High-time-resolution chemical composition and source apportionment of PM_2.5 in northern Chinese cities: implications for policy

Fine particulate matter (PM_2.5) pollution is still one of China's most important environmental issues, especially in northern cities during wintertime. In this study, intensive real-time measurement campaigns were conducted in Xi'an, Shijiazhuang, and Beijing to investigate the chemical characteristics and source contributions of PM_2.5 and explore the formation of heavy pollution for policy implications. The chemical compositions of PM_2.5 in the three cities were all dominated by organic aerosol (OA) and nitrate (NO${}_{\mathrm{3}}^{-}$). Results of source apportionment analyzed by a hybrid environmental receptor model (HERM) showed that the secondary formation source contributed more to PM_2.5 compared to other primary sources. Biomass burning was the dominant primary source in the three pilot cities. The contribution of coal combustion to PM_2.5 is non-negligible in Xi'an and Shijiazhuang but is no longer an important contributor in the capital city of Beijing due to the execution of a strict coal-banning policy. The potential formation mechanisms of secondary aerosol in the three cities were further explored by establishing the correlations between the secondary formation sources and aerosol liquid water content (ALWC) and O_x (O₃+NO₂), respectively. The results showed that photochemical oxidation and aqueous-phase reaction were two important pathways of secondary aerosol formation. According to source variations, air pollution events that occurred in campaigns were classified into three types: biomass-combustion-dominated, secondary-formation-source-dominated, and a combination of primary and secondary sources. Additionally, this study compares the changes in chemical composition and source contributions of PM_2.5 in past decades. The results suggest that the clean-energy replacements for rural households should be urgently encouraged to reduce the primary source emissions in northern China, and collaborative control on ozone and particulate matter needs to be continuously promoted to weaken the atmosphere oxidation capacity for the sake of reducing secondary aerosol formation.

1 Introduction

Fine particulate matter (PM_2.5, aerodynamic diameter ≤2.5 µm) is of large concern because of its adverse effects on both the natural environment (Kuniyal and Guleria, 2019; Kuo et al., 2013) and human health (Pöschl, 2005; Shen et al., 2021; Zeng and He, 2019). With the soaring economic growth and urbanization in China, PM_2.5 pollution has been a very serious environmental issue in recent decades (Chan and Yao, 2008; He et al., 2002; Pui et al., 2014; Zhang et al., 2013). The most impressive case is that an extremely severe haze pollution episode occurred in eastern and central China in January 2013, with a peak PM_2.5 concentration value of over 500 µg m⁻³. This month had been reported as the haziest month in the past 60 years in Beijing, China (Wang et al., 2014; Huang et al., 2014). Thereafter, aiming to improve air quality, the Chinese central government issued the Air Pollution Prevention and Control Action Plan (APCAP) in September 2013 (http://www.gov.cn/zwgk/2013-09/12/content_2486773.htm, last access: 12 September 2013; in Chinese) and the Three-year Action Plan to Fight Air Pollution (TAPFAP) in June 2018 (http://www.gov.cn/zhengce/content/2018-07/03/content_5303158.htm, last access: 3 July 2018; in Chinese). With the implementation of strict pollution controls, air quality in northern China has improved significantly over the past decade (Wang et al., 2020a, 2017; Li et al., 2020). Previous studies show that PM_2.5 concentration decreased notably in the past 2 decades, and the composition of organic aerosol (OA), black carbon (BC), and sulfate (SO${}_{\mathrm{4}}^{\mathrm{2}-}$) decreased as well, while ammonium (NH${}_{\mathrm{4}}^{+}$) slightly increased, and nitrate (NO${}_{\mathrm{3}}^{-}$) increased obviously. In terms of PM_2.5 sources, the contribution of secondary sources increased obviously, while the contribution of industrial emissions and coal combustion decreased due to elimination of industries and enterprises with high pollutant emissions, promotion of desulfurization in industrial facilities, replacement of clean energy, and optimization of industrial and energy structures (Lu et al., 2021; Ma et al., 2022; Tao et al., 2017; Y. Wang et al., 2019). However, there is still a significant gap between the PM_2.5 concentration in northern China and its latest recommendations on air quality guidelines (5 µg m⁻³) by the World Health Organization (https://apps.who.int/iris/bitstream/handle/10665/345329/9789240034228-eng.pdf, last access: 26 November 2018; p. 78). In addition, severe PM_2.5 pollution still frequently occurred in northern China during wintertime (Guo et al., 2021; J. Li et al., 2017; X. Li et al., 2021). To figure out the causes behind the pollution and further improve air quality in northern China, it is essential to use online high-time-resolution source apportionment technology to understand the chemical composition and source contribution of PM_2.5 in those pollution events.

Recently, more research on measurements of PM_2.5 and its source apportionments were conducted using online high-time-resolution technologies (Y. Li et al., 2017; F. Wang et al., 2021; Elser et al., 2016). Compared to traditional offline filter-based approaches, online methods characterize the short-time variation in PM_2.5. This allows the rapid changes in and evolutions of chemical components to be distinguished and is particularly beneficial for the purpose of gaining knowledge about the formations of heavy air pollution (Liu et al., 2016; Ouyang et al., 2019; Zheng et al., 2016; Elser et al., 2016). For instance, Lv et al. (2021) employed a positive matrix factorization (PMF) model with high-time-resolution online PM_2.5 data to accurately quantify and distinguish the source distributions in Beijing during two haze episodes in January 2019. Liu et al. (2019) recognized the main drivers of haze events that occurred in Beijing in winter of 2016 according to high-time-resolution source apportionment of PM_2.5 with multiple models. Furthermore, M. Wang et al. (2021) found that vehicle emissions contributed most to PM_2.5 during pollution episodes in downtown Lanzhou based on high-resolution online data source apportionment. Currently, to fully understand and solve heavy-pollution events in winter that trouble local governments in northern cities of China (Z. Wang et al., 2022; Xu et al., 2022; Zhou et al., 2022), more advanced online measurement and source apportionment are better choices (Tao et al., 2015). It should be pointed out that previous research has mainly focused on individual cities, and those results have some limitations in guiding the improvement of air quality in the entire northern region of China. Therefore, it is necessary to conduct comparative research among multiple cities.

