Summertime photochemistry during CAREBeijing-2007 : RO x budgets and O 3 formation

Summertime photochemistry during CAREBeijing-2007: ROx budgets and O3 formation Z. Liu, Y. Wang, D. Gu, C. Zhao, L. G. Huey, R. Stickel, J. Liao, M. Shao, T. Zhu, L. Zeng, A. Amoroso, F. Costabile, C.-C. Chang, and S.-C. Liu School of Earth and Atmospheric Science, Georgia Institute of Technology, Atlanta, USA College of Environmental Sciences and Engineering, Peking University, Beijing, China Institute for Atmospheric Pollution, National Research Council (CNR-IIA), Rome, Italy Research Center for Environmental Changes (RCEC), Academic Sinica, Taipei, China now at: the Pacific Northwest National Laboratory, Richland, Washington, USA


Introduction
Photochemical smog was first documented in 1950s in Los Angeles (Haagen-Smit and Fox, 1954), and is nowadays a prevalent air pollution phenomenon around the world (e.g., Molina and Molina, 2004;Monks, et al., 2010).A major contributor to smog is the production of secondary pollutants Published by Copernicus Publications on behalf of the European Geosciences Union.

Z. Liu et al.: Summertime photochemistry during CAREBeijing-2007
such as O 3 and aerosols from photochemical reactions involving NO x (NO x ≡ NO + NO 2 ) and volatile organic compounds (VOCs), which are emitted from various anthropogenic and natural sources.Over the past decades, continuously improving knowledge of photochemical pollution has successfully served as the basis for formulating the pollution control strategies in the United States (NRC, 1991;NARSTO, 2000).Uncertainties of photochemical modeling in some regions remain large due to the lack of accurate emission inventories (NARSTO, 2005;Liu et al., 2012) and the current incomplete knowledge of chemistry (e.g.Volkamer et al., 2010;Lin et al., 2012).
A region of concern is China.The rapidly increasing emissions of NO x and VOCs over China since 1980s driven by economic growth have been observed by satellites (e.g., Richter et al., 2005) and documented in bottom-up inventories (e.g., Zhang et al., 2009).As an expected consequence, elevated O 3 (e.g., Wang et al., 2006) and peroxy acetyl nitrates (PANs) (e.g., Liu et al., 2010) accompanied by high loadings of aerosols (e.g.Chan and Yao, 2008;Zhang et al., 2008) have been observed in the country.Severe O 3 and aerosol pollution on an unprecedented large regional scale (Zhao et al., 2009a;van Donkelaar et al., 2010) have also drawn attention given the large impact on public health.
Furthermore, some recent observations over China highlighted the complexity of photochemistry that cannot be fully explained by current knowledge.For example, surprisingly high daytime HONO concentrations from unknown sources have been observed in Beijing (An et al., 2009) and the Pearl River Delta (PRD) region (Su et al., 2008(Su et al., , 2011)).At a suburban site in PRD, current standard photochemistry could not explain the observed level of OH, the key oxidant in the troposphere (Hofzumahaus et al., 2009;Lu et al., 2012).Due to the high loading of aerosols, heterogeneous chemistry appears to strongly affect the radical budget (Kanaya et al., 2009) and reactive nitrogen processing (Pathak et al., 2009).A case in point is that we still do not have a clear understanding of how the large emission reductions affected secondary pollutants during the 2008 summer Beijing Olympic and Paralympic Games.The chemical transport modeling study by Yang et al. (2011) demonstrated highly variable chemical sensitivities of O 3 to its precursor emissions due to the uncertain emissions of aromatic VOCs.However, the sensitivity relations are very difficult to derive from observations.For example, Wang et al. (2010) found increases of O 3 , sulfate and nitrate while NO x and VOCs decreased at an urban site in Beijing in the first two weeks after the emission control for the Olympics Game.A similar finding was reported at another urban site in Beijing (Chou et al., 2011).These findings reflect the fact that the effects on O 3 from precursor emission changes can be masked by the variations in the spatial pollutant distribution and meteorological conditions for dispersion, transport, and chemical photolysis.
Given the difficulty of interpreting observations, an alternative approach is through in-depth observation-based mod-eling analyses to probe into the chemical system.In this work, we analyze the O 3 photochemical processes in Beijing in August 2007 during the CAREBeijing (Campaigns of Air quality REsearch in Beijing) Experiment employing the 1-D version of the Regional chEmical and trAnsport Model (REAM-1D) constrained by observed chemical species and physical parameters, including O 3 , NO, PAN, HONO, VOCs, and aerosol surface areas.Through detailed chemical budget analysis, we aim to gain a detailed understanding of the budgets of RO x (OH + HO 2 + organic peroxy radicals (RO 2 )) radicals and formation processes of O 3 .A relatively large number of sensitivity experiments are conducted to address the uncertainties in chemistry and diagnose the response of O 3 formation to emission control strategies.The sensitivity experiments are chosen to understand the impacts of aromatics, the observed high concentrations of HONO in daytime, and aerosol uptake of HO 2 separately, such that the effects of the uncertainties in measurements and chemical understanding are apparent from the model results.
The remainder of the paper proceeds as follows.In Sect.2, we describe the measurement methods and the REAM-1D model (model setup, sensitivity experiments, and diagnostics).Section 3 presents the modeling analysis results.We show the budgets of RO x radicals in Sect.3.1, which will also form the basis for analyzing production and loss rates of O 3 in Sect.3.2.We then further elaborate on the significant roles and uncertainties of aromatics, HONO, and aerosol uptake of HO 2 in the budgets of radicals and O 3 , respectively, in Sect.3.3.The sensitivities of O 3 production to NO x and VOCs are examined in Sect. 3.4. Finally,in Sect. 4, we summarize our findings and discuss the implications for O 3 pollution control strategies over China.

