Oxidative Capacity and Radical Chemistry in a Semi-arid and Petrochemical-industrialized City , Northwest China

Heavy ozone (O3) pollution is often observed in chemically industrialized Lanzhou and other capital cities located in the semi-arid and mountainous provinces of northwestern China. There are large knowledge gaps regarding the relationship between radical budgets and photochemistry in these cities. To gain insights into this relationship, a photochemical box model based on the Master Chemical Mechanism (MCM v3.3) was applied to investigate oxidative capacity and radical chemistry in the city of Lanzhou. The budgets of ROx (OH + HO2 + RO2) radicals were quantified, and the initiation, propagation, and termination process of the diurnal variation in the ozone chemistry were examined. The MCM model was constrained by in situ measurements at two sampling sites in the city, one located in the City downtown area (S1) and the other in the heavy petrochemical industrial area (S2) in the western suburb of the City, characterized by significant differences in volatile organic compounds (VOCs) and NOx concentrations between the two sites. Results showed that during the high O3 episodes in summer, OH initiation was dominated by the reaction of excited oxygen atoms O(D) with water and the photolysis of nitrous acid (HONO) at the S1 site. At the S2 site, the most important production of OH was the reaction of O(D) + H2O, followed by the reaction of O3 with VOCs. HONO photolysis was mostly identified at 7:00– 13:00 local time at the S2 site, which is less important than that at the S1 site during the daytime. The photolysis of HCHO and OVOCs (except for HCHO) were the primary sources contributing to the initiation of HO2 and RO2 radicals at both sites. Results also revealed that the ROx termination could be attributed to the reactions of ROx with NO and NO2. The self-reactions between radicals also played an essential role at the S2 site. Overall OH was found to be the predominant oxidant, and NO3 was a major oxidant in the nighttime chemistry in the city.


