Health effects of PM2.5 constituents and source contributions in major metropolitan cities, South Korea

Ambient PM2.5 is one of the major risk factors for human health, and is not fully explained solely by mass concentration. We examined the short-term associations of cause-specific mortality (i.e., all-cause, cardiovascular, and respiratory mortality) with the 15 chemical constituents and sources of PM2.5 in four metropolitan cities of South Korea during 2014–2018. We found transition metals consistently showed significant associations with all-cause mortality, while the effects of other constituents varied across the cities and for cause of death. Carbonaceous components strongly affected the all-cause, cardiovascular, and respiratory mortality in Daejeon. Secondary inorganic aerosols, SO42− and NH4+, showed significant associations with respiratory mortality in Gwangju. We also found the sources from which species closely linked to mortality generally increased the relative mortality risks. Heavy metal markers from soil or industrial sources were significantly associated with mortality in all cities. However, several sources influenced mortality despite their marker species not being significantly associated with it. Secondary nitrate and secondary sulfate sources were linked to mortality in DJ. This could be attributed to the deep inland location, which might have facilitated formation of secondary inorganic aerosols. In addition, primary sources including mobile and coal combustion seemed to have acute impacts on respiratory mortality in Gwangju. Our findings suggest the necessity of positive matrix factorization (PMF)-based approaches for evaluating health effects of PM2.5 while considering the spatial heterogeneity in the compositions and source contributions of PM2.5.


Introduction
Ambient particulate matter has become one of the most concerning factors affecting human health in numerous metropolitans around the globe. PM 2.5 are particles with aerodynamic diameter less than or equal to 2.5 μm, and have more severe impacts on the human body compared to other larger particles. This is attributed to the unique physiochemical properties of PM 2.5 (Schins et al. 2004).
First, the extremely small size of PM 2.5 makes them both highly mobile and difficult to remove once inhaled into the human body. Unlike larger airborne particulate matter, which is removed by muco-ciliary clearance, PM 2.5 can reach deeper regions of the respiratory tract and reside in the lungs for extended periods of time. Thus, PM 2.5 accounts for considerable proportions of particles observed in human pulmonary parenchyma and is very difficult to remove (Churg and Brauer 1997). Large surface area is another physical property of PM 2.5 that explains the relationship between Responsible Editor: Constantini Samara * Seung-Muk Yi yiseung@snu.ac.kr exposure to PM 2.5 and various health outcomes. The great specific surface area facilitates PM 2.5 to act as a carrier of highly toxic compounds such as polycyclic aromatic hydrocarbons (PAHs) and transition metals (Kong et al. 2010;Pandey et al. 2013). Furthermore, the chemical heterogeneity of PM 2.5 composition also plays an important role in affecting human health because each constituent of PM 2.5 has its own mechanisms and effects in terms of toxicity. These properties of PM 2.5 mostly originate from the formation processes, especially secondary formation among various gaseous precursors in the atmosphere (Tutsak and Koçak 2019;Wang et al. 2021). Due to the complexity of formation processes among precursors, PM 2.5 is composed of various constituents including ionic compounds (e.g., sulfate, nitrate, ammonium), carbonaceous compounds (e.g., organic carbon, element carbon), trace elements (e.g., nicker, cadmium, arsenic), and other substances such as PAHs and volatile organic compounds (Ye et al. 2003;Dai et al. 2013;Amil et al. 2016). In addition, the proportions of each constituent vary considerably according to spatial and temporal conditions of the samples because PM 2.5 are highly sourcedependent secondary aerosols (Cheng et al. 2012;He et al. 2012;Li et al. 2017).
Similarly, the toxicity of PM 2.5 may differ spatiotemporally and should be accurately evaluated while considering chemical composition in the region of interest. This can aid in developing relevant strategies or policies to reduce health effects of PM 2.5 . Nevertheless, administrative plans for PM 2.5 have been mostly focused on its mass concentration. Thus, the aims of this study are to investigate the effects of shortterm exposure to PM 2.5 constituents on cause-specific mortality and find the association of source contributions with mortality in four major metropolitan cities of South Korea. Our findings can help control PM 2.5 from a public health perspective.

