Spatiotemporal Variation, Sources, and Secondary Transformation Potential of VOCs in Xi’an, China

As a critical precursors of ozone (O3) and secondary organic aerosols, volatile organic compounds (VOCs) play a 10 vital role in air quality, human health, and climate change. In this study, a campaign of comprehensive field observations and VOC grid sampling was conducted in Xi’an, China from June 20 to July 20, 2019 to identify the spatiotemporal concentration levels, sources, and secondary transformation potential of VOCs. During the observation period, the average VOC concentrations at the Chanba (CB), Di Huan Suo (DHS), Qinling (QL), and gridded sampling sites were 27.8 ± 8.9, 33.8 ± 10.5, 15.5±5.8, and 29.1±8.4 ppb, respectively. Vehicle exhaust was the primary source of VOC emissions in Xi’an, and the 15 contributions of vehicle exhaust to VOCs at the CB, DHS, and QL sites were 41.3%, 30.6%, and 23.6–41.4%, respectively. While industrial emissions were the second-largest source of VOCs in urban areas, contributions from ageing sources were high in rural areas. High potential source contribution function values primarily appeared in eastern and southern Xi’an near the sampling site, which indicates that Xi'an exhibits a strong local VOC source. Moreover, alkenes, aromatics, and oxygenated VOCs played a dominant role in secondary transformation, which is a major concern in reducing O3 pollution in Xi’an. 20


Photochemical reactive activity parameterization
The loss rates of VOCs that react with OH radicals (LOH) and the O3 formation potential (OFP) can be used to characterize VOC photochemical activity (Carter, 2010;Niu et al., 2016;Wu et al., 2016a). LOH and OFP can be calculated using Eqs. (1) and (2) where n represents the number of VOCs, [VOC]i represents the i th VOC species concentration, KOH represents the rate coefficient for the reaction of the i th VOC species with OH radical (molecule -1 · cm 3 ·s -1 ), and MIRi is the maximum incremental reactivity for the i th VOC species. The KOH and MIR for each VOC specie were taken from the updated Carter research results 100 (http://www.engr.ucr.edu/~carter/reactdat.htm).
Specific VOC ratios are often used to calculate the air mass photochemical age or OH exposure (Jimenez et al., 2009;Roberts et al., 1984). OH exposure can be calculated using Eq. (3).
where OH exposure represents OH radical concentration multiplied by the reaction time (Δt); kvoc1 and kvoc2 represent rate 105 coefficients for the reaction of the VOC species with OH radical (molecule -1 · cm 3 ·s -1 ); [VOC1] [VOC2] | t=0 represents the initial emission ratio of specific VOCs, which can be replaced by the highest concentration ratio in periods where the photochemical reaction is weak; and [VOC1] [VOC2] represents the concentration ratio of the specific VOCs in the atmosphere.

Positive matrix factorization (PMF)
The PMF analysis model was first proposed by Paatero and Tapper (1994). For more than two decades, PMF has been widely 110 used to identify and quantify major sources of VOCs (He et al., 2019;Li et al., 2015;Miller et al., 2002;Pallavi et al., 2019;Song et al., 2007;Yuan et al., 2010). The definition and usage of the PMF model is described elsewhere in detail (Liu et al., 2020;Song et al., 2018;Su et al., 2019), and only a brief description is provided here.
In this study, the PMF 5.0 model (EPA) was used to analyze the VOC sources at the CB, DHS, and QL sites. Based on the potential VOCs tracers of different emission sources in China (He et al., 2015;Liu et al., 2008a;Song et al., 2018), VOC 115 tracers with a data integrity greater than 75% and greater than 65% valid data (concentration ≥ MDL) were selected as the input species. In this study, the number of VOC tracers (input species) at the CB, DHS, and QL sites were 32, 34, and 32, respectively. A concentration (Conc.) file and uncertainty (Unc) file are required by PMF. For the concentration file, data below the detection limit were assigned with MDL/2, and the missing data were substituted with mean concentration. The uncertainty is calculated using Eq. (4) as follows: Conc. ≤ MDL 4 × mean conc. Missing data In this study, the PMF factor numbers were explored from 4 to 8 for the optimal solution in the three sites. Qtrue / Qrobust and Qtrue / Qexpected are two important parameters for characterizing the rationality of the PMF results (Brown et al., 2015). After comparing the PMF results, Qtrue / Qrobust ratio, and Qtrue / Qexpected ratio, a seven-factor PMF solution was selected for VOC source apportionments in the three field observation sites. The Qtrue / Qrobust values at the CB, DHS, and QL sites were all 1.0. 125 The Qtrue / Qexpected values at the CB, DHS, and QL sites were 1.3, 1.2, and 1.0, respectively. In addition, an Fpeak parameter, from -1.0 to 1.0 (step of 0.1), was used to rotate the PMF factors for a superior solution (Sun et al., 2012). In this study, the factor rotating results were not significantly different than the non-rotation results. Thus, the results used in this study were from the runs with zero-Fpeak.
through the grid (i, j) is Mij, and the number of endpoints of all the trajectories falling within the grid (i, j) is Nij, then PSCFij 150 can be defined as the ratio of Mij to Nij (Polissar et al., 1999). The weight function Wij was used to increase the accuracy of the model, and PSCFij can be calculated using Eq. (6) as follows: 3 Results and discussion

