Measurement report: Ambient volatile organic compound (VOC) pollution in urban Beijing: characteristics, sources, and implications for pollution control

. The increasing ozone (O 3 ) pollution and high fraction of secondary organic aerosols (SOA) in ﬁne particle mass has highlighted the importance of volatile organic compounds (VOCs) in air pollution control. In this work, four intensive ﬁeld measurements of VOCs during winter of 2018 (from 1 December 2018 to 17 January 2019), spring (15 April to 27 May), summer (17 June to 13 July), and autumn (22 September to 27 November) of 2019 were conducted at an urban site in Beijing to characterize VOC sources and their contributions to air pollution. The total mixing ratio of the 95 quantiﬁed VOCs (TVOC) observed in this study ranged from 5.5– 118.7 ppbv with the mean value of 34.9 ppbv. Alkanes, OVOCs, and halocarbons were the dominant chemical groups, accounting for 75 %–81 % of the TVOC across the sampling months. The molar ratios of VOCs to NO x indicated that O 3 formation was limited by VOCs during the whole sampling period. Positive matrix factorization (PMF) analysis showed that diesel vehicle exhaust, gasoline vehicle exhaust, and industrial emissions were the main VOC sources during both the O 3 -polluted and PM 2 . 5 -polluted months. On the basis of O 3 formation impact, VOCs from fuel evaporation and diesel exhaust, particularly toluene, xylenes, trans -2-butene, acrolein, methyl methacrylate, vinyl acetate, 1-butene, and 1-hexene, were the main contributors, illustrating the neces-sity of conducting emission controls on these pollution sources and species to alleviate O 3 pollution. Instead, VOCs from diesel exhaust as well as coal and biomass combustion were found to be the dominant contributors to secondary organic aerosol formation potential (SOAFP), particularly the VOC species of toluene, 1-hexene, xylenes, ethylbenzene, and styrene, and top priority should be given to these for the alleviation of haze pollution. This study provides insights for the government to formulate effective VOC control measures for air pollution in Beijing.


Introduction
Ozone (O 3 ) and fine particulate matter (PM 2.5 ) pollution has restricted improvements in air quality in China. Observation data from the Chinese Ministry of Environment and Ecology (MEE) network witnessed an upward trend for O 3 across the country over the period 2013-2019 Shen et al., 2019;Fan et al., 2020). Also, haze pollution occurring at urban sites in recent years has commonly been characterized by enhanced formation of secondary organic aerosol (SOA) in fine particles; e.g., the fraction of SOA in organic aerosols reached 58 % in Xi'an during winter 2018 and 53 % in urban Beijing during winter 2014 (Kuang et al., 2020;Sun et al., 2020;L. Cui et al.: Ambient VOC pollution in urban Beijing Q. . Volatile organic compounds (VOCs) are key precursors for the formation of O 3 via gas-phase reactions (Odum et al., 1997;Atkinson, 2000;Sato et al., 2010;Huang et al., 2014). In highly polluted urban regions, O 3 formation is generally VOC-limited, and it is suggested that VOC emission control is necessary for effective alleviation of photochemical smog Liu and Wang, 2020;Shao et al., 2009;Wang et al., 2020;Xing et al., 2011). In addition, VOCs, including aromatics and biogenic species, have a significant impact on SOA formation, which plays an important role in haze formation Tong et al., 2021). VOC emission abatement is therefore imperative for improving air quality in China.
