Scattered coal is the largest source of ambient volatile organic compounds during the heating season in Beijing

We identified scattered coal burning as the largest contributor to ambient volatile organic compounds (VOCs), exceeding traffic-related emissions, during the heating season in Beijing prior to the rigorous emission limitations enacted in 2017. However, scattered coal is underestimated in emission inventories generally, because the activity data are incompletely 10 recorded in official energy statistics. Results of positive matrix factorization (PMF) models confirmed that coal burning was the largest contributor to VOC concentrations prior to the emission limitations of 2017, and a reduction in scattered coal combustion, especially in rural residential sector, was the primary factor in the observed decrease in ambient VOCs and secondary organic aerosol (SOA) formation potential in urban Beijing after 2017. Scattered coal burning was included in a corrected emission inventory and we obtained comparable results between this corrected inventory and PMF analyses, 15 particularly for the non-control period. However, a refined source sub-classification showed that passenger car exhaust, petrochemical manufacturing, gas stations, traffic evaporation, traffic equipment manufacturing, painting, and electronics manufacturing are also contributors to ambient VOCs. These sources should focus on future emission reduction strategies and targets in Beijing. Moreover, in other region with scattered coal-based heating, scattered coal burning is still the key factor to improve the air quality in winter. 20 https://doi.org/10.5194/acp-2020-168 Preprint. Discussion started: 30 April 2020 c © Author(s) 2020. CC BY 4.0 License.

many researches (Gilman et al., 2015;Redington and Derwent, 2013;Li et al., 2015a), were used to evaluate precursor source contributions and variation within different control periods on SOA formation. The definition of this SOAP method describes the mass of aerosol produced per mass of VOC reacted and expressed relative to toluene, which is different from the absolute SOA formation potential value (the mass of aerosol formed per mass of VOC reacted).
SOA potentials of this method were simulated under test conditions of high anthropogenic emissions of VOCs and NOx 160 (Derwent et al., 1998). Due to the low contribution from natural emissions, anthropogenic SOAs predominant in this scenario.
Toluene was chosen as the basic compound for SOAP estimation because of its well-characterized man-made emission status and importance as an SOA precursor (Ng et al., 2007). The amount of SOAs formed is described using a toluene-equivalent, and SOAPs of each compound are expressed as an index relative to toluene. The SOAP represents the propensity for an organic compound to form SOA when an additional mass emission of that compound is added to the ambient atmosphere expressed 165 relative to that SOA formed when the same mass of toluene is added (Derwent et al., 2010). We hypothesized that all VOC species would have an effect on SOA formation. SOAP-weighted mass contributions were calculated based on PMF results (where VOC units were converted from ppbv to µg m -3 ) and the corrected emission inventory (Gg), respectively. The SOAPweighted mass contribution of each VOC source can be calculated using Eq. (1): where VOCi is the mass contribution of a VOC source to species i (µg m -3 / Gg); SOAPi is the SOA formation potential for species i (unitless). Table S2 shows a listing of the propensities for secondary organic aerosol formation expressed on a mass emitted basis as SOAPs relative to toluene=100 for 113 organic compounds.
This SOAP method removes issues associated with uncertainty in absolute SOA concentrations (Li et al., 2015a). Besides, this SOAP method is appropriate for conditions of high anthropogenic emissions of VOCs and NOx (Derwent et al., 1998). 175 Although highly idealized, these conditions are comparable to those in urban Beijing during control and non-control periods.

Unprecedented air pollution control measures in China
In 2013, The People's Republic of China State Council, in determining that improving air quality was not only a human health issue but was also an important focus of economic growth and security, deployed the Action Plan of Air Pollution Prevention 180 and Control (the Action Plan) (http://www.gov.cn). Emission control measures implemented in the Beijing Action Plan (2013 -2017) were summarized by Cheng et al. (2019). The Action Plan mandated that the average annual concentration of PM2.5 had to be limited to 60 μg m -3 in Beijing, and reduced by over 25%, relative to a 2012 baseline, in BTH by 2017. Since 2013, further plans and laws, namely the Air Pollution Prevention Law of 2016, were released to curb emissions and meet air-quality targets. After 3 years of these efforts (2013 -2016), air-quality improvement was less than satisfactory. Hence, in early 2017, 185 the Chinese government released Ten Heavier Measures to Prevent and Control Air Pollution (the Ten Measures) in Beijing.