Temporal variations
Temporal variations in wind speed, wind direction, temperature, relative humidity, O3, and VOCs at the CB, DHS, and QL sites are shown in Figure 2. During the field observation campaign, 99 VOCs were measured, including 29 alkanes, 11 alkenes, 1 alkyne, 16 aromatics, 28 halohydrocarbons, 13 oxygenated VOCs (OVOCs), and 1 acetonitrile. The CB and DHS sites were located in an urban area of Xi'an, while the QL site was located in a rural area. During the observation period, the average 160 VOC concentrations at the CB, DHS, and QL sites were 27.8 ± 8.9, 33.8 ± 10.5, and 15.5±5.8 ppb, respectively. Due to the existence of more emission sources and lower wind speeds (0-2 m/s) in urban areas, the VOC concentrations were higher than those in rural areas. Overall, the VOC concentrations in Xi'an urban sites were approximately twice that of the rural sites. The observation period occurred during summer in Xi'an, and the temperature was high, with average temperatures reaching 26.2±4.3 °C and 25.6±3.9 °C in urban and rural areas, respectively. Higher temperatures may accelerate a secondary 165 transformation of VOCs into O3, resulting in more frequent O3 pollution incidents in Xi'an; the mean O3 concentrations at the CB, DHS, and QL sites reached 50.2±29.9, 47.6±29.4, and 19.7±8.0 ppb, respectively. Therefore, the increase in the concentration of OVOCs on O3 pollution days may indicate a strong secondary conversion ability of VOCs at this stage. 180

Spatial variations
In this study, the VOC grid sampling was used to investigate the spatial variations in VOCs in Xi'an. A total of 20 sites were selected for grid monitoring that covered the entire city of Xi'an. Therefore, the results of the VOC grid sampling were used to represent the levels of the entire city. In the VOC grid sampling, 106 VOCs were measured, including 29 alkanes, 11 alkenes, 1 alkyne, 17 aromatics, 35 halohydrocarbons, 12 OVOCs, and carbon disulfide. The average VOC concentration at the 20 sites 185 on July 1 and July 14, 2019 was 29.1±8.4 ppb, and high VOC concentrations were clustered at the XF, CT, HC, ZYT, and YT sites (Figures 4a and 4b). Of the sites, XF site exhibited the highest VOC concentration of 54 ppb, followed by CT, HC, with concentrations of 41.4, and 38.2 ppb, respectively. XHT site exhibited the lowest concentration of VOCs at 18.0 ppb, followed by WQ and YST sites with concentrations of 19.6 ppb and 19.9 ppb, respectively. Compared with the results of the field observation campaign, the VOC concentration at the CB site was closer to the overall level in Xi'an, and the VOC concentration 190 at the DHS site was significantly higher than the overall. In terms of the VOC composition at each site, alkanes and OVOCs were dominant, accounting for 25.7-39.7% and 22.8-47.4% of the VOC concentration, respectively. In addition, the contribution of OVOCs at the YT site was significantly higher than that of the other sites, indicating that the YT site may be significantly affected by ageing sources (Figure 4a). The top 10 VOC species at the CB, DHS, QL, and grid sampling sites accounted for 66.1%, 63.4%, 71.1%, and 67.1% of the TVOC concentration, respectively (Table 2). Of these species, ethane, 195 acetone, and propane were the top three contributors in Xi'an, accounting for 34.0-41.3% of the TVOC.