VOCs in ambient air can be emitted by a variety of sources including both anthropogenic and biogenic sources. While biogenic emissions are significantly greater than anthropogenic emissions globally (Doumbia et al., 2021;Sindelarova et al., 2022), anthropogenic emissions play the dominant role in urban and surrounding areas (Warneke et al., 2007;Ahmad et al., 2017;Wu and Xie, 2018). The VOC observations in China showed distinct differences in anthropogenic sources among different regions. For example, solvent use and vehicle exhaust are primary VOC sources in urban Shanghai and urban Guangzhou, while the primary sources of VOCs in Wuhan, Zhengzhou, and Beijing are combustion and vehicle exhaust (Han et al., 2020;B. Li et al., 2019). Apart from the diversity of emission sources, different VOC species exhibited different propensities to form O 3 and SOA. Observationbased studies commonly apply the O 3 formation potential (OFP) and SOA formation potential (SOAFP) scales to quantify the relative effects of specific VOCs and sources on O 3 and SOA formation as well as to aid in the development of efficient control strategies (Carter and Atkinson, 1989;Chang and Rudy, 1990;Han et al., 2020;Zhang et al., 2017). Although there have been many studies on ambient VOCs in various locations (e.g., urban, rural, and industrial areas), most of these measurements were confined to short periods (a few days or a certain season), and the understanding of temporal variations in concentrations, sources, and the influence of photochemical reactions of VOCs on an annual scale is still limited. Most of the available reports on VOC analysis based on online analytical techniques include mainly nonmethane hydrocarbon compounds, and thus the characteristics of VOCs as well as their relationships with PM 2.5 and O 3 cannot be fully revealed since OVOCs also actively participate in chemical reactions related to secondary formation (B. Li et al., 2019;Yang et al., 2018;Pallavi et al., 2019). Therefore, the long-term and comprehensive monitoring of VOCs is desired.
As the capital and one of the largest megacities in China, Beijing has been suffering from severe O 3 pollution due to rapid economic development and increases in precursor emissions (Y. S. Wang et al., 2017;Li et al., 2019b;D. Zhao et al., 2020). According to the Re-port on the State of the Ecology and Environment in Beijing, the average 90th percentile O 3 daily maximum 8 h concentration in Beijing has exceeded the national standards, reaching 193, 192, and 191 µg m −3 in 2017, 2018, and 2019, respectively. In addition, the number of motor vehicles in Beijing reached 6.365 million at of the end of 2019 (http: //tjj.beijing.gov.cn/EnglishSite/, last access: 2 October 2021), making Beijing the top city in China in terms of the number of motor vehicles. The existing field measurements in Beijing were mostly conducted before 2016, and observations in most recent years are quite limited (Li et al., 2015(Li et al., , 2019aYang et al., 2018). In this work, ambient air samples were collected at an urban site in Beijing from December 2018 to mid-January 2019, mid-April to late May 2019, mid-June to mid-July 2019, and late September to late November 2019. Several O 3 and PM 2.5 pollution events were captured during the sampling period. The characteristics and the contribution of specific species and sources of VOCs to O 3 and SOA formation, with a focus on photochemical and haze pollution periods, were analyzed in detail. The results and implications from this study can provide useful guidance for policy makers to alleviate ozone and haze pollution in Beijing.

Field measurement
The sampling site is on the roof of a three-floor building on the campus of Tsinghua University (40.00 • N, 116.33 • E), northwest of a Beijing urban area (Fig. S1 in the Supplement). The altitude of the sampling site is 57 m. This sampling site is surrounded by the school and there are no large emission sources nearby; therefore, it can represent the urban air quality in Beijing. Details of the site description are found in W. .
