Statistical modelling of aerosol particle number size distributions in urban and rural environments – A multi-site study

Abstract

A multiple linear regression approach was used to model the urban atmospheric particle number size distribution (NSD) at 11 different sites (roadside, urban background, rural) in Central European cities for the time period 2008–2010. The same set of 13 model input parameters, consisting of temporal information (daytime, season) and meteorological measurement data, was used at each site.NSD model performance indicates an average deviation between observations and model (Bias) in the order of<10% with respect to total particle number concentration. The most reliable predictions were achieved for roadside sites with correlation coefficients (R) of 0.75 on average and a normalized root mean square error (RMSEn) of 0.79. Limited performance was observed for rural sites (R=0.61; RMSEn∼1.2). The transferability of the model approach to an independent urban background site was tested showing R of 0.5 and normalized RMSEn of 1.4. Although the physical relationship between particle NSD, ambient meteorological conditions and the temporal parameters are extremely complex, the model was able to reproduce the variation in particle concentrations. As a first in the field this study focused on modelling the entire number size distribution in contrast to size integrated particle number concentrations.