Development and back-extrapolation of NO2 land use regression models for historic exposure assessment in Great Britain

John Gulliver*, Kees De Hoogh, Anna Hansell, Danielle Vienneau

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

88 Citations (Scopus)

Abstract

Modeling historic air pollution exposures is often restricted by availability of monitored concentration data. We evaluated back-extrapolation of land use regression (LUR) models for annual mean NO2 concentrations in Great Britain for up to 18 years earlier. LUR variables were created in a geographic information system (GIS) using land cover and road network data summarized within buffers, site coordinates, and altitude. Four models were developed for 2009 and 2001 using 75% of monitoring sites (in different groupings) and evaluated on the remaining 25%. Variables selected were generally stable between models. Within year, hold-out validation yielded mean-squared-error-based R2 (MSE-R2) (i.e., fit around the 1:1 line) values of 0.25-0.63 and 0.51-0.65 for 2001 and 2009, respectively. Back-extrapolation was conducted for 2009 and 2001 models to 1991 and for 2009 models to 2001, adjusting to the year using two background NO2 monitoring sites. Evaluation of back-extrapolated predictions used 100% of sites from an historic national NO2 diffusion tube network (n = 451) for 1991 and 70 independent sites from automatic monitoring in 2001. Values of MSE-R2 for back-extrapolation to 1991 were 0.42-0.45 and 0.52-0.55 for 2001 and 2009 models, respectively, but model performance varied by region. Back-extrapolation of LUR models appears valid for exposure assessment for NO2 back to 1991 for Great Britain.

Original languageEnglish
Pages (from-to)7804-7811
Number of pages8
JournalEnvironmental Science and Technology
Volume47
Issue number14
DOIs
Publication statusPublished - 16 Jul 2013
Externally publishedYes

Fingerprint

Dive into the research topics of 'Development and back-extrapolation of NO<sub>2</sub> land use regression models for historic exposure assessment in Great Britain'. Together they form a unique fingerprint.

Cite this