The air samples were collected using 6 L summa canisters (Entech, USA) with a stable rate of 4.26 mL min −1 . The samples were preprocessed to remove N 2 , O 2 , CO 2 , CO, and H 2 O in the samples and to further concentrate the samples in volume by the cryogenic pre-concentrator (model 7100, Entech Instruments Inc., USA). A pressure gauge was used to test if the canister has air leakage before sampling every time, and blanks were prepared using cleaned canisters to fill with high-purity nitrogen. The cryotraps of the precooling system were baked before analyses each day and between every sample. The VOCs in air samples were analyzed by a gas chromatography system that was equipped with a mass spectrometric detector (GC-MS) (Agilent Tech., 7890/5975, USA). The suitability of this system for VOC measurement is well verified, and it has been used in field campaigns Wu et al., 2016). The oven temperature was programmed at 40 • for 3 min initially, then raised to 90 • C at 8 • min −1 , and later raised to 220 • C at 6 • min −1 , holding for 9 min. In this work, 95 target VOCs, including 25 alkanes, 8 alkenes, 16 aromatics, 34 halocarbons, and 12 OVOCs, were quantified. It should be noted that VOCs (C2-C3) with a low boiling point (i.e., ethane, ethene, acetylene, and propane) were not detected by the GC-MS system. The standard substance (SPECTRA GASES Inc., USA) mentioned for Photochemical Assessment Monitoring Stations (PAMS) and the US EPA TO-15 standard were used to construct the calibration curves for the target VOCs. Quality assurance and quality control, including the method detection limit (MDL) of each compound, laboratory and field blanks, retention time, accuracy, and duplicate measurements of samples, were performed according to the US EPA Compendium Method TO-15 (US Environmental Protection Agency, 1999). The correlated coefficients of the calibration curves for all the compounds were > 0.95. The relative standard deviation (RSD) for all of compounds of triplicates was 0.5 %-6.0 %. Previous field measurements have reported that the precision of a GC-MS system for hydrocarbons and aldehydes was below 6 % and 15 %, respectively Wu et al., 2016). In this work, one kind of aldehyde substance, i.e., acrolein, was detected, with R 2 and RSD of 0.99 % and 4.5 %, respectively.
During the sampling periods, the measurements of PM 2.5 , gaseous pollutants (NO x and O 3 ), and meteorological variables (such as temperature, relative humidity, wind speed, and wind direction) were conducted simultaneously. SO 2 , NO x , and O 3 were analyzed using a pulsed fluorescence SO 2 analyzer (Thermo Fisher Scientific USA, 43I), chemiluminescence NO-NO 2 -NO x analyzer (Thermo Fisher Scientific USA, 17I), and ultraviolet (UV) photometric O 3 analyzer (Thermo Fisher Scientific USA, 49I), respectively. The mass concentration of PM 2.5 was measured using an oscillating balance analyzer (TH-2000Z, China) (Y. S. . The quality assurance of SO 2 , NO 2 , O 3 , and PM 2.5 was conducted based on HJ 630-2011 specifications. Meteorological variables including wind speed (WS), wind direction (WD), relative humidity (RH), air pressure, temperature, and precipitation were measured by an automatic weather monitoring system. The planetary boundary height was obtained from the European Centre for Medium-Range Weather Forecasts (https://www.ecmwf.int/ en/forecasts/datasets, last access: 5 November 2021).

Ozone formation potential (OFP) and secondary formation potential (SOAFP) calculation
The formation potential of O 3 and SOA was used to characterize the relative importance of VOC species and sources in secondary formation, which were estimated using Eqs. (1) and (2): where n represents the number of VOCs, [VOC] i represents the ith VOC species concentration, MIR i is the maximum incremental reactivity for the ith VOC species, and Y i is the SOA yield of VOC i (McDonald et al., 2018). The MIR for each VOC species was taken from the updated Carter research results (http://www.engr.ucr.edu/~carter/reactdat.htm, last access: 24 February 2021). For species lacking yield curves, the fractional aerosol coefficient (FAC) values proposed by Grosjean and Seinfeld (1989) were used.

Deweathered model
In this work, the influences of meteorological conditions on O 3 and PM 2.5 were removed using the random forest (RF) model. The meteorological predictors in the RF model include wind speed (WS), wind direction (WD), air temperature (T ), relative humidity (RH), precipitation (Prec), air pressure (P ), time predictors (year, day of year -DOY, hour), and planetary boundary layer height (BLH). These meteorological parameters have been reported to be strongly associated with PM 2.5 and O 3 concentrations in various regions in China Feng et al., 2020) and contributed significantly in previous PM 2.5 and O 3 prediction models (She et al., 2020;Li et al., 2020). The modeling relates the hourly variability of O 3 and PM 2.5 to that of meteorological variables. The model performance was evaluated through the 10-fold cross-validation (CV) approach, which randomly selects 10 % of the dataset for model testing and trains the model with the remaining data. This process was repeated 10 times, and each record was selected once as testing data.
