Health risks and sources of PM_2.5-bound heavy metals in Southwest China: evidence from a 9-year monitoring program

Abstract

Numerous studies have demonstrated that PM_2.5-bound heavy metals pose non-negligible inhalation carcinogenic risks to the population. As one of the largest megacities in southwest China, Chengdu's industrial activity combined with its geographic features may present ongoing health challenges. By monitoring heavy metals and inorganic ion in atmospheric PM_2.5 in Chengdu City for nine years, the inhalation health risks of 12 metal elements were analyzed by applying a year-weighted health risk assessment model. Afterwards, the sources of heavy metals and ions were resolved by using a positive matrix factorization (PMF) model. Further, 34 socio-economic development indicators in Chengdu were collected and analyzed by Spearman correlation and partial correlation to support the source resolution results. The health risks from different emission sources were further quantified. The median hazard index HI (interquartile range) for all metals measured was 0.26 (0.19, 0.31), indicating a negligible non-carcinogenic risk. The order of male’s carcinogenic risk was As (1.90 × 10^–6 (1.40 × 10^–6, 2.43 × 10^–6)) > Cr (4.43 × 10^–7 (3.21 × 10^–7, 6.21 × 10^–7)) > Cd (1.41 × 10^–7 (9.68 × 10^–8, 2.09 × 10^–7)) > Pb (1.17 × 10^–7 (8.59 × 10^–8, 1.53 × 10^–7)) > Ni (1.81 × 10^–8 (1.27 × 10^–8, 2.46 × 10^–8)) > Be (5.01 × 10^–9 (5.01 × 10^–9, 5.10 × 10^–9)), and similar patterns for female and child. PMF identified 5 possible sources of heavy metals, including the chemical industry, motor vehicles, soil dust, coal combustion, and the nonferrous metals industry. These sources were systematically supported and explained through trend and correlation analysis with Chengdu's socio-economic indicators. The non-ferrous metal industry and the chemical industry had the highest population cancer risk levels of 4.71 × 10^–6 and 2.24 × 10^–6, respectively. As (71.8%) and Cr (17.3%) contributed the most to the total carcinogenic risk and are the metal species requiring more stringent regulation. Heavy metal pollution from Chengdu's fast-growing non-ferrous metal industry, chemical batteries, and new energy vehicle industry requires more concern. Moreover, further policies related to cleaner production, energy conservation, and recycling are needed to address pollution from these sectors.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-24596-x.

Introduction

Heavy metals are one of the most concerned components of PM_2.5 and are the most studied chemicals in the field of environmental science [1–3]. Although the mass concentration of most metal species in PM_2.5 can only be measured at the trace level [4, 5], the heavy metal components have a significant impact on the toxicity of PM_2.5 due to cumulative toxicity, non-degradability, and lifetime exposure [6]. For instance, PM_2.5-bound arsenic (As) is potentially teratogenic to humans [7], and long-term exposure to As may cause an increased incidence of skin tumors, hepatic hemangiosarcoma, peripheral neuropathy, and other diseases [8]. Chromium (Cr) exists in the atmosphere in two valence forms, Cr (VI) and Cr (III), with the former being significantly more hazardous than the latter. Long-term chronic exposure to Cr (VI) leads to bronchial inflammation and chemical pneumonitis, and may even lead to lung cancer [9]. Atmospheric cadmium (Cd) causes chronic damage mainly to the kidneys and bones [10]. Chronic inhalation of lead (Pb) and mercury (Hg) can adversely affect the neurological development of the fetus [11, 12]. Inhalation of nickel (Ni), antimony (Sb), manganese (Mn), and beryllium (Be) into the lungs induces the formation of reactive oxygen species (ROS) [13], which can cause oxidative DNA damage and respiratory inflammation [14].

To address the potential threat of heavy metals to public health and to clean up the atmospheric environment, numerous regional-scale health risk assessment and source resolution studies of PM_2.5 binding heavy metals have been conducted worldwide. A recent study from Bangladesh [15] utilized the United States Environmental Protection Agency (USEPA) health risk assessment model to evaluate the carcinogenic and non-carcinogenic health risks of 6 PM_2.5-bound metals in the mega-city of Dhaka, and allocated the pollution risks from different sources with correlation matrices and Positive Matrix Factorization (PMF) modeling to identify the metal species and industrial sources that were prioritized for intervention in the city. In addition, risk assessment studies from India [16] and Japan [6] also made some contributions to local policy development on ambient air heavy metal pollution. As the world's largest developing nation and comprehensive economic entity, China confronts substantial atmospheric environmental challenges arising from accelerated industrialization. Multiregional health risk assessments [3, 17, 18] have uncovered pervasive airborne carcinogenic threats across the country, particularly associated with PM_2.5-bound As and Cr (VI). A comprehensive national-scale analysis of atmospheric metals (2000–2022) [19] reveals that geometric mean carcinogenic risks from PM_2.5-bound As and Cr (VI) in most provinces persistently exceed the negligible threshold (10^–6), with several regions attaining risk magnitudes even approaching the unacceptable threshold (10^–4). Consequently, sustained advancement in regional atmospheric heavy metal monitoring, systematic health risk quantification, and precision source apportionment remains imperative. These coordinated efforts provide critical scientific foundations and regulatory frameworks for implementing targeted pollution mitigation strategies aligned with China's dual carbon goals (2030/2060).

As the largest mega-city in southwestern China, Chengdu has experienced rapid population growth driven by its robust economic development, emerging as the nation's fourth most populous city by 2023 [20]. The city's fast growth in steel manufacturing (a major source of Cr, Mn, and Ni emissions), coal-fired power generation (contributing Hg and As), and automotive production (linked to Cr and selenium (Se)) [21], coupled with its status as China's largest vehicle ownership hub, contributes significantly to atmospheric metal pollution through industrial emissions, vehicular exhaust, tire wear, and road dust [22]. In addition to these anthropogenic factors, Chengdu's basin topography in the Sichuan Plain creates unfavorable meteorological conditions characterized by frequent temperature inversions, stagnant winds, and high humidity, all of which impede atmospheric dispersion and promote metal deposition [23]. This combination of intensive industrial activity and geographic constraints poses persistent health risks to its substantial population base. Current research [22–24] on PM_2.5-bound heavy metals in Chengdu has primarily focused on short-term (< 2 years) monitoring data for risk assessment and source apportionment. However, brief observation periods are susceptible to observation bias from extreme events, including geological activities, meteorological anomalies, and sudden policy changes. For instance, Chengdu's unprecedented 2022 extreme heatwave and drought significantly reduced atmospheric moisture content and wind speeds [25], subsequently influencing PM_2.5 concentrations [26, 27]. Studies confined to such anomalous year risk misattributing transient fluctuations to long-term trends. In contrast, long-term monitoring (> 5 years) enables more robust risk evaluation by capturing temporal trends in pollution patterns and generating stable source identification results.

In this study, we conducted monitoring of atmospheric PM_2.5-bound heavy metals in Chengdu City for 9 years, aiming to characterize the pollution levels and the health risks of metal species to the population, and resolve robust source profiles of metal contamination profiles via the PMF modeling and correlation analysis. Our work provides comprehensive long-term evidence support for the pollution of atmospheric metal species in Chengdu and delivers policy-relevant insights on priority control targets and mitigation strategies about metal pollution.

Materials and Methods

Sampling information

Chengdu City is located in the western part of the Sichuan Basin and the eastern edge of the Qinghai-Tibetan Plateau, with a longitude of 102°54'—104°53' East and a latitude of 30°05'—31°26' North. Three sites were established across Chengdu, with mean concentrations from these sites representing city-wide values. As shown in Fig. 1, sampling site S1 (104°3' East, 30°40' North) was located next to an elementary school in Qingyang District, representing a residential area; S2 (103°55' East, 30°56' North) was located inside a private middle school in Pengzhou, characterizing an educational environment; S3 (104°5' East, 30°33' North) was located in an industrial park in Gaoxin District, capturing industrial emissions. Geographically, the distances of the three sampling sites (S1, S3, and S2) from the center of Chengdu city varied in a gradient, which systematically covered the atmospheric environmental characteristics across different city circles. Chengdu's predominant wind patterns feature north and northeast directions annually [28], while our sampling alignment follows a precisely ordered north–south transect (S2 northernmost, S1 central, S3 southernmost). This configuration minimizes potential industrial contamination from S3 (southern Gaoxin District) to upwind sites (S1 and S2). Furthermore, these locations capture representative population exposure gradients: S3 (> 1 million residents) represents high-density districts, while Qingyang (S1) and Pengzhou (S2) (500,000–1 million residents) typify medium-density counties according to China's 7th National Population Census. Collectively, this sampling design ensures representative of Chengdu's urban atmospheric conditions.