Measurement methods
During the CAREBeijing-2007 experiment (Zhu et al., 2009), a full suite of trace gases were measured concurrently in August 2007 at an urban site located on a building roof top (∼20 m above the ground level) on the campus of Peking University (PKU) (39.99 • N, 116.31 • E).Nitrogen monoxide (NO) was measured with a custom-made chemiluminescence detector (Ryerson et al., 2000).Total reactive nitrogen compounds (NO y gas-phase only) were measured by the conversion of the NO y species to NO on a molybdenum converter operated at 300 • C. PAN was measured using a chemical ionization mass spectrometer (CIMS) (Slusher et al., 2004).HONO was measured with a liquid coil scrubbing/UV-VIS instrument (Amoroso et al., 2006).O 3 and CO were measured by commercial instruments from the ECOTECH (EC9810 and EC9830).C 3 -C 9 NMHCs were measured with a time resolution of 30 min using two online GC-FID/PID systems (Syntech Spectra GC-FID/PID GC955 series 600/800 VOC analyzer), one for the C 3 -C 5 NMHCs, and the other for C 6 -C 9 NMHCs (Shao et al., 2009).Another automated GC/MS/FID system was deployed to measure NMHCs in daytime (08:00-09:00 and 13:00-14:00) (Hofzumahaus et al., 2009).OVOCs were measured using the PFPH-GC/MS method (Ho and Yu, 2004).Size distributions of aerosols (3 nm-10 µm) measured every 10 min with a Twin Differential Mobility Particle Sizer -Aerodynamic Particle Sizer (TDMPS-APS) were used to calculate aerosol surface areas.The uncertainties (1σ ) for these measurements are estimated to be 5 % for NO, O 3 , CO, 3-5 % for NMHCs, 10 % for NO y , PAN, HONO and OVOCs.More detailed descriptions of the instruments and measurement methods are available in the supplement.

