Characterization and Spatial Source Apportionments of Ambient PM10 and PM2.5 during the Heating Period in Tian’jin, China

Received: July 6, 2019
Revised: September 18, 2019
Accepted: October 3, 2019
Download Citation: RIS|BibTeX|https://doi.org/10.4209/aaqr.2019.06.0281  

Cite this article:

Liu, B., Sun, X., Zhang, J., Bi, X., Li, Y., Li, L., Dong, H., Xiao, Z., Zhang, Y. and Feng, Y. (2020). Characterization and Spatial Source Apportionments of Ambient PM₁₀ and PM_2.5 during the Heating Period in Tian'jin, China. Aerosol Air Qual. Res. 20: 1-13. https://doi.org/10.4209/aaqr.2019.06.0281

HIGHLIGHTS

• Impact of sources was assessed by the relationship between SO₄²⁻/NO₃⁻ and PM.
• Error estimation diagnostics was applied to choose an optimal factor number.
• The contributions of most sources show significant spatial-differences.

Keywords: Chemical species; Source apportionment; Heating period; Error estimation; PMF

INTRODUCTION

Particulate matter (PM) is mainly composed of inorganic, carbonaceous, and elemental matter (Zhang et al., 2013, 2014; Liu et al., 2016a). Understanding PM chemical composition and source is critical to assessing impacts on air quality, climate change and human health, as well as to controlling pollution (Zhang et al., 2014; Li et al., 2016; Peng et al., 2017; Liu et al., 2018). In recent years, many studies have focused on defining the chemical compositions and sources of PM_2.5 (Bahadur et al., 2009; Chen et al., 2010; Yang et al., 2011; Tao et al., 2014; Fang et al., 2017; Dai et al., 2018; Deng et al., 2018). 

Receptor models have been very useful tools for PM source apportionments (Khan et al., 2015). For example, Fang et al. (2017) identified sources of ambient PM in Haikou, China, using the CMB model; Deng et al. (2018) conducted source apportionment of PM_2.5 in Lin’an, China, using the PCA/MLR model; Liu et al. (2017b) apportioned the emission sources of air pollutants using the PMF and ME2 models in Tian’jin, China; and Wang et al. (2015) identified the sources of fine particles in Xi’an, China, using the PMF model. The PMF model does not need input source profiles data in the calculation process, which has been widely used to identify PM sources due to the lack of source profiles and the complexity of the sources (Okuda et al., 2010; Xu et al., 2012; Zhu et al., 2012; Chen et al., 2014). However, the choice of factor number in the PMF calculation process can have a critical impact on the apportionment results. In the PMF 5.0 model, the optimal factor solutions can be identified by three error estimation methods (Liu et al., 2017a). The uncertainty of PMF analysis derived from the rotational ambiguity and random errors can be obtained by error estimation (EE) methods; the uncertainty of each factor’s identifying species is shown to be a useful gauge for evaluating multiple solutions, e.g., with a different number of factors (Brown et al., 2015). Therefore, the EE methods can play an important guiding role in choosing an optimal factor solution. For example, Paatero et al. (2014) and Liu et al. (2017a) had applied the EE method to choose an optimal factor solution and obtained the good results. However, the related research on source apportionments using the PMF model with EE methods is still limited.

Tian’jin, a rapidly developing megacity in China, had a permanent population of over 15 million and over 2.83 million automobiles in 2016 (Tian’jin Statistical Yearbook 2017; http://stats.tj.gov.cn/Category_29/Index.aspx). The heating period in Tian’jin normally begins on 15 November, after which coal combustion for heating can increase the level of air pollution (Yu et al., 2010; Huang et al., 2014), and heavy pollution events become more frequent along with unfavorable meteorological conditions (Yang et al., 2016; Liu et al., 2018). According to 2016 monitoring data, the mean concentrations of PM₁₀ and PM_2.5 during the heating period in Tian’jin reached 97 and 66 µg m^–3, 0.9 and 1.6 times higher than the World Health Organization (WHO) 24-h guideline value of 50 and 25 µg m^–3. Given this serious pollution, understanding the chemical compositions and sources of the PM is very important for implementing effective control measures. Although source apportionments of atmospheric particulates in Tian’jin have been conducted in the past (Gu et al., 2011; Zhang et al., 2011; Wu et al., 2015), the recent implementation of pollution control measures may lead to changes in the compositions and sources of PM, and current knowledge is limited. In addition, the distribution for pollution sources in different regions in one city is obviously different; thus, more detailed spatial source apportionment is also needed in order to control these sources more accurately.

