Modelling and assessing trends in traffic-related emissions using a generalised additive modelling approach

A ueneralised additive modelling (GAM) approach is used to model daily concentrations of nitrogen oxides (NOx), nitrogen dioxide (NO2), carbon monoxide (CO), benzene and 1,3-butadiene at a busy street canyon location in central London. The models were developed for the period July 1998-June 2005 using appropriate meteorological and road traffic covariates. For all models, the complex and localised wind-flow patterns resulting from the street canyon location of the monitoring site, which can be difficult to model deterministically, have a large influence on the model predictions. It is shown that GAMs built using simple covariates explain a large amount of the daily variation for these pollutants (mean r(2) = 0.86). It is found that concentrations of benzene and 1,3-butadiene have declined in line with detailed calculations of emissions trends, with some evidence to suggest that reductions in benzene have been greater than estimated reductions in emissions. Although measured concentrations of NOx have declined from 1998 to 2005, much of the decline appears to be associated with reductions in overall traffic and meteorological factors rather than reduced emissions of NOx. Unadjusted NOx trends show a 28.6% reduction (95% confidence interval from 21.2% to 35.8%) from 1998 to 2005, whereas meteorologically adjusted trends show a 19.3% decline (95% confidence interval from 14.8% to 23.5%) over this period. Analysis shows that there were a higher number of occasions in the early part of the time series that led to strong recirculation of exhaust emissions and higher NOX concentrations at this location, thus affecting observed trends in concentration. (C) 2007 Elsevier Ltd. All rights reserved.