In each round, the training dataset includes ∼ 90 % randomly selected data representing different seasons. After the building of the RF model, the deweathered technique was applied to predict the air pollutant level at a specific time point. The differences in original pollutant concentrations and deweathered pollutant concentrations were regarded as the concentrations contributed by meteorology. Statistical indicators including R 2 , root mean square error (RMSE), and mean absolute error (MAE) values were regarded as the major criteria to evaluate the modeling performance.

Positive matrix factorization (PMF)
In this study, the US EPA PMF 5.0 software was used for VOC source apportionment (Abeleira et al., 2017;B. Li et al., 2019;Xue et al., 2017). The detailed description of the PMF model is found elsewhere (Ling et al., 2011;Yuan et al., 2010). PMF uses both concentration and userprovided uncertainty associated with the data to weight individual points. Species with high percentages of missing values (> 40 %) and with signal-to-noise ratio below 2 were excluded. Based on this, 53 VOC species including source tracers (e.g., chloromethane, trichloroethylene, tetrachloroethylene, and MTBE -methyl tert butyl ether) and SO 2 were chosen for the source apportionment analysis. Data values L. Cui et al.: Ambient VOC pollution in urban Beijing below the MDL were replaced by MDL/2, and the missing data were substituted with median concentrations. If the concentration is less than or equal to the MDL provided, the uncertainty is calculated using the equation Unc = 5/6 × MDL; if the concentration is greater than the MDL provided, the uncertainty is calculated as Unc = [(error faction × mixing ratio) 2 + (MDL) 2 ] 1/2. During the PMF analysis, the bootstrap (BS) method, displacement (DISP) analysis, and the combination of the DISP and BS (BS-DISP) were used to evaluate the uncertainty of the base run solution. A total of 100 bootstrap runs were performed, and acceptable results were obtained for all factors (above 90 %). Based on the DISP analysis, the observed drop in the Q value was below 0.1 %, and no factor swap occurred, confirming that the solution was stable. The BS-DISP analysis showed that the observed drop in the Q value was less than 0.5 %, demonstrating that the solution was useful.

TVOC mixing ratios and chemical composition
The time series of meteorological parameters and concentrations of air pollutants during the measurement period are shown in Fig. 1. The ambient temperature ranged from −13.3 to 38.7 • , and the RH varied between 5 % and 99 % across the sampling months. Prevailing winds shifted between southwesterly and northeasterly with WS of 0-6.8 m s −1 . The mixing ratio of total VOCs (TVOC) ranged from 5.5-118.7 ppbv during the sampling period with relatively higher values during September and November (49.9-51.6 ppbv), while there were relatively lower values (22.2-27.5 ppbv) across the other months. Major VOC compositions were generally consistent during the whole measurement period. Alkanes, OVOCs, and halocarbons were the dominant chemical groups, accounting for 75 %-81 % of the TVOC across the sampling months. In terms of individual species, acetone, dichloromethane, n-butane, toluene, methyl tert butyl ether (MTBE), iso-pentane, propylene, nhexane, 1,1-dichloroethane, benzene, and 1-butene made up the largest contribution, accounting for 50.6 % of the TVOC on average during the whole measurement period.
As shown in Fig. 2, the concentrations of TVOC and major VOC groups including alkanes, alkenes, aromatics, halocarbons, and OVOCs observed in this study were apparently lower than those during the sampling months in 2014 and 2016 in urban Beijing Li et al., 2015), indicating the effectiveness of control measures in most recent years in lowering VOC emission. The composition of major chemical groups also showed remarkable changes, with decreased proportions of alkanes but increased fractions of halocarbons, aromatics, and OVOCs, reflecting the changes in emission source types in most recent years.