Fig. 1.

Spatial geographic location of sampling sites. Sampling sites 1–3 are denoted by S1-S3

PM_2.5 samples were collected from January 2015 to December 2023, averaging seven mid-month days (10th to 16th) each month. Additionally, on days experiencing heavy pollution (PM_2.5 > 200 μg/m³), the sampling frequency within the corresponding month was increased accordingly to ensure sample representativeness. In the event of extreme weather, such as rain, snow, wind, hail, etc., that prevents sampling, the sampling date was postponed to later dates to ensure a 7-day sampling period per month. As documented in Appendix A (Figure S1), sampling covered across all months during the study period (2015–2023), eliminating seasonal bias. Extreme weather events < 6% frequency at three sites during 10th to 16th, confirming temporally representative sampling without significant data gaps across monitored timescales. Sampling took place from 9 a.m. to 8:30 a.m. the next day, for 23.5 h per day. PM_2.5 was sampled by the TH-150C Intelligent Medium-Flow Sampler (Tianhong, Wuhan, China), which was placed 10–15 m above the ground, and set to a flow rate of 100 L/min for particle-size fractionation. A total of 2,804 valid samples were obtained, and the collected samples were stored in a refrigerator at −20 °C for subsequent analysis.

Metal and inorganic ion analysis

Heavy metal measurement

Before and after sampling, the polypropylene filter membranes in the PM_2.5 sampler were equilibrated at the temperature (20 °C ± 1 °C) and humidity (50% ± 1%) for 48 h according to the Chinese national standard (Technical Specifications for gravimetric measurement methods for PM_2.5 in ambient air (HJ 656–2013)). After sampling was completed, the membranes were cut 1/4 with ceramic scissors and placed in a digestion tank, followed by the addition of 5 ml of nitric acid (pH = 5.6) and 0.05 ml of 40% hydrofluoric acid (pH = 5.3). The mixture was then refluxed at 220 °C and fully dissolved. The solution was expanded to 10 ml by continuing to add nitric acid at pH = 5.4. The sample digestion procedures adhered to HJ 657–2013: Ambient air and stationary source emission—Determination of metals in ambient particulate matter—Inductively coupled plasma/mass spectrometry. After completion of the pre-treatment, the mass concentrations of 12 metal elements (aluminum (Al), arsenic (As), beryllium (Be), cadmium (Cd), chromium (Cr), mercury (Hg), manganese (Mn), nickel (Ni), lead (Pb), antimony (Sb), selenium (Se), and thallium (Tl)) were determined using inductively coupled plasma-mass spectrometry (ICP-MS) with the measuring instrument PE NexION 5000G (Perkin Elmer, USA).

Inorganic ion measurement

The ultrasonic method was applied to extract water-soluble inorganic ions from PM_2.5 filter membranes, which can usually extract more than 98% of sulfate, nitrate, and ammonium [29]. First, the filter membrane was sealed and stored in 10 mL of ultrapure water, and the extraction was ultrasonicated for 20 min each time, with the extraction repeated three times. Afterwards, the extraction solution was filtered through 0.22 μm polytetrafluoroethylene and analyzed for the mass concentration of water-soluble inorganic ions (Cl⁻, NO₃⁻, SO₄²⁻, NH₄⁺) by ion chromatography (instrument ICS-600, Thermo Fisher, USA). The ion chromatographic conditions were set separately for anions (Cl⁻, NO₃⁻, SO₄²⁻) and cations (NH₄⁺). For anions, the IonPac AS11-HC column (4 mm × 250 mm) was used, with the AG11-HC as the guard column. The eluent was 40.0 mmol/L potassium hydroxide solution, at a flow rate of 1.0 mL/min. In addition, the injection volume was 25 μL, and the column temperature was maintained at 30 ℃. For cations, the IonPac CS12A column (4 mm × 250 mm) was employed. The eluent was 20 mmol/L methanesulfonic acid (MSA), delivered at a flow rate of 1.0 mL/min. The injection volume was 25 μL, and the column temperature was maintained at 35 ℃.

Quality Assurance/Quality Control (QA/QC)

For determination, each group of samples must contain at least one standard sample, one duplicate sample, and one blank sample. Operate and calculate at 90%−110% recovery and ensure that the relative standard deviation (RSD) is less than 10%. The test value of the blank sample must be less than or equal to twice the detection Limit. In addition, the cascade sampler was cleaned regularly, and the integrity of the filter membrane was checked before each sample. The samples obtained on one day of each sampling month were randomly selected for spiked recovery tests. The recoveries of the 12 heavy metals and 4 inorganic ions ranged from 94.30% to 108.8%, with all average RSDs < 6.23% and standard curve R²s ≥ 0.999. Detailed information on the limits of detection (LODs), RSDs, and spiked recoveries for all metals and ions was recorded in Table S1 of Appendix A.

Year-weighted health risk assessment

Year-weighted concentration

Due to the long-time span (9 years) of species concentration data in this study, time-weighted vectors were applied to balance the concentrations from different years [4, 30]. In general, concentration data from later years are more representative of the current situation.

1

2

3

4

is the information entropy, which indicates the information capacity of the whole system; is the year ordering, is the total number of years (); and is the time-weighted vector corresponding to the th year. is a time scale that indicates how much importance the researcher attaches to the time series. takes on a range of 0 to 1, with values closer to 0 indicating more attention to older data and closer to 1 indicating a focus on newer data. All concentration data involved in the risk assessment were weighted and adjusted with .

Sensitivity Analysis on Year-weighted algorithm

Concentration distribution comparisons were applied to determine the optimal λ-value for the year-weighting algorithm and to validate the representativeness and sensitivity of the weighted concentrations to the most recent conditions. Briefly, we first computed the time-weighted vector w_y for λ at different regular values (0.1, 0.2, 0.3, 0.4, and 0.5). Second, the 2023 concentration data for the 12 measured metals served as the validation set, while 2015–2022 data constituted the training set. Third, arithmetic averaging and year weighting (based on w_y) were employed to process the training set, respectively. Fourth, the arithmetic mean and year-weighted (λ = 0.1 to λ = 0.5) results were compared with the concentration data from the validation set, and frequency distribution curves and Quantile–Quantile (QQ) plots were plotted. The Kolmogorov–Smirnov test and the Anderson Darling test were employed to test the consistency of the arithmetic mean concentrations and the five year-weighted concentrations with the latest concentration distribution. In addition, the T test and Wilcox test were applied to assess the difference between the mean and median of the weighted and actual concentrations. P-values were calculated for all tests and compared at different levels of λ. Concentrations calculated by the method with the largest P-value were closest to the most recent concentration, had the highest representative and minimum sensitivity, and corresponded to the best λ. The time-weighted vector values corresponding to different numbers of years were recorded in Table S2 of Appendix A.