Study population
As of April 2020, the "Special act on the improvement of air quality in air control zone" came in effect to efficiently and systemically manage the air quality on a regional basis in South Korea. According to the law, four air control zones where air pollution was much more severe than other region were designated: Seoul metropolitan area, Middle area, Southern area, and Southeastern area. The study populations were all residents in the four metropolitan cities during 2014-2018 as shown in Fig. 1 located in each air control zone.
Seoul (hereinafter SL) represents Seoul metropolitan area. It is located in the northwestern region of and is adjacent to the Yellow Sea lying between mainland China on the west and north and the Korean peninsula on the east. Its area and population are 605 km 2 and approximately 9.6 million, respectively. Daejeon (hereinafter DJ), the largest city in the Middle area, is located in the central region of the country and had a population of 1.5 million in 2019. Gwangju (hereinafter GJ) of the Southern area is a city in the southwest region of the country. Its population and area were approximately 1.5 million and 501 km 2 , which were similar to those of DJ. Ulsan (hereinafter US) is located in the southeast region of the country and borders the sea. It had a population of 1.1 million in 2019 with total area of 1062 km 2 . US has two huge industrial complexes and is widely known as a city of heavy industry (Kim et al. 2018).

Air pollution and meteorological data
The National Institute of Environmental Research (NIER) under the Ministry of Environment, South Korea (Korea MOE), has been operating several air pollution intensive monitoring stations (APIMS) across the country for PM 2.5 speciation along with monitoring gaseous pollutants including nitrogen dioxide and sulfur dioxide. There are currently a total of ten stations in the country for analyzing the regional characteristics of air pollution including the long-range transport of air pollutants and high-concentration PM 2.5 events. Ambient air samples are taken and measured hourly for PM 2.5 mass and chemical constituents at APIMS. We obtained daily PM 2.5 speciation data during 2014-2018 at the SL, DJ, GJ, and US sites from NIER. Chemical constituents were water-soluble ions (e.g., sulfate (SO 4 2− ), nitrate (NO 3 − ), and ammonium (NH 4 + )), carbonaceous compounds (i.e., organic carbon (OC) and elemental carbon (EC)), and trace elements (e.g., titanium (Ti), vanadium (V), and nickel (Ni)). PM 2.5 mass concentration was monitored by beta-ray attenuation monitor (BAM, Met One Instrument Inc., USA) and ionic compounds were measured by using ambient ion monitor (AIM, URG Corporation, USA). Carbonaceous compounds and trace elements were determined by using semi-continuous field analyzer (Sunset Laboratory Inc., USA) and energy-dispersive X-ray fluorescence spectrometry, respectively. Further details on the measurements of PM 2.5 concentration and components can be found in previous studies (Bae et al. 2019;park et al. 2013).
Meteorological parameters are also important factors affecting human health through morbidity and mortality, and should be included in the health effect model for control (Allen and Sheridan 2014;Wang et al. 2019). Thus, we collected daily average temperature, relative humidity (RH), and barometric pressure (BP) data during 2014-2018 via the Weather Data Service (https:// data. kma. go. kr) operated by the Korea Meteorological Administration.

Health outcomes data
Various health-related parameters including mortality, hospital emergency room visits, and out-of-hospital cardiac arrest (OHCA) have been widely used in various epidemiological studies to estimate the adverse effects of exposure to PM 2.5 on health (Atkinson et al. 2014;Qiao et al. 2014;Pradeau et al. 2015). In the present study, we selected cause-specific daily mortality data as health outcomes. Daily death counts in the four cities were obtained from Micro-Data Integrated Service (https:// mdis. kostat. go. kr) operated by Statistics Korea. We included death counts for residents in each city as mortality data after classifying them into allcause (non-accidental, NA, A00-R99), cardiovascular disease (CVD, I00-I99), and respiratory disease (RD, J00-J99) mortality according to the 10th version of the International Classification of Disease (ICD-10). This study was approved by the Seoul National University Institutional Review Board (IRB no: E2101/003-001).

