Figure 1 shows the temporal trends of the O3 and other pollutant mixing ratios during the sampling period. During the pre-NMG period, the highest O3 hourly concentration was 252 µgm-3 at 15:00 LST on 23 August; meanwhile the max 8 h O3 value also appeared on this day (219 µgm-3). In addition, max 8 h O3 concentrations were present on a total of 7 d, exceeding the ambient air quality standard (GB 3095-2012) Grade II standard of 160 µgm-3. At this stage, the ozone pollution cause concern as the days exceeding the standard accounted for 50 %. The highest hourly concentrations of VOCs and NOx were 1017 and 357 µgm-3, with a mean concentration of 150 ± 93 and 49 ± 46 µgm-3, respectively. Higher O3 precursor concentrations were observed, which may be an important factor leading to serious photochemistry pollution.

During the control period, the O3 precursor concentrations showed a decreasing trend, with a mean concentration of 121 ± 55 µgm-3 VOCs and 39 ± 26 µgm-3 NOx. However, the ozone pollution was not dramatically alleviated; and mild and moderate O3 pollution levels accounted for 52 % and 10 %, respectively, during the NMG period according to GB 3095-2012. The phenomenon has been reported that photochemistry is still severe under emergency emission-reduction strategies (R. Li et al., 2019). Due to the highly nonlinear relationships between ozone and its precursors, it is not straightforward to mitigate ozone pollution by reducing the emissions of VOCs and NOx (Tan et al., 2018; S. Wang et al., 2019). A special phenomenon needs to be pointed out that the ozone concentration during the evening peak is much higher than that in the pre-NMG period. The weak titration effect may be the main reason for the phenomena above (Chi et al., 2018; Zou et al., 2019). At the end of the control period, the concentration of precursors increased rapidly, and NOx concentration increased by nearly 1.6 times compared with the control stage. At the same time, the ozone pollution was still severe; the proportions of mild and moderate pollution days were 83 % and 8 %, respectively.

On the whole, the concentration of ozone precursors decreased during the control period (as shown in Fig.S4), and the ozone pollution was severe during the entire observation period. It should be noted that the maximum value of max 8 h O3 concentrations in the NMG period is a high of 235 µgm-3.

Meteorological conditions can significantly influence pollutant concentrations, which makes it difficult to evaluate the emission reduction brought by emission control. In this paper, the meteorological data throughout the three periods in Zhengzhou were compared, including temperature (T), precipitation, relative humidity (RH), wind speed (WS), and visibility. As shown in Table S5 in the Supplement, the meteorological parameters rarely changed during the three periods (T, 27.4 ± 1.2, 24.2 ± 3.3, and 22.3 ± 1.5∘ for the pre-NMG period, the NMG period, and the post-NMG period, respectively; visibility, 16.7 ± 5.5, 14.1 ± 7.0 and 13.0 ± 2.7 km; WS, 1.7 ± 0.3, 1.7 ± 0.4, and 1.5 ± 0.3 ms-1). However, the precipitation during the pre-control period (236.9 mm) was much higher than during the control (39.8 mm) and post-control periods (1.6 mm). In addition, the RH in the first stage is higher, which is conducive to the dissolving of the air pollutants in water vapor, condensation, and settlement, thereby reducing the concentration of pollutants.

Meteorological conditions can influence the transmission and circulation of regional air pollutants (Ren et al., 2019). In this paper, the air clusters were analyzed using the Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT) to distinguish the differences of potential source contributions in the three periods. In previous studies, it is shown that regional transport has an important influence on Zhengzhou's air quality, especially the air mass from the BTH region (Jiang et al., 2019; P. Wang et al., 2019). Figure S4 shows the 48 h backward trajectory results during the sampling period. The dominant trajectory was from the east or southeast of Zhengzhou in the three periods. For the pre- and post-NMG periods, Zhengzhou is greatly affected by the air mass from the BTH region, where there are high concentrations of air pollutants from anthropogenic emissions. Based on the PSCF results in Fig. 2, the high PSCF values for pollutants (including O3, NO2, and VOCs) were distributed significantly at the southeast of Zhengzhou and near the Zhengzhou area. Therefore, the potential source regions for pollutants during the sampling period were mainly from the southeast of Zhengzhou and local sources within the city. For the record, high PSCF values were also present within northern Hebei Province during the pre-NMG period; this region is a major industrialized area in the BTH. To summarize, in addition to air pollution control measures, changes of meteorological conditions may contribute to the improvement of air quality during the NMG period.

