On the logarithmic relationship between NO2 concentration and the distance from a highroad

    Sci Total Environ

    (2004)
  • M.C. Roorda-Knape et al.

    Air pollution from traffic in city districts near major motorways

    Atmos Environ

    (1998)
  • J.D. Yanosky et al.

    Spatio-temporal modeling of chronic PM10 exposure for the Nurses' Health Study

    Atmos Environ

    (2008)
  • D. Zheng et al.

    Edge effects in fragmented landscapes: a generic model for delineating area of edge influences (D-AEI)

    Ecol Model

    (2000)
  • M. Albani et al.

    The choice of window size in approximating topographic surfaces from digital elevation models

    Int J Geogr Inf Sci

    (2004)
  • M.A. Arain et al.

    The use of wind fields in a land use regression model to predict air pollution concentrations for health exposure studies

    Atmos Environ

    (2007)
  • R. Beelen et al.

    Long-term effects of traffic-related air pollution on mortality in a Dutch cohort (NLCS-AIR Study)

    Environ Health Perspect

    (2008)
  • T. Bellander et al.

    Using geographic information systems to assess individual historical exposure to air pollution from traffic and house heating in Stockholm

    Environ Health Perspect

    (2001)
  • R. Brand et al.

    EPER—European Pollutant Emission Register Review Report

    (2004)
  • M. Brauer et al.

    Estimating long-term average particulate air pollution concentrations: application of traffic indicators and geographic information systems

    Epidemiology

    (2003)
  • D. Briggs

    The role of GIS: coping with space (and time) in air pollution exposure assessment

    J Toxicol Environ Health Part A

    (2005)
  • D.J. Briggs

    The use of GIS to evaluate traffic-related pollution

    Occup Environ Med

    (2007)
  • D.J. Briggs et al.

    Mapping urban air pollution using GIS: a regression-based approach

    Int J Geogr Inf Sci

    (1997)
  • P.A. Burrough et al.

    Spatial analysis using continuous fields

  • C. Carroll et al.

    Using presence–absence data to build and test spatial habitat models for the fisher in the Klamath Region, U.S.A

    Conserv Biol

    (1999)
  • Centraal Bureau voor de Statistiek (CBS). Emissieregistratie. Milieu en Natuurplanbureau (MNP), Wageningen Universiteit...
  • M.C. Cirillo et al.

    Review study on European urban emissions inventories

    (1999)
  • S. Collins

    Modelling spatial variations in air pollution using GIS

  • S. Collins et al.

    A GIS approach to modelling small area variations in air pollution in a complex urban environment

  • B. Denby et al.

    Interpolation and assimilation methods for European scale air quality assessment and mapping, Part I: Review and recommendations, ETC/ACC Technical Paper 2005/7

    European Topic Centre on Air and Climate Change

    (2005)
  • Cited by (80)

    • Incorporating Light Gradient Boosting Machine to land use regression model for estimating NO<inf>2</inf> and PM<inf>2.5</inf> levels in Kansai region, Japan

      2022, Environmental Modelling and Software
      Citation Excerpt :

      These values were employed as predictors (distance to highway, secondary road, primary road, and a coast). The distance decay effect was introduced to the LUR model framework by Vienneau et al. (2009). They applied the focal-sum approach to the potential predictors and effectively modeled NO2 in their study.

    View all citing articles on Scopus
    View full text