Considering the differences in geographical location, population, economy, industry and energy structure, air quality, and depth of air pollution control measures among different cities, three cities in northern China (Beijing, Shijiazhuang, and Xi'an) were chosen as pilot research subjects. The cities of Beijing and Shijiazhuang are located in the North China Plain, which is one of the most polluted regions in China (Chan and Yao, 2008). Beijing is the capital of China, and its air quality has significantly improved under the implementation of the strictest clean-air policy since 2013 (J. Li et al., 2021; Pang et al., 2021; Vu et al., 2019; Zhang et al., 2020). However, the city is still plagued by pollution events in wintertime (Wang et al., 2020b; X. Yang et al., 2022; Zhou et al., 2022). Shijiazhuang has been recognized as one of the cities with the most serious air pollution worldwide (B. Liu et al., 2018; R.-J. Huang et al., 2019). Its air quality had also improved under the implementation of the Clean Air Plan, whereas its annual PM_2.5 concentration was still unable to meet China's National Ambient Air Quality Standards (NAAQS-II) of 35 µg m⁻³ until 2021 (Fig. S1 in the Supplement). Xi'an is located in the Fenwei Plain, which is a region that has suffered from heavy pollution and was designated as a key region for the TAPFAP in 2018 (Cao and Cui, 2021). Compared with Beijing and Shijiazhuang, high-intensity air pollution controls in Xi'an started late due to a lack of financial support. And the annual PM_2.5 concentration in Xi'an could not meet the NAAQS-II until 2021 as well (Fig. S1). Meanwhile, it is still unclear whether the actual causes of the pollution are topography, meteorological conditions, or local emissions (Chen et al., 2021; Tian et al., 2022; Wang et al., 2015; Z. Wang et al., 2022). In this study, we conducted intensive real-time observation of PM_2.5 chemical components in Xi'an, Shijiazhuang, and Beijing during wintertime. The objectives were (1) to determine the characteristics of PM_2.5 and its chemical components in the three typical northern Chinese cities during wintertime, (2) to quantify the source contribution and explore the potential formation mechanism of secondary aerosols, (3) to explore the unique causes of heavy-pollution events in different cities, and (4) to provide suggestions for the establishment of efficient policies for continuous air quality improvement. This study provides scientific guidance for developing policy on air quality improvement for northern Chinese cities.

2 Methods

2.1 Sampling sites and periods

In this study, intensive online measurements of PM_2.5 were conducted in the three pilot cities of Xi'an, Shijiazhuang, and Beijing during wintertime (Fig. 1). The sampling sites in Xi'an and Beijing are located at two Chinese Academy of Sciences (CAS) stations. The one in Xi'an is the Guanzhong Plain Ecological Environment Change and Comprehensive Treatment National Observation and Research Station, Institute of Earth Environment (IEE) (34.24^∘ N, 108.87^∘ E), and the one in Beijing is the Tower Branch of the Institute of Atmospheric Physics (IAP) (39.98^∘ N, 116.39^∘ E). Both sites are surrounded by commercial and residential buildings without intense industrial emissions nearby. Previous studies indicated that these two sites were influenced by biomass and coal burning for heating and cooking during wintertime as well as usual local traffic emissions (Tian et al., 2021; Xu et al., 2021). The sampling site in Shijiazhuang is situated in the courtyard of Hebei Sailhero Environmental Protection High-tech Co., Ltd. (38.04^∘ N, 114.65^∘ E), which is surrounded by pharmaceutical and machine-building industries and close to the streets. The intensive campaigns were continuously conducted for ∼1 month in each city (i.e., 12 December 2020 to 7 January 2021 in Xi'an, 20 December 2021 to 24 January 2022 in Shijiazhuang, and 17 January to 20 February 2021 in Beijing).

Figure 1 Chemical composition and source apportionment results of PM_2.5 in the three pilot cities of northern China during the sampling period.

2.2 Online measurements of PM_2.5 chemical components

2.2.1 Organic aerosol and inorganic ions

Concentrations of OA, NO${}_{\mathrm{3}}^{-}$, SO${}_{\mathrm{4}}^{\mathrm{2}-}$, ammonium (NH${}_{\mathrm{4}}^{+}$), and chloride (Cl⁻) in PM_2.5 at a 15 min time resolution were monitored by a quadrupole aerosol chemical speciation monitor (Q-ACSM; Aerodyne Research Inc., Billerica, MA, USA) equipped with a PM_2.5 lens. The detailed operational principles and calibration method of the Q-ACSM are described elsewhere (Ng et al., 2011; Hu et al., 2017). First, the sampled ambient airstream passed through a PM₁₀ impactor inlet and a Nafion^® dryer (MD-700-24F-3, Perma Pure, Inc., Lakewood, NJ, USA) with a flow rate of 5 L min⁻¹ before entering the Q-ACSM chamber. Then, the pre-treatment particles passed through a 100 µm critical orifice at 0.1 L min⁻¹ and were focused into a narrow beam by an aerodynamic intermediate-pressure lens. The focused particle beam was flash-vaporized by a capture vaporizer (CV) at ∼600 ^∘C. The vaporized compounds were then ionized by an electron impactor (EI) ionization source at 70 eV and subsequently analyzed by the quadrupole mass spectrometer.

The calibration system consists of an atomizer (Model 9302, TSI Inc., Shoreview, MN, USA), a differential mobility analyzer (DMA; TSI model 3080, TSI Inc.), and a condensation particle counter (CPC; TSI model 3772, TSI Inc.), and ammonium nitrate (NH₄NO₃) and ammonium sulfate ((NH₄)₂SO₄) aerosols were used for calibration. The raw data of the Q-ACSM were analyzed by the ACSM local tool (V1.5.3.5, Aerodyne Research Inc., Billerica, Massachusetts, USA) compiled with Igor Pro 6.37 (WaveMetrics, Lake Oswego, OR, USA). The response factors (RFs) for NO${}_{\mathrm{3}}^{-}$ in Xi'an, Shijiazhuang, and Beijing were set at $\mathrm{2.03}\times {\mathrm{10}}^{-\mathrm{11}}$, $\mathrm{5.9}\times {\mathrm{10}}^{-\mathrm{11}}$, and $\mathrm{2.20}\times {\mathrm{10}}^{-\mathrm{11}}$, respectively, and the relative ionization efficiencies (RIEs) for NH${}_{\mathrm{4}}^{+}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ were set at 8.06 and 0.83 in Xi'an, 5.82 and 0.30 in Shijiazhuang, and 6.31 and 0.38 in Beijing, respectively. Other RIEs for NO${}_{\mathrm{3}}^{-}$, OA, and Cl⁻ were set at default values of 1.4, 1.1, and 1.3, respectively (Ng et al., 2011). In addition, the collection efficiency (CE) value of the Q-ACSM equipped with a PM_2.5 lens was recommended as 1 based on laboratory simulation experiments by Xu et al. (2017). Finally, the chemical components monitored by the Q-ACSM were corrected by the results of offline filter sampling experiments during the same periods (Fig. S2).

