Shanghai, China, is one of the largest cities in the world, located at the estuary of the Yangtze River, with more than 24 million people and more than 3 million motor vehicles (National Bureau of Statistics, 2018) The measurements were conducted at the Jiangwan campus of Fudan University in the northeast of Shanghai (121.5∘ E, 31.33∘ N). It is a typical urban environment, surrounded by commercial and residential areas. The campus itself faces relatively clean air conditions without significant sources of air pollutants, mainly affected by traffic emissions from viaducts and residential areas nearby.

O3, HONO, NO2, NO, SO2 and HCHO were monitored in real-time. O3 and NO were measured with a short-path DOAS (differential optical absorption spectroscopy) instrument with a light path of 0.15 km and time resolution of 1 min. The fitting windows of them are 250–266 and 212–230 nm, respectively. HONO, NO2, SO2 and HCHO were measured by the long-path DOAS apparatus with a light path of 2.6 km and time resolution of 6 min. The spectral fitting intervals are 339–371, 341–382, 295–309 and 313–341 nm, respectively. Meteorological parameters, including temperature, relative humidity, wind direction and wind speed, were recorded by the collocated automatic weather station (CAMS620-HM, Huatron Technology Co. Ltd). The photolysis frequency of NO2 (JNO2) was measured with a filter radiometer (Meteorologie Consult Gmbh). CO was measured by a Gas Filter Correlation CO Analyzer (Thermo-Model 48i) with a time resolution of 1 h. Additionally, NMVOCs were monitored using the TH-300B online VOCs Monitoring system that includes an ultralow-temperature (-150 ∘C) preconcentration combined with gas chromatography and mass spectrometry (GC/MS). Under ultralow-temperature conditions, the volatile organic compounds in the atmosphere are frozen and captured in the empty capillary trap column; then a rapid heating analysis is performed to make the mixture enter the GC/MS analysis system. After separation by chromatography, NMVOCs are detected by FID (flame ionization detector) and MS detectors. Typically, the complete detection cycle was 1 h. CH4 was measured with a Methane and Non-Methane Hydrocarbon Analyzer (Thermo-Model 55i) with a time resolution of 1 h.

All of the above techniques have been validated and applied in many previous studies, and their measurement principles, quality assurance, and control procedures were described in detail (Wang et al., 2015; Hui et al., 2018, 2019; Shen et al., 2016; Zhao et al., 2015; Nan et al., 2017).

In this study, the in situ atmospheric photochemistry was simulated using an observation-based model (OBM) incorporating the latest version of the Master Chemical Mechanism (MCM, v3.3.1; http://mcm.leeds.ac.uk/MCM/, last access: 30 January 2020), a near-explicit chemical mechanism which describes the degradation of methane and 142 non-methane VOCs and over 17 000 elementary reactions of 6700 primary, secondary and radical species (Jenkin et al., 2003; Saunders et al., 2003). The model can simulate the concentration of highly active radicals, so that the critical aspects of atmospheric chemistry can be quantitatively evaluated, including secondary product formation (e.g., O3 and PAN), VOC oxidation and radical budgets.

The observed data of O3, NO2, NO, CO, SO2, HONO, CH4, 54 species of NMVOCs, JNO2, water vapor (converted from relative humidity) and temperature were interpolated to a time resolution of 5 min and then input into the model as constraints. The photolysis rates of other molecules such as O3, HCHO, HONO and OVOCs were driven by solar zenith angle and scaled by measured JNO2 (Jenkin et al., 1997; Saunders et al., 2003). Considering the potential impact of cloud cover on the frequency of photolysis, we have discussed the impacts of cloud cover on the scaled photolysis rates in the Supplement. In addition to the chemistry, deposition process within the boundary layer height is also included in the model. The loss of all unrestricted and model-generated species caused by the deposition is set as the accumulation of the deposition velocity of 0.01 m s-1 in the boundary layer (Santiago et al., 2016). Given that the boundary layer height (BLH) varied typically from 400 m at night to 1400 m in the afternoon during summer (Shi et al., 2015), the lifetime of the model-generated species ranged between ∼11 h at night and ∼40 h during the afternoon. We have also carried out a sensitivity study on the deposition velocity and boundary layer height, as referred to the Supplement. The model simulation period for three different ozone levels is 7 d, including 4 d of pre-simulation to allow unconstrained compounds to reach a steady state.