INTRODUCTION
The hydroxyl radical (OH) and hydro/organic peroxy radicals (HO 2 and RO 2 ) collectively known as RO x , are crucial factors in atmospheric chemistry reactions and air pollution processes (Stone et al., 2012).These radicals can oxidize other atmospheric species which strongly affect the quality of ambient air as well as climate (Hofzumahaus et al., 2009;Vullo et al., 2016;Chen et al., 2017).With presence of nitrogen oxides (NO x ) and volatile organic compounds (VOCs), the RO x radicals cycling (e.g., OH-RO 2 -RO-HO 2 -OH) can produce O 3 and oxygenated VOCs (OVOCs) (Sheehy et al., 2010;Briggs et al., 2016;Brandt et al., 2016;Xue et al., 2016).Understanding the budgets of RO x radicals and formation processes of O 3 is very important in air quality forecasting.
The major radical sources include ozone photolysis and a rapid reaction of O( 1 D) with water vapors (Zafonte et al., 1977;Jaegle et al., 1998;Edwards et al., 2011), and other radical photolytic sources like nitrous acid (HONO), formaldehyde (HCHO) and other aldehydes, as well as other OVOCs.The non-photolytic sources are also important in the reactions of ozone with alkenes, and NO 3 radical with unsaturated VOC which can produce peroxy radicals during the nighttime (Geyer et al., 2001;Salisbury et al., 2001;Andréshernández et al., 2013).The predominance of these sources differs in different cases.In a polluted atmospheric environment, HONO photolysis (Aumont et al., 2003;Ren, 2003a;Acker et al., 2006;Elshorbany et al., 2009;Edwards et al., 2011;Michoud et al., 2012) and aldehydes photolysis (Alicke, 2002;Emmerson et al., 2005) are primary sources of radicals.The reactions of ozone and alkenes were also important sources of radicals in urban areas (Donahue et al., 1998;Bey et al., 2001;Kanaya et al., 2007).In China, RO x sources have been extensively studied in many regions.Photolysis of OVOCs and HONO is a primary RO x source in Beijing (Liu et al., 2012;Yang et al., 2018).In Hong Kong and the Pearl Region Delta (PRD) region (Xue et al., 2016), RO x sources during the daytime have been identified to be the photolysis of OVOCs (34-47%), HONO photolysis (19-22%), and O 3 photolysis (11-20%).
Under sunlight, a series of complex atmospheric chemical reactions between VOCs and NO x would occur, which leads to the ground ozone formation (Tan, et al., 2012;Wu et al., 2017).Ambient concentrations of VOCs and NO x as well as VOCs/NO x ratios can exert strong impacts on the rate of O 3 photochemical production (Finlayson-Pitts and Pitts, 1997;Lan et al., 2015;Kanaya et al., 2016).Haagen-Smit and Fox (1954) reported perhaps the first evidence which identified the importance of VOCs/NO x ratios on ozone formation process through ozone-NO x -VOC chemistry.VOCs and NO x levels in the atmosphere can alter atmospheric radicals, particularly the radical cycle and loss.For low VOCs/NO x , O 3 production is limited by "RO x cycle," which is referred to as VOC-limited cases.For high VOCs/NO x , on the other hand, the "NO x cycle" becomes a deterministic factor of O 3 production which is referred to as the NO x -limited O 3 formation regime (Wang et al., 2017).In the conditions of low NO atmospheric levels (NO xlimited), OH tends to react with VOCs, rather than NO x .Under the circumstance of high NO x atmospheric levels (VOC-limited), RO x will be terminated by the OH reaction with NO 2 , which also reduces the ozone production (Mao et al., 2010).In the case of high NO 2 concentrations, the reactions OH + NO 2 , HO 2 + NO 2 , and RO 2 + NO 2 which form HNO 3 , peroxides, and organic nitrates, are likely to be the main route of radical termination (Lu et al., 2013).Also, reaction with NO is another important route of radical termination.In the case of low NO x atmospheric level, the OH, HO 2 , and RO 2 termination reactions are the selfreaction and cross-reaction of radicals (Salisbury et al., 2001;Ren et al., 2006), respectively.These are the key processes that determine the self-cleaning ability in the troposphere.
Lanzhou, the capital city of Gansu Province, is encircled by mountains with a mean altitude of 1520 m a.s.l. in northwestern China (Yu et al., 2017).Lanzhou has been one of the large-scale petro-chemical and oil-refinery industrial bases in China since the 1950s.Over the recent decade, the number of motor vehicles has been increasing rapidly which worsens air quality in the city (Wang et al., 2016).The rapid expansion of energy industry in northwestern China (Ling et al., 2017;Wang et al., 2017), dry climate conditions, and strong UV-B and solar radiation due to high elevations might lead to typical photochemical reaction processes in northwestern Chinese cities.In this sense, the unique valley topography along with the petrochemical industry, and considerable increase in vehicle numbers, make Lanzhou a typical city contaminated by O 3 pollution in summer.In fact, the first photochemical smog event was reported in Lanzhou in the late 1970s (Tang et al., 1989), which triggered ground O 3 and air pollution research in China.Forty years after the Lanzhou photochemical event, ozone is still a primary air pollutant in the city during the summertime.In the past decade, while several studies (Xue et al., 2014b) have been devoted to discerning O 3precursor relationships or local vs.regional contributions to O 3 levels across the city, the atmospheric radical chemistry of O 3 is almost unknown.Large knowledge gaps in understanding of atmospheric chemistry in Lanzhou and other energy industrialized cities in northwestern China still exist.
The present study aims to (1) elucidate the combined effects of complex topography, large-scale petrochemical industry, oil refinery, and motor vehicles on ozone chemistry; and (2) enhance our understanding of the chemistry of RO x in ozone-contaminated Lanzhou city.We examined the atmospheric radical chemistry at the downtown and western suburb of the Lanzhou city in summer 2013.Ambient measurement data at the two locations showed lower VOCs/NO x ratio in downtown and higher VOCs/NO x in the western suburb of the city.Model comparisons with different VOCs/NO x , RO x budgets, and atmospheric oxidative capacity (AOC) at the two sites are quantified and discussed.

Field Sampling and Measurements
Atmospheric pollutants, including O 3 , NO x , CO, and VOCs were measured and collected at two monitoring sites from June 1 to August 31, 2013.The one site (S1) is located in Lanzhou Environmental Monitoring Station (LEMS, S1 site) in downtown Lanzhou, and the other (S2 site) is located in the western suburb of the city, featured by heavy petrochemical industry and about 20 km away from the site S1.More details on the sampling sites are referred to our previous publication (Jia et al., 2016).VOCs air concentrations were measured on-line using A52022 series of Gas Chromatography System (C 2 -C 6 A1100 Analyzers, C 6 -C 12 A21022 Analyzers, Chromatotec, France).The instrument is a high performance gas chromatograph with FlameIonization Detector (FID) applied in the analysis of VOC compounds (C 2 -C 6 hydrocarbons and BTEX) in gaseous samples.The detection limits of the instrument are typically higher in ambient concentration at 100 ppt or emission at ppm.Hourly concentrations of nitrogen oxides (NO x ), carbon monoxide (CO), sulfur dioxide (SO 2 ) and ozone (O 3 ) were monitored on-line using EC9841 Nitrogen Oxides Analyzer, EC9830 Carbon Monoxide Analyzer, EC9850B Sulfur Dioxide Analyzer, and EC9810B Ozone Analyzer (ECOTECH, Australia), respectively.The detection of NO (EC9841B) is based on the chemiluminescence technology, and NO 2 was measured after conversion to NO by the molybdenum oxide converter.The meteorological data (temperature, humidity, wind speed, atmospheric pressure, etc.) were collected from an auto meteorological station operated simultaneously at the two sampling sites.Instruments were calibrated using the standard gases in the United States Environment Protection Agency (U.S. EPA) Method TO-14A Calibration Mix (RESTEK, cat.#34400) before measurement.The single point calibration and peak shift correction (PSC) were also applied at regular intervals.The coefficient of variation (CV) in each species was calculated, and those CV values exceeding 15% were deleted.