2.2.2 Black carbon

BC concentration in PM_2.5 was obtained by an Aethalometer (Model AE33, Magee Scientific Inc., Berkeley, CA, USA) with a 1 min time resolution. The AE33 monitors the light attenuation of seven wavelengths (λ= 370, 470, 525, 590, 660, 880, and 940 nm), and the light attenuation at λ=880 nm was used to calculate BC concentration (Q. Wang et al., 2019; Drinovec et al., 2015). Briefly, the ambient air was first sampled on a filter tape inside the instrument through a PM_2.5 cyclone (SCC-1.829, BGI Inc., USA) at a flow rate of 5 L min⁻¹. The entering particles were divided into two sample spots on the filter through two channels with different flows. Then the light attenuation transmitted through two parallel spots was detected. For high-quality monitoring accuracy, the sampled particles were desiccated with a Nafion^® dryer (MD-700-24F-3, Perma Pure, Inc., Lakewood, NJ, USA) before entering the AE33. Furthermore, a real-time loading effect compensation algorithm based on two spot measurements was used to eliminate the nonlinear loading effects of the Aethalometer. A detailed description of the Model AE33 principle can be found in Drinovec et al. (2015).

2.2.3 Elements

A total of 23 elements, including Si, K, Ca, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Ga, As, Se, Ag, Cd, Sn, Ba, Au, Hg, Th, Pb, and Pd in PM_2.5, were analyzed by an Xact625 ambient-metal monitor (Cooper Environmental Services, Tigard, Oregon, USA) with a 1 h time resolution. Si, K, Ca, Cr, Mn, Fe, Ni, Cu, Zn, As, Se, Ba, and Pb were selected for further analysis in Xi'an and Beijing, while other elements were excluded due to most of their concentration being below the method's detection limit. In Shijiazhuang, S, Cl, and Ti were analyzed by replacement of Ga, Ag, and Au, respectively. Finally, Si, K, Ca, Ti, Cr, Mn, Fe, Ni, Cu, Zn, As, Se, Ba, and Pb were selected for further analysis. The description and detection principles of Xact625 were introduced by Furger et al. (2020) and Rai et al. (2020). In brief, the ambient airstream was first sampled on a Teflon filter tape inside the instrument through a PM_2.5 cyclone inlet at a constant flow rate of 16.7 L min⁻¹, and then the sample was automatically analyzed by nondestructive energy-dispersive X-ray fluorescence (XRF) to determine the mass of the species. For quality control and assurance, the Xact625 performed automatic internal quality control by testing the Pd rod every hour to ensure the stability of the instrument. Energy calibration was performed daily from 00:00 to 00:15 and a range calibration from 00:15 to 00:30 local standard time (LST) to monitor any possible shift in and instability of the XRF (Liu et al., 2019). During our sampling periods, the concentration of Pd varied within 3 standard deviations (Fig. S3), illustrating the reliable and stable performance of the Xact625.

2.2.4 Complementary data

Online hourly concentrations of PM_2.5 and gas pollutants (i.e., NO_x, NO₂, CO, SO₂, and O₃) were acquired from the national air quality monitoring station (https://air.cnemc.cn:18007/, last access: 17 August 2023). Meteorological parameters, including wind speed (WS), wind direction (WD), relative humidity (RH), and temperature (T), were obtained from the national meteorological station (http://data.cma.cn/, last access: 30 September 2022). Detailed information on complementary data is listed in Table S1 in the Supplement.

2.3 Data analysis

2.3.1 PM_2.5 mass reconstruction

A chemical closure was utilized to assess whether chemical compositions can be representative of PM_2.5. The sum of OA, NO${}_{\mathrm{3}}^{-}$, SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NH${}_{\mathrm{4}}^{+}$, Cl⁻, BC, mineral dust (MD), and trace elements (TEs) was considered to be the reconstructed PM_2.5, where MD and TEs were calculated as follows (Chow et al., 2015):

$$\begin{array}{}\text{(1)}& {\displaystyle }\begin{array}{rl}\left[\text{MD}\right]& =\mathrm{2.20}\times \left[\mathrm{Al}\right]+\mathrm{2.49}\times \left[\mathrm{Si}\right]+\mathrm{1.63}\times \left[\mathrm{Ca}\right]\\ & +\mathrm{2.42}\times \left[\mathrm{Fe}\right]+\mathrm{1.94}\times \left[\mathrm{Ti}\right]\end{array}\text{(2)}& {\displaystyle }\begin{array}{rl}\left[\text{TE}\right]& =\left[\mathrm{K}\right]+\left[\mathrm{Cr}\right]+\left[\mathrm{Mn}\right]+\left[\mathrm{Ni}\right]+\left[\mathrm{Cu}\right]+\left[\mathrm{Zn}\right]\\ & +\left[\mathrm{As}\right]+\left[\mathrm{Se}\right]+\left[\mathrm{Ba}\right]+\left[\mathrm{Pb}\right],\end{array}\end{array}$$

where the square brackets represent the chemical species concentration; [Al] and [Ti] were calculated by the concentration of Ca ($\left[\mathrm{Al}\right]=\mathrm{4.3}\times \left[\mathrm{Ca}\right]$ and $\left[\mathrm{Ti}\right]=\mathrm{0.25}\times \left[\mathrm{Ca}\right]$) (Wei et al., 1991). Good correlations between online and reconstructed PM_2.5 mass (slope = 0.87–1.10, R²= 0.82–0.93) in the three pilot cities (Fig. S4) indicated that our measurements could detect major components of PM_2.5. The PM_2.5 concentration used in the following discussion refers to the reconstructed PM_2.5 concentration.

2.3.2 Hybrid environment receptor model

Source apportionment of PM_2.5 was analyzed with a bilinear model referred to as a hybrid environment receptor model (HERM). HERM was developed by the IEE, CAS, and the University of Nevada, Las Vegas (Chen and Cao, 2018). Like other receptor models, the speciation of pollutants at a receptor site can be separated into emission sources and the chemical compositions of the sources. To solve the mass balance of PM_2.5, the bilinear HERM in matrix notation is defined as follows:

$$\begin{array}{}\text{(3)}& C_{mn}={\sum }_{i=\mathrm{1}}^I{F}_{mi}{G}_{in}+Q_{mn},\end{array}$$

where C_mn is the measured concentration of chemical species m during time n; F_mi is the source profile, which is the fractional quantity of species m in source i emissions; G_in represents the contribution of source i during time n; and Q_mn is the model residual for species m concentration measured during time n. Based on an iterative conjugate gradient algorithm, the HERM solves G_in and unknown F_mi by minimizing the Q_mn, which is defined as follows:

$$\begin{array}{}\text{(4)}& Q_{mn}={\sum }_{m=\mathrm{1}}^M{\sum }_{n=\mathrm{1}}^N{\displaystyle \frac{{\left(C_{mn}-{\sum }_{i=\mathrm{1}}^I{F}_{mi}{G}_{in}\right)}^{\mathrm{2}}}{{\mathit{\sigma }}_{C_{mn}}^{\mathrm{2}}+{\sum }_{i=\mathrm{1}}^I\left({\mathit{\sigma }}_{F_{mi}}^{\mathrm{2}}{G}_{in}^{\mathrm{2}}+{\mathit{\delta }}_{mi}{\mathit{\sigma }}_{C_{mn}}^{\mathrm{2}}\right)}},\end{array}$$

where M, N, and I are the number of samples, chemical species, and sources, respectively; ${\mathit{\sigma }}_{F_{mi}}$ represents the error in the variability in the constrained factor profile. δ_mi was set to 0 or 1 depending on whether the ith factor profile is constrained or unconstrained, respectively.

The HERM input data included the concentration and uncertainty data of chemical species. A total of 19 chemical species in Xi'an and Shijiazhuang and 20 chemical species in Beijing were selected for source apportionment, respectively. Details of selected chemical species and their uncertainty calculation are described in Sect. S1 in the Supplement. A range of 2- to 10-factor solutions were investigated by HERM with completely unconstrained factor profiles to search for optimal solutions. The detailed diagnostics can be seen in Sect. S2. A six-factor solution for Xi'an and Shijiazhuang and an eight-factor solution for Beijing were found to be the optimal solutions based on multiple criteria, including (1) variations in $Q/Q_{\exp }$, which can be used to choose the optimal number of resolved factors; (2) physical meaningfulness of distinct factor profiles and explained variation (EV) values of variables; (3) good correlations between source contribution and external and internal tracers; and (4) agreement between the measured and modeled PM_2.5 mass. More detailed information on the final selected factor profiles and contributions is presented in Sect. 3.2.

2.3.3 Aerosol liquid water content

Aerosol liquid water content (ALWC) was calculated by the ISORROPIA II thermodynamic equilibrium model (http://isorropia.eas.gatech.edu/, last access: 17 August 2023) based on data of PM_2.5 chemical species (including NO${}_{\mathrm{3}}^{-}$, SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NH${}_{\mathrm{4}}^{+}$, and Cl⁻) and meteorological parameters, including relative humidity (RH) and temperature (T); more model information can be found in Fountoukis and Nenes (2007). It should be noted that the ISORROPIA II model does not consider the contribution of the organic aerosols, as inorganic aerosols are the most hygroscopic species and the most dominant contributor to ALWC (Huang et al., 2020).

3 Results and discussion

3.1 Characteristics of PM_2.5 and its chemical components

Figure 1 illustrates the mass composition of PM_2.5 in the three pilot cities during the sampling periods, and their concentration levels are summarized in Table S4. The average PM_2.5 concentrations in Xi'an, Beijing, and Shijiazhuang were 77±47, 64±57, and 60±39 µg m⁻³, respectively. It is noted that the average PM_2.5 concentrations in Xi'an, Beijing, and Shijiazhuang did not meet the second level of the NAAQS, indicating that it is necessary to establish more particular and efficient pollution reduction measures. As shown in Fig. 1, the chemical compositions of PM_2.5 were similar in Beijing and Shijiazhuang (Fig. 1b and c) and were mainly composed of OA (26.9 %–34.2 %), followed by NO${}_{\mathrm{3}}^{-}$ (23.6 %–26.5 %), SO${}_{\mathrm{4}}^{\mathrm{2}-}$ (11.8 %–15.0 %), NH${}_{\mathrm{4}}^{+}$ (11.8 %–14.8 %), MD (7.4 %–10.1 %), BC (2.9 %–6.5 %), and Cl⁻ (1.1 %–4.8 %). However, in Xi'an, MD contributed comparatively more to PM_2.5 (17.3 %), while SO${}_{\mathrm{4}}^{\mathrm{2}-}$ had a smaller contribution (6.8 %). This could be explained by more construction activity and MD transport from the Loess Plateau to Xi'an (Long et al., 2016; Yan et al., 2015). Meanwhile, the lowest sulfur oxidation ratio (SOR) was observed in Xi'an (0.18±0.08; see Table S5), indicating weak efficiency of the second generation of SO${}_{\mathrm{4}}^{\mathrm{2}-}$. The sum of SO${}_{\mathrm{4}}^{\mathrm{2}-}$, NO${}_{\mathrm{3}}^{-}$, and NH${}_{\mathrm{4}}^{+}$ accounted for 39.0 %–53.0 % of PM_2.5 in the three pilot cities, highlighting the importance of the secondary inorganic components in northern China. In addition, the fractions of BC, Cl⁻, and TE in PM_2.5 were lower in Beijing than those in the other two cities, which can be explained by the stricter local control policies on solid fuel combustion and tightening of industrial emission standards in and near the capital city of China (J. Li et al., 2021; Pang et al., 2021).

To better understand the impact of the chemical components, the mass fraction of each component was plotted as a function of the PM_2.5 mass concentration (Fig. 2a–c). The two dominant components of PM_2.5 were OA (25.7 %–38.0 %) and MD (19.9 %–37.1 %), while the PM_2.5 concentrations were below 40 µg m⁻³. The fraction of OA in PM_2.5 was the highest in Shijiazhuang and Beijing, while MD contributed most to PM_2.5 in Xi'an. This is potentially related to more emissions and higher backgrounds of local dust. With increases in the PM_2.5 mass concentration, the fractions of chemical components in Xi'an and Shijiazhuang changed notably. The fractions of OA and NO${}_{\mathrm{3}}^{-}$ increased the most and reached peaks of 40.1 % and 28.7 %, respectively, when the PM_2.5 concentration reached ∼ 196 µg m⁻³ in Xi'an. In contrast, NO${}_{\mathrm{3}}^{-}$ and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ were two dominant drivers of increasing PM_2.5 concentrations in Shijiazhuang, showing peak contributions of 32.5 % and 18.7 %, respectively, when the PM_2.5 concentration was over 100 µg m⁻³. Compared to Xi'an and Shijiazhuang, Beijing had relatively stable fractions of each chemical component, with increasing PM_2.5 concentrations. Particularly, the fractions of OA and NO${}_{\mathrm{3}}^{-}$ contributed dominantly, with averages of 33.3±3.0 % and 25.3±2.5 %, respectively, when the PM_2.5>40 µg m⁻³.