According to the definition of AOC, it can be calculated by the Eq. (1) (Elshorbany et al., 2009; Xue et al., 2016): 1AOC=∑ikYiYiX, where Yi are VOCs, CO and CH4, X are oxidants (OH, O3 and NO3) and kYi is the bi-molecular rate constant for the reaction of Yi with X. Atmospheric oxidation capacity determines the rate of Yi removal (Prinn, 2003).

Additionally, another widely used indicator of atmospheric oxidation intensity is OH reactivity, which is defined as the reaction rate coefficients multiplied by the concentrations of the reactants with OH and depends on the abundances and compositions of primary pollutants. As the inverse of the OH lifetime, OH reactivity is calculated by Eq. (2) (Mao et al., 2010): 2kOH=∑ikOH+XiXi, where Xi represents the concentration of species (VOC, NO2, CO etc.) which react with OH and kOH+Xi is the corresponding reaction rate coefficients.

Moreover, the ratio of the OH cycling to OH terminal loss, known as the OH chain length, can characterize atmospheric photochemical activity. The OH chain length can be calculated by Eq. (3) when the reaction between OH and NO2 is the main termination reaction of radicals (Martinez et al., 2003; Mao et al., 2010): 3OHChainLength=kOHOH-kOH+NO2+MOHNO2kOH+NO2+MOHNO2. This is one of several definitions available based on the assumption that OH + NO2 is the main chain termination reaction, which is further discussed in Sect. 3.3.

The AOC, OH reactivity and OH chain length, as well as HOx budget, can be quantitatively assessed by tracking the relative reactions and corresponding rates of the reactions in the OBM simulation.

All the measured data were hourly averaged. Figure 1 shows the observed time series of major pollutant mixing ratios and meteorological parameters during the campaign from 1 May to 30 September 2019 at Jiangwan campus in Shanghai. During the 5-month observation period, the average temperature and humidity levels were 26.4 ∘C and 78.78 %, respectively, while the mean mixing ratios of O3, NO2, NO, HONO and HCHO were 35.14±18.72, 13.0±4.31, 5.30±9.26, 0.29±0.18 and 2.78±1.33 ppbv, respectively. According to the air quality index (AQI) data released by the Shanghai Environmental Monitoring Center (SEMC) and the ozone mixing ratio data observed, the overall air quality in Shanghai was good in the spring–summer season of 2018. Days with good air quality (AQI < 100) accounted for 92.2 % of the experiment period. However, there were occasionally high-ozone pollution days, during which the primary pollutant of 10 d of the residual 12 polluted days was ozone (the average hourly ozone exceeded the Class 2 standard 93 ppbv, GB 3095–2012, China; Ambient air quality standards, 2012)

As indicated by the gray rectangle in Fig. 1, three cases of different ozone levels were selected to study atmospheric oxidation and free radical chemistry. These are the polluted period (Case 1) between 11 and 13 June, the semi-polluted period (Case 2) from 2 to 4 September and the non-polluted period (Case 3) of 12 to 14 July, respectively. As can be seen in Table 1 (also refer to Fig. S1 in the Supplement), the averaged O3 mixing ratios in Case 1, Case 2 and Case 3 were 65.13±27.16, 46.12±21.14 and 23.95±11.89 ppbv, during which the maximum mixing ratios reached 111.87, 80.76 and 50.74 ppbv, respectively. By comparing the meteorological parameters, the wind speed of Case 3 was highest, followed by Case 2 and Case 1, indicating that the unfavorable diffusion conditions are one cause of ozone pollution. Although the ozone mixing ratio of Case 2 was much lower than that of Case 1, the levels of NOx, CO, HONO in Case 2 were also high or close to Case 1. This is explained by the fact that meteorological parameters of Case 1 and Case 2 were quite different (Fig. S2); i.e., higher radiation, greater differences in temperature during day and night and lower humidity and air pressure during Case 1 are conductive to enhancing atmospheric photochemistry and lead to ozone formation. In addition, when the JNO2 value input to the OBM was artificially increased by 40 % for Case 2, simulation results showed that the peak value of ozone increased by 30 %–40 % as a consequence. The observations and simulations suggested that high radiation is an influencing factor in ozone pollution. However, ozone levels were lowest during the most intensive radiation in Case 3. Under such favorable meteorological conditions, the low-ozone mixing ratio was attributed to the low mixing ratios of O3 precursors NOx and VOCs. Therefore, it can be inferred that ozone pollution was caused by the combination of high levels of O3 precursors and strong radiation.