Therefore, in this study we aimed to: (1) better understand pollution levels in Tian’jin during the heating period; (2) characterize the concentration variations of PM₁₀ and PM_2.5 and their chemical species; (3) reveal optimal factor solutions using EE analysis and apportion source contributions using the PMF model; and (4) identify the transport pathway of air masses and potential source areas of PM₁₀ and PM_2.5 using backward trajectory and PSCF analysis, respectively.

These results can provide some data support for Tian’jin’s municipal government departments to control atmosphere PM sources more accurately.

MATERIALS AND METHODS

Site Description

Tian’jin (38.57–40.25° N, 116.72–118.01° E), one of the megacities in the Bei’jing-Tian’jin-He’bei area, has an area of ~12,000 km². The area has a temperate monsoon climate, with an average annual temperature of 11.4–12.9°C and an annual precipitation of 520–660 mm. Southwest, northwest, and northeast winds prevail during summer and autumn, while northwest and northeast winds prevail during spring and winter (Liu et al., 2016b). We collected ambient PM_2.5 samples at six sites (NJ, NK, DZG, ADR, QJ, and FK; Fig. 1) set on building rooftops ~9–15 m above ground level; further details are given in Table S1. The data of gaseous pollutants was collected from nearby National or Provincial Air Quality Monitoring Stations (within 10 m). The mean concentrations of SO₂, NO₂, O₃, and CO were 34 µg m^–3, 52 µg m^–3, 43 µg m^–3, and 1.2 mg m^–3 during the study period, respectively, with clear spatial variations (Table S2). 

Fig. 1. Locations of the sampling sites in Tian’jin. NJ: Nanjing Road, NK: Nankou Road, DZG: Dazhigu Road, ADR: Advancing Road, QJ: Qinjian Road, FK: Fukang Road.

Ambient Sampling and Analysis

From 22 February to 2 March 2016, we used four medium-volume samplers (TH-150C; Wuhan Tianhong Ltd., China) to collect PM₁₀ and PM_2.5 samples at each site for 23 h (from 10:00 a.m. to 9:00 a.m. the next day) at a flow rate of 100 L min^–1. Multiple filters were applied to synchronously collect ambient PM_2.5 samples; parallel samples and field blanks were also collected at each site. Finally, a total of 14 PM_2.5 and 14 PM₁₀ samples (including the parallel and blank samples) were collected at each site, respectively. A total of 84 PM_2.5 and 84 PM₁₀ samples were collected at six sites during the sampling period. In order to better determine the representative of sampling time for the heating period, we analyzed the trends of six air pollutants (i.e., PM_2.5, PM₁₀, SO₂, NO₂, CO, and O₃) during the heating period in 2016, and the results were described in Fig. S1. The sampling period in this study can basically include the peak and trough periods of air pollutant concentration changes; therefore, the sampling period is representative for the heating period.

Polypropylene membrane filters were used to determine chemical elements (Na, Fe, V, Mg, K, Ca, Mn, Ti, Cr, Cd, Ni, As, Cu, Pb, Zn, Al, and Si) by inductively coupled plasma atomic emission spectrometry (ICP-AES) (IRIS Intrepid II; Thermo Electron). Quartz fiber filters were used to determine SO₄^2–, NO₃^–, Cl^–, NH₄⁺, Na⁺, K⁺, Mg²⁺, and Ca²⁺ by ion chromatography (DX-120; Dionex). A circular quartz fiber filter of 0.210 cm² was used to measure elemental carbon (EC) and organic carbon (OC) using a Desert Research Institute Model 2001 thermal/optical carbon analyzer with the IMPROVE A thermal/optical reflectance protocol. The method detection limits of chemical species, water-soluble inorganic ions, and carbonaceous species are listed in Table S3. Further details regarding sampling, measurement, and quality control are given in the supplementary material.

PMF and Error Estimation Model

PMF is a multivariate factor analysis tool widely used for PM source apportionment (Liu et al., 2017a, 2018). This model can decompose a matrix of speciated sample data (X) into two matrices: factor contributions (G) and factor profiles (F) (Paatero and Tapper, 1994). The speciated data are the matrix X of i by j dimensions, in which i number of samples and j chemical species were measured; further context is given in the supplementary material. We used three EE methods to choose optimal factor solutions: displacement of factor elements (DISP), classical bootstrap (BS), and bootstrap enhanced by displacement (BS-DISP). Further details were given in Liu et al. (2017a).