PMF modeling
Among various receptor models based on chemical analyses, positive matrix factorization (PMF) and chemical mass balance (CMB) have been frequently performed for identifying sources and estimating their contributions. Many previous studies showed that results based on both PMF and CMB were comparable (Begum et al. 2007;Ke et al. 2008;Teixeira et al. 2015). However, PMF is more flexible than CMB in terms of applicability since PMF does not require source profiles for model execution. It is difficult to obtain suitable source profiles established for the region of interest although it strongly affects the source apportionment results. Therefore, we selected PMF to investigate the sources and their Fig. 1 Geographical locations the study cities: a SL, b DJ, c GJ, and d US contribution to PM 2.5 in the study cities. Principles and more details of PMF can be found elsewhere (Song et al. 2006;Watson et al. 2008;Manousakas et al. 2017).
As for PMF input data, we constructed the concentrations and their corresponding uncertainty data for chemical species according to the method described in the US EPA PMF 5.0 guidebook (Norris et al. 2008). Data values below the method detection limit (MDL) were replaced with half of the MDL and 5/6 × MDL were used as their corresponding uncertainties. For measured values greater than the MDL, uncertainties were calculated by the following Eq. (1), and we used 10% as error fraction of each PM 2.5 constituent.
Whole datasets on days when the mass balance and ion balance were unsatisfactory were excluded from the input data. Furthermore, after obtaining the best solutions, we conducted error estimation using the displacement method (DISP) to acquire uncertainty estimates for each factor of the solutions (Paatero et al. 2014).

Health effects analyses
Health effects of exposure to PM 2.5 can be expressed as the strength of associations between PM 2.5 constituents and health outcomes via time-series analyses using various statistical models. Previous studies have revealed the effects of PM 2.5 on health-related outcomes including mortality and morbidity using generalized linear model (GLM), generalized additive model (GAM), and other statistical models (Pascal et al. 2014;Phung et al. 2018;Cai et al. 2019).
In the present study, we selected 15 chemical species and adopted GLM with a natural spline for analyzing the adverse effects of PM 2.5 constituent and source contributions on daily mortality. Death counts were modeled with either Poisson distribution or over-dispersed Poisson distribution according to their distribution in the study cities. We applied natural spline (ns) function of time and temperature for controlling both longterm and seasonal trends of death counts. Furthermore, we added PM 2.5 as a covariate for adjusting the effects by PM 2.5 mass concentration. Sensitivity analyses were performed to test robustness of the results by changing the degrees of freedom (df) of calendar time (2-10 per year) and temperature (2-10 per year). Based on AIC values, we selected df = 2 per year for temperature of all cities and df = 2 per year for time used for DJ, GJ, and US, while df = 6 per year was used for SL. Furthermore, we examined the delayed effect and single-lag effect from the day of exposure (lag 0) to 7 days after exposure (lag 7) because lagging effects of the exposure to PM 2.5 were steadily observed (Janssen et al. 2013;Chai et al. 2019;Li et al. 2021). The final model we constructed followed Eq. (2). (1) where Y t is the observed daily death counts on day t; X t-l represents the daily mean concentration of PM 2.5 chemical constituents at day t to l (l = 0, 1,2 …, 7); RH, BP, and DOW stand for daily average relative humidity, barometric pressure, and day of week, respectively; and β 1 -β 5 is the coefficient of X t , PM 2.5 , RH, BP, and DOW. After drawing β 1 from Eq. (2), the relative risk (RR) was calculated by the following Eq. (3).
All analyses were performed by using "mgcv" package in statistical software R version 4.0.1 (http:// www.R-proje ct. org).

Descriptive statistics of death, meteorological data, and PM 2.5 composition
During the study period, a total of 195,649 people died in SL due to non-accidental causes, while CVD and RD deaths were 42,707 and 19,911 among them, respectively. Death cases of NA/CVD/RD were 30,996/6721/3309 in DJ; 34,012/7680/4343 in GJ; and 22,003/5824/2485 in US, respectively. Tables 1 and  2 show the descriptive statistics of explanatory variables for the four cities. Due to the largest population in SL, daily death counts were much larger in SL, while those in the other cities were relatively similar to each other. The proportions of CVD and RD deaths for total death in SL and DJ were comparable to those of the whole country (21.4 and 10.8%), whereas the CVD and RD deaths in GJ (26.3 and 15.8%) and US (30.8 and 15.4%) accounted for higher proportions of total death.
The average concentrations of PM 2.5 mass at all sites exceeded both the World Health Organization guidelines (5 µg/m 3 of annual mean concentration) and U.S. Environment Protection Agency standards (15 µg/m 3 of annual mean concentration). This suggests that populations in the four cities have been exposed to potential health risks during 2014-2018. Notably, US showed approximately 25-30% lower PM 2.5 concentration than the other cities. (2) (3) RR = e β1 * IQR PM 2.5 chemical compositions showed considerable spatial variability. Secondary inorganic aerosols (SIA; i.e., SO 4 2− , NO 3 − , NH 4 + ) constructed the PM 2.5 mass at a range of 46.9 (US)-55.6% (SL), which indicates that secondary aerosol formation was an important factor in PM 2.5 . In addition, the SO 4 2− /NO 3 − ratio was much lower in SL (0.89) than in DJ (0.96), GJ (1.17), and US (1.37), which implies that dominant gas-phase pollutants were different across the cities. Carbonaceous species were reported to have the second largest proportions in PM 2.5 mass in South Korea (Son et al. 2012;Bae et al. 2019), and accounted for 18.2 (SL), 20.1 (DJ), 19.6 (GJ), and 18.4% (US), respectively. The proportions of trace elements were much higher in US (27.4%) than in SL (12.5%), DJ (14.5%), and GJ (10.0%), and these were considered to reflect the industrial characteristics of US. Correlations among PM 2.5 components (Fig. S1) varied across the cities but were obviously high between SO 4 2− and NH 4 + as well as between NO 3 − and NH 4 + due to their formation mechanism in the atmosphere ).