Statistical information of each species group of VOCs classified based on their chemical nature and composition is also shown in Table 1. In general, the mixing ratios of VOCs were highest in Case 2, followed by Case 1 and Case 3, with average total VOC mixing ratios of 25.31±6.16, 29.73±12.10 and 12.18±3.69 ppbv, respectively. During Case 1, OVOCs and alkanes accounted for the vast majority of total NMVOCs, reaching 36.3 % and 36.4 %, followed by alkenes (12.8 %), other VOCs (8.7 %) and aromatics (5.8 %). For Case 2, alkanes and OVOCs also dominated total NMVOCs (35.5 % and 31.6 %), followed by alkenes (12.1 %), other VOCs (11.1 %) and aromatics (9.7 %). During Case 3, OVOCs represented the largest contribution to total NMVOCs (33.8 %), followed by alkanes (30.1 %), other VOCs (14.4 %), alkenes (11.6 %) and aromatics (10.1 %). Table 2 shows the average mixing ratios and standard deviation of 54 VOCs including methane during the three cases. The key species in different groups were consistent in three cases; for example, ethane and propane were the two highest mixing ratio alkanes, the main alkene species were ethylene and acetylene, the highest concentrations in aromatics were benzene and toluene and HCHO and acetone were the dominant fraction in OVOCs.

According to Eq. (1), AOC during the three case periods was quantified based on the OBM, as shown in Fig. 2. The calculated maximum AOC for the three Cases was 1.0×108, 9.1×107 and 8.8×107 molecules cm-3 s-1, respectively. Comparatively, these are much lower than those computed for Santiago de Chile, Chile, with a peak of 3.2×108 molecules cm-3 s-1 (Elshorbany et al., 2009), but much higher than that in Berlin, Germany with 1.4×107 molecules cm-3 s-1 (Geyer et al., 2001). It can be seen from Fig. 2 that the time profile of the AOC exhibits a diurnal variation, which is the same as the time series of the model calculated OH concentration and the observed JNO2, with a peak at noon. Daytime average AOC values were 3.96±2.32×107, 3.54±2.24×107 and 3.59±2.51×107 molecules cm-3 s-1, while nighttime average AOC value were 3.11±1.15×106, 2.38±0.57×106 and 4.30±0.53×105 molecules cm-3 s-1, for the three cases, respectively. These values were in line with the ozone levels, suggesting that atmospheric oxidation capacity during the ozone pollution period is greater than under clean conditions.

As expected, OH was calculated to be the main contributor to AOC. In the three cases, the average contribution of OH to AOC during the daytime was over 96 %. O3, as the second important oxidant, accounted for 1 %–3 % of the daytime AOC. The contribution of NO3 to nighttime AOC was 1.50±0.52×106, 1.24±0.38×106 and 3.02±1.94×104 molecules cm-3 s-1, respectively (or see Fig. S3). During Case 1 and 2 with relatively polluted conditions, NO3 became the primary oxidant in AOC, accounting for 48.3 % and 52.3 % of the nighttime AOC, respectively. It is worth noting that the chlorine atom produced by the photolysis of ClNO2 may also contribute to AOC (Bannan et al., 2015) , but unfortunately it has not been quantitatively characterized in this study. In general, OH dominated AOC during daytime and NO3 is the main oxidant at night, which is consistent with previous studies (Asaf et al., 2009; Elshorbany et al., 2009).