During the measurement period, 14 O 3 pollution episodes (days with maximum 8 h average O 3 exceeding 160 µg m −3 ) were observed on 17 April, 3-4 May, 16 May, 19-20 June, 24-25 June, 2 July, 5 July, 13 July, 25-26 September, and 28 September 2019. The comparison of meteorological parameters and air pollutants on O 3 pollution and compliance days (days with maximum 8 h average O 3 below 160 µg m −3 ) during the 5 O 3 -polluted months (i.e., April, May, June, July, and September of 2019) is discussed here. As shown in Fig. 3, the WS on O 3 pollution days (1.31 ± 0.90 m s −1 ) was slightly lower than that on O 3 compliance days (1.47 ± 1.10 m s −1 ), indicating that precursors were more conductive to being diluted on O 3 compliance days. The variation trend of O 3 and temperature displayed a negative correlation, and the linear correlations between O 3 and temperature on O 3 pollution days (R 2 = 0.63) were stronger than that on O 3 compliance days (R 2 = 0.35). The mean TVOC concentration on O 3 pollution days (32.3 ppbv) was higher than that on O 3 compliance days (29.6 ppbv), which was mainly attributed to higher concentrations of MTBE, acrolein, and trans-2-butene on pollution days. MTBE is widely used as a fuel additive in motor gasoline (Liang et al., 2020), and trans-2-butene is the main component of oil and gas evaporation (B. Li et al., 2019). Such a result suggests enhanced contribution of traffic emissions on O 3 pollution days. Also, the concentration of isoprene, which is primarily produced by vegetation through photosynthesis, increased significantly on O 3 pollution days, probably due to stronger plant emissions at elevated temperatures (Guenther et al., 1993(Guenther et al., , 2012Stavrakou et al., 2014). The ratio of m/p-xylene to ethylbenzene (X/E) measured can be used as an indicator of the photochemical aging of air masses because of their similar sources in urban environments and differences in atmospheric lifetimes (Carter, 2010;Miller et al., 2012;H. L. Wang et al., 2013). The mean X/E value on O 3 compliance days (1.41) was higher than that on O 3 pollution days (1.17), indicating enhanced secondary transformation of VOCs on O 3 pollution days.
The daily PM 2.5 concentrations ranged from 9-260 µg m −3 with the mean value of 88.5 µg m −3 during the measurement period. 15 PM 2.5 pollution days (daily average PM 2.5 exceeding 75 µg m −3 ) were observed on 3 January, 12-13 January, 22-23 April, 29 April, 12 May, 15 May, 19 October, 21-23 November 2019, 1-2 December, and 5 December 2018. During the six PM 2.5 -polluted months (i.e., December 2018, January, April, May, October, and November 2019), the WS on PM 2.5 pollution days (1.05 ± 1.06 m s −1 ) was lower than that on PM 2.5 compliance days (1.43 ± 1.06 m s −1 ), indicating the weaker ability of winds to dilute and diffuse precursors on PM 2.5 pollution days. The mean X/E value on PM 2.5 compliance days (1.47) was slightly higher than that on PM 2.5 pollution days (1.44), indicating enhanced secondary transformation of VOCs on PM 2.5 pollution days.