Specific VOC Ratios
Different VOC species may have different sources; hence, the ratio of different species can be used preliminarily analyze the difference in VOC sources at each site. The ratios of specific species that are often used are the toluene/benzene (T/B), m/p-200 xylene/ethylbenzene (X/E), iso-pentane/n-pentane, benzene/ethyne, toluene/ethyne, and m/p-xylene/ethyne ratios.
The T/B ratio is clearly different for various source profiles. In industrial region research, the ratio of T/B ranged from 3.0 to 6.9 when the VOCs were heavily impacted by industrial sources (Chan et al., 2006;Li et al., 2019a;Zhang et al., 2015). In roadside research, the T/B ratio was approximately 1.52, indicating that the VOCs were more impacted by vehicle emissions (Liu et al., 2008a). Research demonstrates that the T/B ratio ranged from 0.2 to 0.6 when the VOCs were significantly impacted 205 by coal and biogenic burning sources (Akagi et al., 2011;Liu et al., 2008a;Wang et al., 2009). Paint solvent usage was more dominant when the T/B ratio was greater than 11.5 (Yuan et al., 2010). Regarding the spatial variations of the T/B ratio in Xi'an ( Figure 5), most of the T/B ratios at the various stations in Xi'an were distributed near the tunnel experiment results, and only a few samples were affected by paint solvent usage. Linear correlation coefficients between toluene and benzene at the https://doi.org/10.5194/acp-2020-704 Preprint. Discussion started: 27 August 2020 c Author(s) 2020. CC BY 4.0 License. CB, DHS, QL, and gridded sampling sites were 0.7 (RPearson=0.5), 2.3 (RPearson=0.6), 0.5 (RPearson=0.6), and 1.2 (RPearson=0.9), 210 respectively. In the CB site, the T/B ratio exhibited a large distribution span and was primarily concentrated between 0.2-6.9, implying that vehicle exhaust, combustion, and industrial sources primarily contributed to the VOCs (Figure 5a). In the DHS site, the T/B ratio was primarily concentrated between 0.6-6.9, indicating that vehicle exhaust and industrial sources primarily contributed to the VOCs (Figure 5b). In the QL site, the T/B ratio was mainly concentrated between 0.2-3, implying that vehicle exhaust and combustion sources primarily contributed to the VOCs (Figure 5c). In the gridded sampling sites, the T/B 215 ratio was predominately concentrated around 1.52, indicating that vehicle exhaust sources greatly contributed to the overall VOCs in Xi'an ( Figure 5d).
The reactivity of m/p-xylene with OH radicals was 2.7 times that of ethylbenzene (Carter, 2010); therefore, a lower X/E ratio represents a higher degree of air mass ageing. In areas with high air mass ageing, the contribution of external source transport to VOCs increased significantly. The diurnal variation of the X/E ratios at the CB, DHS, and QL sites demonstrates that the 220 X/E ratio at the three sites all significantly decreased from 9:00 to 13:00 (Figures 6a-c). This indicated that there was a significant photochemical consumption effect on VOCs between 9:00 and 13:00. In addition, the X/E ratios at the CB and DHS sites significantly increased after 13:00, while variations in the X/E ratio were not clear at the QL site, indicating that there were more primary emissions from anthropogenic sources at the urban sites (CB and DHS sites). The diurnal variations in OH exposure exhibit an inverse correlation with the X/E ratio, reaching a maximum daily value between 12:00-15:00 225 (Figures 6d-f). The average OH exposures of the CB, DHS, and QL sites were 5.1×10 10 , 1.7×10 10 , and 3.1×10 10 molecule cm -3 s, respectively.
The ratio of iso-pentane/n-pentane is clearly different for various sources. Recent studies have found that the iso-pentane/npentane ratio was 2.93 for vehicle exhaust sources (Liu et al., 2008a) and 0.56-0.8 for coal burning (Yan et al., 2017). In this study, linear correlation coefficients between iso-pentane and n-pentane at the CB, DHS, QL, and gridded sampling sites were 230 1.8 (RPearson=0.7), 1.1 (RPearson=0.8), 1.5 (RPearson=0.9), and 3.2 (RPearson=1.0), respectively (Figure 7a). These results indicated that propane sources in Xi'an are greatly affected by vehicle emissions. Propane and ethane are the main components of liquefied petroleum gas (LPG) and natural gas (NG) (Blake and Rowland, 1995;Katzenstein et al., 2003). The propane/ethane (P/E) ratio in LPG vehicle exhaust was approximately 3, which is significantly higher than that in gasoline and diesel vehicles Biogenic sources are characterized by high concentrations of isoprene and the oxidation products of isoprene (methacrolein and methyl vinyl ketone) (Gong et al., 2018;Ling and Guo, 2014;Ling et al., 2019). Factors that meet these characteristics were identified as biogenic sources in this study. Biogenic sources were identified at the DHS and CB sites, accounting for 7.2% in both sites (Figure 8). 280 Biogenic burning sources are characterized by high concentrations of acetylene, methyl chloride, benzene, and toluene (Liu et al., 2008a). The fifth factor of the QL site met this characteristic and was identified as a biogenic burning source, accounting for 8.2% (Figure 8).
Fuel evaporation sources are characterized by high concentrations of iso-butane, n-butane, iso-pentane, n-pentane, 2methylpentane, and 3-methylpentane (Liu et al., 2017;Zheng et al., 2020). The seventh factor of the CB site met this 285 characteristic and is identified as a fuel evaporation source, accounting for 9.8% (Figure 8).
Ageing sources are characterized by high concentrations of OVOCs (Li et al., 2015;Zhu et al., 2018). As important tracers of ageing sources, OVOCs include both primary and secondary sources and have a longer lifetime in the atmosphere (Derstroff et al., 2017). The fourth factor of the QL site met this characteristic and was identified as an ageing source, accounting for 19.2% (Figure 8). The CPF results illustrate that ageing sources at the QL site exhibited a high potential (CPF > 0.8) of source 290 transport from the east when the wind speed exceeded 1 m/s (Figure 9).

