Using the Generalised Additive Model to model the particle number count of ultrafine particles

In this paper, we compare the Generalised Linear Model (GLM) and Generalised Additive Model (GAM) for modelling the particle number concentration (PNC) of outdoor, airborne ultrafine particles in Helsinki, Finland. We examine temporal trends in PNC and examine the relationship between PNC and rainfall, wind speed and direction, humidity, temperature and solar insolation. Model choice is via the Akaike Information Criterion (AIC). e shown that the Generalised Additive Model provides a better fit than the equivalent Generalised Linear Model (ELM) when fitting models with the same covariates with equivalent degrees of freedom (AIC and BIC for the GAM are 10266.52 and 10793.04, AIC and BIC for the ELM are 10297.19 and 10885.97, both have an R2 value of 0.836). We also present results that show that modelling both temporal trends and the effect of rainfall, wind speed and direction, humidity, temperature and solar insolation yields a better fitting model, according to the AIC, than either temporal trends or meteorological conditions by themselves. del is applicable to any longitudinal monitoring-type measurement campaign where long time series are recorded. Use of this technique may be inappropriate for very short measurement campaigns. Attempting to fit a representative daily trend to one or two days’ measurements may lead to a high degree of uncertainty; inclusion of a yearly trend requires having at least a year’s worth of data with few gaps, particularly large gaps. In such a situation, the temporal trends may end up being penalised to zero and the model reverts to one largely influenced by meteorology.