Exposure and Risk

Based on USEPA's Human Health Risk Assessment Framework [31, 32], exposure levels, non-carcinogenic health risks, and carcinogenic risks for inhaled PM_2.5-bound heavy metals were estimated. The equations utilized for the assessment are listed below:

5

6

7

8

9

Subscript refers to metal species that pose a non-carcinogenic health risk, including As, Be, Cd, Cr, Hg, Mn, Ni, Sb, and Se; subscript is the number of non-carcinogenic metal species, i.e. . refers to metals posing carcinogenic risks, including As, Be, Cd, Cr, Pb, and Ni; and refers to the number of carcinogenic metal species, i.e., . Non-carcinogenic and carcinogenic metal species were determined based on the latest toxicological profile from USEPA's Integrated Risk Information System (IRIS), the Agency for Toxic Substances and Disease Registry (ATSDR), and the Ministry of Environmental Protection of the People's Republic of China (MEPPRC). is the exposure concentration of the given metal species; is the year-weighted concentration with the optimal λ of PM_2.5-bound heavy metals. For Cr, ICP-MS determines the total Cr concentration in PM_2.5, while the species that poses a threat to human health is Cr(VI). Therefore, a multiplication by a factor of 1/7 was employed to adjust the Cr concentration [33–35]. is the exposure time in 8 h/d; is the exposure frequency (250 d/year applied); is the exposure duration (child: 6 years; adult: 24 years) [18]; and is the average exposure time. For non-carcinogenic effects, ; for carcinogenic effects, [19]. per capita was extracted from the Statistical Bulletin of Chengdu Municipal Health Commission 2024 (6224d4fa8b1646eabc7e80f9be27bb45.pdf), i.e., 79.25 years for males and 84.99 years for females, with an average of 82.00 years. is the hazard quotient indicating the intensity of non-carcinogenic health risk due to one single species; is the reference concentration in mg/m³. is the hazard index indicating the sum of non-carcinogenic risks for all species. is the carcinogenic risk; is the inhalation unit risk indicating the intensity of human carcinogenicity per unit of concentration of the pollutant in units of (mg/m³)⁻¹. is the Sum of carcinogenic risks for all species. Table S3 records the and for all species.

Source Apportionment

Source Analysis

The sources of atmospheric PM_2.5 were resolved using the Positive Matrix Factorization (PMF) model, a receptor model developed by USEPA based on the least squares algorithm. PMF is widely applied to the source analysis of ambient mixed pollutants [33, 36], as it requires no prior measurement of source profiles and effectively resolves source contributions [37]. The specific algorithms and principles of PMF were documented by Niu et al. [38]. Given PMF's requirement for normally distributed input data [39], we applied log₁₀(x + 1) transformation to the 9-year concentration dataset to ensure positivity and improve normality. Table S4 (Appendix A) presents probability density curves and Shapiro–Wilk normality test results for all species pre- and post-transformation. Briefly, the transformed concentration distributions are more uniform in scale and the normality tests have larger P-values, indicating closer convergence to the normal distribution. Afterwards, the transformed data of (including 12 metals and 4 ions) were imported and modeled in the PMF using the EPA PMF 5.0 software. The signal-to-noise (S/N) ratios for almost all species in the PMF exceeded 1, indicating strong robustness of the samples. We conducted scenario analyses ranging from 2 to 8 factors, calculated Q values for each scenario and ran the 3 error analyses built into the EPA PMF 5.0 (Displacement (DISP), Bootstrap (BS), and Bootstrap-Displacement (BS-DISP)) to determine the optimal number of factors. PMF model setup information (missing data, detection limit, seed, block size, operating time, etc.) for all sampling data was documented in Table S5 of Appendix A. The results of all scenarios, including details of the receptor factor profiles, model diagnostics, and error analyses, were encapsulated in Appendix B. Twenty model iterations per scenario achieved solution convergence. Finally, the factor-specie concentration assignment at the scenario of optimal factor numbers reverted from a logarithmic scale to a linear scale.

Correlation and partial correlation analysis

As a typical receptor model, PMF is prone to cause the source resolution results disconnected from actual regional dynamics (regional industrial development, pollution emission, cleaner production, etc.) due to its inability to form a linkage with the sector-specific source profile data [40]. To address this disconnect, we develop the correlation and partial correlation analysis between the species concentration data in PM_2.5 and the socio-economic development of Chengdu City. This integrated approach validates the authenticity of PMF-derived source apportionment through empirical alignment with real-world development patterns.

Specifically, socio-economic development indicators covering our sampling years (2015–2023), including industrial production, energy consumption, transportation capacity, pollution and environmental protection, are first retrieved from the Chengdu Statistical Yearbook (https://cdstats.chengdu.gov.cn/cdstjj/c178732/list.shtml). Since some categories of indicators may not include the source-related industrial sectors, for example, the production of non-ferrous metals is not recorded in the industrial production dataset, we additionally collected documentation on industrial output value and sector-energy consumption. A total of 34 indicators in 6 categories were collected that objectively reflect Chengdu's social development, and the annual trends and interpretations of these indicators were recorded in Figure S2-S5 of Appendix A. Data on species concentrations in PM_2.5 were matched with socio-economic development data by year. Afterwards, Spearman correlation coefficients between each species and each indicator were calculated, matrixed and then plotted into correlation heat maps. Given that various social sectors may directly affect the PM_2.5 species profile through industrial dust emissions, annual dust emissions were controlled as a key covariate for partial correlation analysis. Of note, the period of the COVID-19 epidemic in China (2020–2022) significantly influenced the social development [41], so the epidemic time factor was also employed as one of the covariates.

Health risk based on source apportionment

After identifying all the sources by PMF, the non-carcinogenic and carcinogenic risks posed by different sources are quantified based on the proportion of source apportionments. The health risk calculation equations corresponding to the different sources [18] are as follows.

10

11

is the hazard quotient of source , where is any of the five sources identified. is the contributed concentration of the source to the th noncarcinogenic metal. is the carcinogenic risk of the source ; and is the concentration contribution of the identified source to the th carcinogenic metal. The contributed concentrations of PM_2.5-bound metals from all PMF-identified sources were recorded in Table S6 of Appendix A.

Software

All simulations and calculations in this study were performed in the R-4.1.3 environment (Ross Ihaka, New Zealand). The source analysis was based on EPA PMF 5.0 software. ArcGIS 10.8 (ESRI, USA) was used to map the data, and data visualization was performed using the "ggplot2", "pheatmap", and "circlize" packages. All R code for this study is attached in Appendix C.

Results and discussion

Level and distribution of specie concentrations

Table 1 illustrates the statistical characteristics of the concentrations of all species measured throughout the sampling period from January 2015 to December 2023. The skewness of the concentrations of all species was greater than 0, indicating a high degree of positive skewness of the data, and the mean and standard deviation were not appropriate to describe the distributional characteristics of the concentration levels. In addition, the Shapiro Wilk test showed that all concentrations did not follow normal distributions (P < 0.01, Table S4), so the median (M) and interquartile range (Q1, Q3) were adopted to characterize the concentration level. The concentration of PM_2.5 fluctuated between 2 and 560 μg/m³, with the median level of 72 (46, 108.5) μg/m³. The median concentration levels of the four ions (SO₄²⁻, NO₃⁻, Cl⁻, NH₄⁺) were 7.01 (4.37, 11.35) μg/m³, 6.68 (3.33, 13.95) μg/m³, 1.01 (0.48, 2.1) μg/m³, and 4.72 (2.34, 9.29) μg/m³. The concentration ranking of the 12 heavy metals is Al (141.38 (86.86, 227.89) ng/m³) > Pb (27.01 (14.91, 48.96) ng/m³) > Mn (25.01 (16.89, 35.60) ng/m³) > As (7.5 (4.71, 12.69) ng/m³) > Se (2.58 (1.00, 4.05) ng/m³) > Cr (2.58 (1.00, 4.01) ng/m³) > Sb (2.11 (1.28, 3.55) ng/m³) > Cd (1.26 (0.64, 2.47) ng/m³) > Ni (0.60 (0.60, 1.89) ng/m³) > Tl (0.47 (0.28, 0.80) ng/m³) > Hg (0.10 (0.10, 0.28) ng/m³) > Be (0.03 (0.03, 0.03) ng/m³).

Table 1.

