Characteristics of one-year observational data of VOCs, NOx and O3 at a suburban site in Guangzhou, China

Abstract. Guangzhou, one of China's megacities, is beset with frequent occurrence of high-concentration ozone events. In this study, online instruments were used to simultaneously monitor ozone, nitrogen oxides (NOx) and volatile organic compounds (VOCs) at GPACS (the Guangzhou Panyu Atmospheric Composition Station) of the China Meteorological Administration, from June 2011 to May 2012, in order to determine their characteristics, the effect of VOCs on ozone photochemical production and the relationship between VOC / NOx ratio and ozone formation. The results showed that during the observation period, the seasonal variation of ozone concentration was lower in spring and winter compared to summer and autumn, which is opposite that for VOCs and NOx. In terms of VOCs, aromatics had the largest ozone formation potential, among which toluene, xylenes, ethylbenzene, 1,2,4-trimethylbenzene and 1,3,5-trimethylbenzene were the most important species, with a total contribution of about 44%. As the VOC / NOx ratios were very high during high-concentration ozone events that occur all year round, we speculate ozone production was likely to be NOx-limited regime (12:00–16:00 LT) in Guangzhou. Further investigation based on numerical models is needed in the future to obtain more detailed and robust conclusions.


Introduction
Along with its rapid economic development and urbanization, the Pearl River Delta (PRD), whose major cities include but are not limited to Hong Kong, Guangzhou, Dongguan and Shenzhen, has become one of the most polluted areas in China (Chan and Yao, 2008). Unlike other regions in China, such as the region surrounding Beijing and the Yangtze River Delta region, where the main air pollutant is particulate matter Wang et al., 2012), high-concentration ozone events occur frequently in the PRD  due to its unique geographical location and climate, as well as a rapid increase in the emission of ozone precursors (volatile organic compounds (VOCs) and NO x ) from industrial activities and motor vehicles.
Tropospheric ozone is a secondary pollutant generated by the photochemical reaction of VOCs and NO x in the presence of sunlight (Sillman, 1995). Smog chamber studies have been widely used to investigate the gas-phase photochemical transformations of NO x and VOCs, as well as their roles in ozone production. It has been found that VOCs and NO x have no linear relationship with ozone formation. Instead, their impact on ozone formation can be described by VOC-or NO x -limited regimes Tie et al., 2007;Geng et al., 2008). Numerical simulations in polluted regions, such as Los Angeles in the 1980s, have shown ozone formation to be NO x -limited when the VOC / NO x ratio is greater than 8 : 1 for VOC expressed as the concentration of carbon atoms (Seinfeld, 1989), a finding which many re-searchers later used to explain photochemical ozone formation in other areas, where specific atmospheric situations may differ (Seinfeld, 1991;Trainer, 1993;Sillman, 1999;Ran et al., 2009Ran et al., , 2011"Li, 2011). Additionally, there is increasing evidence showing that ozone formation may change with the time of day (Kanaya et al., 2009;Sillman and West, 2009;. For example, a recent study reported that ozone formation was VOC-limited in the morning and became NO x -limited during peak ozone hours in most of the PRD region . Such temporal variations in ozone formation have also been documented for other places in the world (Kanaya et al., 2009;Sillman and West, 2009).
In order to understand and therefore reduce the formation of ozone, it is critical to evaluate the relative importance of various VOCs in terms of ozone production potential because individual VOCs exhibit a wide range of reactivity with respect to ozone photochemical formation. Several reactivity scales have been used to assess the relative ozone forming capacity of individual VOCs species (Atkinson, 1990;Carter, 1994;Shao et al., 2009;Li and Wang, 2012). One widely used scaling method is maximum incremental reactivity (MIR), defined as the maximum amount of ozone formed per unit of VOCs added or subtracted from the whole gaseous mixture in a given air mass, which considers the chemical mechanism and the impact of the VOC / NO x ratio on ozone production (Atkinson, 2000). Another common approach is the OH-reactivity scale (e.g., propylene-equivalent concentration), which is based on assumptions that the reactions with OH radicals are dominant in the tropospheric atmosphere and that the mechanistic factor and its associated uncertainties are negligible (Darnall et al., 1976;Middleton et al., 1990).