We now evaluate the loss frequency of the different reactants to OH using the indicator of OH reactivity according to Eq. (2). The diurnal variations of OH reactivity calculated via the OBM are presented in Fig. 3, including the contribution from measured VOCs, NOx, CO and model-generated intermediate species during three cases. It is evident that the OH reactivity peaked in the morning, with maximum values of 19.61, 24.55 and 13.32 s-1 for three cases, respectively. This is due to the increased NOx during rush hour traffic (Sheehy et al., 2010). The average values in the three cases were 11.72±2.84, 13.45±4.25 and 7.56±1.52 s-1, respectively. The OH reactivity of Case 3 in the clean environment was significantly lower than that of Case 1 and Case 2, which is consistent with previous studies (Mao et al., 2010; Li et al., 2018). In general, the OH reactivity assessed in Shanghai was in the range of 4.6–25.0 s-1 under different air quality conditions, which was at a relatively low level compared to that calculated for other big cities in China such as Guangzhou (20–30 s-1), Chongqing (15–25 s-1) and Beijing (15–25 s-1; Tan et al., 2019b), reflecting that the abundance of pollutants in Shanghai is relatively lower compared to other metropolitan areas in China.

Total OH reactivity has been measured in many urban areas over the past two decades. Compared to studies in other regions, the estimated average OH reactivity in Shanghai was much lower than that in Paris (Dolgorouky et al., 2012), New York (Ren et al., 2003, 2006) and Tokyo (Yoshino et al., 2006), and was equivalent to Nashville (Kovacs et al., 2003), Houston (Mao et al., 2010) and London (Whalley et al., 2018). In addition, there are some differences between the actual measured values and the estimated values of OH reactivity as mentioned in previous studies, which may be attributed to missing OH reactivity that originates from secondary products such as other OVOCs and nitrates produced by photochemical reactions (Di Carlo et al., 2004; Yoshino et al., 2006; Dolgorouky et al., 2012). We also calculated the OH reactivity only considering the measured species, and the contribution of OVOCs to OH reactivity was 1.28, 1.43 and 0.82 s-1, while the OH reactivity of OVOCs was calculated by considering the simulated intermediate species was 1.77, 2.05 and 1.26 s-1 in three cases, respectively. These differences indicates unmeasured species and unknown secondary products contributed considerably to the actual OH reactivity.

Figure 4a shows the average contribution of major groups of reactants to the total OH reactivity for three cases, including NMVOCs, NO2, NO, CO and CH4. Overall, NMVOCs, CO and NO2 are major contributors to OH reactivity, in line with past studies carried out in urban environments (Ling et al., 2014; Gilman et al., 2009). The remarkable contribution of CO to the total OH reactivity in Case 1 points to effective CO+OH and its significant contribution to ozone formation (Ling et al., 2014). The main difference in the composition of OH reactivity was that the absolute contribution of NMVOCs in Case 1 was about 1.45 times than that of Case 2, while the absolute contributions of CO and NOx to OH reactivity in Case 1 were comparable to those of Case 2. This is caused by the higher VOCs levels of 29.73±12.10 ppbv during Case 2 as compared to Case 1 of about 15 % lower. Since the mixing ratios of pollutants in Case 3 were quite low, the contribution of each reactant component to OH reactivity was much lower than the other cases.

Figure 4b also presents the detailed contribution of each NMVOC group to the total OH reactivity. It can be seen that the contribution of OVOCs to OH reactivity is predominant, accounting for 46.87 %, 40.79 % and 43.03 % of the total OH reactivity of NMVOCs in the three cases. The contribution rate of OVOCs to OH reactivity in Case 1 was 3 to 6 percentage points higher than Case 2 and Case 3, illustrating the importance of OVOCs in atmospheric photochemistry and ozone generation (Fuchs et al., 2017). The contribution of alkenes to OH reactivity was important in three cases, reaching about 36 %, which may be caused by the relatively higher contribution of alkenes emitted by motor vehicles at the urban site, indicating that ozone pollution was severely affected by vehicle emissions in Shanghai (Ling et al., 2014; Guo et al., 2013). The contribution of aromatics and alkanes to OH reactivity were comparable in the three periods, both in the range of 0.3–0.6 s-1, accounting for 10 %–20 %. The contribution of other VOCs to OH reactivity was negligible, with contributions of only 0.4 % or less. Tan et al. (2019b) also reported a comparable average OH reactivity of about 13.5 s-1 (kOH=13.45 s-1 in Case 2 this study) and a similar contribution distribution of OH reactivity during summer in Shanghai.