Observation-based Model
An observation-based chemical box model was applied to quantify the chemical budgets of OH, HO 2 , and RO 2 .This model is built on the MCM (available on-line http://mcm.leeds.ac.uk/MCM/).The model includes a near-explicit chemical mechanism describing the detailed gas-phase degradation of a series of emitted VOC, and the formation of O 3 and other secondary pollutants, under the local environmental conditions.The MCM describes the degradation of 143 primary VOCs (Jenkin et al., 2003;Saunders et al., 2003).The resultant mechanism comprises 12691 reactions in 4351 organic species, and 46 associated inorganic reactions.The model has been successfully used in air quality modeling investigations in some previous research projects in China (Xue et al., 2014a(Xue et al., , b, 2016;;Li et al., 2018).
The measurement data of O 3 , NO, NO 2 , SO 2 , CO, C 2 -C 10 HCs, temperature, and relative humidity (RH) were averaged or interpolated to a 10-min time resolution subject to the model constraints.Averaged or linearly interpolated approaches depending on their time resolution were employed to create a 10-min database for the model constraints.
In the present study, since the measured HONO concentrations were not available, we simulated HONO formation in the model.In this model, vehicle emission, the homogeneous source from the OH + NO reaction, and the heterogeneous sources from reactions of NO 2 on the ground and aerosol surfaces were taken into account.HONO can be directly emitted from vehicles with an emission ratio HONO/NO x = 0.008 based on a tunnel measurement by Kurtenbach et al. (2001).The most significant gas-phase reaction that leads to HONO formation is the reaction of NO with OH.In addition, HONO is also formed from the conversion of NO 2 on aerosol and the ground surfaces.The reaction rates can be assumed to be the first order of the NO 2 concentrations (Aumont et al., 2003).According to Xue et al. (2014), the first order rate of the aerosol (k1) and the ground (k2) surface reactions can be calculated as: where C NO 2 is the mean molecular speed of NO 2 ; γ NO 2 ,aerosol and γ NO 2 ,ground are the uptake coefficient of NO 2 on the aerosol and the ground surfaces; A is the aerosol surface area concentration, S/V is the effective surface density of the ground.The values of these parameters were given by Li et al. (2010) and Xue et al. (2014).
The secondary VOCs might have affect on the RO x formation, propagation, and termination rates (Emmerson et al., 2005;Volkamer et al., 2010;Kato et al., 2011;Liu et al., 2012), and were estimated following the method proposed by Michoud et al. (2012).The formation of secondary species from the oxidation of the measured compounds up to a daily stationary state on a daily basis (from midnight to midnight) was repeatedly simulated, constraining each model run with the same measured data until the daily stationary concentrations of the secondary species were reached.These daily steady state secondary compounds were reached after 5-day model integration.After that, the daily secondary species concentrations were simulated by the MCM with zero initial concentration at the beginning of each day.This suggests that the concentration on next day is not associated with its level on a previous day.Such an approach has often been used in previous studies (Carslaw et al., 2001;Emmerson et al., 2007;Kanaya et al., 2007;Elshorbany et al., 2009;Bloss et al., 2010;Wang, 2010;Michoud et al., 2012).Hereafter, we shall refer this model configuration as the 5-day spin up model run.The model simulation starts from 00:00 local time (LT) and runs for a 24-h period.
In the present study, the photolysis frequencies were calculated as a function of solar zenith angle under clear and cloudless sky conditions (Saunders et al., 2003).All photolysis frequencies (J values) were calculated by (Jenkin et al., 1997): where x is the solar zenith angle, and L i , M i , N i are chemical-specific parameters.These parameters were derived via a fitting procedure for each chemical, developed by Hough (1988).Following Xue et al. (2014a), we also implemented the dilution mixing and dry deposition processes into the model.These two processes are associated with the mixing layer height, which varies from 300 m during the night to 1500 m in the afternoon.