Backward Trajectory and PSCF Models

The 72-h backward trajectories from the center of Tian’jin (39.13°N, 117.20°E) were calculated at 1-h intervals during the sampling period. The final global analysis data were produced by the National Centers for the Global Data Assimilation System (GDAS; http://www.arl.noaa.gov/). The backward trajectory model was performed four times per day at starting times of 0:00, 06:00, 12:00, and 18:00 LT at a starting height of 100 m above the ground (Byčenkienė et al., 2014; Liu et al., 2016a, b). The PSCF model combines backward trajectory calculation and air pollutant concentrations for identifying potential source areas. All hourly endpoints from the backward trajectories were classified into 1° × 1° latitude and longitude grid cells; further details are given in the supplementary material.

RESULTS AND DISCUSSION

PM₁₀ and PM_2.5 Concentrations

During the heating period, the concentrations of PM₁₀ and PM_2.5 ranged between 37–262 µg m^–3 and between 17–214 µg m^–3, respectively (Fig. 2); the mean PM_2.5/PM₁₀ ratio was 0.67. The average PM₁₀ and PM_2.5 concentrations of 98 and 71 µg m^–3 exceeded Chinese air quality standards by 140% and 203%, respectively, as well as the stricter World Health Organization air quality guidelines of 20 and 10 µg m^–3, respectively (World Health Organization, 2006). The PM₁₀ and PM_2.5 concentrations showed significant spatial differences (t-test, p < 0.05) with standard deviations of up to 12 and 6 µg m^–3, respectively. A higher PM₁₀ concentration was observed at NK (108 µg m^–3), with a relatively low PM_2.5/PM₁₀ ratio (0.60), likely indicating that crustal dust had an important influence here (Shakya et al., 2017). Lower PM₁₀ (94 µg m^–3) and higher PM_2.5 (78 µg m^–3) concentrations were observed at QJ along with a higher PM_2.5/PM₁₀ ratio (0.78), likely indicating important contributions of fine-particle emission sources such as vehicles and secondary inorganic aerosol, as QJ was characterized by traffic and residential areas where the influence of vehicle emissions would be relatively high (Table S1). The mass concentrations of PM₁₀ and PM_2.5 at each site showed increasing trends from February to March, likely due to the heavy pollution event in March (Fig. 2). Compared with other regions around the world (Table S4), we found that the mean PM_2.5 and PM₁₀ concentrations in Tian’jin were usually higher than some cities in Europe and Korea, but apparently lower than many cities in Beijing-Tianjin-Hebei and surrounding areas. From 2008 to 2013, the concentration of PM_2.5 in Tian’jin showed an increasing trend (Gu et al., 2011; Wu et al., 2015). Subsequently, the PM_2.5 concentration decreased significantly (Tian et al., 2018; Yan et al., 2018), which might be closely associated with the Air Pollution Prevention and Control Action Plan issued by China in 2013 (http://www.gov.cn/jrzg/2013-09/12/content_2486918.htm). The control measures of PM sources were tightened in Tian’jin so that the PM concentrations can decrease significantly.

Fig. 2. Temporal variations of PM₁₀ and PM_2.5 concentrations at six sampling sites during the heating period. The grey shadows represent the pollution episode.

Characteristics of Chemical Species in PM

Water-soluble Ions

The average concentrations of the SNA (the sum of SO₄^2–, NO₃^–, and NH₄⁺) in PM₁₀ and PM_2.5 reached 23.4 and 19.9 µg m^–3, respectively, accounting for 72.4% and 77.1% of the total measured ions, and 23.9% and 28.0% of PM₁₀ and PM_2.5. The mean concentrations of NO₃^–, SO₄^2–, Cl^–, and NH₄⁺ in PM₁₀ were 12.4, 7.1, 3.8, and 3.9 µg m^–3, accounting for 12.6%, 7.2%, 3.9%, and 4.0% of PM₁₀ concentrations, respectively. The mean concentrations of NO₃^–, SO₄^2–, Cl^–, and NH₄⁺ in PM_2.5 were 10.5, 5.7, 3.1, and 3.7 µg m^–3, accounting for 14.8%, 8.0%, 4.3%, and 5.2% of PM_2.5 concentrations, respectively. The concentrations and percentages of NO₃^– in PM_2.5 and PM₁₀ were significantly higher than other measured ions (t-test, p < 0.01). The nitrate is mainly produced from secondary reactions of nitrogen oxide emitted from motor vehicles, coal combustion, and industrial emissions (Liu et al., 2016a; He et al., 2017), likely implying that these sources had larger contributions to PM_2.5 and PM₁₀ during the heating period. The ratio of SO₄^2– in PM_2.5 to that in PM₁₀ (0.80), along with those of NO₃^– (0.85), NH₄⁺ (0.94), and Cl^– (0.80), suggested that these measured ions were mainly enriched in fine particles.