In summary, the mixing ratios of ozone precursors and their contribution to OH reactivity were found to be different in the three cases. To further investigate these differences, HOx budget, OH chain length and OFP (ozone formation potential) are discussed in depth in the following sections.

The OH chain length serves as an indicator for evaluating HOx cycling and is closely related to ozone production efficiency. The OH concentration and the terminal loss rate of OH by the reaction with NO2 were simulated by the OBM. The longer chain length means that more OH radicals are generated in the HOx cycling and more O3 is produced before the OH terminal reaction occurs (Mao et al., 2010; Ling et al., 2014). As a previous studies showed, the OH chain length began to rise in the morning and peaked at noon (Mao et al., 2010; Ling et al., 2014; Emmerson et al., 2007). As illustrated in Fig. 5, the OH chain lengths were all less than 8, with a peak at noon. Interestingly, it was found that the OH chain length peak in Case 1 appeared around 14:00 LT (UTC+8), coinciding with the observed NOx variability (see Fig. S1). The OH chain lengths for the three cases peaked at 6.3 in Case 3, followed by Case 2 (peak of 5.5) and Case 1 (peak of 4.1), the opposite of O3 levels (Table 1). This is due to the relatively higher NOx level in Case 1 (see Fig. S1), resulting in a relatively bigger sink of OH+NO2. In summary, the longer OH chain length in Case 3 indicated per NOx converted into HNO3 produces more O3, whereas the NOx mixing ratio in Case 3 is almost half that of Case 1 and 2 during daytime (see Fig. S1), causing the ozone mixing ratio to be lower than Case 1 and 2. In addition, previous studies also found that the OH chain length was the opposite of the ozone level, and gave the possible explanation also due to the lower NOx concentrations (Mao et al., 2010; Ling et al., 2014).

We calculated the primary sources of HOx, including the photolysis of O3, HONO, HCHO and other OVOCs, as well as the ozonolysis of alkenes, excluding parts (i.e., H2O2, CH3OOH) that contribute less to HOx (Mao et al., 2010; Ling et al., 2014; Sommariva et al., 2004) and any reactions in HOx cycling such as HO2+NO that dominate OH generation that are just the cycling between OH and HO2 (Mao et al., 2010). At the same time, the sinks of HOx were also simulated, including the reactions of OH+NO2, HO2+HO2 and HO2+RO2, and any reactions of HOx cycling as well as smaller contributing reactions were also excluded. These HOx production and loss pathways have been considered and well investigated in other studies and locations (Mao et al., 2010; Ling et al., 2014; Wang et al., 2018).

Figure 6 shows the diurnal variability of the main generation and loss pathways of HOx. It can be seen that the intensity of the sources and sinks of HOx was different, but the primary contributions to HOx budget of three cases were consistent, i.e., O3 photolysis and reaction of OH with NO2, respectively. The average generation rates of HOx were 1.51±0.92×107, 1.10±0.70×107 and 1.05±0.71×107 molecules cm-3 s-1, while the average loss rates were 1.34±0.74×107, 1.00±0.55×107 and 0.8±0.52×107 molecules cm-3 s-1, respectively. During the daytime, the biggest contribution to HOx production was ozone photolysis, around 40 % in Case 1 and Case 2, while HONO photolysis contributed 41.1 % in Case 3. This indicates that ozone photolysis dominates the production of HOx under high-ozone conditions, whereas photolysis of HONO is important at lower ozone concentrations (Wang et al., 2018; Ling et al., 2014; Ren et al., 2008). Additionally, the model results show that the photolysis of HCHO was also an important contributor to HOx production in the three cases, reaching 25.9 %, 22.9 % and 21.0 %, respectively (Ling et al., 2014; Liu et al., 2012; Lu et al., 2012; Mao et al., 2010).

