Sampling site and PM 10 collection
PM10 sampling was conducted at six sites located in the rural-fringe areas of Changhua, Yunlin, and Chiayi Counties in central Taiwan (Fig. 1), where the air quality is reported to be among the worst in Taiwan [60]. Several obvious anthropogenic sources of PM exist, such as a coal-fired plant (one of the largest in the world) in Taichung City, Changhua coastal industrial park in Changhua County, and a petrochemical complex (including a coal-fired power plant) in Yunlin County. Moreover, river dust in the Choshui River catchment is the major source of PM derived from natural materials. Details about the sampling sites, sampling period, wind speed, wind direction, temperature, precipitation, and relative humidity are summarized in Table 1. A total of five sampling campaigns were conducted during the northeast monsoon season to collect PM10 samples under different types of events, including three local events in 2018, one LRT event, and one dust storm event in 2019. Each event type was defined based on meteorological data, announcements provided by the TEPA, air mass back trajectory analysis (Hybrid Single Particle Lagrangian Integrated Trajectory [HYSPLIT]), and the global reanalysis dataset (EAC4). Note that the PM10 samples were only collected at site #5 during the dust storm event because only one sampler was available.
PM10 samples were collected following the TEPA PM10 sampling protocol (NIEA A208). In brief, PM10 samples were collected at a flow rate of 66 ± 3 m3·h− 1 on a polytetrafluoroethylene (PTFE) filter using a high-volume sampler (Tisch Environmental., Cleves, OH, USA) with a size-selective inlet (10 µm) attached on the rooftop (approximately 10 m above ground level). The sampling was conducted for 8–28 h, except for site #2 (2 h) on Nov 25, 2018. After collection, the filter was transferred into a plastic envelope and delivered to the conditioning room within a few hours. Furthermore, a total of 13 river dust samples were collected from eight sites (Fig. 1) to constrain the chemical and isotope signals of the natural source. The dust samples were collected in both upstream and downstream segments of the Choshui River catchment with plastic bags, dried at 45ºC in an oven, ground and sieved through a 75 µm sieve, and stored in a desiccator.
Chemical analysis
The PTFE filter was measured for particle mass concentration with a microbalance after equilibration at 25 ± 1.5ºC under a relative humidity of 40% ± 5% for 24 h. Next, 1/9 of the filter was cut using a ceramic cutter, and the aliquot was then digested with an acid mixture of 9 mL of concentrated HNO3, 3 mL of concentrated HCl, and 3 mL of concentrated HF using a microwave (CEM Corp., Matthews, NC, USA). A two-stage heating procedure was employed; first, the mixture was heated to 170ºC over 20 min and maintained at this temperature for 10 min, followed by another round of heating to 200ºC over 7 min and maintaining for 10 min. After cooling, the solution was transferred into a PFA beaker and evaporated. The dried sample was re-dissolved with an acid mixture of 2 mL of concentrated HNO3 and 1 mL of concentrated HCl, and finally diluted to 50 mL with Milli-Q water. Ultrapure concentrated acids and Milli-Q water were used for sample preparation in this study. For dust samples, 100 mg of dust was weighed and digested following the aforementioned procedure. A total of 23 elements, including major and trace metal concentrations, were analyzed by using Q-ICP-MS (Agilent 7700X) with internal standards (Sc, Y, Rh, Tb, Lu, and Bi) to monitor the instrumental drift and matrix effect (NIEA M105); the analytical precision was greater than 10% (RSD). The method detection limit for each element was determined by field blanks, as presented in Supplementary Table S3. For each batch of the sample digestion, the accuracy of metal analysis was examined according to two international standards, NIST SRM 1648a (urban PM) and NIES No. 30 (Gobi Kosa dust). The results of metal concentrations and recoveries for NIST SRM 1648a and NIES No. 30 are summarized in Supplementary Tables S4 and S5, respectively.
For the ion concentration, a sample aliquot of the filter was extracted with 50 mL of Milli-Q water in an ultrasonic bath for 30 min, followed by settling for 30 min. The solution was then filtered through a 0.45 µm PVDF filter (Merck KGaA, Darmstadt, Germany). Ion concentrations (SO42−, NO3−, NH4+, and Cl−) were measured using ion chromatography (Thermo-Fisher Scientific, Waltham, MA USA) with analytical precision greater than 10% (RSD).
Pb isotope analysis
An aliquot of the digested sample was transferred into an acid-cleaned PFA beaker. This solution was evaporated and re-dissolved with 2 mL of 2M HNO3 and 0.07M HF. The Pb fraction was extracted using Sr-spec ion exchange resin (Eichrom Technologies Inc., Lisle, IL, USA) following the steps modified from Pin et al. [61]. An international reference material (NIST 1648a) was also processed for each purification batch to assess the column chemistry performance. Pb was purified under a Class-10 laminar flow bench in a Class-10,000 clean room. The total procedural blanks for Pb were < 80 pg. Pb isotope ratios were measured using HR-MC-ICP-MS (Neptune PLUS, Thermo-Fisher Scientific) at the Institute of Earth Sciences, Academia Sinica (AS-IES). Standards and samples were doped with thallium (Tl) (NIST 997; 203Tl/205Tl ratio of 0.418673) to correct for instrumental mass fractionation [62]. Two standard reference materials (NIST SRM 981 and NIST SRM 1648a) were analyzed to assess the accuracy and long-term precision of the analytical protocol developed at AS-IES. The measured Pb isotope ratios for these international reference materials were in good agreement with the recommended values and are listed in Supplementary Table S6. The analytical uncertainties (2SD) for 206Pb/207Pb and 208Pb/207Pb were ± 0.0002 (n = 69) and ± 0.0003 (n = 69), respectively. Pb isotope ratios were reported as 206Pb/207Pb and 208Pb/207Pb in this study.
Enrichment factor
The EF has been widely used to examine the contributions from natural and anthropogenic sources in aerosols [63–65]. The EF of elements was calculated using Eq. (3):
where (Xi/Al)PM is the concentration ratio of element X to Al in PM, and (Xi/Al)Crust is the concentration ratio of element X to Al in the upper continental crust [42]. In general, EF values close to unity indicate the predominance of crustal sources; EF values ≥ 5 indicate a significant contribution from noncrustal sources; and EF values higher than 10 indicate essentially anthropogenic origins [24, 64, 66].
Reanalysis datasets
The EAC4 global reanalysis dataset provided by the Copernicus Atmosphere Monitoring Service (CAMS) was applied to provide additional constraints on the sources and transportation of pollutants in this study. EAC4 reanalysis combines model data with global observations into a globally complete and consistent dataset; the dataset used in this study was generated using CAMS information (2020) [67]. The spatial resolution of the dataset was 0.75° latitude by 0.75° longitude, with a temporal resolution of 3 h.
In addition, air mass back trajectory analysis was used to track the origins of the air parcels transported to the study sites. Back trajectories were calculated using the HYSPLIT model maintained by the US National Oceanographic and Atmospheric Administration with a spatial resolution of 1° latitude by 1° longitude in the meteorological dataset [68].