As illustrated in Fig. 1, the mixing ratios of hourly total VOCs (TVOCs) show an average value of 150 ± 93 µgm-3, ranging from 41 to 1017 µgm-3, before the control period. During control, this reduced to an average of 121 ± 55 µgm-3, with a range from 37 to 333 µgm-3. After the control period, the average VOC concentration increased to 136 ± 60 µgm-3. Overall, it is clear that the emission control policies were beneficial in reducing VOC concentration, decreasing by about 19 % and 11 % before and after emission reduction.

The percentage distributions of VOC groups were similar in the three sampling periods (Fig. S6 in the Supplement). Alkanes were the dominant group, accounting for 37 %, 35 %, and 33 % of the total VOC concentration for the three periods, respectively, followed by halocarbons. Notably, OVOCs slightly decreased in the entire sampling period, comprising 17 %, 16 %, and 15 %, respectively. However, the active components of aromatics increased over time. In addition to the impact of emission sources, meteorological conditions and transport might be key factors that can influence VOC compositions (Su et al., 2021).

The top 20 VOC species are summarized in Table 1. The top 20 substances were similar in the three stages, but the concentration levels were quite different. Tracers of solvent sources including hexane and dichloromethane (Huang and Hsieh, 2019; Wei et al., 2019) decreased in the control period, reducing by 42 % and 47 %, respectively. The reduction of vinyl acetate and tetrachloroethylene is relatively large, which may be attributed to industrial emission reduction (Hsu et al., 2018; Zhang et al., 2015). In addition, the concentration of acetylene is reduced by 55 % compared with the pre-NMG period, as a potential result of the control of combustion sources (Liu et al., 2020; F. Wu et al., 2016).

The mean diurnal variations of TVOCs and their compounds before, during, and after the control period are shown in Fig. S7 in the Supplement. Clearly, the diurnal variations of TVOCs during the three periods are similar, showing higher values from evening till morning rush hours, while they are lowest in the afternoon. The composition of alkanes, alkenes, alkynes, and aromatics shows similar daily variations. Previous studies have suggested that VOCs can be oxidized by O3, OH radicals, and NO3 radicals (Carter, 2010b). In short, the reactions with O3 and OH radicals are the most important chemical reactions during daytime, and the reactions with NO3 radicals and O3 are dominant reactions for VOCs occurring at night (Atkinson and Arey, 2003). However, the reaction rate of the OH radical is much higher than that of the NO3 radical, and thus, the concentration of TVOCs and its compounds at night is generally higher than that during the daytime (Atkinson and Arey, 2003). It should be noted that the average VOC mixing ratio at midnight during the control period was significantly lower than that in the other two periods; in particular, the concentrations of aromatics compounds were significantly decreased.

As each source type has its own fingerprint, the mean diurnal variations of tracers during the three periods are presented in Fig. 3. Isopentane and n-pentane in the three periods have several minor peaks (e.g., 02:00 and 18:30 LST). The 02:00 LST peak is mainly from the freight trucks, and the 07:00 LST peak is most likely from traffic rush-hour emissions (Gentner et al., 2009; K. Li et al., 2019; Zheng et al. 2018). It should be noted that during the control period, the nighttime concentration significantly decreased, while the daytime concentrations of the three stages are close. Toluene, ethylbenzene, and xylenes (Fig. 3) as well as the tracer gases of NO and NO2 (Fig. S8 in the Supplement) in the three periods had similar diurnal patterns to those of pentane. All of the species mentioned above are tracers of traffic emission (Brito et al., 2015; Dörter et al., 2020; Yenisoy-Karakas et al., 2020). It is speculated that the control effect on muck truck is significant during the control period. As shown in Fig. 3, tracers of solvent utilization, such as hexane and dichloromethane, had different diurnal patterns to those of the vehicle source. During the pre-NMG period, the solvent tracer emissions are so strong that they almost offset the daytime trough caused by photochemical reactions and boundary layer height (K. Li et al., 2019). The daytime levels of the NMG period are lower than those of the first stage, which might be attributed to the intensive control over the use of outdoor solvents. Chloromethane and acetylene are tracers of biomass burning and combustion, respectively (McRoberts et al., 2015; Schauer et al., 2001). During the control period, acetylene decreased significantly, while chloromethane remained higher levels.

