Elsevier

Atmospheric Environment

Volume 44, Issue 5, February 2010, Pages 688-696
Atmospheric Environment

Comparison of land-use regression models between Great Britain and the Netherlands

https://doi.org/10.1016/j.atmosenv.2009.11.016Get rights and content

Abstract

Land-use regression models have increasingly been applied for air pollution mapping at typically the city level. Though models generally predict spatial variability well, the structure of models differs widely between studies. The observed differences in the models may be due to artefacts of data and methodology or underlying differences in source or dispersion characteristics. If the former, more standardised methods using common data sets could be beneficial. We compared land-use regression models for NO2 and PM10, developed with a consistent protocol in Great Britain (GB) and the Netherlands (NL).

Models were constructed on the basis of 2001 annual mean concentrations from the national air quality networks. Predictor variables used for modelling related to traffic, population, land use and topography. Four sets of models were developed for each country. First, predictor variables derived from data sets common to both countries were used in a pooled analysis, including an indicator for country and interaction terms between country and the identified predictor variables. Second, the common data sets were used to develop individual baseline models for each country. Third, the country-specific baseline models were applied after calibration in the other country to explore transferability. The fourth model was developed using the best possible predictor variables for each country.

A common model for GB and NL explained NO2 concentrations well (adjusted R2 0.64), with no significant differences in intercept and slopes between the two countries. The country-specific model developed on common variables for NL but not GB improved the prediction.

The performance of models based upon common data was only slightly worse than models optimised with local data. Models transferred to the other country performed substantially worse than the country-specific models. In conclusion, care is needed both in transferring models across different study areas, and in developing large inter-regional LUR models.

Introduction

Various studies have documented health effects related to current day levels of outdoor air pollution (Brunekreef and Holgate, 2002). Health effects related to long-term exposure to outdoor air pollution have been less studied than health effects of short-term exposure, partly because of the difficulty in generating valid estimates of long-term exposure over large study areas. A variety of methods have been used in recent years to represent spatial variation of outdoor air pollution, including indicator variables, interpolation of measurements, dispersion modelling and land-use regression models (Jerrett et al., 2005, Briggs, 2005). Land-use regression (LUR) models were initially devised as a means of modelling spatial variations in urban air pollution as part of epidemiological studies of traffic-related pollution and health (Briggs et al., 1997). They involve developing multiple regression models by analysing associations between measured pollutant concentrations at a number of monitoring sites spread over the study area and predictor variables relating to nearby emission sources and the local dispersion environment (e.g. traffic flows, road length, land cover, topography). Since then the approach has been tested and applied in a range of different urban environments (Briggs, 2005, Hoek et al., 2008), and it has been used for exposure assessment in a number of epidemiological studies (Brauer et al., 2003, Beelen et al., 2007, Morgenstern et al., 2007, Jerrett et al., 2009). Recently, Hoek et al. (2008) have reviewed the methods and applications, and shown that, although performance of the method is generally good and remarkably consistent in different study areas, the resulting models vary considerably. Whether these differences are artefacts of the available data, are due to arbitrary choices made by the researchers, or are a reflection of underlying differences in the configuration of cities and their effects on air pollution processes is not clear.

Most studies to date have focused on the use of LUR for local (usually city-scale) applications. Few attempts have been made to apply the technique at a broader, for example national or trans-national, scale (Stedman et al., 1997, Beelen et al., 2007, Beelen et al., 2009, Janssen et al., 2008). This poses a dilemma because the small relative risks generally associated with ambient air pollution imply the need for large population studies in order to provide sufficient statistical power to obtain reliable risk estimates. At the same time, these studies need to be undertaken at the individual level. Increasing attention is therefore being given to the implementation of large multicenter studies, often across different countries. There is also a growing demand for consistency in methods of exposure assessment between different studies, in order to facilitate comparison and pooling of risk estimates (HEI, in press). In addition, there is interest in using relatively simple screening methods to map air pollution at the national and continental scale, both to inform policy and as a basis for risk assessment. LUR techniques would seem to have great potential in this context. To be effective, however, it must be possible to apply the method in a consistent form, with readily available and uniform data, across relatively large study areas – and to do so without paying an unacceptable penalty either in terms of spatial resolution or model performance.

This paper explores the capability to achieve this. It tests the application of LUR techniques to map air pollution in Great Britain (GB) and the Netherlands (NL), under different data conditions: using only consistent data available in both countries, and using country-specific data to derive locally optimised models. It thus assesses the impact both on model performance and exposure estimates of restricting the method to the use of common data sets, and evaluates the transferability of models from one country to another.

Section snippets

Methods

Models were developed in the two countries for two pollutants – NO2 and PM10. These were selected both because they are recognised as priority pollutants under EU air quality regulations, and because of their widespread use in epidemiological studies either as important exposures in their own right, or as markers for traffic-related pollution more generally (Brunekreef and Holgate, 2002). Models were constructed on the basis of monitored data representing annual mean concentrations for the year

Model structure and performance

The regression models obtained for each pollutant and each country, together with their performance statistics are presented in Table 3, Table 4. The locally optimised model (model 3) for NO2 is shown in Fig. 1.

For NO2 (Table 3), the combined model (model 0) comprises three variables in addition to an indicator variable for country: population within 3000 m, major road length within 200 m, and urban influence (total built-up land within 20 km). The overall R2 during model building is 0.64 and

Discussion

The growing use of land-use regression models for air pollution mapping has resulted in models of differing structure and performance in different study areas. Developing locally optimised models may improve the estimates of air pollution, but it also increases the difficulty of comparing results from different studies. The observed inconsistencies in the models may be due to artefacts of data and methodology or may reflect underlying differences in source or dispersion characteristics. If the

Conclusion

This paper provides evidence of the capability to use LUR models as a basis for air pollution mapping and exposure assessment at the national level. This potential is substantially better for NO2 than for PM10, probably related to the more limited density of particulate monitoring networks and the smaller spatial variations in PM10 concentrations. In the countries studied here, LUR models developed on the basis of widely available and common data do not perform markedly worse than models

Acknowledgements

The work was supported by grant RGI-137 from the Dutch program Ruimte voor Geoinformatie supported by the Ministry of Housing, Spatial Planning and the Environment, and also partly supported though the EU 6th Framework Programme INTARESE project (018385-2).

References (19)

There are more references available in the full text version of this article.

Cited by (96)

  • The impacts of land supply on PM2.5 concentration: Evidence from 292 cities in China from 2009 to 2017

    2022, Journal of Cleaner Production
    Citation Excerpt :

    In fact, the impact of land use on air quality has been widely recognized by the academic community. A typical example is that many studies have developed and applied the Land Use Regression (LUR) model at different scales (Briggs et al., 1997; Clougherty et al., 2009; Tripathy et al., 2019; Vienneau et al., 2010). The LUR model combines the air pollutant concentration data of a limited number of monitoring sites, with land use characteristics, and other influencing factors.

View all citing articles on Scopus
View full text