The concentrations of the predominant water-soluble ions in PM₁₀ and PM_2.5 at all sampling sites showed the same tendency and decreased in the order: NO₃^– (8.2–17.4 µg m^–3) > SO₄^2– (4.7–9.5 µg m^–3) > NH₄⁺ and Cl^– (2.0–5.0 µg m^–3) (Fig. S2). The concentrations of NO₃^– (15.7–17.4 µg m^–3), SO₄^2– (8.0–9.5 µg m^–3), and Cl^– (3.9–5.0 µg m^–3) were highest at QJ, consistent with higher PM_2.5 concentrations at that site. SO₄^2– and NO₃^– are mainly formed by secondary reactions of sulfur dioxide and nitrogen oxide emitted from coal combustion and vehicles (Almeida et al., 2015; Liu et al., 2016a). Wet-process desulfurization from coal-fired power plants can also directly discharge abundant sulfate particles into the air (Ma et al., 2015; Liu et al., 2017a). Cl^– is another indicator of coal combustion (Wu et al., 2015), suggesting that QJ was primarily affected by coal combustion and vehicle emissions. By contrast, the concentrations of SO₄^2– (4.7–5.6 µg m^–3), Cl^– (2.8–3.7 µg m^–3), and NO₃^– (8.2–9.0 µg m^–3) were lowest at ADR, implying lower contributions of coal combustion and vehicle emissions, perhaps because that site was characterized by residential, commercial, and scenic areas (Table S1).

The sulfur and nitrogen oxidized ratios (SOR and NOR, respectively) were calculated by the molar ratios of [SO₄^2–] to [SO₄^2– + SO₂] and [NO₃^–] to [NO₃^– + NO₂], respectively (Dai et al., 2018). The average values of SOR and NOR in PM₁₀ during the heating period were 0.17 and 0.14, respectively, while those for PM_2.5 were 0.13 and 0.12, respectively. All were higher than 0.1, indicating the formation of secondary species (Truex et al., 1980; Zhang et al., 2013). Higher average values of SOR (0.20–0.22) and NOR (0.16) and a higher O₃ concentration (39 µg m^–3) were observed at QJ (Table S2), likely implying stronger oxidation of SO₂ to SO₄^2– and NO₂ to NO₃^–, which could have an important influence on the high concentrations of SO₄^2– and NO₃^– at this site (Fig. S2). The mean values of SOR (0.05–0.12) and NOR (0.04–0.08) were lower at ADR, likely due to the lower O₃ concentration (25 µg m^–3) (Table S2). Gao et al. (2011), Zhou et al. (2016), and Zhang et al. (2013) found that average winter SOR values in Ji’nan, Shang’hai, and Fu’zhou were 0.17, 0.20, and 0.22–0.27, while those for NOR were 0.12, 0.12, and 0.05–0.08, respectively—all higher than those in Tian’jin.

The average SO₄^2–/NO₃^– ratios were 0.57 and 0.54 for PM₁₀ and PM_2.5 during the heating period, respectively. The mass ratio of SO₄^2–/NO₃^– can reflect the relative importance of stationary versus mobile sources of sulfur and nitrogen in the atmosphere (Liu et al., 2017a; Murillo et al., 2012). The mean ratio of SO₄^2–/NO₃^– (1.0) was highest at NK, and the SO₄^2– concentration was higher (Fig. S2), suggesting that the influence of stationary sources might be larger here (Table S1). The average ratio of SO₄^2–/NO₃^– (0.5–0.6) at QJ was lower, and the NO₃^– concentration was higher, indicating that the contribution of mobile sources was larger at this site (Table S1). The SO₄^2–/NO₃^– ratios showed a decreasing tendency with increasing PM₁₀ and PM_2.5 concentrations (slope = –0.004 and r = –0.70 for PM₁₀; slope = –0.005 and r = –0.77 for PM_2.5) (Fig. S3), likely implying that mobile sources could have had an important impact on the increase of PM₁₀ and PM_2.5 concentrations (Liu et al., 2017a).