During the sampling period, the great changes in the mixing ratios of VOCs may be caused by the altered contribution of emission sources. Ratios of specific VOCs have commonly been used to identify emission sources.

Because i- and n-pentane have similar atmospheric lifetimes, these i/n ratios are widely used to examine the impact of vehicle emissions and combustion emissions, and the values varied according to sources (i.e., 0.56–0.80 for coal combustion, 1.5–3.0 for liquid gasoline, 2.2–3.8 for vehicle emissions; Yan et al., 2017; Zheng et al., 2018). Isopentane and n-pentane showed highly significant correlations during the three periods (R2 > 0.6), suggesting the source of these two species was similar (Fig. S9 in the Supplement). In this study, the ratios of i/n-pentane during the three periods were 1.5, 1.7, and 1.5, indicating that the VOCs originated from the mixed sources of coal combustion and vehicle emissions.

The toluene / benzene (T/B) ratio has also been widely applied to be an indicator of sources. A previous study reported that these two species are most probably from biomass burning, coal combustion, vehicle emissions, and solvent use, with the T/B ratios ranging between 0.2–0.6, 0.6–1.0, 1.0–2.0, and > 4, respectively (Hui et al., 2018; Kumar et al., 2018; M. Song et al., 2019). As shown in Fig. S9, low correlations (R2 = 0.3–0.5) were found during the three periods, suggesting a more complex set of sources for the two species. The T/B ratio in the three periods was 0.78, 0.75, and 0.92, respectively, indicating that the VOCs were greatly influenced by the mixed source of coal combustion and vehicle emissions. The ratio was lower in the control period. Acetylene concentration was low but the chloromethane concentration was high during the control period, indicating that biomass burning had a greater impact during this period.

The 42 most abundant species, accounting for almost 90 % total VOC concentrations, were selected to be applied in the PMF receptor model to analyze the relative contribution of each potential source. The factor profiles of seven emission sources, namely, liquefied petroleum gas (LPG) evaporation, industrial processes, vehicle exhausts, biomass burning, biogenic source, solvent usage, and coal combustion, are identified in Fig. 4.

Source 1 was characterized by both high proportions and high abundances of ethane (38 %), ethene (53 %), propane (43 %), and other C3–C5 alkanes. A high proportion of short linear alkanes, such as ethane and propane, was likely released from the use of LPG (Yadav et al., 2019; Zhang et al., 2015). Consequently, factor 1 was assigned to LPG.

Source 2 accounts for larger percentages of carbon disulfide and halohydrocarbon, such as cis-1,2-dichloroethylene, chlorotrifluoromethane, trichloromethane, dichloromethane, 1,2-dichloroethane, and 1,2-dichloropropane, followed by some aromatics. Some previous studies indicated that these species were related to industrial processes (Hui et al., 2020; Zhang et al., 2018). Furthermore, ethylene, considered the blood of the industrial development (C. Song et al., 2019; Zheng et al., 2020), accounted for almost 65 % of the TVOCs in this factor. Therefore, this source is considered to be related to industrial processes.

Source 3 was characterized by a high percentage of some C4–C5 alkanes and aromatics. Toluene, ethyl-benzene, m,p-xylene, o-xylene, i,n-butane, and i,n-pentane are all associated with vehicle exhausts (Huang and Hsieh, 2019; Liu et al., 2019). Furthermore, the T/B ratio was approximately 1.6, suggesting that this source was significantly affected by vehicle exhaust emissions. Factor 3 also includes high proportions of methyl tert-butyl ether, which is a common gasoline additive in China (Yang et al., 2018). Therefore, this factor can be labeled as vehicle exhaust.