A detailed description of these enhanced control measures is shown in Table S3. Neighboring provinces, including Tianjin, Hebei, Shandong, Shanxi, and Henan, cooperated with Beijing to increase the effectiveness of these measures. Further, the Beijing Municipal Government promoted a 2017 revision of the Emergency Plan for Heavy Air Pollution in an effort to confront future heavy pollution periods. These enhanced measures had demonstrable effects on air quality and PM2.5,and 190 Beijing has since met the targets laid out in the Action Plan. In 2017, the mean concentration of PM2.5 was 58 μg m -3 , with a year-on-year reduction of 20.5%. This effort was a huge success, and a series of researches associated with the impact of these clean air actions have been launched (Zheng et al., 2018a;Li et al., 2019a;Xue et al., 2019;Zhang et al., 2019).
Most of them paid attention to the improvement of air quality and health benefits, the transition of PM2.5 chemical composition and contributors, and the trend of anthropogenic emissions. But nonetheless sufficiently detailed information on ambient VOC 195 mixing ratios and chemical compositions, as well as variation in emission sources after controls were established, have not been reported.
Vehicle exhaust, gasoline evaporation, fuel combustion, solvent utilization, and industrial production are the most prevalent sources of VOCs in Beijing, particularly during hazy days (Wu et al., 2016b;Guo et al., 2012;Sheng et al., 2018). These five https://doi.org/10.5194/acp-2020-168 Preprint. Discussion started: 30 April 2020 c Author(s) 2020. CC BY 4.0 License. sources were all indicated as controlled objects under the Ten Measures, and SC burning and high-pollution industries were 200 the most stringent control objects. Compared to 2016, the reduction of civil SC consumption in Beijing exceeded 2 million tons in 2017. Rural and urban residential sectors contributed 74% and 15% of reduction, respectively, followed by commercial and public sector (6%) and rural production sector (5%). SC is consumed more, and has greater VOC emissions per unit combustion, than other fuel types (Fig. 1). Also, a large proportion of civil SC (> 90%) is used for heating in winter. As for industrial sector, sustained clampdown of the coal-fired boilers was put into action in Beijing from 2013, 99.8% of the boilers 205 associated with nearly 9 million tons of SC consumption annually were banned and more than half of that were emerged in 2017. All the industrial scattered coal was eradicated by the end of 2017. Therefore, a lot of VOC emissions from SC burning would be prohibited during the heating season in Beijing.
From 2013 to 2017, 1,992 high-pollution industries were phased out in Beijing, including chemical engineering, furniture manufacturing, printing, and non-metal mineral product industries. At the municipal level, 11,000 "small, clustered, and 210 polluting" factories that did not meet efficiency, environmental, or safety standards were either regulated or closed by the end of 2017, according to the Beijing Municipal Bureau of Economy and Information Technology. Meanwhile, a large part of high-pollution enterprises (those heavy polluting industries as stipulated by the state environmental protection department) were removed out of Beijing year by year. The annual variation of designed size enterprises, high-pollution enterprises, and the annual benefits from industry in Beijing are summarized in Fig. 2. In addition, efforts to increase the quality of gasoline 215 and diesel fuels began in January 2017; these efforts could lead to a marked decrease in traffic emissions.
Over the course of the study period, control measures were differentially enacted. December 2017 was the most tightly controlled period, wherein SC burning and substandard coal-fired boilers were forbidden in an effort to meet the targets identified in the Action Plan. In January 2018, residents were allowed to burn some SC to ensure their well-being. Therefore, we divided our study into three time periods: non-control (December 2016 -January 2017), strict-control (December 2017), 220 and eased-control (January 2018). Cold temperatures and a low mixing layer are related to increased emissions and the https://doi.org/10.5194/acp-2020-168 Preprint. Discussion started: 30 April 2020 c Author(s) 2020. CC BY 4.0 License. accumulation of gaseous pollutants during winter months (Zhang et al., 2015). Thus, this time period represents an ideal opportunity to assess the contributions and effects of various emission sources in Beijing by analyzing ambient VOCs.