The formation and levels of ozone pollution have been investigated in many megacities around the world (Qin et al., 2004;Stephens et al., 2008) and a number of studies have also been conducted to explore the formation of ozone in the PRD (Wang et al., 2005;Guo et al., 2009;Cheng et al., 2010). Most of these studies have shown that ozone formation was VOC-limited in the PRD region (Zhang et al., 2008;Guo et al., 2009). However, few long-term observational data are available for ozone, NO x and VOCs in the PRD, in particular, continuous VOC data Shao et al., 2009), which only serves to limit our understanding of ozone production.
This study aimed to explore the characteristics of VOCs, NO x and ozone and to examine the possible seasonal and diurnal variations in ozone formation in the PRD based on 1 year of observational data for ozone and its precursors collected in Guangzhou, one of the PRD's megacities. In the present paper, we document general information about the sampling site and the data collection procedure (Sect. 2), as well as the seasonal and diurnal characteristics of ozone and its precursors (Sect. 3.1). The effects of VOCs on ozone formation, based on estimated MIRs and propylene-equivalent concentrations, are also presented (Sect. 3.2) and finally,  1). It is a suburban site (about 15 km south of downtown Guangzhou) where high-concentration ozone events often occur, and it is also one of the main observation sites for atmospheric composition in the PRD. At the measurement site, the relative contributions of the three categories of VOCs (i.e. alkanes, alkenes and aromatics), remained fairly uniform throughout the observational period regardless of the large variation in their concentrations (see Fig. 2). Such uniformity implies that the air at the GPACS site was sufficiently homogenous with respect to various sources near the surface of the ground. The prevailing wind, wind speed and temperature at the sampling site during different seasons are shown in Fig. 3 and Table 1. At the sampling point, northeasterly and southwesterly winds prevailed in spring (March, April and May), southwesterly winds in summer (June, July and August), northeasterly winds in autumn (September, October and November) and northeasterly winds in winter (December, January and February). During the different seasons, the average wind speed remained close to 1.4 m s −1 , while the average temperature When the prevailing wind was northeasterly in December, the difference between weekend and weekday VOC concentrations was very clear, indicating that large amounts of pollutants were emitted from downtown Guangzhou City. When the prevailing wind was southwesterly in July, the difference between weekend and weekday VOC concentrations was mild, indicating that only small amounts of pollutants were emitted from the outer suburban areas.

Instrument description
Ozone was measured using an EC9810B ozone analyzer (Ecotech Co., Australia) based on the UV-absorption method and the Lambert-Beer law. NO x was measured by an EC9841B nitrogen oxide analyzer (Ecotech Co., Australia) with a heated molybdenum NO 2 -to-NO converter, and gasphase chemiluminescence technology was used to quantify NO concentrations. It was possible that the measured NO x may have included some oxidized reactive nitrogen that was converted by the molybdenum, and thus, the NO x concentrations given below are considered the upper limits of their actual values (Dunlea et al., 2007;Ran et al., 2011). VOCs were continuously monitored at a 1 h sampling frequency using the GC5000 analysis systems coupled with flame ionization detectors (FID) from AMA Instruments GmbH (AMA, Germany). This system has been described in detail elsewhere (An et al., 2014;Zou et al., 2013). In brief, it consisted of two sets of sampling systems and two sets of chromatography column systems, which includes a low boiling point VOC analyzer (GC5000VOC) for C 2 -C 6 VOC species and a high boiling point VOC analyzer (GC5000BTX) for C 6 -C 12 VOC species. Air samples were enriched in the GC5000VOC analyzer through two-stage trap, and then thermally desorbed when the temperature increased to 200 • C, followed by separation with two-dimensional chromatography. The chromatographic columns consisted of an Al 2 O 3 / Na 2 SO 4 plot column (60 m × 0.32 mm inner diameter × 5 µm thickness) and a CARBOWAX ™ back-flushing column (30 m × 0.32 mm inner diameter × 0.25 µm thickness). The back-flushing column was first used to remove the moisture component and high-boiling VOC species, and then the plot column was followed to separate the low-boiling VOC species. The GC5000BTX pre-concentrated the VOCs at 30 • C and then thermally desorbed them at 270 • C; afterwards, the analyzer separated the compounds in the DB-1 column (60 m × 0.32 mm inner diameter × 10 µm thickness), in order to achieve optimum separation and prevent interference from related compounds.