Source 4 has high concentrations of chloromethane, which is a typical tracer of biomass burning (Ling et al., 2011; Zhang et al., 2019). The percentages of benzene and toluene were lower, but they could still not be neglected, and the T/B ratio was < 0.5 (B. Li et al., 2019). Thus, source 4 was identified as biomass burning.

Source 5 accounts for larger percentages of isoprene, accounting for 86 % of the TVOCs in the source. Isoprene is an indicator of biogenic emissions and is emitted from many plants (Guenther et al., 1995 and 1997). This factor also included a considerable proportion of intermediate products (Liu et al., 2019), such as acetone, 2-hexanone, and 2-butanone. Therefore, this source is considered to be biogenic emissions.

Source 6 was differentiated by C6–C8 alkanes, such as n-hexane, methyl cyclopentane, 3-methylpentane, 2-methylpentane, and 2,3-dimethylbutane. The percentages of ethyl acetate, tetrachloroethylene, carbon tetrachloride, chloroform, and dichloromethane were high. These substances are important organic solvents and typical tracers of solvent usage (Hui et al., 2018, 2020). Meanwhile, there are virtually no other short-chain hydrocarbons in this source. Therefore, source 6 was primarily attributed to solvent usage rather than industrial processes or the vehicle emission source.

Source 7 was dominated by acetylene, which accounted for 75 % of the TVOCs in the source. Acetylene is a typical tracer of combustion emission (Hui et al., 2019; R. Wu et al., 2016). Some of the VOC species, such as alkanes and benzene, are the main components in emissions from coal burning (Liu et al., 2019; M. Song et al., 2019; Yang et al., 2018). Thus, factor 7 was assigned to combustion emission.

The concentrations of hourly mixing ratio and the relative contributions of each VOC sources are illustrated in Figs. 5 and S10 in the Supplement. Compared with the non-control periods, the contributions of coal combustion, vehicle exhausts, and solvent utilization are significantly reduced during the control period.

Conversely, the mixing ratios of LPG showed higher values during the control period. Peak values of biomass combustion were frequently present during the second period, and biomass combustion accounts for a relatively high proportion in this stage. The highest concentration was observed in the afternoon of the 18 September. Zhengzhou and its surrounding areas are in the harvest period of crops in September, so the emissions of biomass combustion need to be considered. Figure S11 in the Supplement shows the hotspots diagram of Zhengzhou and its surrounding areas during the observation period, and the number of fire spots in September was significantly higher than that in August.

Time series of each identified source contributions are shown in Fig. 6. During the first period, solvent utilization (33 µgm-3) was the largest contributor and accounted for 30 % of TVOCs, followed by industrial processes (26 µgm-3, 23 %) and vehicle exhausts (24 µgm-3, 21 %). Although it was not a typical time for heating, coal combustion still accounted for 10 % of the TVOCs, probably due to several coal-fired power plants around Zhengzhou (B. Li et al., 2019). In contrast, the proportion of biomass combustion was very low during this period, accounting for only 2 % of the TVOCs.

During the control period, solvent utilization made the largest contribution (23 %) to atmospheric VOCs, with the concentration of 23 µgm-3, followed by industrial processes (22 %), vehicle emissions (22 %), and LPG (11 %). Biomass burning should not be ignored in this period, accounting for 10 % of total VOCs. The contribution from coal combustion was relatively low (3.5 µgm-3), accounting only for 4 % of TVOCs.

For the third period, the largest contributor was fuel combustion related to vehicle exhausts, with 30 µgm-3, accounting for 28 % of total VOCs. Industrial processes (23 µgm-3), solvent utilization (20 µgm-3), biomass burning (12 µgm-3), coal combustion (11 µgm-3), LPG (5.7 µgm-3), and biogenic emissions (5.6 µgm-3) accounted for 21 %, 19 %, 11 %, 10 %, 5 %, and 5 % of total VOCs, respectively.