Ambient VOC concentrations and source contributions
The implementation of federal and municipal control policies led to significant changes in emission intensity for several 225 sources, which was reflected in ambient VOC concentration characteristics. Fig. 3 shows the ambient VOC concentrations, reported from multiple studies, across seasons in Beijing, where winter generally has the highest values (Liu et al., 2005;Wang et al., 2014;Wei et al., 2018;Li et al., 2019b). Mixing ratios and the chemical composition of VOC groups, as well as the average volume mixing ratios of 91 measured species at PKU, are summarized in Table S4. closer to the characteristic values for traffic exhaust and evaporation, and those in winter are closer to the characteristic value for coal burning. Results from the strict-control period may represent illegal SC, coal-to-gas, and coal-to-electricity use, and we observed an increase in B:T during the eased-control period relative to the strict-control period.
Analyses of variation in tracers can reflect changes in emission sources. We observed a significant decline in methyl tertiary butyl ether (MTBE, a common gasoline additive), 2,2-dimethylbutane, 3-methylpentane, methyl cyclopentane, 2-245 methylhexane, and 3-methylhexane (all common components in gasoline evaporation and tailpipe exhaust) during the control period (Chang et al., 2006). Acetonitrile, an inert tracer, can reflect the intensity of biomass burning (Sinha et al., 2014). We estimated no significant changes in biomass burning by comparing acetonitrile concentrations between the three periods. Freon 113 is typically used to estimate background levels. We found that the concentration of Freon 113 was constant around 0.09 -0.11 ppbv, indicating a consistent background concentration. 250 The top 20 most-decreased VOC species after control measures are listed in Table 1. During the strict-control period, tracers of incomplete burning (e.g., ethylene, acetylene, benzene, styrene, and 1,2-dichloropropane) decreased by > 60%. Tracers of industrial and vehicle-related sources decreased by 50%, including some chlorinated hydrocarbons, esters and aromatics (Li et al., 2016c;Hellen et al., 2006;Barletta et al., 2009). We observed a precipitous decline in methacrolein (MACR) and methyl vinyl ketone (MVK), which are the major oxidation products of isoprene (Xie et al., 2008). Terrestrial vegetation is typically 255 the main contributor of isoprene in the environment. However, heavy traffic in megacities contributes to a large proportion of isoprene emissions, particularly after leaf-drop (Song et al., 2007). Ethyl acetate is a widely used industrial solvent, and propene is characteristic product of internal combustion engines (Scheff and Wadden, 1993). Some aromatics, such as styrene and benzene, are found in high concentrations in petrochemical plants (Liu et al., 2008a). Benzene, toluene, ethylbenzene, and xylenes (BTEX) are also major components of vehicle and solvent utilization (Seila et al., 2001). During the eased-control 260 period, we observed differences in the top declining species and their respective reductions, particularly for tracers of vehicle exhaust, which had relatively small reductions. Indicator species of oil-refining and fuel burning emissions became more prevalent during this period, including styrene, C2-C4 alkenes, C3-C10 alkanes, and acetylene. Fuel evaporation is often indicated by iso-/n-pentane and cyclopentane (Zheng et al., 2018b), both of which showed an obvious decline during the easedcontrol period. In both the strict-and eased-control periods, acetylene, a tracer for vehicular and other combustion processes (Baker et al., 2008), decreased by > 60%. Secondary products from primary anthropogenic VOCs, including ketones and aldehydes, were also reduced (Yuan et al., 2012).
PMF, a receptor-based source apportionment method, was used to estimate temporal variation in source contributions. Eight appropriate factors were determined. Profiles from the literature were referenced in identifying the factor profiles, which were recognized as: (1) coal burning, (2) fuel oil and gas usage, (3) traffic exhaust, (4) petroleum-related evaporation, (5) VOC-270 related industry, (6) VOC-product utilization, (7) biomass burning, and (8)  Source contributions (ppbv) and proportions, determined by PMF analyses, are shown in Table 2. Source reduction contributions during strict-and eased-control periods relative to the non-control period are shown in Fig. 5.