The correlations between NH₄⁺ and both NO₃^– and SO₄^2– (r = 0.64–0.81, p < 0.01) were higher than that between NH₄⁺ and Cl^– (r = 0.37–0.43, p < 0.01) (Table S5), indicating that NH₄⁺ might exist in the form of ammonium sulfate, ammonium hydrogen sulfate, and ammonium nitrate during the heating period (Wang et al., 2005). There were significant correlations between NO₃^– and SO₄^2– (r = 0.73–0.82, p < 0.01), reflecting higher homology. The correlation between Ca²⁺ and Mg²⁺ was higher in PM₁₀ (r = 0.71, p < 0.01) than PM_2.5 (r = 0.54, p < 0.05), suggesting that they might be derived mainly from sources of coarse particles such as dust. In addition, the correlation between K⁺ and Cl^– was higher in PM_2.5 (r = 0.81, p < 0.01) than PM₁₀ (r = 0.56, p < 0.01), probably indicating that they mainly originated from sources of fine particles such as biomass burning. Meanwhile, potassium chloride might be an important form of Cl^– during the heating period.

Carbonaceous Materials

The average concentrations of OC and EC in PM₁₀ were 11.8 and 3.9 µg m^–3 during the heating period, respectively—significantly higher than those in PM_2.5 (9.4 and 3.0 µg m^–3; t-test, p < 0.01). The percentages of OC and EC in PM₁₀ were 12.0% and 4.0%, respectively, lower than those in PM_2.5 (13.2% and 4.2%), indicating these carbonaceous materials were mainly enriched in PM_2.5. Note that the concentrations and percentages of SNA in PM were apparently higher than those of total carbon (TC; the sum of OC and EC). The concentrations and percentages of OC in PM were close to those of NO₃^– but significantly higher than other water-soluble ions (t-test, p < 0.01). The NO₃^– and OC have become the dominant components of particulate matter at present. The higher correlations between OC and EC (R² = 0.87–0.93) (Fig. S4) suggest that they might be derived from the same sources (Liu et al., 2016c). The correlations between OC and EC did not differ statistically between the sampling sites, except for NK (Fig. 3), likely implying similar sources of carbonaceous materials. The correlations between OC and EC were higher at NJ (r = 0.98–0.99) and lower at NK (r = 0.44–0.94), indicating the high homology of OC and EC at the former. The sources of OC and EC at NK might be complex because this site was located in industrial and traffic areas (Table S1). The OC/EC average ratios in PM₁₀ and PM_2.5 during the heating period were 3.1 and 3.2, respectively. The mean OC/EC ratios by site showed little spatial differences (Table S6), ranging from 2.8 (FK) to 3.3 (NJ and DZG) for PM₁₀ and 2.8 (NK) to 3.4 (ADR) for PM_2.5. Other studies have reported that the OC/EC ratios from biomass burning, coal combustion, and vehicle exhaust were 4.1–14.5, 0.3–7.6, and 1.0–4.2, respectively (Watson et al., 2001; Schauer et al., 2002; Liu et al., 2016a; Zhang et al., 2007). Therefore, the contributions of coal combustion and vehicle emission might be relatively high during the heating period. 

Fig. 3. Correlations between OC and EC in (a) PM₁₀ and (b) PM_2.5 at different sampling sites.

An OC/EC ratio larger than 2.0–2.2 has been used to identify and evaluate secondary organic aerosols (SOAs) (Turpin and Huntzicker, 1991; Chow et al., 1996). In Tian’jin, the average OC/EC ratio during the heating period was 3.1 and 3.2 for PM₁₀ and PM_2.5, respectively, indicating the possible presence of SOAs. The SOC is calculated using the EC-tracer method (Gu et al., 2010):

Linear regression parameters of OC and EC based on different percentiles of OC/EC ratios are shown in Tables S7–S13. The regression of OC and EC data below 10^th percentile of OC/EC ratios shows the best solution to calculate the (OC/EC)_min ratio. In this study, the observed (OC/EC)_min values during the heating period were 1.8 and 2.2 in PM₁₀ and PM_2.5, respectively, and the average SOC concentrations were 4.8 and 2.8 µg m^–3, respectively (Table S6). The fractions of SOC to OC mass were 39.8% and 28.0% in PM₁₀ and PM_2.5, respectively (Table S6), which could indicate an important contribution to particle matter. Relatively high SOC concentrations in PM₁₀ were observed at ADR (4.0 µg m^–3) and NK (3.8 µg m^–3), while the SOC concentration at DZG (1.6 µg m^–3) was relatively low. A higher SOC concentration in PM_2.5 was observed at NK (3.8 µg m^–3), and a lower SOC concentration was found at QJ (0.5 µg m^–3).