In summary, the concentrations of solvent utilization were reduced to the greatest extent during the control period, with the value of 10 µgm-3 in the pre-NMG period, followed by coal combustion (7.1 µgm-3), industrial processes (4.0 µgm-3), and vehicle exhausts (2.2 µgm-3). Reductions of solvent utilization, coal combustion, industrial processes and vehicle exhausts were responsible for 80 %, 57 %, 32 %, and 18 % of the reductions in ambient VOCs, indicating that the control measures on solvent utilization and coal combustion were the most effective. In contrast, due to weak control on biomass burning and LPG, contributions from these sources were elevated. September is a harvest month in northern China, which means that biomass burning contributions might increase with time. Meanwhile, the peak contribution of this source occurred during the control period because of a lack of relative control measures on LPG.

In addition to the impacts on ambient air quality, some VOC species are also toxic with various health impacts. In this paper, the non-carcinogenic risk (expressed by HQ) and carcinogenic risk (expressed by lifetime cancer risk, LCR) of hazardous VOC species were investigated, and the acceptable safety thresholds were 1 and 1 × 10-6, respectively (US EPA, 2009). On the whole, the HQ of almost all substances is far below the safety threshold, indicating no chance of non-carcinogenic risk. However, only the HQ of acrolein (1.8) exceeded the value of 1, suggesting the obvious chance of non-carcinogenic effects (Fig. 7). As for LCR, six species were above 1 × 10-6 in this study, including 1,2-dichloroethane (2.5 × 10-5), chloroform (1.1 × 10-5), 1,2-dibromoethane (8.1 × 10-6), naphthalene (6.4 × 10-6), benzene (5.2 × 10-6), and tetrachloromethane (3.3 × 10-6).

During the entire observation period, a total of seven VOCs may pose potential risks to human health. Health risk assessment in Zhengzhou was compared with other cities, as shown in Table S3. Overall, the values of risk assessment in this study are evidently lower than those reported in Beijing (Gu et al., 2019) and Langfang (Yang et al., 2019a), whereas they are higher than the summer of 2018 in Zhengzhou (Li et al., 2020). Evaluated health risk assessment before, during, and after the control period shows cumulative LCR was 5.8 × 10-5, 5.1 × 10-5, and 6.3 × 10-5, respectively, during the three periods (Fig. S12 in the Supplement). Control measures can reduce non-carcinogenic risk. On the other hand, values of six substances still exceed the acceptable safety threshold of LCR during the control period. As for non-carcinogenic risk assessment, the HI was 1.6, 2.1, and 2.1, respectively. Noticeably, the HQ of acrolein (1.9) during the control period was higher than the other two periods, which should be paid more attention to. In summary, VOC concentrations decreased significantly during the control period but still posed health risks to humans. Therefore, we need to focus on the targeted emission reduction of these characteristic substances to protect human health.

The OFP and their compositions during the three periods are shown in Fig. S13 and Table S6 in the Supplement. The total OFP during the control period was 183 ± 115 µgm-3, which was 0.77 and 0.83 times lower than before and after the control period, respectively. As shown in Fig. S13, the aromatics were the dominant contributors to total OFP in all three periods, comprising 42 %, 50 %, and 56 %, respectively, followed by OVOCs, alkanes, alkenes, halohydrocarbon, and acetylene. Aromatics played a key role in ozone formation, which is similar to the results of many previous reports (Wang et al., 2013; Zou et al., 2015).

The source contributions to OFP were calculated using the PMF model (Table 2). The most important source to ozone formation was traffic emissions. Industrial emissions and solvent usage were the second and third sources of photochemical ozone formation. Among them, solvent use has the greatest contribution to the OFP reduction with the emission reduction during the control period, explaining a 48 % reduction in OFP. Although combustion contributes only 10 % of the total OFP, this source played an important role in the reduction in OFP, explaining 33 % of the OFP reduction. At the same time, control of traffic and industry also reduced the OFP during the games. Thus, solvent utilization and combustion controls were the most important measures taken to reduce OFP during the National Minority Games 2019 in Zhengzhou. However, the current knowledge about formation mechanisms of O3 is still very limited, and the next section discusses the sensitivity of ozone.