PMF is a widely used method to identify emission sources and their contributions (Yuan et al., 2009;Simayi et al., 2020), but its results have some subjectivity and cannot be determined to be absolutely accurate. For this reason, PMF results are usually mutually corroborated with the actual situation, which is the implementation of control measures in this study: during the 280 strict-control period, coal burning had the greatest reducing contribution (54.33%) relative to the non-control period, followed by petroleum-related evaporation (31.49%) and VOC-related industry (16.25%). During the eased-control period, coal burning contributed 49.33% of total reduction relative to the non-control period, as did petroleum-related evaporation (24.03%) and composed of coal combustion, was the largest VOC contributor in winter. The contribution proportions of fuel combustion in winter ranged from 45% -55% (Li et al., 2015a;Yang et al., 2018;Li et al., 2019b), which are even higher than, but still 290 comparable with that of non-control period in this study (37%). Other studies in Table S7 show that vehicle-related source is the largest VOC contributor in Beijing, especially in summer and autumn, with the contribution ranged from 50% -57%, and 33% -49%, respectively. And the smaller, and comparable contribution of vehicle-related source in winter is reflected among other studies and this study. During the non-control period, coal burning contributed 37% (33.5 ppbv) of the total ambient VOCs, far surpassing the contributions of other emission sources, even the combined influence of traffic exhaust and 295 petroleum-related evaporation (33%, 30.0 ppbv).
The terminal sectors of the burned coal include centralized coal burning (power generation, heat supply, large-scale industrial boilers), scattered coal burning (rural and urban residential consumption, rural production, commercial and public consumption, small-scale industrial boilers). Of the overall coal burning in PMF results, SC burning, whose emission factors are far above centralized coal burning, could contribute much higher emissions than centralized part. Residential and industrial sectors were 300 the majority part of scattered coal burning in Beijing, both of them contributed more than 90% of all. Generally, residential SC burning is mostly concentrated during heating season and most of urban families are centrally supplied heating without SC consumption. It makes residential sectors, especially rural residential sector, more significant in winter than industrial sector.
Monthly distribution of industrial sector is relatively average throughout the year, therefore, SC used in small-scale industrial boilers and furnaces is much less than residential sectors in winter, Furthermore, higher combustion efficiency and lower VOC 305 emission factors of industrial sector than civil utilization may make its emissions contribution lower (Bo et al., 2008;Cheng et al., 2017). A small contribution of coal burning in summer, which is held up by the PMF results of other studies in Table S7, corroborates the important effect of residential SC burning. Detailed estimation of emissions from coal burning is given in section 3.3.

Corrected emission inventory for VOCs
The reduction in SC burning in December 2018 was reflected in both the emission inventory and PMF results. Industrial SC consumption was estimated in the emission inventory but with high uncertainty (Table S5), however, its relatively small proportion of the total SC consumption in Beijing during heating season would largely reduce its influence on the estimations of total emissions from SC burning.
Based on the PMF results and the estimated emissions of other sources, coal burning contributed 36.2 ± 10.4 Gg and 14.7 ± 325 8.6 Gg of anthropogenic VOC emissions in Beijing during the non-control and control periods, respectively. However, at least 80% of the VOC emissions from coal burning were not considered in most existing emission inventory researches (Li et al., 2019c;Wu and Xie, 2018;Li et al., 2019b). Calculated emissions of SC burning in the corrected emission inventory were 23.7 Gg and 10.5 Gg during the non-control and control periods, respectively, which are comparable to, but lower than, the estimations calculated using PMF. We note that uncertainty in heating demand and coal quality is a potential reason for these 330 differences.
The corrected emission inventory had a more comparable proportional distribution to that of the PMF results during the noncontrol period. However, during the strict-and eased-control periods, we observed poor consistency between the PMF and emission inventory results, particularly for VOC-related industry and petroleum-related evaporation. This could be the result of neglected factors, such as changes in industrial petrochemical and chemical production during air pollution alerts, and the 335 shut-down of many polluting industries in 2017. We therefore suggest that the emission inventory results for the non-control period are reliable, but VOC-related industry and petrochemical evaporation are overestimated for the control periods.