The 56 VOCs designated as photochemical precursors by the US EPA were used as the target compounds. The gas standard used was the same as that used by the US EPA Photochemical Assessment Monitoring Stations (PAMS). Zero and span gas checks (using PAM calibration gases) were conducted monthly throughout the observation period, together with adjustment of the retention time. Five-level calibration curves were used to quantify the VOC concentrations during the online monitoring. The calibration curves and detection limits of individual VOCs are listed in Table 2. The correlation coefficient ranged from 0.984 to 0.999, and the detection limits ranged from 0.03 to 0.09 ppbv. Outliers (about 8 %) were eliminated before data processing, to guarantee the validity of the data. The outlier was identified by the VOC data validation program recommended by US EPA PAMS (Main et al., 1999). However, the instruments were not intercompared with other VOC measurement systems, and such an intercomparison is an important priority for future work.

Characteristics of ozone, NO x and VOCs
The seasonal and diurnal variation characteristics of ozone, VOCs and NO x are shown in Fig. 4. The seasonal variation of ozone is readily apparent, being lower in the spring and winter, and higher in the summer and autumn. Furthermore, the range of increase is much larger in late spring and early summer than the other periods, which is mainly due to the sunlight, ambient temperature, other meteorological  factors and emission strength of air pollutants being different during each season (Bloomer et al., 2009(Bloomer et al., , 2010). In contrast, seasonal variation in the concentration of ozone precursors was opposite to that of ozone concentration, with NO x concentrations being higher in spring and winter and VOC concentrations higher in winter compared to summer. Average concentrations of VOCs, NO x and ozone in each season are shown in The diurnal variation of ozone concentration was unimodal, reaching its maximum at 14:00 LT, while the ozone precursors (VOCs and NO x ) showed bimodal variation, with a peak during rush hours, although these variations were not always apparent for some seasons. These variations in diurnal concentrations indicate that photochemical precursors gradually decreased, while their products increased. It should be noted that the measured NO x concentrations are in fact upper limits because the NO x probably includes some oxidized reactive nitrogen that was converted by the molybdenum. In order to further understand the variation in ozone concentrations and reveal its dependence on photochemical reactions, it is necessary to analyze the change rates of ozone with respect to time in each season, as follows: In the equation, [O 3 ] t represents the ozone concentration at time t, and [O 3 ] t+1 is the ozone concentration for the next   hour after time t. In the absence of transport effects, a negative rate of change in ozone concentration indicates that the chemical loss of ozone plays a dominant role in the variation of ozone concentration, while the opposite indicates that the generation of ozone photochemical reactions plays a key role. Although wind speed is low at this site, the effect of horizontal transport on daily variations in ozone concentrations is difficult to estimate and cannot be ruled out. Here, we interpret all variations as due to chemical production and loss, understanding that such interpretation is an upper limit to the effect of chemistry, due to the likelihood of transport also influencing daily ozone variation. Although diurnal variations in the change rate of ozone for different seasons show minor differences, the results indicate that a negative change rate in ozone concentrations occurs at about 15:00 LT. At this time, there was a decrease in OH radicals due to the fact that sunlight, and its associated photochemistry, began to decrease. At about 19:00 LT, there was no sunlight present to generate OH radicals, however the titration of ozone by emission of NO x could still consume ozone, and it was not until 00:00-07:00 LT that the ozone concentration was seen to stabilize. After this, the concentration of ozone began to show a positive change at 08:00 LT, due to the breakup of the nocturnal boundary layer and increased photochemistry (Fig. 5).  (Carter, 1994). b k OH denotes rate constant of VOCs react with hydroxyl radicals at 298 K (Atkinson and Arey, 2003).