Uncertainty Analysis for Model Parameters
Modeling results are likely subjected to some uncertainties and limitations.In this study, a first-order error propagation derived by Macleod et al. (2002) was used to calculate the uncertainties in input parameters (Cf i ) of the model, given by   2 2 exp ln where Cf out and Cf i are the confidence factors that span the 95% confidence interval around the median of a lognormally distributed variable, and S i is the relative sensitivity of the model output toward a change in input parameter i.
The Cf of input parameter uncertainties can be calculated as follows: where CV is coefficient of variation, defined as the ratio of the standard deviation to the mean.The sensitivity (S) is calculated by where ΔI and ΔO are the relative changes in input (I) and output (O) parameters of interest, respectively.The average sensitivity of increasing and decreasing input parameter was calculated.Each input parameter varied individually by ±10%, respectively.Due to a large number of complex atmospheric chemical reaction mechanisms in the MCM model, detailed distributions for all input parameters and reaction process included in the uncertainty analysis could not be obtained readily.In the present study, only those parameters significantly affecting modeling results were taken into consideration.The uncertainties in RO x concentrations simulated by the photochemical box model come from concentrations of their precursors (NO x and HCs), and primary factors affecting the atmospheric chemical reaction, such as relative humidity, photolysis rates, and mixing height.The uncertainties of these input parameters were collected from other studies and listed in Table 1.By using above equations, we estimated the uncertainties of RO x , OH, HO 2 , and RO 2. Results indicated that the CV of OH, HO 2 , and RO 2 at the S1 site was 7.42%, 13.88%, and 16.10%, respectively.Compared to that at the S1 site, the OH, HO 2 , and RO 2 exhibited the larger uncertainties, which varied at 10.35%, 14.63%, and 16.82%, respectively.
It should be noted that there were also some the uncertainties in the present modeling study which were mainly from the treatment of OVOCs and HONO simulations and the clear-sky assumptions.Unfortunately, these uncertainties are difficult to be evaluated by Monte Carlo or other uncertainty analysis because the coefficients of variation of these chemicals and assumption are not readily determined.

Field Measurements
Sampled concentrations of VOCs and criteria air pollutants and meteorological parameters during the summertime illustrated marked differences at the two sites.The atmospheric levels of VOCs at the western suburb site of the city (S2) were 3-11 times higher than that at the downtown site (S1).On the contrary, the average concentration of NO x at the downtown (S1) site was 30% higher than that at the industrial suburb (S2) site.The measured seasonal mean NO 2 and NO levels were 23.5 ± 3.1 and 11.1 ± 1.8 ppbv at the S1 site, and 14.2 ± 1.9 and 5.67 ± 0.92 ppbv at the S2 site, respectively.Similarly, the seasonal mean CO concentration at the downtown site (875 ± 43.8 ppbv) was also higher than its level at the suburb site (643 ± 32.2 ppbv).However, the seasonal mean SO 2 concentration at the S1 site (8.2 ± 1.1 ppbv) was lower than its level at the S2 site (9.2 ± 1.7 ppbv).
The MCM model simulations were performed for high O 3 episodes occurring at the two sites.These high O 3 episodes were chosen when O 3 maximum hourly concentrations exceeded 100 µg m -3 (~50 ppbv), which was set by the World Health Organization (WHO) and regarded as a threshold of ozone level imposing risks to human health.Eleven high O 3 episode days were selected at each site.It should be pointed out that at the S1 site, there were ten O 3 episodes with its maximum hourly concentration exceeding 50 ppbv, so we also selected an additional relatively high O 3 concentration event at the maximum hourly concentration of 48 ppbv (Table 2).The maximum hourly O 3 concentration was 98 ppbv at the S1 site and 186 ppbv at the S2 site, which was about the twice higher than that at the S1 site.Photochemical pollution with an hourly O 3 concentration higher than 100 ppbv was observed several times at the S2 site.But such a high O 3 episode was seldom occurring at the S1 site where O 3 levels were always lower than that at the S2 site.The date and ambient concentrations of O 3 , HCs, and NO x during the high O 3 episodes at the two sampling sites are presented in Table 2.The daytime hourly temperatures during high O 3 episode days were as high as 35°C.The daily mean RH levels were lower than 50%.Clear sky conditions prevailed in these selected high O 3 episode days.Other detailed information of VOCs and high O 3 episodes were referred to our previous study (Jia et al., 2016).

