Developing intra-urban land. use regression air pollution models with enhanced geographical variables

摘要

Background Land use regression (LUR) modelling has been widely used to provide long term exposure assessments of air pollutants for epidemiological studies. Quite often LUR models lack variables on site characteristics such as street canyons and building densities, which may affect the dispersion environment, as air pollutants can be trapped by buildings, especially where continuous street canyons are found with high buildings (e.g. urban centres). Aims To improve LUR models, and characterisation of local pollution microenvironments, using 'enhanced' geographical variables extracted from newly available city-wide datasets on building heights and geometry. Methods Models were developed for NOX, NO2 and PM10 for 2008-2011 for London, UK. A separate set of models, which only used 'traditional' land use and traffic indicators (e.g. traffic intensities and distances from road), were also developed and compared with the enhanced models. Results Considerable improvements to performance were seen in enhanced NOX and NO2 models compared to corresponding traditional ones (26.8% and 15.3% increase in adjusted R2, and 24.2% and 22.2% decrease in RMSE for NOX and NO2 respectively). Some improvements were found for PM10 (4.8% increase in adjusted R2 and 7.1% decrease in RMSE). New geographical variables provided greater levels of improvements for NOX and NO2, which was expected as these have higher proportions of local air pollution than PM. A new variable developed from building area × building height within 50m explained highest levels of spatial variability for NOX and NO2. Conclusions This study demonstrated that enhanced geographical datasets can provide substantial improvements to model capabilities, in particular predictions of pollutant spatial variability at road- and kerb-side locations for traffic-related pollutants. This may help to reduce exposure misclassification in epidemiological studies and merely improve estimation of health risks.