Source apportionment of PM 2.5
A total of 1438, 1278, 1419, and 1211 PM 2.5 speciation datasets were used for identifying PM 2.5 sources and their contributions for SL, DJ, GJ, and US, respectively. A number of factors were chosen based on the evaluation of model results including Q value, residual distribution, and the coefficient of determination, and we confirmed that source profiles in the study cities were physically meaningful and understandable. A nine-factor solution for DJ and ten-factor solutions for SL, GJ, and US were drawn, while source contributions varied among cities despite resolved factors being identical. Source profiles for SL, DJ, GJ, and US obtained from PMF modeling are displayed in Figs. 2, 3, 4, and 5, respectively, along with the monthly source contributions (Figs. S2-S5).
The first source, secondary nitrate, was characterized by high loadings of NO 3 − and NH 4 + and showed higher contributions in winter. The formation of nitrate is favored by low temperature and high relative humidity (Behera and Sharma 2010;Wang et al. 2012). However, the contributions of secondary sulfate, identified by high concentrations of SO 4 2− and NH 4 + , were much higher during summer seasons. High solar radiation and hydroxyl radical enhance the conversion of SO 2 to SO 4 2− and result in much larger contributions of secondary sulfate to PM 2.5 in summer (Behera and Sharma 2010;Lei and Wuebbles 2013). The third source was interpreted to be mobile considering high loadings of OC, EC, NO 3 − , and NH 4 + and medium loadings of Ti, Fe, Zn, and Mn. The monthly variations of mobile sources in the contributions to PM 2.5 were lower compared to those of secondary nitrate and secondary sulfate and did not show obvious seasonal trends. The fourth source was biomass burning, and was explained by high K + concentrations. Its monthly contribution peaked at the beginning or end of the year. Additionally, open field burning of agricultural residue before cultivation (Feb-Mar) and after harvest (Oct-Nov) has been widely done in East Asia for enhancing land productivity (Ryu et al. 2007;Jung et al. 2014). The fifth source, incinerator, was identified by high loadings of Cl − , NO 3 − , and NH 4 + and its contributions to PM 2.5 were larger during winter seasons. The sixth source was soil and could be distinguished by high concentrations of crustal elements such as Ca, Ti, and Fe. The contributions to PM 2.5 from soil source considerably increased during spring season (Mar-May) in all cities, and was affected by yellow sand events mainly originating from the Gobi Desert and/ or the Loess plateau located in China and Mongolia . Industry, the seventh source, was represented by high concentrations of heavy metals including Fe, Mn, Cu, and Zn. Seasonal trends in monthly contributions of the source were absent, which elucidates that PM 2.5 emission  from industry was dependent on the region. Coal combustion and oil combustion, the eighth and ninth source, respectively, were determined by their own marker species contained in the fuel. These two sources were characterized by high concentrations of As and Pb, and Ni and V, respectively (Bie et al. 2021;Wojas and Almquist 2007). Apparently, the tenth source of aged sea salt was distinguished by high Na + concentrations, which implies the depletion of Cl − due to the long retention time.
In SL, secondary nitrate, secondary sulfate, mobile, and coal combustion source constituted the majority (75.1%) of PM 2.5 mass concentration, which is consistent with past studies conducted for SL (Heo et al. 2009). Mobile was the largest contributor followed by secondary sulfate, secondary nitrate, and coal combustion, which implies that sources related to secondary formation were dominant for PM 2.5 . Specifically, sulfur-related sources, secondary sulfate, and coal combustion were thought to be affected by several huge Fig. 2 Source profiles of PM 2.5 in SL during 2014-2018; gray bar, black dot, and hollow dot indicate mass fraction, percentage contribution, and average error estimates of each species, respectively coal-fired power plants (e.g., Taean 6100 MW, Dangjin 6040 MW, Yeongheung 5080 MW) located in the southwest part of the city. Sulfate is largely formed by oxidative reactions of sulfur dioxide in the atmosphere, which is the major gaseous precursor (Bao et al. 2018;Zhang et al. 2020). Some previous studies also identified these coal power stations as attributable factors to PM 2.5 in SL by analyzing polycyclic aromatic hydrocarbons (Kang et. al. 2020) and executing dispersion modeling (Kim et. al. 2016).
Similar to SL, PM 2.5 mass concentration in DJ was mostly contributed by four factors-mobile (21.7%), secondary sulfate (19.7%), secondary nitrate (19.5%), and coal combustion (10.7%)-while the aged sea salt source was not solely resolved due to its location in deep land. In GJ, secondary sulfate (21.3%) was the largest contributor followed by mobile (20.5%) and secondary nitrate (19.5%) sources, and these major factors constituted 61.3% of PM 2.5 mass concentration.
In US, the two largest factors were mobile (24.9%) and secondary sulfate (23.9%) sources which explained the higher PM 2.5 mass compared to other cities. This is assumed to reflect the characteristics of the city where both huge heavy industrial complex and industrial ports exist. The annual emission of SO x in the city during 2014-2018 was 47,305 ton on average, which is more than double the national average annual emission (19,045 ton) during the same period (Korea MOE 2022). In the same context, the contribution of industry (8.8%) was much larger than that in other cities and was the fourth-largest factor in the city.

