Properties of aerosol signature size distributions in the urban environment as derived by cluster analysis

Abstract

A cluster analysis algorithm was applied to reduce the amount and complexity of 30 min aerosol number size distributions in a three year data set (2006–2008) at a urban background station in Helsinki, Finland. Only after objective validity tests to determine the appropriate number of clusters, a k-means cluster algorithm was applied to extract seven characteristic size distributions from the data set. The average total number concentrations of the clustered size distributions range from 6067 cm−3 to 12,818 cm−3 with modal diameters between 5 and 193 nm. The clustered size distributions were analyzed in terms of their physical properties (shape, log-normal modes, mode diameter), temporal occurrence (e.g. time of day, season) and their relation to local meteorology.Three different types of cluster distributions being represented by either three or two log-normal modes (in only one case) were characterized at the site: four clusters that were indicative for urban-type size distributions with different influence of anthropogenic and traffic activities occurring 69% of the study time, two maritime-type distributions (29% occurrence) and one nucleation-type size distribution (2% occurrence). We were able to relate the clusters to characteristic modal diameters, different temporal occurrence on the daily and annual cycle, e.g. urban clusters that occurred year round and those that were attributed to winter daytime situations specifically. Analysis of the daily patterns clearly reveals the influence of local traffic activity on three of the four urban-type cluster size distributions. The method offers the chance for a simple kind of source apportionment by establishing signature size distributions based on the physical properties of the aerosol size spectra.