Total emissions during the non-control period amounted to 82 Gg, and decreased to 59 Gg during the control period. The largest sublevel source, coal burning, had the largest reduction (19 Gg), and the reduction was mostly contributed by rural residential SC burning. Other sublevel sources also contributed to the overall decline in emissions, such as fuel oil and gas 340 usage (1 Gg), traffic exhaust (4 Gg), petroleum-related evaporation (3 Gg), and VOC-related industry (2 Gg). Emissions from VOC-product utilization and biomass burning did not change significantly between the non-control and control periods. Of the total coal burning during non-control and control period, rural residential SC burning contributed 60% and 68%, respectively; urban residential SC burning contributed 17% and 25%, respectively; industrial SC burning contributed 16% and 0%, respectively; centralized coal burning only contributed 1% and 2%, respectively. Emissions of seven sub-level 345 anthropogenic sources of level 1, major refined sub-contributors of each anthropogenic source, and reduction contribution of each refined sub-contributor from non-control to control period are shown in Fig. 7.

SOAP calculations based on PMF results and the corrected emission inventory
In urban areas, anthropogenic-produced VOCs are major precursors to SOAs (Lin et al., 2009). The formation of SOAs from VOCs occurs through varied, complex, physical and chemical processes. These are broadly categorized under three main 350 theories within the literature: mechanisms related to photooxidation, nucleation processes, and condensation, gas/particle partition, and heterogeneous reactions (Hallquist et al., 2009;Kroll and Seinfeld, 2008).
Organic gaseous compounds can condense on primary particles, of which the greatest number are within 0.1 -1 μm. Primary particles rarely coagulate, but do undergo species (including VOC species) exchange in the gas phase. Transformation of https://doi.org/10.5194/acp-2020-168 Preprint. Discussion started: 30 April 2020 c Author(s) 2020. CC BY 4.0 License. organic vapors to a liquid or solid phase is promoted when the equilibrium vapor pressure is above that of the aerosol surface 355 (Raes et al., 2000). Generally, molecular clusters tend to evaporate owing to the stronger Kelvin effect, but fulminic nucleation will occur under suitable conditions. Zhang et al. (2004) suggested that nucleation (new particle formation, NPF) is greatly enhanced by an interaction between organic and sulfuric acids, particularly in an atmosphere polluted by heavy coal burning.
Two types of NPF events, sulfates-dominated and organics-dominated, have been identified on the North China Plain (Ma et al., 2016). Condensation and self-coagulation begin, and thus promote growth, around 0.1 μm. Continuous growth of 360 nucleation-mode particles over several days would lead to haze in Beijing, which has more abundant precursors in the atmosphere (Guo et al., 2014;Wang et al., 2013b). NPF has been recognized as an important process contributing to the formation of cloud condensation nuclei (CCN), concentrations of which have increased by 0.4 -6 times in and around Beijing (Yue et al., 2011). Typically, organic matter contributes significantly to the mass growth that is characteristic of newly formed SOAs (Pennington et al., 2013). In Beijing, organic matter is likely the dominant chemical contributor facilitating the 365 conversion of newly formed particles to CCN.
We used the SOAP approach to determine the effectiveness of the air-quality control period as it pertained to a reduction in VOC emissions. We note that because SOA formation processes are poorly understood, SOAP was computed to understand the potential for SOA formation from VOC species, but we could not estimate the actual formation under specific atmospheric conditions. The concentration of PM2.5 was reportedly reduced in 2017 (Beijing Municipal Environmental Protection Bureau, 370 http://www.bjepb.gov.cn/). The observed large reduction in VOCs may have resulted in the reduction of SOAs and thereby contributed to the reduction in PM2.5. SOAP-weighted mass contributions of each VOC source were used to estimate the influence of precursor emissions on SOAs. SOAP-weighted mass contributions based on PMF results and corrected emission inventory are shown in Figs. 8 and 9. According to the factor profiles of PMF results, styrene, toluene, benzene and xylene, as the major contributors of SOAP, were largely attributed by coal burning (about 40% on average), which accounted for 27% of 375 ambient VOCs but contributed 40% of total SOAP during the whole study period. Besides, according to the corrected emission inventory, 47% of benzene, 27% of toluene and 10% of xylene were contributed by coal burning as well.
The greatest contributor to SOAP reduction was coal burning, according to the results of both the PMF and the emission inventory, despite differences in their respective raw values. PMF results indicated that coal burning was the greatest contributor to SOAP at 18.81 μg m -3 , accounting for 47% during the non-control period. After control measures were enacted, 380 the contribution of coal burning decreased to 4.56 μg m -3 , and accounted for 30% of the total. The reduction of coal burning contributed approximately 55% of the total reduction. Petroleum-related evaporation and VOC-related industry contributed 25% and 10% of the total SOAP reduction prior to the establishment of control measures, respectively, and emissions from both sources were reduced over the control period.