Association ofPM 2.5 chemical constituents and source contributions with mortality
We analyzed the effects of chemical constituents of PM 2.5 and sources on cause-specific mortality using PMF-modeled source contributions coupled with PM 2.5 speciation data. Overall, the IQR increase in the concentration of each constituent influenced all cause-specific mortality, whereas only a few constituents were significantly linked to mortality with different strengths depending on city and cause of death, which shows risk heterogeneity among various PM 2.5 constituents. In addition, the impacts of source contributions on mortality also varied among the cities (Fig. 6). The associations of cause-specific mortality Fig. 4 Source profiles of PM 2.5 in GJ during 2014-2018; gray bar, black dot, and hollow dot indicate the mass fraction, percentage contribution, and average error estimates of each species, respectively with both PM 2.5 constituents and sources are summarized in Tables S1 and S2, respectively, and Table 3 shows the highest RR of mortality significantly associated with chemical constituents.
In SL, the chemical constituents showing significant associations were as follows: K + (RR 1.016, 95% CI 1.002-1.031), EC (1.015, 1.000-1.031), Ti (1.011, 1.003-1.019), and Fe (1.010, 1.000-1.021) with NA mortality; Ti (1.016, 1.001-1.032), Fe (1.023, 1.003-1.044), and As (1.030, 1.003-1.058) with CVD mortality; and Pb (1.071, 1.013-1.128) with RD mortality. Ti and Fe mostly originate from soil dust and are known to generate reactive oxygen species (ROS) which leads to cell damage (Moreno et al. 2019). As and Pb, markers of coal combustion, are toxic even at low levels and are carcinogenic to humans according to the International Agency for Research on Cancer. Significant associations of mortality with sources were mostly found for CVD mortality: incinerator (1.032, 1.007-1.057), The chemical constituents with their RR and sources of PM 2.5 significantly associated with cause-specific mortality in SL, DJ, GJ, and US  (Mills et al. 2009). In addition, the respiratory system is subject to damage by those metals triggering oxidative stress in lung alveoli (Donaldson et al. 1996), which can lead to impact on the cardiovascular system through circulation of ROS and inflammatory cytokines. Peculiarly, the mobile source showed no significant association only in SL despite being significantly associated with mortality in the other cities. This might be due to the intensive policies implemented in the Seoul Capital Area for improving air quality since the late 2000s. These policies included various programs such as the supply of low-pollution motor vehicles and promoting early scrapping of rundown vehicles for reducing air pollutants from them.
In DJ, we found that PM 2.5 constituents increased the RR of all types of mortality, but most of the significant associations were related to CVD mortality.  (Huang et al. 2012;Schneider et al. 2010;Wu et al. 2013) which can lead to cardiovascular health outcomes (Bell et al. 2009;Ito et al. 2011).
The associations between PM 2.5 source and CVD mortality were also consistent with the PM 2.5 constituentmortality relationship. Sources whose markers showed significant associations with CVD mortality (e.g., mobile, industry) also increased the RR of CVD mortality in the city. However, secondary nitrate and secondary sulfate were closely linked to RD mortality in the city although their marker species (e.g., NO 3 − , SO 4 2− , and NH 4 + ) showed no significant associations. This result may have been affected by their deep inland location. The percentage of calm conditions (less than 0.5 m/s) in DJ during the study period was 12.5%, which is much larger than the values in SL (3.8%) and US (6.2%). These meteorological conditions seemed to facilitate the formation of SIA in DJ.
Significant associations in GJ were mostly found for RD mortality, which was uniquely observed among the four cities. Chemical species significantly associated with RD mortality were ionic species, SO 4 2− (1.085,  (Ueda et al. 2016). Nevertheless, some plausible theories have been suggested to explain the positive association of SO 4 2− with adverse health effects. First, particle acidity by SO 4 2− may change the pulmonary toxicity of other PM 2.5 constituents or physical properties from their own toxicity (Dreher 2000). Second, catalyzation of metals into more bioavailable forms is another possible explanation for the strong association of SO 4 2− with health effects (Lay et al. 2001). Regarding source contributions, sources of the species with significant association (e.g., coal combustion and sources mobile) were revealed to increase the RD mortality in GJ. In addition, primary sources including mobile and coal combustion had acute impacts on respiratory mortality in the city although their representative markers were not significantly associated with mortality. The percentage of calm conditions in GJ during 2014-2018 was as high as that in DJ, which might have hindered the dispersion of air pollutants and resulted in acute adverse effects on the respiratory system.