Estimating O 3 and PM 2.5 levels contributed by emissions
O 3 and secondary aerosols are primarily formed via photochemical reactions in the atmosphere, concentrations of which could be largely influenced by meteorological conditions Feng et al., 2020;Zhai et al., 2019). In this work, the respective contributions of meteorology and emissions to PM 2.5 and O 3 variations were determined using the RF model as described in Sect. 2.3. The coefficients of determination (R 2 ) for the RF model in predicting PM 2.5 and O 3 are 0.85 and 0.91, respectively (shown in Fig. S2). The respective contributions of anthropogenic emissions and meteorology to O 3 and PM 2.5 during each period is shown in Fig. 4. During the O 3 -polluted months, the meteorologically driven O 3 level on O 3 pollution days (72.5 µg m −3 ) was significantly higher than that on O 3 compliance days (35.3 µg m −3 ). After removing the meteorological contribution, the residual emission-driven O 3 level on O 3 pollution (45.3 µg m −3 ) and compliance days (44.9 µg m −3 ) of the O 3 -polluted months was almost identical and was significantly higher than that during the non-O 3 -polluted months (23.8 µg m −3 ). The emission-driven PM 2.5 level was in the order of: PM 2.5 pollution days of the PM 2.5 -polluted months (55 µg m −3 ) > PM 2.5 compliance days of the PM 2.5polluted months (44 µg m −3 ) > non-PM 2.5 -polluted months (29 µg m −3 ). These results suggest that apart from meteorological factors, emissions also play a role in deteriorating PM 2.5 and O 3 pollution, and reducing anthropogenic emissions is essential for improving air quality. The VOCs/NO x ratio has been widely used to distinguish whether O 3 formation is VOC-limited or NO x -limited (B. Li et al., 2019). Generally, a VOC-sensitive regime occurs when VOCs/NO x ratios are below 10, while an NO x -sensitive regime occurs when VOCs/NO x ratios are higher than 20 (Hanna et al., 1996;Sillman, 1999). In this study, the values of VOCs/NO x (ppbv ppbv −1 ) were all below 3 during both the O 3 -polluted and non-O 3 -polluted months (Fig. S3), suggesting that O 3 formation was sensitive to VOCs, and thus the reductions of the emissions of VOCs will be beneficial for O 3 alleviation.

Indication from tracers
The great changes in the mixing ratios of different species are mainly affected by photochemical processing, emission inputs, and ratios of VOC species having similar atmospheric lifetimes can reflect the source features (B. Li et al., 2019;Raysoni et al., 2017;Song et al., 2021). The ratio of ipentane to n-pentane is widely used to examine the impact of vehicle emissions, fuel evaporation, and combustion emissions within the i/n-pentane ratios of ranging 2.2-3.8, 1. 8-4.6, and 0.56-0.80, respectively (McGaughey et al., 2004;Jobson et al., 2004;Russo et al., 2010;Yan et al., 2017). As shown in Fig. 6, the i/n-pentane ratios during the PM 2.5 -polluted months were mostly within the range of 0.3-2.0, suggesting the pentanes were from mixed sources of coal combustion and fuel evaporation. During the non-PM 2.5 -polluted months, the i/n-pentane ratios were distributed in the range of 1.3-3.4, indicating strong impacts from vehicle exhaust and fuel evaporation. During the O 3polluted months, most of the i/n-pentane ratios (1.5-2.5) were distributed within the reference range of vehicle exhaust and fuel evaporation, whereas most of the i/n-pentane ratios during the non-O 3 -polluted months ranged 1.7-2.1, suggesting the significant impact of fuel evaporation.
The toluene / benzene (T/B) ratio is a widely used indicator for sources of aromatics. In areas heavily impacted by vehicle emissions, the T/B ratio lies in the range of 0.9-2.2 (Qiao et al., 2012;Dai et al., 2013;Yao et al., 2013;Zhang et al., 2013;Yao et al., 2015;Mo et al., 2016;Deng et al., 2018). Higher T/B ratios were reported for solvent use (greater than 8.8) (Yuan et al., 2010;Zheng et al., 2013) and industrial processes (1.4-5.8) (Mo et al., 2015;Shi et al., 2015). In burning source emission studies, the T/B ratio was below 0.6 in different combustion process and raw materials (Tsai et al., 2003;Akagi et al., 2011;Mo et al., 2016). Most of the T/B ratios during the PM 2.5 -polluted and non-PM 2.5 -polluted months were within the range of 1.1-1.8 and 0.8-2.2, whereas the T/B ratios were mostly distributed within the range of 0.8-2.2 and 0.9-1.9 during the O 3 -polluted and non-O 3 -polluted months, respectively, suggesting the significant impact of vehicle and industrial emissions.

