*2.4. Cluster Analysis of Back Trajectories*

The three-dimensional (latitude, longitude, and altitude) cluster analysis of back trajectories shows us the dominant transport pathways of air mass, which could reflect the potential influence of sources from a large scale during the sampling campaign. Trajectories were generated with version 4.9 of the NOAA Air Resources Laboratory's Hybrid Single-

Particle Lagrangian Integrated Trajectory (HYSPLIT) model [39]. Archived meteorological data from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis project data set with 6 h intervals and a spatial resolution of 2.5 × 2.5◦ in longitude and latitude were used as input data to calculate 36 h trajectories starting at 20 m and 380 m, separately. During the observation periods, back trajectories were computed every three hours for each sampling day (00 h, 03 h, 06 h, 09 h, 12 h, 15 h, 18 h, and 21 h in UTC). Then, all the trajectories were clustered with HYSPLIT clustering algorithm. The clustering of back trajectories is based on the total spatial variance (TSV) method [40], which minimizes the inter-cluster differences among trajectories.

Hourly weather data such as temperature (T), relative humidity (RH), wind speed (WS), and wind direction (WD) were obtained from an automatic weather station on the roof of a 20 m high building in Gulou campus, Nanjing University.