The effect of VOCs on ozone formation
VOCs exhibit a wide range of reactivity, and their concentrations are not proportional to ozone photochemical formation. The average concentration of VOCs (ppbv) in each season has been discussed in Sect. 3.1. In order to adequately understand the properties of VOCs observed in the ambient atmosphere and to identify their critical role in ozone photochemical production, VOC composition and reactivity have long been of particular concern. VOCs have two main characteristics that determine their ozone formation potential: kinetic reactivity and mechanism reactivity. To estimate the reactivity of VOCs according to their kinetic reactivity, we calculated the propylene-equivalent concentration (Chameides et al., 1992;Lawrimore et al., 1995): where j represents a species of VOC, C j,C represents the carbon atom concentration (ppbC) of this species and k j,OH and k Propy,OH denote the chemical reaction rate constant in the free radical reaction of species j and propylene with OH. k j,OH is obtained from a study by Atkinson and Aery (2003).
To estimate for the reactivity of VOCs by the mechanism reactivity, the MIR-weighted concentration, which represents the maximum ozone concentration generated by this species based on estimated MIR, was calculated as per the equation below: where C j,ppbv represents the actual concentration by volume (ppbv) for species j , M represents the molecular mass of ozone, and m j represents the relative molecular mass of species j in the VOCs. The MIR j was estimated by selecting the specific MIR value for each of the VOCs from published studies which were conducted based on modeled scenarios for Los Angeles in the 1980s (Dodge, 1984;Carter, 1994). The OH reaction rate constants and estimated MIR coefficients, as well as their concentrations, for each VOC species are shown in Table 4. Figure 6 shows the characteristics of each VOC category obtained at the sampling point, with concentrations expressed on different scales. As can be seen from the nonweighted concentration by volume and by carbon atom, alkanes occupied the largest proportion, accounting for 59 and 53 % of the VOC concentration, respectively, followed by aromatics (24 and 36 %, respectively) and lastly alkenes (17 and 11 %, respectively). As can be seen from the Propy-Equiv and MIR-weighted concentrations, the alkenes and aromatics were dominant, accounting for 73 and 83 %, respectively. Total Propy-Equiv concentration accounted for nearly half of the carbon atom concentration, indicating that the reactivity of major VOCs species was lower than propylene at the sampling point. In summary, during the monitoring period, alkanes and aromatics were shown to have the highest concentration based on the concentrations by volume and carbon atoms. However, based on ozone formation potential, aromatics and alkenes were found to make the largest contributions. While the alkane content was high, because of their low reactivity, they contribute less to the reactivity of VOCs and ozone formation potential. Although alkene concentrations were smaller than those of alkanes, due to their high reactivity, alkenes made a greater contribution to ozone than alkanes, which is consistent with findings from other locations (Russell, 1995). Table 5 shows the ranking of ozone formation potential of VOC species, as calculated by the Propy-Equiv concentration and MIR factor methods, which both yielded similar results. Among the top 10 species, 8 species were the same, differing only in terms of rank order. As a result, both methods can be used to reflect the ozone formation potential of each VOC species, to some extent, especially for those which make a greater contribution to ozone formation. However, since these two methods differ in principle, the calculated ranks of ozone formation potential are also different. The Propy-Equiv concentration method only considers kinetic activity and ignores the different mechanism activities of the reaction between peroxide radicals and NO, thus when assessing ozone formation potential, the species with a faster OH reaction rate, such as isoprene, may be overestimated. On the other hand, the MIR method takes into consideration the chemical mechanisms