Elements

Seventeen elements (Na, Fe, V, Mg, K, Ca, Mn, Ti, Cr, Cd, Ni, As, Cu, Pb, Zn, Al, and Si) in PM₁₀ and PM_2.5 were determined during the heating period; the concentrations of the total detected elements in PM₁₀ and PM_2.5 were 12.4 and 6.4 µg m^–3, respectively, accounting for 12.7% and 9.0% of the PM₁₀ and PM_2.5 mass, respectively. Crustal elements (Mg, Al, K, Si, Na, Ca, and Fe) were the most abundant elements, accounting for 73.2–84.2% of the total detected elemental mass, similar to the results of Mohammed et al. (2017) and Liu et al. (2017a). The average concentrations of Mg, Al, K, Si, Na, Ca, and Fe in PM₁₀ were 0.9, 1.3, 1.0, 3.2, 0.3, 3.4, and 1.9 µg m^–3, respectively—significantly higher than those in PM_2.5. The mass ratios of Mg, Al, Si, Ca, and Fe in PM_2.5 to those in PM₁₀ were 0.44, 0.48, 0.47, 0.45, and 0.44, respectively, indicating that these elements were mainly enriched in PM₁₀. The total concentrations of crustal elements at different sites showed apparent differences (t-test, p < 0.05). The higher concentrations were observed at NK (7.3–13.4 µg m^–3), likely suggesting that the influence of crustal sources around that site was higher than at other sites. The lower concentrations were found at FK (3.5–10.6 µg m^–3), probably because this site was situated in the city center and crustal sources were relatively scarce (Fig. 1). The trace elements (V, Mn, Ti, Cd, Ni, As, Cu, Pb, and Zn) accounted for a small part of the total detected elements.

The enrichment factors (EFs) relative to the composition of Earth’s upper crust were used to identify anthropogenic influences on the PM-related elements (Rogula-Kozłowska et al., 2012; Zhang et al., 2015), using Al as a reference element (Liu et al., 2016a). The concentrations of elements in the crust refer to their content in the topsoil in China (SEPA, 1990). The EFs were calculated as follows:

where (C_X/C_R)_aerosol is the ratio of X to R in the aerosol and (C_X/C_R)_crust is the ratio of X to R in the crust. The average EFs of elements in PM₁₀ and PM_2.5 at different sampling sites are shown in Fig. 4 and Figs. S5–S6. The EF level of selected trace metals varied similarly at each site. The EF values for As, Ca, Cd, Cu, Ni, Pb, and Zn were > 10, indicating that they were closely associated with human activities such as vehicle emissions and metal manufacturing (Zhang et al., 2010, 2015). The mean EF values for Na and Ti were close to 1 at each site, suggesting that they originated mainly from crustal sources (Xu et al., 2013). The EF values for Ca varied from 14–21 at different sites, which might be affected by diverse sources such as construction activity and crustal sources (Almeida et al., 2015; Liu et al., 2016a). The EF value of K in PM_2.5 was less than 5 at each site, likely due to the influence of crustal sources, while the impact of anthropogenic sources such as biomass burning might be relatively low. 

Fig. 4. Enrichment factors of trace elements in PM₁₀ and PM_2.5 during the heating period.

Source Apportionment

EE Analysis

EE diagnostics are summarized in Tables S14–S15. In moving from two to eight factors, the Q/Q_except of PM₁₀ and PM_2.5 decreased from 8.1 to 3.3 and 9.0 to 3.4, respectively, along with a smaller decrease when moving from five to six factors (4.5 to 4.2 for PM₁₀ and 5.1 to 4.5 for PM_2.5). When changes in Q values become small with increasing factors, it can suggest that there may be too many factors being fit, indicating that five factors can be an optimal solution (Brown et al., 2015). Our results were generally stable at five factors, with all factors mapped in BS in 100% of runs and no swaps occurring with DISP. Thus, five factors (F_peak = 0, seed = 16 for PM₁₀, seed = 2 for PM_2.5) were selected for PMF analysis. The range of all scaled residuals was between –2 and 2. The calculated PM mass concentrations from the PMF model and observed PM mass concentrations showed higher correlations (slope = 0.95 and R² = 0.94 for PM₁₀; slope = 0.96 and R² = 0.95 for PM_2.5) (Fig. S7). 

