Using cluster algorithms with a machine learning technique and PMF models to quantify local-specific origins of PM2.5 and associated metals in Taiwan.

The influence of long-range transport (LRT) of air pollutants on neighboring regions and countries has been documented. The magnitude of LRT aerosols and related constituents can misdirect control strategies for local air quality management. In this study, we aimed to quantify PM 2.5 (diameter less than 2.5 μm, PM 2.5) and associated metals derived from local sources and LRT in different geographic locations in Taiwan using advanced receptor models. We collected daily PM 2.5 samples (n = ∼1000) and analyzed 28 metals every three days from 2016 to 2018 in the northern, central-south, eastern, and southern areas of Taiwan. We first used a machine learning technique with a cluster algorithm coupled with a backward trajectory to classify local, regional, and LRT-related aerosols. We then quantified the source contributions with a positive matrix factorization (PMF) model for Taiwan weighted by region-specific populations. The northern and eastern regions were found to be more vulnerable to LRT-related PM 2.5 and metals than the central-south and southern regions in Taiwan. The LRT increased Pb and As concentrations by 90–200% and ∼40% in the northern and central-south regions. Ambient PM 2.5 -metals mainly originated from local traffic-related emissions in the northern, central-south, and southern regions, whereas oil combustion was the primary source of PM 2.5 -metals in the eastern region. By subtracting the influence from the LRT, the contributions of domestic emission sources to ambient PM 2.5 metals in Taiwan were 35% from traffic-related emission, 17% from non-ferrous metallurgy, 13% from iron ore and steel factories, 12% from coal combustion, 12% from oil combustion, 10% from incinerator emissions, and <1% from cement manufacturing emissions. This study proposed an advanced method for refining local source contributions to ambient PM 2.5 metals in Taiwan, which provides useful information on regional control strategies. [Display omitted] • To quantify PM 2.5 -metals derived from domestic sources in different geographic locations. • A machine learning technique with a cluster algorithm coupled with a backward trajectory was used. • LRT aerosols remarkably increased PM 2.5 in Taipei and Hualien. • The LRT increased Pb and As by 90–200% in the northern regions. • Traffic-related emissions were the most important sources of PM 2.5 -metals in Taiwan. [ABSTRACT FROM AUTHOR]