and impacts of VOC / NO x ratios on ozone production, which may also introduce some mechanism and simulation uncertainties (Dimitriades, 1996;Ran et al., 2009). In addition, it should be note that the MIR determined in one location may be different in another, such that the application of MIR may be only approximate for Guangzhou. In summary, aromatics were found to be the species with the highest reactivity at the sampling site, among which toluene, m-Xylene, p-Xylene, o-Xylene, ethylbenzene, 1,2,4-trimethylbenzene and 1,3,5trimethylbenzene made a total contribution of approximately 44 % to ozone formation potential, followed by alkenes. Toluene, m-Xylene, p-Xylene and 1,3,5-trimethylbenzene are mainly from large factories and industrial activities (Liu et al., 2008). Located in the eastern part of the sampling site, the industrial city of Dongguan (circle on the map in Fig. 1) are presumed to have made some part of the contribution to these species at the sampling site in the autumn and winter, when northeasterly winds prevailed at the sampling site. Moreover, while isoprene concentrations were relatively low, it ranks at first and third in terms of OH reactivity and MIR, respectively. Therefore, isoprene emissions also need to be considered with respect to the control of ozone in Guangzhou.

Relationship between the variation of VOC / NO x ratios and ozone formation regime
The impact of ozone precursors on ozone formation can be described as either VOC-or NO x -limited. Assessing whether an area is VOC-or NO x -limited is an important step towards reducing regional ozone pollution. The ratios of VOC / NO x have been widely used to help determine the ozone formation regime. For example, numerical simulations of ozone pollution in Los Angeles in the 1980s indicated a transition from VOC-limited to NO x -limited regime at a VOC / NO x ratio of approximately 8 : 1 (Seinfeld, 1989). In this study, we did not perform any modeling work, but used the approximate VOC / NO x ratio to determine if ozone formation was VOC-or NO x -limited in Guangzhou. It is important to note that the actual VOC / NO x ratio may be higher than the ratios reported here, as the measured NO x may include some oxidized reactive nitrogen, due to the application of a molybdenum converter during the measurement phase. Figure 7 shows the diurnal variations of VOC / NO x ratios, as well as ozone concentrations for each season during the measurement period. It can be seen that, in the summer and autumn, the average VOC / NO x ratio was about 7 : 1 from 06:00 to 09:00 LT, and much higher than the 8 : 1 threshold for NO x -limited formation from 12:00 to 16:00 LT, when ozone concentration reached a relatively high level. This indicates that ozone formation was likely to be VOClimited during the morning and NO x -limited during peak ozone hours, respectively. However, in the spring and winter, VOC / NO x ratios were much lower than the ratio of 8 : 1, with an average value of about 7 : 1 and 5 : 1, respectively. The results suggest that ozone formation was more likely to be VOC-limited during the spring and winter. Figure 8 presents the frequency distributions of hourly ozone concentrations for each season. As seen in the figure, high hourly ozone concentrations (i.e., > 80 ppbv) were observed more frequently in summer and autumn, accounting for about 5 and 3 %, respectively, followed by 2 % in spring. Hourly ozone concentrations did not exceed > 80 ppbv in winter. As a result, ozone formation was more likely to be NO x -limited during high-concentration ozone episodes in the PRD, a finding which is similar to that reported by . Considering that the impact of VOCs on ozone formation was more closely related to the reactivity of individual VOC species than to the amount of total VOCs, a VOC (reactivity) / NO x ratio was used to analyze ozone formation (Fig. 7). The reactivity-based VOC / NO x ratio was consistent with the non-reactivity-based VOC / NO x ratio, due to the fact that, regardless of large variations in the concentration of the different types of VOCs (i.e., alkanes, alkenes and aromatics), the relative contribution of each type remained fairly uniform throughout the observation period, as shown in Fig. 2.