Figure 2 Mass fractions of chemical components (a–c) and source contribution (d–f) with reconstructed PM_2.5 concentration in Xi'an, Shijiazhuang, and Beijing.

3.2 Source apportionment of PM_2.5

Six potential sources, including biomass burning, fugitive dust, industrial emissions, coal combustion, vehicle emissions, and secondary formation sources, were resolved by the HERM analysis. In Beijing, secondary formation sources were further divided into secondary nitrate plus OA and secondary sulfate plus OA. Fireworks as a special pollution source were separated due to the Chinese Spring Festival (from New Year's Eve to 3 January on the lunar calendar). Figures S6–S8 present the source profiles and contributions in Xi'an, Shijiazhuang, and Beijing, respectively. Biomass burning features high EV for the two tracers Cl⁻ (33 %–58 %) and K (30 %–44 %) in the three cities (Ni et al., 2017; Zhao et al., 2021). The fugitive dust is characterized by high EV values for Si (60 %–90 %) and Ca (34 %–54 %), which are the dominant chemical species in the fugitive dust profiles in northern China (Shen et al., 2016; Zhao et al., 2006). The fractions of industrial emissions vary among the cities, showing high EV for Ni (55 % and 87 %) and Cr (25 % and 70 %) in Xi'an and Shijiazhuang and high EV for Cr (26 %), Mn (40 %), and Pb (27 %) in Beijing. Ni is possibly emitted from the semiconductor industry (Simka et al., 2005). Cr, Mn, and Pb could originate from steel manufacturing and incinerator fly ash (Duan and Tan, 2013; Ledoux et al., 2017). Coal combustion is characterized by high EV values for As (38 %–75 %), Se (40 %–50 %), and Pb (31 %–57 %). These elements are enriched in coals, which are reliable indicators of coal combustion (Tian et al., 2013; Xu et al., 2012). The non-exhaust vehicle emissions could be identified by the elements Ba, Cu, Ca, Fe, and Mn. Cu and Ba can be released from brake and tire wear of vehicles (Adachi and Tainosho, 2004; Thorpe and Harrison, 2008). Moreover, Fe and Mn could be emitted from the combustion of lubricating oil and fuel additives (Ålander et al., 2005; Lewis et al., 2003). Relatively high EV values for Ba (68 %), Cu (36 %), and Ca (35 %) are seen in Xi'an; significantly high EV values of Mn (68 %), Fe (65 %), Cu (53 %), and Ba (80 %) are characterized in Shijiazhuang; and relatively high EV values of Fe (34 %) and Cu (39 %) are featured in Beijing. Moreover, moderate EV values for BC (18 %–27 %) and OA (13 %–22 %) are commonly regarded as contributions of vehicle engine exhaust, while the temporal variations in the source concentration are well correlated with gaseous NO_x or NO₂ in the three cities (R²= 0.45–0.78), which is a good tracer of traffic-related emissions (Huang et al., 2017; L. Li et al., 2017). The secondary sources resolved by HERM are different among the three cities. In Xi'an and Shijiazhuang, this factor is characterized by high EV values for SO${}_{\mathrm{4}}^{\mathrm{2}-}$ (62 %–75 %), NO${}_{\mathrm{3}}^{-}$ (55 %–53 %), and NH${}_{\mathrm{4}}^{+}$ (60 %–56 %) and a medium EV value for OA (23 %–29 %), which showed good correlations with SO${}_{\mathrm{4}}^{\mathrm{2}-}$ (R²= 0.85–0.90) and NO${}_{\mathrm{3}}^{-}$ (R²= 0.85–0.92) (Dai et al., 2020; Tian et al., 2022). In addition, the OA concentration of this factor was calculated by the EV value of OA, which was close to the secondary OA (SOA) concentration estimated by the BC-trace method (see Sect. S3 and Table S6). This means that SOA was mixed in this factor; therefore, this factor was identified as a secondary formation source. In Beijing, two secondary sources were resolved. The first one was characterized by a high EV value for NO${}_{\mathrm{3}}^{-}$ (58 %) and NH${}_{\mathrm{4}}^{+}$ (42 %) and medium values for OA (21 %); another one was characterized by a high EV value for SO${}_{\mathrm{4}}^{\mathrm{2}-}$ (58 %) and medium values for OA (16 %) and NH${}_{\mathrm{4}}^{+}$ (30 %). The OA concentration in those two factors was also comparable to that estimated by the BC-trace method (see Sect. S3). So those two sources were identified as secondary nitrate plus OA and secondary sulfate plus OA, respectively. The combination of secondary nitrate plus OA and secondary sulfate plus OA is equivalent to the secondary formation sources for the next discussion. Additionally, the source of firework emissions is characterized by high EV values of Ba (83 %), Cu (45 %), and K (38 %), which are recognized as common indicators of fireworks (Rai et al., 2020; Tian et al., 2014).

The modeled PM_2.5 mass was well correlated with the reconstructed PM_2.5 mass (R²=0.99, slope = 0.90–1.01; Fig. S10) in the three pilot cities, indicating that the established models are reasonable. As shown in Fig. 1d and e, the contributions of primary sources (i.e., the sum of biomass burning, fugitive dust, industrial emissions, coal combustion, and vehicle emissions) in PM_2.5 were significantly higher than those of the secondary formation sources in Xi'an and Shijiazhuang, indicating that the PM_2.5 in these two cities is mainly influenced by the primary source emissions during wintertime. Particularly, biomass burning and coal combustion were two dominant contributors to PM_2.5, with contributions of 24.6 % and 15.1 %, respectively, in Xi'an and 24.4 % and 16.0 %, respectively, in Shijiazhuang. These suggest that controls of solid fuel combustion are critical to reducing PM_2.5 pollution in these cities. In contrast, the contribution of secondary formation sources to PM_2.5 in Beijing was highly dominant (>50 %), potentially attributed to the strict control of primary emissions under the execution of a series of pollution control policies (Lv et al., 2016; Pang et al., 2021) and more regional transportation of secondary pollutants (Liu et al., 2019; Wang and Zhao, 2018). Among the primary sources, the contributions of biomass burning and vehicle emissions were only 18.4 % and 11.3 %, respectively, further reflecting the benefits of reductions in all primary emissions. Due to the Chinese Spring Festival, the contribution of fireworks (7.9 %) to PM_2.5 ranked second in primary sources (Fig. S11), which indicates that more refined control schemes need to be encouraged to deal with such special events in the future. It should be noted that the contribution of fugitive dust was lower than the fraction of mineral dust in the three pilot cities (Fig. 1). This is because fugitive dust defined here mainly refers to road and construction dust emissions, while mineral dust represents material assumed to be oxides of mineral elements such as Al, Si, Ca, Ti, and Fe (Chow et al., 2015). These mineral elements in PM_2.5 come from more emission sectors, including industry, the Earth's crust, transportation, construction, and combustion (J. Liu et al., 2018; Lu et al., 2014; Pant and Harrison, 2013; Shen et al., 2016).

