Generalised additive modelling of air pollution, traffic volume and meteorology

We present a general model where the logarithm of hourly concentration of an air pollutant is modelled as a sum of non-linear functions of traffic volume and several meteorological variables. The model can be estimated within the framework of generalised additive models.Although the model is non-linear, it is simple and easy to interpret. It quantifies how meteorological conditions and traffic volume influence the level of air pollution. A measure of relative importance of each predictor variable is presented.Separate models are estimated for the concentration of PM10PM10, PM2.5PM2.5, the difference PM10PM10–PM2.5PM2.5, NO2NO2 and NOxNOx at four different locations in Oslo, based on hourly data in the period 2001–2003. We obtain a reasonably good fit, in particular for the largest particles, PM10PM10 and PM10PM10–PM2.5PM2.5, and for NOxNOx. The most important predictor variables are related to traffic volume and wind. Further, relative humidity has a clear effect on the PM variables, but not on the NO variables. Other predictor variables, such as temperature, precipitation and snow cover on the ground are of some importance for one or more of the pollutants, but their effects are less pronounced.