Given that high-concentration ozone is particularly harmful to human health, many regulators are focusing on reducing emissions during peak ozone hours (12:00-16:00 LT) (Castellanos et al., 2009). In order to further study the relationship between the variation of VOC / NO x ratios and ozone formation during high-concentration ozone events, 36 days with an hourly ozone value higher than 93 ppbv (i.e., 200 µg m −3 ), a threshold set by a new Chinese ambient air quality standard (GB 3095-2012(GB 3095- , 2012 were selected for analysis during the monitoring period. Figure 9 shows that the VOC / NO x ratio during peak ozone hours (12:00-16:00 LT) was twice as high as the average ratio of 8 : 1 for summer and autumn, suggesting that ozone formation was likely to be NO x -limited during peak ozone hours (12:00-16:00 LT) in Guangzhou. However, when ozone concentrations were relatively low in the morning (06:00-09:00 LT) and at night (20:00-00:00 LT), the ratio of VOC / NO x was about 6 : 1 and 7 : 1, respectively, which indicates that ozone formation was VOC-limited. It should be noted that, since we used the VOC / NO x ratio of 8 : 1 from Los Angeles to determine ozone formation regimes in Guangzhou, the analysis is only approximate in terms of the assessment of VOC or NO x limitations for ozone control. A more accurate analysis would require a photochemical model and more accurate measurements of NO x in the Guangzhou region.

Conclusions
Ground-level ozone and its precursors (i.e., VOCs and NO x ) were monitored over a 12-month period (from June 2011 to May 2012) at GPACS, which is located in a suburban area of Guangzhou, where high ozone events often occur. Observation-based analysis was performed to investigate the characteristics of VOCs, NO x and ozone in this highly populated region. Ozone concentrations were significantly higher in summer and autumn, while the opposite was observed for VOCs and NO x , which were higher in winter and spring. Ozone concentration began to show a net increase at 08:00 LT, likely due to the breakup of the nocturnal boundary layer and increased photochemistry, while a net decrease in ozone concentration occurred at about 15:00 LT, due to the fact that sunlight, and its associated photochemistry, begin to decrease, leading to low OH radicals and the titration of ozone by emissions of NO x . In terms of reactivity-based concentration of VOCs, aromatics had the largest ozone formation potential, among which toluene, m-, o-and p-Xylene, ethylbenzene,  Figure 9. The variation patterns of VOC / NO x ratio and ozone concentration at a high ozone episode at GPACS. High ozone episode refers to the day with an hourly ozone value higher than 93 ppbv. A black level line represents the VOC / NO x of 8 : 1.
1,2,4-trimethylbenzene and 1,3,5-trimethylbenzene made a total contribution of approximately 44 % to ozone formation potential. It should be noted that while the concentration of isoprene emitted by plants is not high, it makes a large contribution to ozone formation (about 8 %), due to its high MIR reactivity.
We also examined the temporal characteristics of ozone formation regimes according to the ratio of VOC / NO x , which indicated that ozone formation was likely to be NO xlimited during peak ozone hours (12:00-16:00 LT) in summer and autumn, as well as during high-concentration ozone events (with an hourly ozone value higher than 93 ppbv) throughout the year. However, in spring and winter, ozone formation was more likely to be VOC-limited for an extended period of time.
It should be noted that the results presented above are based on observational data, and the discussion of ozone formation regimes in Guangzhou is only based on a comparison with the VOC / NO x ratios reported for Los Angeles. Further investigations using numerical simulations are needed to obtain more detailed and robust conclusions. The application of various numerical models (e.g., observationbased model (OBM), Weather Research and Forecasting-Chemistry Model (WRF-Chem) and the U.S. Environmental Protection Agency's Community Multi-scale Air Quality (CMAQ)) to simulate ozone pollution in the atmosphere is imperative to fully understand the formation of ozone from VOCs and NO x .