RO x Sources OH Sources
The maximum OH peak values were about 5 × 10 6 molecules cm -3 at both sites.This value was comparable to the estimated value over Mountain Tai in eastern China in June 2006 (~6 × 10 6 molecules cm -3 ) (Kanaya et al., 2009), but lower than the observed (13 × 10 6 molecules cm -3 ) and simulated (7 × 10 6 molecules cm -3 ) values at a sampling site in the PRD (Hofzumahaus et al., 2009), a well-developed and one of most heavily ozone-polluted area in China.The simulated OH value in our study was only the half of the simulated OH level in Beijing (9 × 10 6 molecules cm -3 ) (Liu et al., 2012), but was comparable to that measured at a suburban site in Europe (Michoud et al., 2012).The diurnal cycle of OH with the maximum daytime level of 5 × 10 6 molecules cm -3 was also comparable with a daily variation of OH with daytime maximal between 1.2-7.5 ×  for the same year.The summer relative humidity (RH) was also lower in Lanzhou (52%) compare to Beijing (70%) and Guangzhou (86%) (NBSC, 2014).Two OH radicals were formed by reaction of O( 1 D) with ambient water vapor.This reaction was a primary OH radical source in either clean or polluted atmosphere (Michoud et al., 2012).The OH production rate by the reaction of O( 1 D) with water vapor in Lanzhou was about 0.25 ppbv h -1 , much lower than that in Beijing (0.4 ppbv h -1 ) (Liu et al., 2012) and Hong Kong (0.9 ppbv h -1 ) (Xue et al., 2016).Also, the different OH loss processes were important reasons causing the different radical values in different cities. Fig. 2 shows the contributions of OH initiation by different species in the summertime at two different sites.At the S1 site, the reaction of O( 1 D) + H 2 O and the photolysis of nitrous acid (HONO) contributed 52.0% and 46.7% to the OH production, respectively.The same result was also reported in Hong Kong and the PRD region (Xue et al., 2016).In this case, the dominant OH sources were HONO photolysis and O 3 photolysis.At the S2 site, the reactions of O( 1 D) + H 2 O and O 3 + VOCs contributed 37.4% and 31.5% to the OH production, respectively.The other reactions contributing to OH radical include the photolysis of HONO (17.3%),H 2 O 2 (6.6%),OVOC (6.08%), and HNO 3 (0.61%).Particularly, at the S2 site where both VOC and O 3 concentrations were higher than that measured at the S1 site, O 3 + VOC reaction and OVOC photolysis were much more important to the OH radical formation.In summer, reactions of O 1 D + H 2 O, O 3 + VOC, and HONO  photolysis become the major source of OH, although these reactions were subject to different significance levels on different days.Similar results have also been reported in previous studies.For example, O 3 photolysis was the major radical source in New York, California, and Nashville (Martinez et al., 2003;Zhou et al., 2007;Volkamer et al., 2010).HONO photolysis has been identified as the main contributor of OH production in New York City, USA; Hohenpreissenberg, Germany; and Paris, France (Ren et al., 2003b;Acker et al., 2006;Michoud et al., 2012).In downtown Birmingham, ozone + alkene reactions as the main origin contributed 46% to OH radical in the summertime (Emmerson et al., 2005).
Recent studies have further revealed that HONO photolysis exerts a significant influence on radicals throughout a day (Ren, 2003b;Ren et al., 2006;Kleffmann et al., 2007;Dusanter et al., 2009;Elshorbany et al., 2009).The same result was also found in the present study at the downtown site.However, at the petrochemical industrialized suburb site, OH formation by the HONO photolysis mainly took place in the morning time (7:00-12:00 LT).Given that HONO measurement data were not available in the present study, in our model investigation, homogeneous reactions of NO + OH and heterogeneous processes on the particle and ground surfaces have been considered as vital HONO sources.This approach has also been used in previous studies, particularly in the elevated NO x emission area (Qin et al., 2009;Hendrick et al., 2014;Xu et al., 2015).In the early morning, with accumulated HONO during the nighttime (Hendrick et al., 2014), the dominate source of OH radicals was HONO photolysis.HONO also can be directly derived from fossil fuel combustion, such as vehicular emissions (Kurtenbach et al., 2001;Kleffmann, 2007).Compared with the downtown site (S1), lower population and a small number of vehicles at the suburb site (S2) resulted in lower NO x concentration.As a result, HONO photolysis occurred mainly in 7:00-13:00 local time at this site, which made a less significant contribution to the OH production at the S2 site.