Moreover, the diurnal profile of the HOx budget was explored. Before 09:00, 09:30 and 11:00 during the three cases, respectively, HONO photolysis dominated the production of HOx in the morning due to the accumulation of HONO at night. This is consistent with a previous report in Shanghai in July 2014 which found that the contribution of HONO photolysis could reach up to 80 % of HOx production in the morning (Chan et al., 2017). In addition, ClNO2 photolysis is also reported to be an important source of radicals in the morning (Young et al., 2012). In the afternoon, the HONO mixing ratio decreased with photolysis, O3 levels increased with the enhancement of photochemical intensity and O3 photolysis becomes the main contributor to HOx production. Note however that the contribution of HONO and HCHO photolysis are not negligible in the afternoon. The other two HOx formation pathways, OVOCs photolysis and alkenes ozonolysis, accounted for less than 5 % in the three cases.

For the HOx sink, the reaction of OH and NO2 was dominant all day, and its average contribution reached 1.20±0.67×107, 0.84±0.45×107 and 0.71±0.40×107 molecules cm-3 s-1, accounting for 89.11 %, 84.56 % and 83.29 % in three cases, respectively. In Case 2 and Case 3, the reaction of OH and NO2 dominates the sinks of HOx before 09:00 when NOx was at a high level due to rush hour traffic. However, the reaction of OH and NO2 completely dominated the HOx sinks from 05:30 to 11:00 in Case 1, almost constituting the entire HOx sinks, which indicates that the rush hour traffic was prolonged and the NOx was maintained at a high concentration. This is consistent with the fact that the peak of the OH chain length appears at 14:00 in Case 1, as mentioned above. The reaction with NO2 was the main sink of HOx, confirming that Eq. (3) of the OH chain length chosen in this study is appropriate. The reactions between radicals themselves such as HO2+HO2 and HO2+RO2 became more important for the contribution of HOx sinks in the afternoon for the three cases, in agreement with previous studies in other regions (Guo et al., 2013; Ling et al., 2014; Mao et al., 2010).

Regarding the model-simulated concentrations of OH and HO2, as shown in Fig. S4, the maximum concentrations of OH for three cases were 9.97×106, 8.34×106 and 10.3×106 molecule cm-3, respectively, and the maximum concentrations of HO2 for three cases were 4.06×108, 3.84×108 and 3.41×108 molecule cm-3, respectively. The previous simulated maximum concentrations of OH and HO2 for the urban site in Shanghai were 6.9×106 and 1.9×108 molecule cm-3 in summer, lower than the simulated results here probably because of the different atmospheric conditions (Tan et al., 2019b). Due to lack of measured values of HOx in Shanghai, we compared the measured values of other places in China. For instance, daily maximum concentrations were in the range of (4–17)×106 molecule cm-3 for OH and (2–24)×108 molecule cm-3 for HO2 at both the suburban site Yufa and rural site Wangdu during summer in the North China Plain (Lu et al., 2013; Tan et al., 2017). In autumn, maximum median radical concentrations of4.5×106 molecule cm-3 for OH at noon and 3×108 molecule cm-3 for HO2 were reported for the Pearl River Delta in the early afternoon (Tan et al., 2019a). The simulated HOx concentrations in this study were comparable with the measured results of other places in China, suggesting the moderate abundance of the HOx radical in Shanghai.

Different VOC species have a wide range of reactivity and different potentials for O3 formation, which can be calculated by the maximum incremental reactivity (MIR; Carter, 2010). The calculated ozone formation potential (OFP) of each VOC species is used to characterize the maximum contribution of the species to ozone formation (Bufalini and Dodge, 1983). The following equation is used to calculate the OFP for each VOC species (Schmitz et al., 2000; Ma et al., 2019): 4OFPi=MIRi×VOCi×MiMozone, where OFPi (ppbv) is the ozone formation potential of VOC species i, [VOCi] (ppbv) is the atmospheric mixing ratio of VOC species i, MIRi (g O3/g VOC, as listed in Table 1) is the ozone formation coefficient for VOC species i in the maximum increment reactions of ozone, Mozone and Mi are the molar mass (g mol-1) of O3 and VOC species i, respectively.