Cluster and PSCF results
The 24-h backward trajectories from Xi'an for the cluster and PSCF analysis are shown in Figure 10. Based on the figure, air mass back trajectories can be clustered into the eastern trajectories (Cluster 1), southeastern trajectories (Cluster 2+4), and northeastern trajectories (Cluster 3+5) at the CB site ( Figure 10a). It is evident that the proportion of southeastern trajectories 295 to the total trajectories and that of the southeastern pollution trajectories to the total pollution trajectories were significantly higher than those of the other cluster trajectories, accounting for 58.7% and 60.8%, respectively (Table 3). There were two trajectory clusters from the southeast direction, the southeast short distance trajectories (Cluster 2) and southeast medium-long distance trajectories (Cluster 4), accounting for 35.2% and 23.5%, respectively. This result indicated that the VOC concentration in the CB site was significantly affected by the southeast trajectory from the junction of the Shaanxi Province, 300 Hubei Province, and Henan Province in addition to local sources. In addition, although the proportion pollution trajectories from the northwest (LDT) was small (Cluster 3), the concentrations in these pollution trajectories were the greatest, reaching 41.5 ppb (Table 3). Thus, attention should be paid to the long-distance transmission of highly polluting air masses from Inner Mongolia. Regarding the VOC composition (Figure 10b), alkanes with lower activities accounted for a larger proportion in the long-distance trajectories (Cluster 3). 305 The air mass back trajectories can be clustered into the eastern trajectories (Cluster 1), southeastern trajectories (Cluster 2+4), and northeastern trajectories (Cluster 3+5) at the DHS site (Figure 10c). It is evident that the proportion of southeast trajectories to the total trajectories and that of the southeast pollution trajectories to the total pollution trajectories were significantly higher than those of the other cluster trajectories, accounting for 68.9% and 73.3%, respectively (Table 3). This result indicated that https://doi.org/10.5194/acp-2020-704 Preprint. Discussion started: 27 August 2020 c Author(s) 2020. CC BY 4.0 License.