PMF
The factor profiles given by PMF and the contribution of each source to ambient VOCs during each period are presented in Figs. 7 and 8, respectively. Six emission sources were identified: coal and biomass burning, solvent use, industrial sources, oil gas evaporation, gasoline vehicle emission, and diesel vehicle emission based on the corresponding markers for each source category. In general, diesel vehicle exhaust, gasoline vehicle exhaust, and industrial emissions were the main VOC sources during both the O 3 -polluted and PM 2.5 -polluted months, with total contributions of 62 % and 62 % on O 3 pollution and compliance days of the O 3polluted months, as well as 66 % and 59 % on PM 2.5 pollution and compliance days of the PM 2.5 -polluted months, respectively. The O 3 -polluted months exhibited higher proportions of diesel (24 % on O 3 compliance days and 27 % on O 3 pollution days) and gasoline vehicle emission (17 % on O 3 Figure 6. Ratios of ito n-pentane and toluene to benzene at different PM 2.5 and O 3 levels. compliance days and 16 % on O 3 pollution days) compared with the non-O 3 -polluted months (8 % and 13 %, respectively). During the O 3 -polluted months, the contributions of industrial emissions (22 %) and fuel evaporation (18 %) on O 3 pollution days were much higher than those on O 3 compliance days (18 % and 13 %, respectively). Figure 9 presents the relative contributions of individual VOC sources from PMF to OFP. On the basis of O 3 formation impact, diesel and gasoline vehicle exhaust were major contributors. During the O 3 -polluted months, vehicle emissions and fuel evaporation showed higher OFP values on O 3 pollution days (93.9 and 35.5 µg m −3 ) compared with those on O 3 compliance days (88.0 and 25.8 µg m −3 , respectively). Although industrial emissions act as an important source for VOC concentrations on O 3 pollution days (shown in Fig. 8), the potential to form O 3 is limited, accounting for 11 % of the total OFP. As illustrated in Fig. 7, the industrial source was distinguished by high compositions of alkanes but relatively lower compositions of alkenes and aromatics, resulting in low O 3 formation potentials. Such results suggested that the fuel use and diesel vehicle exhaust should be preferentially controlled for O 3 mitigation.
The PM 2.5 -polluted months showed higher proportions of industrial (29 % on both PM 2.5 compliance and PM 2.5 pollution days) as well as coal and biomass combustion emissions (16 % on PM 2.5 compliance days and 18 % on PM 2.5 pollution days) compared with the non-PM 2.5 -polluted months (17 % and 10 %, respectively). The PM 2.5 pollution days were dominated by industrial emission (29 %), diesel vehicle exhaust (24 %), and combustion sources (18 %). During the PM 2.5 -polluted months, the contribution of diesel vehi- cle exhaust on PM 2.5 pollution days (24 %) was higher than that on PM 2.5 compliance days (16 %). On the basis of PM 2.5 formation impact, diesel vehicle exhaust and combustion were two major contributors on PM 2.5 pollution days (shown in Fig. 9), and these two sources showed obvious higher SOAFP on PM 2.5 pollution days (0.30 and 0.32 µg m −3 , respectively) compared with those on PM 2.5 compliance days  of the PM 2.5 -polluted months (0.15 and 0.14 µg m −3 , respectively). Although industrial emissions act as an important source for VOC concentrations on PM 2.5 pollution days, the potential to form PM 2.5 is limited, accounting for 16 % of the total SOAFP. The above results suggested that diesel vehicle exhaust and combustion should be preferentially controlled to alleviate PM 2.5 pollution.