HO 2 Sources
During the daytime, simulated HO 2 concentrations ranged from 0.6 × 10 8 to 1.5 × 10 8 molecules cm -3 (or ~2-8 pptv) at the S1 site.The peak HO 2 concentrations as high as 18 × 10 8 molecules cm -3 at the S2 site were 10 times higher than that at the S1 site.Compared to other urban areas in China, the simulated HO 2 concentrations at the urban S1 site was lower than Beijing and the PRD region (Hofzumahaus et al., 2009;Liu et al., 2012).At the S2 site, the simulated HO 2 concentrations were comparable to that in the populated and industrialized PRD region (Hofzumahaus et al., 2009).High HO2 concentrations at the noontime were simulated at the S2 site, which yielded a HO2/OH ratio of about 100.The same result was also found in the Mexico City Metropolitan area in April 2003(Shirley et al., 2006).In Mexico City, the peak OH was 0.35 pptv (~7 × 10 6 cm -3 ) which was similar to OH levels observed in a number of cities in the USA, but HO 2 (40 pptv) was much higher than most USA cities.Such feature was usually observed in the relatively clean air with low concentrations of NO x .Although the level of VOCs at the S2 site was higher than that at the S1 site, the NO level was lower at the S2 with mean concentration at about 6 ppbv which was about the half of its mean level (11.1 ppbv) at the S1 site.The smaller value of HO 2 at the S1 site (or urban area) could be attributed to higher NO concentrations, which can convert HO 2 to OH rapidly (Ren, 2003b).Accordingly, higher levels of VOCs at the S2 site yielded higher HO 2 by a series of RO x reactions.The concentration discrepancy of HO 2 at the two sites can also be attributed to the contribution of different sources.
Since the key reactions of OVOCs for HO 2 were rather complex, the reactions were grouped to reduce complexity.The reaction rate for the production of HO 2 from the photolysis of HCHO was studied as a single reaction to emphasize the importance of HCHO in the HO 2 formation.Here, the production of HO 2 from the photolysis of all other aldehyde species was considered as a group.The other OVOCs were grouped to another category (Carslaw et al., 2005).The HO 2 formation by initiation reactions is shown in Fig. 3.The major reactions that produce HO 2 did not differ significantly during day and night at the two sites.The photolysis of HCHO plays a most important role in the HO 2 formation in the daytime at the two sites, contributing 63.7% at the S1 and 45% at the S2 site.The reaction of NO 3 + VOC was the second source in the nighttime at the S1 site with the absence of photolysis reactions.At the S2 site, the contribution of other aldehydes (except HCHO) to HO 2 initiation (28%) was also critical during the daytime, followed by the reactions of O 3 + VOC (13.9%), and OVOC photolysis (12.5%), respectively.In the nighttime, the reaction of O 3 + VOC was the major source of HO 2 at the both sites.This is likely attributed to the differences of the precursor chemicals measured at the two sites.Ambient concentration data showed the relatively high concentration of NO x at the S1 site and high VOC concentration at the S2 site (Jia et al., 2016).

RO 2 Sources
The averaged peak value of RO 2 radical was 6.92 × 10 8 molecules cm -3 , which was similar to that of HO 2 radical at the S1 site.The simulated maximum RO 2 concentration at the S2 site (20 × 10 8 molecules cm -3 ) was almost identical to that in the populated PRD region (Hofzumahaus et al., 2009).
Fig. 4 shows the percentage contribution of OVOCs photolysis (except for HCHO) and reactions of O 3 + VOCs and NO 3 + VOCs to the production of RO 2 .As seen, OVOCs photolysis was the largest source of RO 2 during the daytime, which accounted for 82.2% and 63.4% of total production of RO 2 at the S1 and S2 site, respectively.In the nighttime, O 3 + VOC was the main source of RO 2 radicals, contributing 64% and 86% to the production of RO 2 at the S1 and S2 site.Our modeling result also revealed that the reaction of NO 3 radical with VOC contributed 5.23% in the daytime and 32.7% in the nighttime to RO 2 production at the S1 site, and 10.9% and 10.7% during the day and nighttime at the S2 site, respectively.This result is somewhat different from many previous studies which showed a negligible role of NO 3 during the daytime (Monks et al., 1998;Fuentes et al., 2000;Kleinman, 2000).NO 3 is of vital importance during the daytime chemistry in the polluted atmosphere in Lanzhou, because VOCs can be oxidized by the NO 3 and result in the formation of RO 2 and nitrogen-containing organic aerosols (Saunders et al., 2003;Rollins et al., 2012;Xue et al., 2016).