Statistical description of species concentration data in Chengdu City and comparison with other cities

City	Statistic	PM2.5	SO42−	NO3−	Cl−	NH4+	Sb	Al	As	Be	Cd	Cr	Hg	Pb	Mn	Ni	Se	Tl	Reference
Chengdu, China (2015–2023)	min	2.00	0.13	0.14	0.01	0.03	0.10	3.00	0.14	0.01	0.03	0.28	0.02	1.09	0.60	0.10	0.14	0.01	This study
	max	560	69.5	70.5	57.4	49.9	81.0	6600	150	1.77	55.5	50.1	10.0	928	310	53.6	36.5	16.0
	mean	85.1	9.00	10.4	1.69	6.67	3.01	195	10.7	0.07	2.25	3.29	0.32	40.7	29.2	1.78	3.14	0.64
	SD	57.6	6.95	10.2	2.56	5.91	3.60	274	10.9	0.10	3.62	3.69	0.59	51.7	20.5	3.21	2.96	0.68
	CV	0.68	0.77	0.99	1.51	0.89	1.20	1.41	1.02	1.41	1.61	1.12	1.86	1.27	0.70	1.80	0.94	1.08
	median	72.0	7.01	6.68	1.01	4.72	2.11	141	7.50	0.03	1.26	2.58	0.10	27.0	25.0	0.60	2.58	0.47
	Q1	46.0	4.37	3.33	0.48	2.34	1.28	86.9	4.71	0.03	0.64	1.00	0.10	14.9	16.9	0.60	1.00	0.28
	Q3	109	11.4	14.0	2.10	9.29	3.55	228	12.7	0.03	2.47	4.01	0.28	49.0	35.6	1.89	4.05	0.80
	skewness	2.06	2.05	2.00	8.47	1.55	8.40	13.8	4.23	3.67	6.41	4.08	8.04	6.31	3.59	6.99	4.44	7.43
	kurtosis	8.06	9.96	8.25	135	5.84	135	280	33.3	39.0	64.7	30.0	91.2	72.9	30.5	72.0	34.9	119
	log_mean	69.2	6.95	6.62	0.75	4.36	2.17	140	7.72	0.04	1.27	2.20	0.19	27.1	24.0	1.09	2.34	0.45
Chengdu, China (2017–2018)	mean	51.4							13.5		10.0	8.3	1.70	40.0	48.0	2.00	2.70		[24]
Beijing, China (2021)	mean	37.2					0.35	211	0.77		0.11	13.1		3.61	9.34	8.93			[42]
Shanghai, China (2021)	mean	20.9					0.30	143	0.91		0.09	11.8		3.84	9.69	13.1
Shenzhen, China (2021)	mean	16.0					0.14	196	0.71		0.07	13.2		2.21	4.87	8.01
Samut Prakan, Thailand (2011–2021)	mean	28.5							1.76		0.43	2.62		12.1		3.33			[1]
Seoul, South Korea (2020–2022)	mean	22.3							3.95		2.28	2.85		15.2	12.6	1.32			[43]
Los Angeles, USA (2018–2019)	mean	11.8							0.08		0.02	1.82		1.10		1.00			[44]
Košetice, Czech Republic (2009–2010)	mean								0.70					4.50	1.80	0.20	0.40		[45]

Q1: first quartile or lower quartile. Q3: third quartile or upper quartile. Concentrations of PM_2.5 and inorganic ions are in μg/m³ and concentrations of mental species are in ng/m³. Bold indicates species exceeding China National Ambient Air Quality Standards (NAAQS)

SD Standard deviation, CV Coefficient of variation

By comparing with other studies (Table 1), the average concentration of PM_2.5 in Chengdu was slightly higher compared to that of 2017–2018 [24], while it was notable that the concentration of PM_2.5-bound metals (As, Cd, Cr, Hg, Mn, and Ni) decreased significantly, which emphasized the efficacy of the prevention and control of heavy metals in Chengdu in recent years. In the cross-sectional comparison with other megacities in China (Beijing, Shanghai, Shenzhen) [42], Chengdu had higher mass concentrations of As, Cd, Pb, and Mn, and lower concentrations of Cr. Furthermore, comparative analyses revealed that Chengdu demonstrated significantly higher concentrations of PM_2.5 and PM_2.5-bound heavy metals than both developed nations/regions (USA, Korea, Europe) [43–45] and tourism-dependent economies (Thailand) [1]. These findings highlight the urgency to strengthen atmospheric purification measures and enhance heavy metal pollution mitigation strategies to align with international environmental quality standards.

Annual trend of specie concentration

On the basis of horizontal species comparisons, longitudinal comparisons of annual concentrations were further performed (Figure S6). In general, the concentrations of almost all species, including inorganic ions, low-level metals (< 1 ng/m³), medium-level metals (1−10 ng/m³), and high-level metals (> 10 ng/m³), showed a downward trend over year, with the largest decline in 2015 to 2016. This trend is closely associated with environmental protection policies and supervision measures implemented in Chengdu. On the one hand, to implement China’s newly enacted Environmental Protection Law (implemented in 2015), the municipal government promptly conducted a major environmental protection inspection (January 2015−January 2016), integrated environmental administrative law enforcement with criminal justice, and significantly enhanced regulation for pollution-intensive industries. On the other hand, the local environmental protection department had implemented a five-year initiative to strengthen environmental supervision and successfully established a regional grid-based supervision system.

It is worth noting that a few species (NO₃⁻, Cr, and Al) concentrations showed anomalous trends. NO₃⁻ concentrations increased from 3.18 μg/m³ in 2015 to 7.88 μg/m³ in 2016, representing a nearly two-fold increase, and then fluctuated around 8 μg/m³ through 2023 (Figure S6(b)). Isotope source-tracing studies [46, 47] revealed that NO₃⁻ was primarily derived from fossil fuel combustion and vehicle exhaust emissions. Therefore, this anomalous rise is likely attributable to the rapid growth of the automobile industry (Figure S2(c) and S2(d)) and expanding fossil energy consumption (e.g., gasoline, kerosene, diesel, fuel oil, and liquefied petroleum gas) (Figure S3(c)-S3(g)) in Chengdu during 2015–2016. Cr concentrations exhibited significant year-to-year fluctuations between 2016 and 2018: decreasing from 3.01 ng/m³ to 1.76 ng/m³, rising sharply to 4.55 ng/m³, and then fluctuating slightly around 2.5 ng/m³ (Figure S6(k)). Industrially, Cr is primarily used in electroplating processes and the manufacturing of special steel [21, 48]. According to our statistics (Figure S2(a) and S2(d)), steel production and ferrous metal smelting exhibited a similar pattern in industrial production and output value time series−an initial decrease, followed by an increase, and eventual stabilization. This suggests a close relationship between Cr concentration fluctuations in Chengdu and steel production activities. The concentration of Al has consistently fluctuated around 150 ng/m³ during 2015–2023 (Figure S6(n)), largely because in air pollutant source analysis, Al is typically associated with natural and dust sources [49] and exhibits low sensitivity to industrial production and energy consumption.