In this study, OFP was introduced to estimate the photochemical reactivity of VOCs. The comparison of the average mixing ratios of the five VOC groups and their OFP during three cases is shown in Fig. 7. VOC mixing ratios of Case 2 were higher than in Case 1 and Case 3, as was the OFP level of Case 2. However, it is obvious that the mixing ratio of the VOC group was not proportional to its OFP. The biggest contribution to VOCs mixing ratios here was alkanes (36.4 %) and OVOCs (36.3 %) in Case 1, while OVOCs (45.4 %), alkenes (25.2 %) and aromatics (18.6 %) were the top three contributing to OFP. In Case 2, the mixing ratio of total NMVOCs reached 29.73 ppbv, the main contributors of which were alkanes (accounting for 35.5 %) and OVOCs (31.6 %), while the top three contributions to total OFP (96.16 ppbv) were OVOCs (accounting for 36.1 %), aromatics (30.4 %) and alkenes (21.8 %). Our results are consistent with those reported for Beijing in summer 2006 where OVOCs (40 %), aromatics (28 %) and alkenes (20 %) were also the top three contributors (Duan et al., 2008). In Case 3, the NMVOCs mixing ratio (12.2 ppbv) and the corresponding OFP (53.7 ppbv) were both at a relatively lower level.

According to the comparison between VOC groups mixing ratios and their OFP in Case 1 and Case 2 with relatively high-ozone mixing ratios, alkanes and OVOCs were the most important contributors to NMVOCs in both cases. Although the mixing ratios of these two groups were comparable in both Case 1 and 2, the contribution of OVOCs to OFP was about 3.5 times that of alkanes, indicating that the reactivity of alkanes is so low that it contributes less to the formation of ozone than other groups. Conversely, OVOCs show significant contributions to ozone formation with higher mixing ratios leading to higher OFP. The contribution of aromatics to OFP reached 30.2 % in Case 2. At the same time, the contribution of alkenes to ozone generation cannot be ignored, and for example, it reached 26.7 % in Case 1. Due to the different composition profile of VOCs, the contribution of VOC to OFP is quite different in the other areas of China. For example, in Shenyang the top three contributors were aromatics (31.2 %), alkenes (25.7 %) and OVOCs (25.6 %; Ma et al., 2019); OVOCs (34.0 %–50.8 %) dominated OFP in Guangzhou (Yuan et al., 2012); alkenes (48.34 %) were the main contributor in Wuhan (Hui et al., 2018), while alkanes, alkenes and aromatics accounted for 57 %, 23 % and 20 % in Lanzhou, respectively (Jia et al., 2016).

The top 12 NMVOCs in OFP and their average mixing ratios during the three cases are shown in Fig. 8. These 12 species accounted for 50.90 %, 41.63 % and 36.33 % of the total NMVOCs observed and contributed about 79.57 %, 76.55 % and 75.73 % to the ozone formation in the three cases, respectively. As mentioned above, not all high-concentration species had substantial OFP contributions. As shown in Fig. 8, acetone was the third most abundant species in total NMVOCs, accounting for 14.6 % of the total NMVOCs mixing ratio, but it only contributed 2.2 % to total OFP in Case 1. And m/p-xylene ranked second in the contribution of OFP, accounting for 12.1 %, while it represents only 1.8 % of total NMVOCs mixing ratio in Case 2. The results show that HCHO was the most important OFP contributor, accounting for 35.6 %, 23.6 % and 22.1 % in each of the three cases, respectively. Under high-ozone mixing ratios during Case 1 and Case 2, four of the top five species contributing to OFP were the same, i.e., HCHO, toluene, ethylene and m/p-xylene, while the mixing ratio and OFP of these four species were at a lower level under the clean conditions in Case 3, indicating that these four species can play a very different role in ozone formation under different chemical conditions. These results are similar to the research in the Pearl River Delta region in 2006 where the top four contributions to OFP were isoprene, m/p-xylene, ethylene and toluene (Zheng et al., 2009). Additionally, it was found that the total mixing ratios of HCHO, toluene, ethylene and m/p-xylene accounted for only 23.5 %, 22.6 % and 26.0 % of the total NMVOCs, whereas the overall contribution of these four species to OFP was 55.7 %, 55.3 % and 49.8 % in the three cases, respectively. This suggests that controlling different key VOC components is effective in preventing ozone pollution episodes. For instance, by controlling the concentration of these four species in Case 1 to the level of Case 3 (reduced by 2.78 ppbv), the contribution of NMVOCs to OFP would be reduced by nearly 20 %.