Emission inventory results indicated that VOC-related industry was the largest contributor to the total SOAP-weighted mass 385 during both the non-control and control periods. SC burning was the next largest contributor during the non-control period, contributing 64% of the total.

Evaluation of control policies
The idea that reducing residential coal burning will improve air quality in Beijing has been well accepted within the scientific community (Cheng et al., 2016). Liu et al. (2016) proposed that reductions in residential emissions may have greater benefit 390 to air quality in Beijing than reductions from other emitters during the heating season, and that promoting alternative fuels may be an effective solution. Our study confirmed that coal burning is the greatest contributor to VOC emissions (see PMF results), and we inferred that a large proportion of emissions were the result of SC burning during the heating season in noncontrol periods (see corrected emission inventory). The contributions of identified VOC sources decreased significantly after the control period, which means that VOC-related control measures were highly effective. The sharp decrease in SC burning 395 between the non-control and control period related to a reduction in the contribution of coal burning to ambient VOC concentrations. Multiple related measures were enacted during the control period, including the prohibition of SC burning in the countryside, deactivation of coal-fired units in thermal power plants, and conversion of decentralized coal-fired boilers to gas-fired boilers. Coal burning increased slightly between the strict-and eased-control periods, reflecting the allowance of residential SC burning in January 2018. Petroleum-related evaporation in PMF results was sharply reduced by controlling high-emission vehicles and reducing leakage from petrochemical industries. We accordingly confirmed that the contribution of coal burning to ambient VOC concentrations exceeded that of traffic-related sources prior to the strict-control period, and that coal burning was the greatest contributor to higher VOC concentrations observed in winter. During and after the strictcontrol period, VOC concentrations decreased and vehicle exhaust became the main contributor again.
The limitations set by federal and municipal governments on coal burning played a significant role in improving air quality 405 during Beijing's winters. According to the statistics of CESY, the total coal consumption of Beijing reached its highest point of over 30 million tons in 2005, then it kept going down to 4.9 million tons in 2017. And the percentage of coal occupied in primary energy consumption dropped sharply from at least 30% to less than 6% in the past decade. It is reported by State Grid Beijing that Beijing had implemented "switching from coal to electricity" project for residential heating in winter since 2003, and carried out electric heating exceeding 1.2 million families by 2018. Although significant effects have been achieved in 410 Beijing, civil SC consumption is still widely used nationwide.
For Beijing city, we corrected the SC consumption of both civil and industrial sectors, of which emission factors and monthly activity data are presented in Tables S1 and S5, respectively. But for the nationwide SC estimation, the consumption from industrial sector was difficult to count in most areas, and only data of civil sectors were available. Hence, only civil SC (rural residential, urban residential, rural production, commercial and public sectors) was discussed (Huo et al., 2017). In 2016, total 415 civil SC consumption was 311.4 million tons in China, and near 65% of them came from rural residential sector. Due to the unprecedented air pollution control measures, civil SC consumption in 2017 achieved a reduction of 18.7 million tons, and rural residential sector took up over 74% of all (Tables S8 and S9). BTH and surrounding provinces, including Beijing, Tianjin, Hebei, Shanxi, Shandong, Henan and Neimenggu, contributed 95% of total reduction (17.8 million tons). However, after the centralized limitation, BTH and surrounding provinces is still areas with the largest civil SC consumption; Heilongjiang, 420 Guizhou, Hunan and Xinjiang are also provinces with large numbers of civil SC consumption (Fig. 10). To achieve global Sustainable Development Goals (SDGs) (Carter et al., 2019), the control of civil SC consumption is a significant topic to improve air quality, and emission reduction of industrial SC combustion is also worth attention.
https://doi.org/10.5194/acp-2020-168 Preprint. Discussion started: 30 April 2020 c Author(s) 2020. CC BY 4.0 License. Table captions:   Table 1. The top 20 major VOC species with the highest decreasing ratios compared with non-control period. Table 2. Source contributions to ambient VOCs concentration (ppbv) and their contributions to total reduction compared with noncontrol period (ratio) derived by PMF analysis.