Year-weighted health risk by heavy metals

Non-carcinogenic risk

Figure 2 presents the data distribution, proportion, spatial characteristics, and species-specific risks of the 9 noncarcinogenic metals. For male, female, and child, the total noncancer risk HI was below 1, with a median level of 0.26 (0.19, 0.31), indicating that the noncancer risk of all metals was at a negligible level (Fig. 2(a)). This suggests that the heavy metal elements in PM_2.5 may not cause significant cumulative non-carcinogenic damage in humans, as confirmed by recent studies [1, 50]. Further analysis of the risk share of different metal species revealed that Mn and As contributed the most to the non-carcinogenic risk with 47.74% and 38.34%, respectively (Fig. 2(b)). The risk ranking of all non-carcinogenic metals was Mn (1.17 × 10^–1 (9.10 × 10^–2, 1.47 × 10^–1)) > As (9.70 × 10^–2 (7.15 × 10^–2, 1.25 × 10^–1)) > Cd (2.59 × 10^–2 (1.78 × 10^–2, 3.84 × 10^–2)) > Cr (2.71 × 10^–3 (1.97 × 10^–3, 3.80 × 10^–3)) > Ni (2.55 × 10^–3 (1.79 × 10^–3, 3.47 × 10^–3)) > Sb (2.20 × 10^–3 (1.61 × 10^–3, 3.11 × 10^–3)) > Be (3.45 × 10^–4 (3.45 × 10^–4, 3.51 × 10^–4)) > Se (2.05 × 10^–4 (1.44 × 10^–4, 2.65 × 10^–4)) > Hg (1.15 × 10^–4 (8.25 × 10^–5, × 10^–4)) (Table 2 and Fig. 2(d)). Most of the previous risk assessment studies [17, 51] indicated that As was the largest contributor to the noncarcinogenic effect of PM_2.5, surprisingly our results showed Mn-related noncarcinogenic risk is slightly higher than As (Table 2 and Fig. 2(d)), which may be associated with the high level of Mn environmental loading and Mn-related industrial emissions in Chengdu city. On the one hand, the atmospheric Mn load (29–48 ng/m³) in Chengdu (the capital city of Sichuan Province) is much higher than that in other parts of China (< 10 ng/m³) (e.g., Beijing, Shanghai, and Shenzhen) (Table 1). On the other hand, as a critical deoxidizer and desulfurizer in metallurgical processes, Mn demonstrates strong geochemical affinity with iron [52]. Notably, the recent China Steel Industry Yearbook shows that Sichuan is the largest iron ore producer in Southwestern China, with an output of more than 10,000,000 tons [53]. As the premier industrial and commercial Hub of Sichuan Province, Chengdu accounts for at least 50% of the province’s steel production. Crucially, energy consumption in Chengdu’s ferrous metal smelting industry (Figure S3(h)) is strikingly correlated with annual variations in Mn concentrations (Figure S6(p))—exhibiting a year-on-year decrease from 2015 to 2017, followed by an upward fluctuation. This partially supports the association between Mn concentrations and steel production activities. Intensive metallurgical activities involving high-temperature iron smelting likely drive Mn volatilization, providing a plausible explanation for Chengdu’s elevated non-carcinogenic Mn risk levels. Figure 2(c) illustrates the spatial characteristics of the non-carcinogenic risk HQs for the nine metals at the sampling sites in the three districts and counties (Pengzhou, Qingyang, and Gaoxin). The total non-carcinogenic risk HI values of the three sampling sites were 0.28, 0.26, and 0.25, respectively. Additionally, the non-carcinogenic risk rankings of the three sampling sites were almost identical, and the differences of the priority metals As or Mn between these sites were less than 0.01. This indicated that the differences in the non-carcinogenic metal pollution profiles across the different regions of Chengdu were small, further implying negligible spatial factors for non-carcinogenicity.

Fig. 2.

Non-Carcinogenic Risks of 9 Metals. a Data distribution of logarithmic hazard index (HI) across different populations. b Percentage contribution of 9 metals to total non-carcinogenic risk. c Spatial distribution of species-specific non-carcinogenic risks. Bar charts were utilized to characterize the non-carcinogenic risk of different metals in different sampling locations. d Data distribution of the logarithmic hazard index (HQ) for 9 metals

Table 2.

Non-carcinogenic and carcinogenic risks due to different species of heavy metals

Specie	HQ (Median (Q1, Q3))	CR (Median (Q1, Q3))
		Child	Female	Male
As	9.70E-02 (7.15E-02, 1.25E-01)	4.58E-07 (3.38E-07, 5.88E-07)	1.77E-06 (1.30E-06, 2.27E-06)	1.90E-06 (1.40E-06, 2.43E-06)
Be	3.45E-04 (3.45E-04, 3.51E-04)	1.21E-09 (1.21E-09, 1.23E-09)	4.68E-09 (4.68E-09, 4.75E-09)	5.01E-09 (5.01E-09, 5.10E-09)
Cd	2.59E-02 (1.78E-02, 3.84E-02)	3.41E-08 (2.34E-08, 5.06E-08)	1.32E-07 (9.03E-08, 1.95E-07)	1.41E-07 (9.68E-08, 2.09E-07)
Cr	2.71E-03 (1.97E-03, 3.80E-03)	1.07E-07 (7.77E-08, 1.50E-07)	4.13E-07 (3.00E-07, 5.79E-07)	4.43E-07 (3.21E-07, 6.21E-07)
Hg	1.15E-04 (8.25E-05, 1.65E-04)
Mn	1.17E-01 (9.10E-02, 1.47E-01)
Ni	2.55E-03 (1.79E-03, 3.47E-03)	4.37E-09 (3.06E-09, 5.94E-09)	1.69E-08 (1.18E-08, 2.29E-08)	1.81E-08 (1.27E-08, 2.46E-08)
Pb		2.84E-08 (2.08E-08, 3.69E-08)	1.10E-07 (8.01E-08, 1.42E-07)	1.17E-07 (8.59E-08, 1.53E-07)
Sb	2.20E-03 (1.61E-03, 3.11E-03)
Se	2.05E-04 (1.44E-04, 2.65E-04)

Q1: first quartile or lower quartile. Q3: third quartile or upper quartile

HQ Hazard quotient, CR Carcinogenic risk

Carcinogenic risk

For the total carcinogenic risk TCR, male > female > child, the median level was 2.68 × 10^–6 (1.99 × 10^–6, 3.35 × 10^–6), 2.50 × 10^–6 (1.85 × 10^–6, 3.13 × 10^–6), and 6.48 × 10^–7 (4.80 × 10^–7, 8.10 × 10^–7), respectively (Fig. 3(a)). Population differences in carcinogenic risk primarily stem from variations in exposure duration and Life expectancy across demographic groups. Adults face significantly elevated lifetime cancer risks compared to children, attributable to a fourfold longer exposure period. Furthermore, males generally have shorter life expectancies than females, resulting in higher lifetime carcinogenic risk despite equivalent exposures. The latest Chinese study on cancer burden has confirmed population differences in carcinogenic risks: cancer incidence rates increase with age in both genders, with a faster increase among males, which exhibit consistently higher rates than females after age 55 [54].

Fig. 3.

Carcinogenic Risks of 9 Metals. a Data distribution of logarithmic total carcinogenic risk (TCR) across different populations. b Percentage contribution of 6 metals to total carcinogenic risk. c Spatial distribution of species-specific carcinogenic risks. Bar charts were utilized to characterize the carcinogenic risk of different metals in different sampling locations. d Data distribution of the logarithmic carcinogenic risk (CR) for 6 metals

Among the six carcinogenic metals, As contributed the highest (71.76%) to the total carcinogenic risk, followed by Cr (17.31%) (Fig. 3(b)). For adult males, the order of carcinogenic risk was As (1.90 × 10^–6 (1.40 × 10^–6, 2.43 × 10^–6)) > Cr (4.43 × 10^–7 (3.21 × 10^–7, 6.21 × 10^–7)) > Cd (1.41 × 10^–7 (9.68 × 10^–8, 2.09 × 10^–7)) > Pb (1.17 × 10^–7 (8.59 × 10^–8, 1.53 × 10^–7)) > Ni (1.81 × 10^–8 (1.27 × 10^–8, 2.46 × 10^–8)) > Be (5.01 × 10^–9 (5.01 × 10^–9, 5.10 × 10^–9)), and similar patterns for females and children (Table 2 and Fig. 3(c)). Consistent with our results, numerous regional risk assessment studies have demonstrated that As and Cr are the metals with the highest carcinogenic risk in the atmosphere [18, 33, 55]. Therefore, in order to mitigate these risks, it is necessary to impose stricter restrictions on relevant industries, to utilize cleaner energy sources, and to strengthen atmospheric heavy metal monitoring. Figure 3(c) illustrates the species characterization of carcinogenic risk at the three sampling sites. Similar to the non-carcinogenic risk (Fig. 2(c)), the metal species ordering of the risk at the three sampling sites was essentially identical, and the TCRs of these sites were similar (2.97 × 10^–6 (Pengzhou), 2.79 × 10^–6 (Qingyang), and 2.63 × 10^–6 (Gaoxin)). Consistent with non-carcinogenic risks, the spatial distribution of carcinogenic risks also characterized the cross-regional robustness of the metal measurements.