Source Apportionment

Five factors were chosen using the EE analysis and field investigation results during the heating period (Figs. 5–6), and then the PMF model was run based on the chemical species concentration data in PM₁₀ and PM_2.5. The main species defining Factor 1 were SO₄^2– (49.8–52.1% of the total SO₄^2–), NO₃^– (59.6–77.6%), and NH₄⁺ (71.3–83.2%), so this factor was identified as secondary inorganic aerosol (Wang et al., 2006; Contini et al., 2010; Tao et al., 2013; Liu et al., 2016a). Factor 2 was represented by high loadings of OC (24.1–28.5%), EC (3.8–5.3%), K (58.2–74.7%), and Cl^– (17.7–27.8%), which are generally indicators of biomass burning (Heo et al., 2009; Hleis et al., 2013; Lan et al., 2016), so this factor was identified as biomass burning. The chemical profile of Factor 3 was mainly defined by Si (61.6–70.8%), Ca (39.9–66.7%), Mg (47.6–76.8%), Al (52.5–64.1%), and Fe (33.3–65.5%), which are generally derived from crustal dust (Tullio et al., 2008; Almeida et al., 2015; Liu et al., 2016a), so this factor was identified as crustal dust. Factor 4 was represented by high loadings of As (44.2–49.6%), Cl^– (43.3–43.5%), OC (18.1–19.3%), NO₃^– (16.3–27.3%), SO₄^2– (16.9–24.8%), Si (21.3–27.7%), Ca (31.6–32.1%), Al (13.7–23.3%), and Mg (23.0–36.9%), which are closely related to coal combustion (Bhangare et al., 2011; Cao et al., 2011; Schleicher et al., 2011; Liu et al., 2016a), so this factor was identified as coal combustion. Factor 5 had relatively high loadings of OC (36.2–43.5%), EC (65.9–66.7%), Cu (35.4–40.4%), and Zn (33.0–51.6%); OC and EC are major pollutants from gasoline and diesel combustion (Liu et al., 2016a), while Zn is usually used as an additive in lubricating oil in two-stroke engines and Cu is linked to brake wear (Begum et al., 2004; Canha et al., 2012; Shafer et al., 2012; Lin et al., 2015). Thus, this factor was identified as vehicle exhaust. 

Fig. 5. Factor profile (% of species) of each source for PM₁₀.

Fig. 6. Factor profile (% of species) of each source for PM_2.5.

The source apportionment results for PM₁₀ and PM_2.5 are shown in Fig. 7 and Figs. S8–S9. The contributions of secondary inorganic aerosol, coal combustion, crustal dust, vehicle exhaust, and biomass burning to the ambient PM₁₀ and PM_2.5 were 28–30%, 20–21%, 18–21%, 17–20%, and 4%, respectively. Compared to PM_2.5, the contribution of crustal dust to the ambient PM₁₀ was relatively high, while the contribution of vehicle exhaust to PM₁₀ was relatively low (Fig. 7). The contributions of secondary inorganic aerosol to the ambient PM₁₀ and PM_2.5 showed little difference. The secondary sulfate was mainly derived from the secondary reaction of SO₂ emitted from coal combustion (Russell et al., 1983; Liu et al., 2013), suggesting that the contributions of coal combustion to the ambient PM₁₀ and PM_2.5 were higher than other sources during the heating period, likely due to space heating in winter. The higher contributions of vehicle exhaust were probably associated with the growing number of automobiles in Tian’jin. Some sources (contributions of 8–10%) were not resolved by the PMF, likely because emission sources are complicated and diverse and there is a lack of actual source profiles, especially for open sources such as garbage combustion and other unorganized sources (Liu et al., 2016a). 

Fig. 7. The contribution of each source to PM₁₀ and PM_2.5 at the urban area during the heating period.

Figs. S8–S9 show the results of source apportionment for ambient PM_2.5 and PM₁₀ at different sampling sites during the heating period. The contributions of coal combustion and vehicle exhaust at ADR were lowest, probably because this site was located in scenic and commercial area (Table S1). The contributions of coal combustion and crustal dust at NK were the highest (22–23% and 21–25%, respectively). This site was situated in an industrial area with ongoing construction activity (Table S1). The contributions of vehicle exhaust at QJ and FK were higher than at other sites, likely because these were located in traffic, residential, and scenic areas (Table S1) with higher traffic volume. The contributions of crustal dust at QJ and FK were significantly lower than at other sites. The contributions of secondary inorganic aerosol showed minor differences between sites, with higher contributions at QJ and lower ones at DZG. The contributions of biomass burning were apparently lower than other sources at most sites; the contribution at DZG was significantly higher than at other sites, whose contributions did not differ much. 