Based on the mass concentrations of individual species in each source, the following were the dominant species contributing to photochemical O 3 formation: m/p-xylene, oxylene, methyl methacrylate, vinyl acetate, 1-hexene, and acrolein in gasoline and diesel vehicular emissions; toluene, trans-2-butene, and 1-pentene in fuel evaporation and diesel vehicular emissions; and acrolein in solvent, gasoline vehicular, and diesel vehicular emissions (Fig. 10). The following were the dominant contributors to SOA formation during the PM 2.5 pollution periods (Fig. 10): toluene, m/p-xylene, o-xylene, styrene, ethylbenzene, 1-pentene, and 1,2,3-trimethylbenzene from combustion and diesel vehicular emissions; 1-hexene from diesel vehicular emissions; and methyl cyclopentane from combustion, industrial, and diesel vehicular emissions.

Limitation
This study analyzed the VOC sources and their contributions to O 3 and SOA formation across different seasons. It should be pointed out that the sampling campaign for VOC measurement was not conducted continuously during December 2018 and November 2019. For instance, the air samples were not collected in August and February-March 2019, during which pollution events of O 3 and PM 2.5 occurred, respectively. The variations, sources, and secondary transformation potentials of VOCs, particularly for O 3 and PM 2.5 pollution periods, cannot be fully depicted. Despite the uncertainties that remain, the results obtained in this study provide useful information for VOC emission control strategies and assist in overcoming air pollution issues in Beijing.

Conclusions
In this work, a field sampling campaign of VOCs was conducted in urban Beijing from December 2018 to November 2019. The VOC concentrations ranged from 5.5 to 118.7 ppbv with a mean value of 34.9 ppbv. Alkanes, OVOCs, and halocarbons were the dominant chemical groups, accounting for 75 %-81 % of the TVOC across the sampling months. By excluding the meteorological impact, the emission-driven O 3 levels during the O 3 -polluted months were higher than during the non-O 3 -polluted months, and a similar pattern was found for PM 2.5 . The molar ratio of VOCs to NO x indicated that O 3 formation was limited by VOCs during both the O 3 -polluted non-O 3 -polluted months, and thus reducing VOC emissions is essential for allevia- tion of O 3 pollution. The contributions of coal and biomass combustion, solvent use, industrial sources, oil and gas evaporation, gasoline exhaust, and diesel exhaust were identified based on PMF analysis. Considering both the concentration and maximum incremental reactivity of individual VOC species for each source, fuel use and diesel exhaust sources were identified as the main contributors of O 3 formation during the O 3 -polluted months, particularly the VOC species of toluene, xylenes, trans-2-butene, acrolein, methyl methacrylate, vinyl acetate, 1-butene, and 1-hexene, illustrating the necessity of conducting emission controls on these pollution sources and species to alleviate O 3 pollution. VOCs from diesel vehicles and combustion were found to be the dominant contributors for SOAFP, particularly the VOC species of toluene, 1-hexene, xylenes, ethylbenzene, and styrene, and top priority should be given to these for the alleviation of haze pollution. Data availability. Meteorological data and concentrations of air pollutants including PM 2.5 , O 3 , and NO x are available from the authors upon request. The daily mixing ratio of individual VOC species can be accessed through https://doi.org/10.5281/zenodo.6888723 (Cui et al., 2022).
Author contributions. DW designed the study and performed the VOC measurements. QX and RH assisted in air sampling and data collection. LC performed the data analysis and wrote the paper with contributions from all co-authors. SW and JH reviewed the paper and provided comments for improving the paper.
Competing interests. The contact author has declared that none of the authors has any competing interests.
Disclaimer. Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Financial support. This work was supported by the National Natural Science Foundation of China (grant no. 92044302) and the Beijing Municipal Science and Technology Project (grant nos. Z211100004321006 and Z191100009119001).
Review statement. This paper was edited by Eleanor Browne and reviewed by two anonymous referees.