Sensitivity analysis

The results of the sensitivity analysis for year-weighted concentrations (Al, As, Be, Cd, Cr, Hg, Mn, Ni, Pb, Sb, Se, Tl) are shown in Figures S7-S18 in Appendix A. Panel (a) of Figures S7-S18 presents the data distribution of average concentrations for the years 2015–2022, year-weighted concentrations (λ = 0.1− 0.5), and the most recent concentrations in 2023. Overall, all three metal concentration metrics including latest concentration, series-weighted concentration, and arithmetic mean—exhibit similar distribution patterns: the latest concentration peaks on the left, the series-weighted concentration is centered, and the arithmetic mean lies on the right. This indicates that the year-weighted concentration is closer to the latest concentration and offers greater representativeness than the arithmetic mean. Moreover, the weighted data distribution is more concentrated (exhibiting lower sensitivity to extreme values), as outliers are down-weighted rather than entirely ignored. Additionally, in the year-weighted concentration series, a smaller λ value (which places greater emphasis on more recent times) shifts the concentration distribution peak further left, bringing it closer to that of the most recent concentration. Notably, Table S7 shows that distribution tests (Kolmogorov–Smirnov and Anderson Darling) both indicate the annual weighted algorithm (λ = 0.1) yields the highest P-values among the six algorithms, meaning its values are closest to the latest year's concentration distribution. T tests and Wilcox tests confirm that mean and median concentrations for nearly all species using λ = 0.1 are closest to the latest levels. Therefore, considering both algorithm accuracy and data utilization efficiency, we determined the optimal year-weighted λ value to be 0.1. Panel (b) of Figures S7-S18 illustrates the quantile matching performance between the weighted concentration and the latest concentration for this optimal λ. For most metals, the high degree of overlap between the quantiles of the annual weighting algorithm and those of the latest concentrations further validates the effectiveness of the weighting approach.

Source apportionment of PM_2.5-bound species

PMF source identification

The selection of the optimal number of factors for source apportionment was informed by the trend of the Positive Matrix Factorization (PMF) objective function Q and comprehensive error diagnostics. Figure S19 (Appendix A) illustrates the variation of Q values for the four categories (Q_expected​, Q_true​, Q_robust​, and Q_true/Q_robust) with increasing factor numbers. Q_expected​ decreased linearly as the number of factors increased, whereas Q_true​, Q_robust​, and Q_true/Q_robust ratio declined gradually. For factor numbers exceeding 5, the reduction in Q_true​​ became negligible (< 20,000), and the Q_true/Q_robust ratio fell below 2. These results suggest that the PMF model adequately captures the concentration variability of species when using 5 or more factors [55]. Additionally, the PMF error diagnosis (Table 3) summarizes the key results of three error analyses. For the displacement analysis (DISP), the percentage change in Q-value (% dQ) before and after substituting strong species was extremely small (< 0.1%) across all scenarios, with no factor swapping observed. This indicates that the DISP analysis alone deems model stability acceptable for all scenarios. In the bootstrap analysis (BS), the factor mapping rate was 100% for all scenarios except the 3-factor case. Notably, BS-DISP (Bootstrap-Displacement) is designed to identify destabilizing factors in the model maximally [39]. For factor numbers exceeding 5, the case acceptance rate decreased gradually, while the proportion of cases involving factor swapping increased, which indicated that the model began to exhibit instability when using more than 5 factors. In summary, the optimal number of factors for PMF analysis in this study was determined to be 5.

Table 3.

Summary of PMF error estimation diagnosis

Diagnostic	Scenario
	2 factors	3 factors	4 factors	5 factors	6 factors	7 factors	8 factors
Qexpected	38,790	36,001	33,212	30,423	27,634	24,845	22,056
Qtrue	173,502.85	120,591.40	82,226.97	55,231.99	40,191.91	27,974.84	20,115.02
Qrobust	141,722.65	106,152.40	76,065.76	53,086.94	39,173.13	27,518.65	19,926.33
Qrobust/Qexpected	4.47	3.35	2.48	1.82	1.45	1.13	0.91
DISP % dQ	< 0.1%	< 0.1%	< 0.1%	< 0.1%	< 0.1%	< 0.1%	< 0.1%
DISP % swaps	0	0	0	0	0	0	0
Factors with BS mapping < 100%	0	Factor 2: 95%	0	0	0	0	0
BS-DISP % cases accepted	100%	100%	85%	90%	70%	70%	55%
BS-DISP % cases with swaps	0%	0%	3%	0%	2%	5%	11%

Q The objective function in the Positive Matrix Factorization model (PMF), DISP Displacement, one of the classic methods for estimating PMF error, BS Bootstrap

Figure 4 presents the main resolution results of the PMF model, including the inter-assignment relationships between species concentrations in PM_2.5 and emission factors (Fig. 4(a)), the contribution of factors to each species (Fig. 4(b)), the contribution of species to each factor (Fig. 4(c)), and the contribution of factors to the total species concentration in the 9-year time series (Fig. 4(d)). The 5 PM_2.5 emission factors contributed 14.7%, 15.0%, 52.2%, 5.9%, and 12.1% to the total concentration loadings (Fig. 4(a)). Factor 3 had the highest concentration loading because the vast majority of this factor was contributed to atmospheric Al, which is the most abundant of the trace metals (61.1% of the total species concentration) (Fig. 4(a)).

Fig. 4.

Source resolution of PM_2.5-bound heavy metals based on positive matrix factorization (PMF) modeling. a The inter-assignment relationships between the 16 species and the fitted five source factors were characterized by chord diagrams. b The contribution of factors to species is depicted in a circular stacked percentage bar plot. c Species contributions to factors are presented as a stacked percentage bar graph. d Time series of normalized contributions of five factors to total species concentrations

Factor 1 is characterized by a high contribution to Cl⁻ (89.3%), Hg (33.3%), Pb (43.6%), and Cd (52.3%) (Fig. 4(b)). Industrial sources of atmospheric Cl⁻ are commonly found in salt production and volatilization from electrolytic cells [56]. Since China completely abandoned the use of Pb-containing gasoline in 2000, the share of Pb in transportation emissions has declined precipitously [57]. Currently, atmospheric Pb and Hg originate mainly from the well-developed storage battery industry, including lead-acid batteries and lead-mercury primary batteries [58]. In addition, Cd is also involved in photovoltaic or nickel–cadmium battery manufacturing [59]. Taken together, all of the above species are primarily from the battery-related chemical industry, and therefore, Factor 1 is identified as the chemical industry. In addition, the high contribution of Pb to Factor 1 (38.3%) also reflects the significant chemical industry characteristics (Fig. 4(c)). Furthermore, we found that Factor 1 had a unique temporal profile compared to other factors (Fig. 4(d)), i.e., the contribution of Factor 1 to the total species concentration peaked at the end of 2016, after which it began to decline abruptly. This timing corresponds to the period of key events in Chengdu City, including the reform of the battery industry and the prevention of air pollution. Briefly, China's Ministry of Industry and Information Technology promulgated the “Lead Storage Battery Industry Standardized Conditions (2015 version)” and “Lead Storage Battery Industry Standardized Bulletin Management Measures (2015 version)”, which greatly improved the industry's entry standards and led to the emergence of new structures of lead batteries that are cleaner and emit less pollution [60]. This key event caused chemical industry-related emissions to fall off a cliff from one year later (end of 2016). In particular, in response to the reform of the environmental protection law (2015) and the Central Environmental Protection Inspection (August 7 to September 7, 2017), the Chengdu Municipal Government had already adopted a series of unprecedentedly stringent air pollutant emission measures before this date. Therefore, the above events may be the key reason for the special temporal change of Factor 1, which also supports the possibility of Factor 1 as the chemical industry.

Factor 2 has high loadings on NO₃⁻, NH₄⁺, Cr, and Se, with 83.5%, 40.7%, 51.4%, and 32.8%, respectively (Fig. 4(b)). Numerous studies [18, 38, 61] have shown that these 4 species are closely associated with motor vehicle emissions, thus Factor 2 was identified as a motor vehicle source. The low contribution of factor 2 to Pb is also mainly attributed to the widespread banning of Pb-containing gasoline. Moreover, the contributing concentration of Factor 2 keeps increasing from 2019 to 2024 (Fig. 4(d)), which is consistent with the trend of automobile ownership (Figure S4(a)) in Chengdu.