Similar studies have recently been conducted in many other Chinese cities. For example, Qiu et al. (2016) found that dust (26.9–29.8%), secondary aerosols (26.9–28.3%), coal burning (15.7–17.3%), traffic (15.1–16.7%), and industry (8.2–9.0%) were major contributors to the ambient PM₁₀ and PM_2.5 in Lan’zhou. Fang et al. (2017) found that resuspended dust (17.5–35.0%), vehicle exhaust (14.9–23.6%), and secondary particulates (20.4–28.8%) were major sources for PM_2.5 and PM₁₀ in Hai’kou. Gu et al. (2014) reported that coal combustion and biomass burning (38.0–46.2%), secondary sulfate (35.0–36.9%), and crustal dust (10.2–16.9%) were major sources for PM_2.5 in Ji’nan. Zhang et al. (2017) summarized and analyzed receptor-based source apportionment research on fine-particulate matter in China, finding that secondary inorganic aerosol and traffic emissions had higher contributions in South China while the percentages contributed by coal combustion, dust, and biomass burning to the total PM_2.5 were higher in North China. Although our results were broadly consistent with Zhang et al. (2017), major sources and contributions of ambient particulate matter still differ between cities, likely due to different economic development patterns, individual industrial and energy structures, and patterns of human activity (Liu et al., 2016a, 2017a).

Backward Trajectory and PSCF Analysis

We used backward trajectory analysis to reveal the transport pathways of the air masses affecting Tian’jin (Fig. 8) and used the PSCF model to identify potential source areas of PM_2.5 and PM₁₀ (Fig. S10). In terms of direction and travelled region, all trajectories were divided into three groups during the heating period. Trajectory Clusters (1), (2), and (3) accounted for 40.0%, 47.5%, and 12.5% of the total trajectories, respectively. Clusters (1) and (2) represented long-range transport patterns, being mainly derived from Mongolia before crossing Inner Mongolia, He’bei, and Bei’jing to reach Tian’jin. Cluster (3) represented a shorter-range transport pattern, mainly originating in Inner Mongolia before passing through Shan’xi and He’bei Provinces to reach Tian’jin. Overall, Tian’jin was mainly influenced by the transport of air masses originating to the northwest during the heating period. Northeastern He’nan, southwestern Shan’dong, Bei’jing, and Tian’jin itself were identified as the potential source areas of PM₁₀ and PM_2.5 using the PSCF model (Fig. S10). 

Fig. 8. Analytical results of the 72-h air mass back trajectories at 100 m elevation during the heating period. Three trajectories were calculated based on the trajectory clustering results at Tian’jin.

CONCLUSIONS

The mean PM₁₀ and PM_2.5 concentrations in Tian’jin, China, during the 2016 heating season were 98 and 71 µg m^–3, and the mean PM_2.5/PM₁₀ ratio was 0.67. The concentrations for both fractions exhibited significant spatial differences, with standard deviations of 6–12 µg m^–3. NO₃^– and OC were dominant in the ambient PM_2.5 and PM₁₀, with concentrations and mass percentages significantly exceeding those of the other components. The average SNA concentrations were 19.9–23.4 µg m^–3, accounting for 72.4–77.1% of the total measured ions. The SO₄^2–/NO₃^– ratio showed a decreasing tendency as the particle concentrations increased, implying a strong influence from mobile sources. The mean OC/EC ratios for PM₁₀ and PM_2.5 were 3.1 and 3.2, respectively, with small spatial differences. The most abundant elements were crustal and were mainly enriched in the PM₁₀.

Based on EE diagnostics, five factors for the PM_2.5 and PM₁₀ were selected via PMF analysis: secondary inorganic aerosol, coal combustion, crustal dust, vehicle exhaust, and biomass burning, which contributed 28–30%, 20–21%, 18–21%, 17–20%, and 4%, respectively. Secondary inorganic aerosol displayed less spatial heterogeneity than the other sources in its contributions. Backward trajectory and PSCF analysis showed that air masses affecting Tian’jin mainly originated in the northwest during the heating period, and northeastern He’nan, southwestern Shan’dong, Bei’jing, and Tian’jin itself were major potential source areas.

ACKNOWLEDGMENTS

This study was financially supported by the National Key Research and Development Program of China (2016YFC0208501), Tianjin Science and Technology Foundation (16YFZCSF00260), the Tianjin Science and Technology Plan Program (18ZXSZSF00160), and Fundamental Research Funds for the Central Universities. The authors thank Tian’jin Environment Monitoring Center for their participation in the sampling campaign.