Figure 2d–f show variations in source contribution with increases in PM_2.5 mass concentrations in the three pilot cities. The two most dominant sources were secondary formation sources (32.1 %) and fugitive dust (31.4 %) in Xi'an, coal combustion (24.9 %) and vehicle emissions (21.3 %) in Shijiazhuang, and secondary formation sources (24.3 %) and fugitive dust (23.8 %) in Beijing when the PM_2.5 mass concentration was <40 µg m⁻³. In Xi'an, when the PM_2.5 mass concentrations exceeded 180 µg m⁻³, the contribution of biomass burning rose the most and reached a peak of 38.4 %, demonstrating that biomass burning plays an important role in the worsening of air quality in Xi'an. In contrast, the contributions of secondary formation sources increased the most in comparison to other sources in Shijiazhuang and Beijing, indicating that the PM_2.5 pollution was mainly dominated by secondary aerosol formation during the wintertime. And the peak contributions of secondary nitrate and sulfate were 66.5 % and 74.7 %, while the PM_2.5 mass concentration increased to 113 and 223 µg m⁻³ in Shijiazhuang and Beijing, respectively.

3.3 Formation of secondary aerosols

Using the high-time-resolution data, we further explored the possible formation mechanisms of secondary formation sources. The concentration of secondary formation sources is standardized by dividing background-corrected CO (ΔCO) to weaken the impact of planetary boundary layer height (PBLH) (DeCarlo et al., 2010). In this study, ΔCO is defined as the 1.25th percentiles of CO concentration during the campaign, which are 0.17, 0.15, and 0.16 ppm in Xi'an, Shijiazhuang, and Beijing, respectively. O_x (NO₂+O₃) is an indicator of the photochemical oxidation degree (Wood et al., 2010). The function between the secondary formation source $/$ ΔCO ratio and O_x during the daytime (i.e., 08:00–17:00 LST) (Fig. S12) was plotted to explain the effect of photochemical oxidation in the three pilot cities. As shown in Fig. 3, good linear correlations of the secondary formation source $/$ ΔCO ratio and O_x (R²= 0.83–0.99) suggest that photochemical oxidations play an important role in the formation of secondary aerosol during the daytime. Compared to the low-level O_x, formation of secondary aerosol was significantly enhanced at high O_x levels (>50 ppb) in Xi'an and Beijing, characterized by larger slopes of 17.2 and 38.9, respectively (Fig. 3a and c). Furthermore, the highest atmospheric oxidation capacity was found in Beijing, inferred by the highest fraction of O₃ to O_x. This is consistent with the highest contribution of secondary formation sources to PM_2.5 in Beijing during the daytime (Fig. S13a–c).

Figure 3 Correlations of the secondary formation source $/$ ΔCO ratio and O_x in (a) Xi'an, (b) Shijiazhuang, and (c) Beijing. Each point and its error bar represent the mean and standard deviation in each bin (ΔO_x=5 ppb).

The aqueous-phase reaction is another important pathway for secondary aerosol formation in the atmosphere (Wang et al., 2018; Xue et al., 2014). ALWC is considered an indicator of an aqueous-phase reaction (Ervens et al., 2011). As shown in Fig. S14, the aqueous-phase reaction occurred during both daytime and nighttime and was characterized by good correlations between the secondary formation source $/$ ΔCO ratio and ALWC (R²= 0.81–0.98). The correlations of the secondary formation source $/$ ΔCO ratio and ALWC during all sampling periods were re-established in the three pilot cities to assess the implications of aqueous-phase chemistry for secondary aerosol production. As shown in Fig. 4, the secondary formation source $/$ ΔCO ratio showed a significant linear correlation with ALWC (R²= 0.92–0.99) when RH <80 %, indicating an obvious effect of the aqueous-phase reaction on the secondary aerosol formation during the sampling periods. However, when RH >80 %, the secondary formation source $/$ ΔCO ratio showed no notable increase with ALWC in Shijiazhuang (Fig. 4b), whereas it showed a tiny increase with ALWC in Beijing (Fig. 4c). The higher ALWC at RH >80 % probably inhibits secondary aerosol formation due to the decrease in aerosol acidity (X. Huang et al., 2019; Meng et al., 2014). Khan et al. (2008) found that NO₃ radicals can be rapidly generated from the reaction between NO₂ and O₃ with unsaturated organic species during nighttime. The value of O₃ × NO₂ can thus represent its production reaction rate or be used as a proxy for the NO₃ radical. The highest NO₃ radical production rate was found in Beijing, followed by Xi'an and Shijiazhuang, when RH <80 %. This could be used to explain the highest contribution of secondary formation sources to PM_2.5 in Beijing during the daytime and nighttime (Fig. S13). Moreover, the results showed that both photochemical oxidation and aqueous-phase reaction play more important roles in Beijing, where the primary sources have been better controlled. This reflects that pollution control policies need to focus on the suppression of secondary formations.

Figure 4 Correlation of the secondary formation source $/$ ΔCO ratio and ALWC during sampling periods in (a) Xi'an, (b) Shijiazhuang, and (c) Beijing, respectively. The points and error bars represent the mean values and standard deviation values of the secondary formation source $/$ ΔCO ratio and ALWC in each bin. In Xi'an, each bin is 5 µg m⁻³ (ΔALWC=5 µg m⁻³). In Shijiazhuang, each bin is 5 µg m⁻³ (ΔALWC=5 µg m⁻³) when ALWC ranged from 0 to 75 µg m⁻³ but 25 µg m⁻³ (ΔALWC=25 µg m⁻³) when ALWC ranged from 75 to 200 µg m⁻³ due to limitations in data. In Beijing, each bin is 5 µg m⁻³ (ΔALWC=5 µg m⁻³) when ALWC ranged from 0 to 50 µg m⁻³ but 100 µg m⁻³ (ΔALWC=100 µg m⁻³) when ALWC ranged from 50 to 900 µg m⁻³ due to limitations in data.