Factor 3 has an extremely high atmospheric Al loading (79.2%) (Fig. 4(a) and 4(b)) and its main contributing species is Al, with a contributing concentration of 92.7% (Fig. 4(a) and 4(c)). Since Al is one of the abundant metallic elements in soil [62] and the key identifying component of dust sources [63], Factor 3 was identified as a soil dust source. Additionally, the concentration contribution of Factor 3 exhibited relatively low variability over the sampling period (Fig. 4(d)). Compared to the other factors, the seasonal variation was less prominent, which fits well with the temporal variation pattern of soil dust sources.

The contribution of Factor 4 to SO₄²⁻ (69.6%) and As (25.0%) in PM_2.5 was at a high level (Fig. 4(b)), and SO₄²⁻ had the highest contribution (35.8%) to Factor 4 (Fig. 4(c)). SO₄²⁻ and As are the identifying components of coal combustion sources, and the historical emission inventories revealed that approximately more than 50% of China's atmospheric As originated directly from coal combustion [21], so Factor 4 was considered a coal combustion source.

Factor 5 contributed highly to almost all heavy metals Sb, As, Be, Cd, Cr, Hg, Pb, Mn, Ni, Se, and Tl with 38.3%, 44.2%, 45.6%, 44.9%, 27.1%, 22.4%, 49.8%, 28.1%, 40.2%, 25.1%, and 52.8%, respectively. Therefore, Factor 5 is mainly the non-ferrous metal industry [18].

Source-related social development indicators

Figure 5 presents Spearman correlation and partial correlation analyses of 34 socio-economic indicators in Chengdu with the concentrations of all species. First, most heavy metals (Hg, Tl, Sb, As, Cd, Pb) exhibit strong, statistically significant positive correlations with industrial dust emissions—this species-source covariation supports the plausibility of the data. For Factor 1 (chemical industry) identified via PMF, the four marker species (Cl⁻, Hg, Pb, and Cd) received partial support from the correlation analysis with Chengdu’s socio-economic indicators. Specifically, Spearman correlation results (Fig. 5(a)) revealed significant positive relationships (P < 0.05) between Cl⁻, Hg, and Pb concentrations and chemical materials production, with correlation coefficients (r_s) of 0.73, 0.97, and 0.93, respectively. Cd also showed a positive correlation with chemical industry activity (r_s = 0.67), though this association was not statistically significant (P = 0.06). This lack of significance may stem from low statistical power due to the limited number of years included in the analysis.

Fig. 5.

Relationships between 34 socio-economic development indicators and species concentrations in Chengdu. Spearman correlation and partial correlation analyses were conducted to examine relationships between six categories of 34 socio-economic development indicators and species concentrations. The six categories are industrial production (annual production steel, chemical material, and vehicles), industrial output value/revenue (ferrous metal, non-ferrous metal, chemical material sectors, automotive manufacturing, and manufacturing industry of other transportation vehicles (aircraft, trains, subways, etc.) above designated size), energy consumption (coal, electricity, gasoline, kerosene, diesel, fuel oil, and liquefied petroleum gas), sectoral energy consumption (total energy consumed by the sector of ferrous metal, non-ferrous metal, chemical material, automotive manufacturing, and manufacturing industry of other transportation vehicles), transport capacity (vehicle ownership, freight turnover volume of highway, railway, airline, and freight passenger volume of highway, railway, airline), and pollution and environmental protection (wastewater discharge, exhaust gas emissions, industrial dust emissions, solid waste generation, comprehensive utilization of solid waste, solid waste storage, and solid waste disposal). r_s: Spearman correlation coefficient; r_p: partial correlation coefficient

Species with high contributions from motor vehicle sources include NO₃⁻, NH₄⁺, Cr, and Se. Correlation analysis revealed that only NH₄⁺ exhibited a statistically significant positive association with automobile industry output value (P < 0.05), whereas NO₃⁻, Cr, and Se showed non-significant positive trends with automobile production (Fig. 5(a)). The lack of significance for these latter species may again be attributed to limited statistical power from the small number of years analyzed, as NO₃⁻, Cr, and Se demonstrated weak or non-significant associations with most socio-economic indicators. Notably, in the partial correlation analysis (controlling industrial dust emissions and the COVID-19 pandemic), NO₃⁻ displayed a significant positive association with airline cargo turnover (r_p = 0.79, P = 0.04). This finding can be explained by the large-scale fuel oil consumption (Figure S3(f)) driven by the rapid growth in air traffic (Figure S4(d)) as transportation and trade resumed post-pandemic (2019–2022).

Factor 3 (soil dust) was characterized by its dominant contribution to Al. In both correlation and partial correlation analyses, Al exhibited non-significant associations with nearly all industrial, energy, transportation, and environmental sectors, consistent with its role as a marker species for natural sources. Notably, even after covariate adjustment, Al maintained a significant negative correlation (P < 0.05) with the industrial output value of non-ferrous metal smelting (a major source of atmospheric heavy metal pollution) (Figs. 5(a) and 5(b)). This inverse relationship further substantiates that atmospheric Al likely originates predominantly from resuspended soil dust rather than anthropogenic activities.

The PMF results indicate that SO₄²⁻ and As receive substantial contributions from Factor 4 (coal combustion). Correlation analysis revealed a positive but non-significant association between SO₄²⁻ and annual coal consumption (r_s = 0.13), which strengthened after covariate adjustment (r_p = 0.58), though statistical significance was not achieved (P > 0.05) due to limited temporal coverage. Factor profiles (Table S6 and Fig. 4(b)) demonstrate that As exhibits high loadings not only from coal combustion but also from chemical production, motor vehicles, and non-ferrous metal industries. This multi-source profile explains its significant positive correlations with chemical material production (r_s = 0.81, P < 0.01) and gasoline consumption (r_s = 0.76, P = 0.02) in Fig. 5(a). Conversely, As showed non-significant negative correlations with coal combustion and non-ferrous metal smelting, likely masked by its stronger associations with chemical and vehicular sources.

For Factor 5 (non-ferrous metal industry), conventional correlation analysis revealed an unexpected pattern: most heavy metal concentrations showed non-significant negative correlations with non-ferrous metal smelting output (Fig. 5(a)). This apparent paradox may stem from masking effects, as these metals exhibit stronger associations with other sectors (e.g., chemical production, motor vehicles, industrial dust). Supporting this interpretation, partial correlation analysis (controlling industrial dust) showed increased positive associations between heavy metals and non-ferrous smelting activity (Fig. 5(b)). Notably, nickel (Ni) constituted an exception, demonstrating significant positive correlations with energy consumption in both ferrous (r_s = 0.81; r_p = 0.87) and non-ferrous smelting (r_s = 0.74; r_p = 0.80) before and after covariate adjustment (Figs. 5(a) and 5(b)). This aligns with atmospheric emission inventories from Japan, where Ni predominantly originates from iron/steel and other metal smelting operations [48]. As the only metal showing consistent and statistically robust links to metal smelting sectors, Ni demonstrates potential as a marker species for this industrial source [18].

Source-apportioned health risks

Figure 6 illustrates the health risks due to the five sources identified by PMF (chemical industry, motor vehicles, soil dust, coal combustion, and non-ferrous metal smelting). For non-carcinogenic risks, the HQ values for the above sources are all below 1 (Fig. 6(a)), indicating that non-carcinogenic risks from these sources are negligible.

Fig. 6.