RO x Propagation and Loss
The main propagation routes and loss of OH, HO 2 , and RO 2 at the S1 and S2 sites were shown in Fig. 5.The total initiation rates and loss rates for the combined OH, HO 2 , and RO 2 were approximately in balance, which was 9.3 × 10 6 molecule cm -3 s -1 at the downtown site (S1) and 27.2 × 10 6 molecules cm -3 s -1 at the suburb site, respectively.
It has been widely known that the radical propagation is effective in producing new radicals under the condition of abundant NO x and VOCs (Elshorbany et al., 2009;Liu et al., 2012;Xue et al., 2016).Our simulations showed that OH was transformed into HO 2 and RO 2 by oxidation of CO and VOCs at the average rates of 2.96 and 7.32 × 10 6 molecule cm -3 s -1 during the daytime at the S1 site.At the petrochemical industry S2 site, OH -HO 2 also has a rate of 2.97 × 10 6 molecule cm -3 s -1 whereas the rate of OH -RO 2 at 13.3 × 10 6 molecule cm -3 s -1 was almost twice as high as the rate at the S1 site.These were the major reasons for the comparable OH concentrations at both sites, although the OH sources at S2 site were several times more than that at S1 site.These reactions, in turn, led to the production of RO and OH, with O 3 formed as a by-product.Overall the rates of these reactions were higher at the petrochemical industry site than that at the downtown site.The rates of RO 2 + NO and HO 2 + NO were 12.7 and 6.45 × 10 6 molecule cm -3 s -1 at the S2 site and were 10.5 and 5.93 × 10 6 molecule cm -3 s -1 at the S1 site, respectively.
In the termination processes, the radical loss is controlled by the NO x + RO x reactions only under high NO x conditions, and the self-reactions of radicals are important under low NO x conditions (Xue et al., 2016).In our case, at the S1 site with a high level of NO x , the main terminations were the reactions with NO x species.This is because reactions with NO x usually dominate the radical sink in the high-NO x environment.The loss rate of OH, HO 2 , and RO 2 at the S1 site was 3.15, 2.44, and 2.59 × 10 6 molecule cm -3 s -1 , respectively.The reactions OH + NO 2 , HO 2 + NO 2 and RO 2 + NO 2 together contributed more than 99% to the termination processes.
At the S2 site, the reactions with NO x species were also important processes for the loss of RO x , whereas the selfreactions between radicals (HO 2 + HO 2 , HO 2 + RO 2 , and RO 2 + RO 2 ) are not negligible because of significantly  high HO 2 and RO 2 levels at this industrial site in the summertime.The loss rates of OH, HO 2 , and RO 2 were 2.34, 11.4, and 15.1 × 10 6 molecule cm -3 s -1 , respectively.The main radical loss was governed by HO 2 and RO 2 , which was significantly different from that at the S1 site.Compared to the S1 site, the self-reactions between radicals (HO 2 + HO 2 , HO 2 + RO 2 , and RO 2 + RO 2 ) at the S2 site accounted for a relatively high fraction of radical loss, namely the HO 2 loss process.The reactions HO 2 + RO 2 and HO 2 + HO 2 were the main loss processes of HO 2 , accounting for 45.8% and 25.3% of the total loss of HO 2 , respectively, attributable to the different levels of VOCs and NO x and different VOCs/NO x ratios at the two sites (Jia et al., 2016).

Atmospheric Oxidative Capacity
Atmospheric Oxidative Capacity (AOC) was calculated by summing the oxidation rates of VOCs, CO, and CH 4 molecules by OH, O 3 , and NO 3 (Geyer et al., 2001;Li et al., 2018).To highlight the significance of AOC in a polluted ambient environment, we selected a high O 3 pollution episode occurring on July 6, 2013, with maximum hourly O 3 concentrations of 84 and 186 ppbv at the S1 and S2 sites, respectively.High NMHCs levels were also observed during this episode with peak concentrations at 110 ppbv at the S1 site and 281 ppbv at the S2 site.Fig. 6 illustrates estimated AOC subject to three oxidants OH, O 3 , and NO 3 .As shown, the maximum AOC at the S2 site was 1.7 × 10 8 molecules cm -3 s -1 , which was much higher than that estimated at the S1 site (0.37 × 10 8 molecules cm -3 s -1 ).Mean AOC averaged over the daytime (08:00-20:00 LT in summer) was 2.60 × 10 7 and 6.23 × 10 7 molecules cm -3 s -1 at the S1 and S2 sites, respectively.Higher AOC at the S2 site is likely associated with higher concentration of VOCs and O 3 at this petrochemical industrial site.
The AOC components during this episode were almost identical.In the daytime, OH was the predominant oxidant contributing 91.2% and 85.9% to AOC at the S1 and S2 sites, respectively.O 3 was the second most important oxidant contributing 7.86% and 12% to AOC at these two sites, respectively.NO 3 only accounted for a small portion of AOC at 0.89%, and 2.17% at the downtown and suburban sites, respectively.In the nighttime, NO 3 became a principal oxidant contributing 67.5% and 72.8% to AOC at the S1 and S2 sites, respectively.O 3 was again the second important oxidant contributing 14.6% and 21.6% to AOC at these two sites, followed by OH radical (17.9% at the S1 site and 5.67% at the S2 site).
Our results agree, in general, with many previous studies for AOC (Elshorbany et al., 2009;Xue et al., 2014b).Since OH was the predominant oxidant and could contribute ~90% to the AOC, NO 3 and O 3 play a less important role.At the downtown S1 site, AOC in summer 2013 was much higher than those estimated at a rural site in Germany, with the daytime average AOC at 2.06 × 10 7 molecules cm -3 s -1 (Geyer et al., 2001), but lower than that at some urban sites, such as the downtown Santiago and Hong Kong (Elshorbany et al., 2009;Xue et al., 2016).At the petrochemical industrial suburban S2 site, the calculated AOC level reached 1.7 × 10 8 molecules cm -3 s -1 .This level was also lower than that at TC (2.04 × 10 8 molecules cm -3 s -1 ) (Xue et al., 2016) and Santiago (3.4 × 10 8 molecules cm -3 s -1 ) (Elshorbany et al., 2009).Averaged AOC over the daytime was 6.23 × 10 7 molecules cm -3 s -1 , which was comparable to 7.26 × 10 7 molecules cm -3 s -1 estimated in downtown Hong Kong.
As aforementioned, the model simulations were performed under the condition of high O 3 concentrations (up to 180 ppbv), the oxidation capacity of O 3 may be, to some extent, overestimated, especially during the daytime.Xue et al. (2016) suggested that the NO 3 radical may play an important role in the daytime oxidation under certain ambient environmental conditions.Our result also revealed that NO 3 contributed 18% to AOC during the nighttime at the S1 site, which was higher than that at the S2 site (5.76%).This can be attributed to high NO x level at the downtown S1 site.

