Source attribution of air pollution using a generalized additive model and particle trajectory clusters

Highlights: • Trajectory Cluster Contribution Function quantifies impacts of source regions. • Long-range transport estimated to increase local air pollution by 50%. • Variability in aerosol concentrations is about half local, half long-range. • NO₂ and elemental carbon are dominated by local sources. • COVID-19 pandemic had distinct impacts on local NO₂ and long-range PM_2.5. Speciated hourly measurements of fine aerosols were made for more than two years at an urban, an industrial and a port site in Busan, Korea. A Generalized Additive Model (GAM) was designed to deconvolve factors contributing to the pollutant concentrations at multiple scales. The model yields estimates of source contributions to pollution by separately identifying the signals in the time series due to meteorology, vertical mixing, horizontal wind transport and temporal variations such as diurnal, weekly, seasonal and annual trends. The GAM model was expanded to include FLEXPART back trajectory clusters generated using fuzzy c-means clustering. This made it possible to quantify the impact of long-range transport using the Trajectory Cluster Contribution Function (TCCF). TCCF provides a development of methods such as Concentration Field Analysis and Potential Source Contribution Function by providing numerical estimates of concentration changes associated with different air mass transport patterns while accounting for possible confounding factors from meteorology. The GAM simulations identified the importance of local transport for primary pollutants and long-range transport from China for secondary pollutants. Local factors accounted for up to 72% of the variance in concentrations of NO₂ and elemental carbon whereas large-scale/seasonal factors accounted for up to 56% of PM_2.5 and 80% of inorganic species. The algorithm further identified the importance of the weekend effect and the holiday effect at the different sites in Busan. The residual from the analysis was used to estimate the impact of the COVID-19 pandemic. The signature of the pandemic was different between the pollutants as well as from site to site. The model was able to distinguish small impacts from local pollutants at the residential site; short-lived acute impacts from industrial changes; and longer-term changes due to the early pandemic response in China.