3.4 Elaborations of different episodes

During the sampling periods, the concentration of PM_2.5 and its chemical components accumulated within a short period in a few cases (Fig. S15a–c). We define such a rapid rise in PM_2.5 mass concentration as a pollution episode. As shown in Table 1, meteorological conditions, the concentration levels of gaseous pollutants, chemical compositions, and source contributions of PM_2.5 during pollution episodes in the three pilot cities are summarized. The episodes were accompanied by low wind speed (<2 m s⁻¹), leading to weak dispersions of fresh emissions and accumulated pollutants (Z. Chen et al., 2020). OA and NO${}_{\mathrm{3}}^{-}$ were the two dominant chemical components in PM_2.5 during all pollution episodes, with fractions of 26 %–40 % and 23 %–32 %, respectively. Their high abundances could be explained by the significant reduction in SO₂ emissions because of the prohibition of burning bulk coals and enforcement of the “Coal-to-Natural Gas” policy in recent years (Meng et al., 2022). In this study, eight pollution episodes (denoted as EP1–EP8) were classified into three types: the first type was dominated by biomass burning (30 %–40 %) (EP1, EP4, and EP8); the second type was dominated by secondary formation sources (61 %–70 %) (EP5, EP6, and EP7); and the two remaining pollution episodes were mutually contributed by both primary and secondary sources (EP2 and EP3), in which secondary formation sources (34 %–39 %) and biomass burning (23 %–24 %) were the two dominant contributors to PM_2.5.

Table 1 Meteorological conditions, gas pollutants, chemical composition, and source contribution of PM_2.5 during pollution episodes in Xi'an, Shijiazhuang, and Beijing.

^a WSW: west-southwest; WNW: west-northwest; NNW: north-northwest; NNE: north-northeast; ENE: east-northeast. ^b BB: biomass burning; SF: secondary formation source; CC: coal combustion; VE: vehicle emissions.

To gain insights into the process of pollution episodes, three typical pollution events were chosen for detailed discussion (i.e., EP2 in Xi'an, EP4 in Shijiazhuang, and EP7 in Beijing) based on the validity and integrity of the data and the representativeness of the selected pollution events. The two-stage evolution was distinguished for EP4 as an example of the first type of episode (Fig. S16). In Stage 1, the PM_2.5 mass concentrations rapidly increased from 7 to 82 µg m⁻³ under stable weather conditions, inferred by low wind speed (1.8±0.8 m s⁻¹; Fig. S16b), in which the concentrations (fractions) of biomass burning increased from 0.6 µg m⁻³ (7 %) to 36.7 µg m⁻³ (55 %) due to heating activity during nighttime. Meanwhile, the chemical composition was relatively stable and dominated by OA (31±5 %) and NO${}_{\mathrm{3}}^{-}$ (21±5 %). In Stage 2, the PM_2.5 mass concentration continuously increased to 105 µg m⁻³ in a few hours, and the concentration (contribution) of secondary formation sources rapidly increased from 2.3 µg m⁻³ (4 %) to 54.4 µg m⁻³ (52 %) (Fig. S16g and h). This is due to the aqueous-phase-reaction effect inferred by the rapid increase in ALWC (from 16 to 78 µg m⁻³; Fig. S16c) and RH (from 51 % to 79 %; Fig. S16a).

In contrast, a three-stage evolution was distinguished for the second type of episode, using EP7 as an example (Fig. S17). In Stage 1, the PM_2.5 concentration gradually increased from 11 to 30 µg m⁻³, as did NO₂ (from 15 to 59 µg m⁻³; Fig. S17d), due to the boosts of the predominant contributions of vehicle emissions and biomass burning (Fig. S17g and h). In the meantime, the contribution of coal combustion also slowly increased along with SO₂ (Fig. S17d and h). In Stage 2, under the lowest average wind speed in the study period (0.7±0.4 m s⁻¹; Fig. S17b), the PM_2.5 mass concentrations moderately increased from 30 to 91 µg m⁻³ with relatively stable chemical composition and source contribution (Fig. S17f and h). Compared to Stage 1, the fractions of NO${}_{\mathrm{3}}^{-}$ increased mostly from 9±3 % to 23±3 %; this is probably influenced by photochemical oxidation, inferred by relatively high O_x and NO₂ concentration (Fig. S17c and d). In Stage 3, the PM_2.5 mass concentration rapidly rose to 142 µg m⁻³ and then remained stable. Furthermore, the greatest increase in source concentration (contribution) was secondary formation sources, from 18.9 µg m⁻³ (48 %) to 120.6 µg m⁻³ (80 %). This might be due to the occurrence of an aqueous-phase reaction, which was indicated by the elevation of RH and ALWC (Fig. S17a and c).

Figure S16 illustrates the third type of episode, in which a four-stage evolution was resolved, using EP2 as an example. In Stage 1, the PM_2.5 mass concentration (14±3 µg m⁻³) was relatively low and was dominated by the contributions of secondary formation sources (43±17 %) and fugitive dust (24±8 %), as well as MD (28±7 %) and OA (26±7 %). In Stage 2, the PM_2.5 mass concentrations promptly increased from 21 to 82 µg m⁻³ with the two dominant chemical components of OA (21.7 µg m⁻³) and NO${}_{\mathrm{3}}^{-}$ (17.1 µg m⁻³). The PM_2.5 increases can also be attributed to the rise in secondary formation sources (25.3 µg m⁻³) and biomass burning (14.4 µg m⁻³). The enhancement of secondary aerosol was probably generated through the aqueous-phase reaction evidenced by the increase in ALWC and NO₂ (Fig. S18c and d). In Stage 3, PM_2.5 mass continuously increased to 139 µg m⁻³ with a dominant increase in primary source emissions, including biomass burning (29.0 µg m⁻³), vehicle emissions (21.5 µg m⁻³), and coal combustion (16.5 µg m⁻³), along with the increases in SO₂ and NO₂ as well (Fig. S18d). The three primary sources contributed >60 % of the total resolved sources. Meanwhile, the secondary formation sources also increased slowly through aqueous-phase reaction, inferred by the increase in ALWC (Fig. S18c). In the final stage, Stage 4, the PM_2.5 mass concentration remained relatively stable, with an average of 142±11 µg m⁻³, dominated by secondary formation sources (34±6 %) and biomass burning (28±6 %) as well as chemical components of OA (36±4 %) and NO${}_{\mathrm{3}}^{-}$ (25±1 %).

In summary, the pollution events that occurred in Xi'an were mainly derived by stronger emissions of primary sources under adverse meteorological conditions, even though the aqueous-phase reaction also contributed to secondary aerosol formation. In contrast, pollution events that occurred in Shijiazhuang and Beijing were mainly influenced by the formation of secondary aerosols through both aqueous-phase reaction and photochemical oxidation. Moreover, aqueous-phase reaction plays a more important role than photochemical oxidation. Hence, to further improve the air quality in the north of China, primary source emissions should be prioritized for control in the northwestern region, with a focus on biomass burning and coal combustion. In the North China Plain, priority should be given to the reduction in emissions of the precursors from secondary sources, with a focus on NO_x and volatile organic compounds (VOCs).