Conclusion
In the present study, we investigated the associations of cause-specific mortality with the chemical constituents and sources of PM 2.5 in four metropolitan cities in South Korea. The significance and strength of the associations varied across the study cities, which implies that adverse health effects of short-term exposure to PM 2.5 were spatially heterogeneous. In SL, significant associations were found for both NA and CVD mortality mostly related to transition metals and relevant sources including soil and coal combustion. In DJ, most of the significant associations were found for CVD mortality with carbonaceous and heavy metallic components of PM 2.5 . Sources closely linked to CVD mortality in DJ were mobile, incinerator, and industry, which is generally consistent with the constituent-mortality relationship. However, many significant associations in GJ were found for RD mortality with PM 2.5 constituents related to secondary formation (e.g., SO 4 2− , NH 4 + ) and heavy metals. Lastly, we found that significant associations in US were closely related to SO 4 2− , OC, and Mn in line with the characteristic large heavy and chemical industry complexes located in the city.
From our results, we concluded that the risks of mortality increase by short-term exposure to PM 2.5 were heterogeneous depending on regions. In addition, the PM 2.5 constituent-mortality relationship was generally in line with the association of source mortality. Therefore, identifying PM 2.5 constituents and source contributions significantly affecting health outcomes in the region of interest should take priority over designing policies for controlling PM 2.5 efficiently. In addition, policies currently focused on PM 2.5 mass concentration need to be shifted to health risk-based ones for effectively protecting public health against air pollution.
However, the present study also has some limitations which can be improved in further studies. Although ambient PM 2.5 compositional data was employed as a proxy of personal exposure to PM 2.5 , we acquired the data from one site for each city due to the insufficient number of APIMS. As the area covered by an APIMS is broad, the exactness of estimating the level of PM 2.5 exposure decreased, which may result in inevitable errors while evaluating the health effects. Accordingly, more sites for PM 2.5 speciation need to be established considering relevant indexes such as population and administrative area for a better assessment of PM 2.5 exposure.

Supplementary Information
The online version contains supplementary material available at https:// doi. org/ 10. 1007/ s11356-022-21592-1. Data availability All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations
Ethics approval and consent to participate Not applicable.

Competing interests
The authors declare no competing interests.