CONCLUSIONS
In this work, a photochemical box model was used to quantify the key reaction pathways between OH, HO 2 , and RO 2 radicals at an urban site and a petrochemical industrial site in Lanzhou in northwestern China.The maximum OH peak values were similar at both sites, but the sources of OH at the two sites differed.At the downtown site, the reaction of O( 1 D) + H 2 O and the photolysis of nitrous acid (HONO) were the dominant OH sources, while at the petrochemical industrial site, the reactions of O( 1 D) + H 2 O and O 3 + VOCs made the main contributions to OH.At both the S1 and S2 sites, the photolysis of HCHO and reaction of O 3 + VOC dominated the initiation of HO 2 radicals in the daytime and nighttime, respectively.Photolysis of OVOCs (except for HCHO), reactions of O 3 + VOCs, and NO 3 + VOCs were the main sources of RO 2 in Lanzhou.
The total initiation rates and loss rates for the combined OH, HO 2 , and RO 2 were approximately balanced at the S1 and S2 sites.At the S1 site, the reactions OH + NO 2 , HO 2 + NO 2 , and RO 2 + NO 2 collectively accounted for over 99% of the termination processes; the self-reactions among radicals (HO 2 + HO 2 , HO 2 + RO 2 , and RO 2 + RO 2 ) contributed a large portion of the radical loss at the S2 site.OH was the predominant oxidant during the daytime; NO 3 became a principal oxidant during the nighttime.Although the lack of OVOCs and HONO data in the present study might create uncertainties in and bring difficulties to the interpretation and verification of the model results, the results can enhance the understanding of radical chemistry in typical industrial cities in northwestern China.

Fig. 1
illustrates the hourly variation of mean OH levels at the two sites in summer 2013.As shown in the figure, the peak value of OH at the S1 site was found at the local solar noon (about 15:00 LT) whereas the OH concentration reached the peak at about 12:00 LT at the S2 site.This difference was caused by the different source of OH at the different sites.In general, at the S1 site, the reaction of O( 1 D) + H 2 O and the photolysis of nitrous acid (HONO) were the dominate OH sources.At the S2 site, the reactions of O( 1 D) + H 2 O and O 3 + VOCs made a main contribution to OH.

Fig. 1 .
Fig. 1.Hourly variations of mean OH at two sites averaged during the high O 3 episode in summer (June 1-August 31), 2013.The vertical bars show the hourly standard deviations.

Fig. 4 .
Fig. 4. Percentage contributions of OVOC photolysis and reactions of O 3 + VOC and NO 3 + VOC to the production of RO 2 in the daytime (8:00-20:00) and nighttime (21:00-7:00) during the high O 3 episodes for the summertime at the two sites in Lanzhou City.

Table 1 .
The CV of input parameters in the MCM model

Table 2 .
Date and peak concentration of O 3 (ppb) during high O 3 episodes at two sampling sites, Lanzhou.
(Emmerson et al., 2007) the summer heat waves in 2003 at Writtle College, United Kingdom(Emmerson et al., 2007). T relatively lower OH values at our sampling sites comparing to eastern China were likely attributed to very low RH in Lanzhou in arid and semi-arid northwestern China with annual precipitation < 300 mm.The summer mean temperature over Lanzhou was 23.2°C in 2013, which was lower than Beijing (26.4°C) andGuangzhou (27.5°C)