Figure 5 Summary of PM_2.5 and its composition (a, b, c) and source contribution (d, e, f) in Xi'an, Shijiazhuang, and Beijing in winter in past decades, where ^* represents the results of this study. The data and references used for this figure are listed in Tables S7 and S8.

3.5 Policy implications

In past decades, the air quality in China improved notably under the implementation of air pollution control policies, including the APCAP and TAPFAP. The PM_2.5 mass in Xi'an, Shijiazhuang, and Beijing was the lowest during the campaigns compared with those in the last few decades (Table S7). The variations in the chemical composition and the source contribution of PM_2.5 in the three pilot cities are displayed in Fig. 5. As shown, the dominant chemical components of PM_2.5 changed from OA and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ to OA and NO${}_{\mathrm{3}}^{-}$ (Fig. 5a–c). This could be attributed to the reduction in coal consumption due to clean-energy replacement and the increase in vehicle ownership, which led to a decrease in SO₂ and an increase in NO₂ (Wang et al., 2013). Since the atmospheric oxidation reaction (i.e., aqueous-phase reaction and photochemical oxidation) of the precursors ( i.e., NO₂, VOCs) is the primary source for the OA and NO${}_{\mathrm{3}}^{-}$ in the atmosphere (Feng et al., 2018; Li et al., 2022; Tao et al., 2016; S. Yang et al., 2022; Ziemann and Atkinson, 2012), and it is impossible to avoid, the precursors of OA and NO${}_{\mathrm{3}}^{-}$ should be reduced from the combustion and transportation sectors (Fermo et al., 2021; Liu et al., 2022; Y. Wang et al., 2021; Zhang et al., 2019). In addition, the fraction of NH${}_{\mathrm{4}}^{+}$ in PM_2.5 increased at an alarming rate. Coincidentally, this follows a similar trend to NH₃. Studies have reported that controls of NH${}_{\mathrm{4}}^{+}$ are more effective than those of NO_x in the reduction in PM_2.5 mass concentrations (Gu et al., 2021; Zheng et al., 2022). Therefore, collaborative control measures for the emissions of precursors, including NO_x, VOCs, and NH₃, are necessary.

As shown in Table S8 and Fig. 5d–f, coal combustion decreased remarkably due to the coal-related policies implemented, including the strength of emissions standards for coal-fired power plants, the change in energy source from coal to natural gas in some industries, and the ban on coal burning in the main urban areas (Shen, 2016; Yang and Teng, 2018). A similar trend was also found in the results of PM_2.5 source apportionment in Beijing released by the Beijing Municipal Ecology and Environment Bureau (Fig. S19). Meanwhile, the contribution of industrial emissions and vehicle emissions decreased slightly because of the improvement of industrial emission standards (He et al., 2020; Wang et al., 2020a) and traffic-related policy implementation such as strengthening of vehicle emission standards, improvement of fuel quality, and elimination of high-emission vehicles. This resulted in a reduction in precursor gases and PM_2.5 from vehicles (Feng et al., 2021; Fontaras et al., 2012; Jin et al., 2012). However, the emissions of biomass burning did not show a significant reduction in recent years, and its contribution increased from 9 % in 2014 to 25 % in 2020 (Xi'an), from 3 % in 2015 to 24 % in 2022 (Shijiazhuang), and from 6 % in 2013 to 18 % in 2021 (Beijing) (Fig. 5d–f). This is likely because biomass burning is an open source, which makes it more difficult to control compared to other primary sources. Biomass used for residential heating in rural areas is still common (Ren, 2021; Tian et al., 2022; D. Yang et al., 2022; Zhang et al., 2017). Hence, the clean-energy revolution should be promoted urgently, especially in northwestern China. Moreover, the contributions of secondary formation sources increased, which is potentially explained by the high reduction rate of primary emissions and the improvement of atmospheric oxidation capacity (S. Chen et al., 2020; Feng et al., 2020). Therefore, more control measures should focus on weakening the atmospheric oxidation capacity, such as reduction in O₃ formation, to reduce the formation of secondary pollutants, which are now identified as the most critical drivers of pollution. Considering those factors, it is also important to promote the mitigation of both PM_2.5 and O₃.

4 Conclusion

Intensive real-time measurement campaigns related to PM_2.5 chemical components were conducted in Xi'an, Shijiazhuang, and Beijing during the wintertime. Chemical compositions of PM_2.5 in the three cities were all dominated by OA (26.9 %–34.2 %) and NO${}_{\mathrm{3}}^{-}$ (23.6 %–26.5 %). Six sources of PM_2.5 in Xi'an and Shijiazhuang were resolved by HERM, and their contributions were similar. In descending order, they were secondary formation sources (32.2 %–37.6 %), biomass burning (24.4 %–24.6 %), coal combustion (15.1 %–16.0 %), vehicle emissions (12.2 %–12.5 %), industrial emissions (5.5 %–7.7 %), and fugitive dust (4.4 %–7.8 %). However, the secondary nitrate (29.0 %) and the secondary sulfate (23.0 %) were separately resolved and were relatively more important in Beijing. In addition, the contribution of fireworks (7.9 %) to PM_2.5 was found during the Chinese Spring Festival.

The possible formation mechanism of secondary formation sources in the three pilot cities was explored. The results showed that secondary aerosols were generated by both photochemical oxidation and aqueous-phase reaction. Meanwhile, the formation rate of secondary aerosols in Beijing was higher than that in Xi'an and Shijiazhuang. Furthermore, the eight pollution episodes within the sampling periods were categorized into three types and characterized. The dominant chemical compositions of PM_2.5 were OA (26 %–40 %) and NO${}_{\mathrm{3}}^{-}$ (23 %–32 %) during all pollution episodes. Furthermore, secondary formation sources and biomass burning were two major drivers of the pollution.

The dominant chemical components of PM_2.5 in the pilot cities changed from OA and SO${}_{\mathrm{4}}^{\mathrm{2}-}$ to OA and NO${}_{\mathrm{3}}^{-}$ under the implementation of a clean-air plan in the past decades. This indicates that a reduction in precursors, including NO₂ and VOCs, should be a key task in the future. In addition, the contribution of biomass burning increased, especially in Xi'an. This indicates that clean energy for heating activity in rural areas in northwestern China is still insufficient. Furthermore, to weaken the atmospheric oxidation capacity for reducing the contribution of secondary formation sources, it is necessary to promote the collaborative control on ozone and particulate matter.