Non-carcinogenic and carcinogenic risks due to five PMF-identified sources. Stacked bar plots were employed to characterize source-specific HQ and CR

As for the carcinogenic risk, the CRs for chemical industry, motor vehicles, soil dust, coal combustion, and nonferrous metal smelting were 2.24 × 10^–6, 1.44 × 10^–6, 5.19 × 10^–7, 2.17 × 10^–6, and 4.71 × 10^–6, respectively (Fig. 6(b)). Except for the soil dust, all the sources posed non-negligible carcinogenic risks. The nonferrous metal industry is the most urgent heavy metal pollution-related industry to be regulated in Chengdu, with the highest population carcinogenic risk, which is closely associated with the restructuring of Chengdu's industrial layout and policy orientation. According to the Chengdu Metallurgical Industry ‘Twelfth Five-Year Plan’ (2011–2015) [64], it is essential to vigorously develop the metal smelting economy, realize the leap of non-ferrous metal production capacity, and build Chengdu into one of the crucial new iron and steel material bases in China and a demonstration base for waste metal resources processing during the ‘12th Five-Year Plan’ period. The rise of the non-ferrous metals industry during the ‘12th Five-Year Plan’ period may have exacerbated heavy metal pollution in the atmosphere, consistent with the fact that the contribution of non-ferrous metals peaked at the end of 2015 (Fig. 4(d)). Subsequently, the Chengdu municipal government took into account the environmental problems posed by metal smelting and implemented a series of preventive and control measures, including cleaner production, energy saving, and utilization of recycling [22], which resulted in an overall downward trend in ambient heavy metal loads (Fig. 4(d)). Nevertheless, our results still indicate that the non-ferrous metal industry poses an excess carcinogenic risk (4.71 × 10^–6) to Chengdu citizens. Therefore, the non-ferrous metal industry remains a priority to be restricted and regulated in the next urban development plan for Chengdu.

The chemical industry, especially the battery industry, and the motor vehicle industry are closely linked in China, and their contributions to PM_2.5 are approximately equal, at 14.7% and 15.0%, respectively (Fig. 4(a)) in this study. Notably, the former has the second-highest cancer risk (Fig. 6(b)), while the latter has a cancer risk slightly higher than the safety threshold (10^–6). The chemical industry's metal contamination problems have been further exacerbated in recent years, largely due to a burst of capacity incentivized by China's new energy vehicle policy. Under the backdrop of the “dual-carbon” strategy, to combat climate change and promote green development, China's State Council first issued the “Energy Saving and New Energy Vehicle Industry Development Plan (2012–2020)” in 2012, which vigorously promotes the transition from traditional fuel vehicles to purely electric-powered vehicles. As disclosed by China's central government, from 2014–2024, the ownership of new energy vehicles grew from 0.12 million to 31.4 million, an increase of 260 times. Additionally, 11.25 million new-energy vehicles were newly registered, even accounting for 41.83% of the total number of newly registered vehicles. Trend analysis of Chengdu’s socio-economic indicators reflects this shift: electricity consumption has risen steadily (Figure S3(b)) while gasoline use has declined (Figure S3(c)), indicating the city’s ongoing automotive energy transition from fossil fuels to electricity. Additionally, chemical raw material production in Chengdu has decreased (Figure S2(b)) even as industrial output value has grown consistently since 2018. This suggests a transformation of the chemical industry from high-volume, low-value production to high-quality, high-value output—likely driven by the rapid development of key chemical batteries for new-energy vehicles, a critical component of the automotive energy transition. However, these transformations have inevitably introduced new environmental risks, particularly heavy metal pollution from the battery industry [65, 66]. Carcinogenic hazards to Humans from heavy metal pollution from the battery industry were reported in 2015 [67], and significantly elevated blood metal levels were also found in populations around battery plants. Of concern, recent Pb isotope traceability studies [68] have shown that the penetration of electric vehicles in India is insufficient to curb atmospheric heavy metal emissions. This suggests that the strategy of transforming electric drives does not seem to be able to reduce environmental heavy metal loads, although it can significantly ameliorate pollution due to motor vehicle exhaust. In summary, China's environmental authorities must devote more effort to limiting heavy metal emissions from the chemical battery industry.

Coal combustion is also one of the main sources of heavy metals in the atmosphere, which is continuously confirmed by studies [18, 21, 22]. In our study, the coal-fired sector in Chengdu presents a low level of both PM_2.5 contribution (Fig. 4(a)) and health risk (Fig. 6) compared to other industrial cities [18, 33], which is mainly determined by the energy structure of the city. According to official data [69], installed hydropower in Sichuan Province (the province to which Chengdu belongs) accounts for 78% of the total installed capacity in the province, while thermal or coal-fired power generation accounts for only 16%. The low level of coal-fired industry may be the primary reason for the low cancer risk. In addition, Chengdu City has been vigorously promoting a green economy in recent years, strictly controlling new industrial projects with coal-fired boilers, with the proportion of clean energy accounting for more than 60% of total energy consumption [70]. This clean energy policy further reduces the carcinogenic risk from the coal-fired industry.

Limitations

Nevertheless, this study involves uncertainties that resist quantification. First, the sampling period for this study (10th to 16th of each month) included days with heavy pollution, which may have resulted in an overestimation of the assessed health risks. Second, the risk assessment exclusively addressed PM_2.5-bound heavy metals, excluding other particle size fractions (e.g., PM₁₀, ultrafines), potentially resulting in underestimated health risks. Third, we did not integrate combined indoor-outdoor exposure scenarios. Given demonstrated concentration disparities between microenvironments and differential exposure durations [19], this omission introduces exposure misclassification uncertainty. Fourth, approximating carcinogenic Cr (VI) concentration as 1/7 of total Cr, though methodologically common, may bias risk estimates. Fifth, while the year-weighting algorithm optimized temporal representativeness, residual gaps in capturing extreme concentration events persist. Sixth, the retrospective design of this study means the health risk and source analyses may always remain gaps to capture the current air pollution situation. Seventh, although the integration of socio-economic trend analysis and correlation modeling partially mitigates PMF's inherent disconnection from sector-specific source profiles, the 9-year sampling period in this study may limit statistical power for detecting source-receptor relationships.

Conclusion

This study investigated the mass concentration, health risk, and source apportionment of heavy metals in PM_2.5 in Chengdu City from 2015 to 2023. Moreover, we assessed the non-carcinogenic and carcinogenic risks due to different emission sources. The total non-carcinogenic risk due to PM_2.5-bound heavy metals is at a negligible level, while the total carcinogenic risk requires more attention from the environmental authorities, particularly the risks associated with As and Cr. Five sources of heavy metals and ions in PM_2.5 were revealed via PMF: chemical industry, motor vehicles, soil dust, coal combustion, and non-ferrous metal industries. These sources were systematically supported and explained through trend analysis and correlation studies with Chengdu's socio-economic indicators. All sources had negligible non-carcinogenic risks, while carcinogenic risks associated with the non-ferrous metals and chemical industries may require increasing concern.

Future regional air pollutant health risk assessment and source apportionment studies should consider integrating atmospheric transport modeling to overcome the limitations of retrospective data. In addition, policies for cleaner production, energy conservation, and recycling require further development to address pollution issues in non-ferrous metal industry. Moreover, in view of the excessive growth rate of China's new energy vehicle and battery chemical industries, the relevant authorities should pay more attention to the environmental problems arising from them.

Authors’ contributions

WH: Methodology, Validation, Formal analysis, Investigation, Writing—original draft. DK: Data curation, Validation, Investigation, Writing—review & editing. FQ: Data curation, Validation, Investigation. LL: Data curation, Validation. JY: Data curation, Validation. SS: Validation, Methodology. SJ: Data curation, Validation. JH: Data curation, Validation. CW: Supervision, Project administration. RL: Validation, Methodology, Writing—review & editing, Supervision, Project administration. XG: Conceptualization, Investigation, Writing—review & editing, Supervision, Project administration.

Funding

This study belongs to the medical research projects of Sichuan Provincial Health Commission and Chengdu Municipal Health Commission, with the project name and project number of “Research on the Establishment and Application of Health Risk Assessment System for Environmental Hazard Factors in Chengdu” (S23028) and “Research on Health Risk Assessment of Air Pollution in Chengdu” (2023360), respectively.

Data availability

Data and material can be obtained according to the research purpose by contacting the corresponding author, Dr. Xufang Gao (E-mail: 15688415@qq.com).

Declarations

Ethics approval and consent to participate

The data in this study were based on environmental monitoring and did not involve any human or animal participants. Therefore, the ethical review process and consent to participate was not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.