Cluster analysis of atmospheric particle number size distributions at a rural site downwind of Seoul, Korea

We identified seven clusters from SMPS measurements ranging from 10 to 291 nm using k-means clustering, representing background (BG), local emissions (LocEm), two types of traffic emissions (TR), and three types of nucleation (NC). BG and LocEm differed in that pollutant concentrations were low and high, respectively, but both exhibited less pronounced diurnal variations in the frequency of occurrence. TR was divided into larger particles peaking at 72 nm and smaller particles with a primary peak at ~10 nm and a secondary peak at 72 nm. Because the main road is more than 1 km away from the study site, the primary peak consisting of smaller particles in the nucleation mode disappeared in most cases. However, when being affected directly from vehicle emissions, the primary peak size was reduced by the volatilization of semi-volatile substances. NC was divided into NC (fresh), NC (aged), and NC (trans) denoting nucleated far away and transported. NC (fresh) had a peak size of 21 nm and a peak time of 13:00, along with typical conditions favorable for new particle formation (NPF). The highest ozone and lowest PM10 concentrations for NC (aged) were conducive to particle growth, because semi-volatile substances were actively produced, while the condensation sink of these substances onto pre-existing particles was unavailable. Many characteristics of NC (trans) were similar to BG and LocEm, indicating NPF over a wider area. Because of the distance to the main road, the contribution to the nucleation mode was 44% for NC, higher than 26% for TR. •Optimal number of clusters by reviewing the separation of clusters.•Variation in the first mode of the bimodal distribution of vehicle-emitted particles.•Distinguishing favorable conditions for fresh nucleation and subsequent growth.•Contribution of each cluster to the number concentrations by mode.