The REAM-1D model and sensitivity experiments
The 3-D version of the Regional chEmical and trAnsport Model (REAM-3D) has been applied in a number of studies on O 3 photochemistry and transport at northern mid-latitudes (Choi et al., 2005(Choi et al., , 2008a, b;, b;Wang et al., 2007;Zhao et al., 2009aZhao et al., , b, 2010;;Zhao and Wang, 2009;Yang et al., 2011;Liu et al., 2012).The REAM-1D model shares the modules for O 3 -NO x -hydrocarbon photochemistry, vertical diffusion, convective transport, and wet/dry deposition (Liu et al., 2010) with the REAM-3D model.The chemical kinetics data were updated with the latest compilation by Sander et al. (2011), and the VOC chemistry in REAM-3D is expanded to include the chemistry of aromatics based on the SAPRC-07 chemical mechanism (Carter, 2009).Vertical transport is driven by WRF assimilated meteorological fields based on the NCEP reanalysis data (Zhao et al., 2009a).Meteorological parameters (i.e.water vapor concentrations, temperature, pressure, and diffusion coefficient) in the model are taken from WRF outputs.Photolysis rates are dependent on cloud fraction and optical depth calculated based on WRF meteorological fields (Choi et al., 2008b).
The standard 1-D model is constrained with measured CO, O 3 , NO, HONO, NMHCs (C 2 -C 9 ), OVOCs (acetone, acetaldehyde and formaldehyde) and aerosol surface areas in the first layer.With these tracers transported vertically (mostly via eddy diffusion), such a setup is similar to a 1-D model with specified emissions from the surface (Trainer et al., 1991), except that the model in this study is constrained by available observations.3-D REAM model simulated chemical tracer concentrations (Zhao et al., 2010) in the column over Beijing were used as initial and boundary conditions at upper model layers.For measurements made with a time resolution longer than 1 min (e.g.NMHCs and OVOCs, aerosol surface areas), constant measured values were assigned during the measurement period.Missing data points on some days due to instrumental issues were replaced with the corresponding value in the overall average diurnal profile at the time of missing data.The 1-D model was run with a 1-min time step, continuously from 1 August to 30 August 2007 and the results for the last 20 days were analyzed after a spin-up time of 10 days.The REAM-1D model with such a setup is able to reproduce the observed PAN near the surface in Beijing, which has been shown to have equal contributions by chemistry near the surface and from downward vertical transport (Liu et al., 2010).In addition to resolving vertical mixing in the planetary boundary layer (PBL), the 1-D model allows for explicit computation of dry deposition rates, which are assumed or estimated as lifetime parameters in box models.For additional information of the REAM-1D model and its performance, we refer the readers to the supplement and Liu et al. (2010).
OVOC and HONO both have photochemical sources and sinks.Their concentrations in the model cannot be specified using observed values when the sensitivities of RO x budgets and O 3 production to these species are analyzed.Model simulated OVOCs, including formaldehyde, acetaldehyde, and acetone, methylglyoxal, and glyoxal, agree with the observations within 20 % in terms of daytime average concentrations (Table S3 in the Supplement), indicating that secondary production is their predominant source.Removing the constraint of observed OVOCs in the standard model did not lead to notable changes in the simulated RO x concentrations or O 3 production/loss rates.Exceptionally high levels of HONO were observed at daytime (∼1ppbv in the afternoon) during the study period.The gas-phase source from the NO+OH reaction alone could only explain a minor portion (∼10 %) of the observed HONO concentrations.In order to investigate the photochemical impact of this missing source of HONO, we introduce a pseudo-reaction of NO 2 →HONO in the model (at a rate of 6.4×10 −5 s −1 on average during daytime, in the first model layer near the surface) to reproduce the observed daytime HONO and quantitatively estimate the primary radical source from the heterogeneous HONO production pathway.Replacing the observed HONO with the pseudo HONO production does not lead to notable changes in RO x concentrations.
Due to the large aerosol surface areas (∼1000 µm −2 cm −3 ), the uptake of HO 2 by aerosols may become a large HO 2 sink.The HO 2 aerosol reactive uptake coefficient, γ , is still quite uncertain and may be a function of temperature and aerosol composition (Thornton and Abbatt, 2005;Thornton et al., 2008;Kanaya et al., 2009;Mao et al., 2010).In this work, we chose a moderate value of γ (HO 2 ) = 0.02 in the standard model (S0 in Table 1) (Thornton et al., 2008), and we evaluate model sensitivities by varying the value of γ (HO 2 ) from 0 to 0.2, a range covering the general range of γ (HO 2 ) (e.g., Thornton et al., 2008).
Besides the standard model S0, we designed and conducted a number of sensitivity simulations summarized in Table 1.S0 is the standard model, while S0a and S0b are S0 with varied γ (HO 2 ) values, 0 in S0a and 0.2 in S0b, respectively.S1 is S0 without the "excess" HONO that cannot be explained by gas-phase HONO production; S2 is S0 without ) senst S0 with γ (HO 2 ) = 0, 0.02, 0.05, 0.1, 0.15, 0.2 aromatics; S3 is S0 without excess HONO or aromatics, and S3a further removes the aerosol HO 2 uptake (γ (HO 2 ) = 0) in S3.These experiments (S1-S3, γ (HO 2 ) senst ) further examine the sensitivities of radical budgets and O 3 production to those key factors identified in S0, including aromatic VOCs, daytime HONO concentrations, and aerosol uptake of HO 2 , which have considerable uncertainties due to the current incomplete knowledge in the related processes (Sect.3.3).The P(O 3 ) senst simulations compare O 3 production rates under varied NO x and VOCs conditions.In these sensitivity simulations, we did not constrain HONO or OVOCs to the observations in order to retain the feedback from them.

Model diagnostics
The 1-D modeling is constrained by surface observations.Vertical profiles of O 3 and its precursors are also simulated by turbulence transport from the surface.We can diagnose the radical and O 3 budgets near the surface and in the PBL.We focus the model diagnostics on the budgets near the surface for two reasons.Ozone and its precursors are simulated in the PBL and are specified to the observations near the surface.The near-the-surface model diagnostics are therefore less uncertain than in the PBL.Heterogeneous processes of excess HONO production and aerosol uptake of HO 2 are only simulated near the surface since we only have the observed HONO and aerosol size distribution near the surface.The radical budgets cannot be properly analyzed in the PBL given the importance of these heterogeneous processes.We will, however, present the O 3 budget analysis in the PBL.The concentrations of O 3 near the surface are affected by the production rates in the PBL because of the relatively long chemical lifetime of O 3 .The integration of O 3 budget terms is described in the supplement.In the following sections, all the PBL analysis results are specifically noted.If not noted, the analysis is for the surface layer.

Results and discussions
O 3 formation and the chemical dependence on the emissions of O 3 precursors involve production and recycling of a variety of RO x radicals.In this section, we first present the abundance and budgets of RO x radicals calculated with the observation-constrained 1-D model, to identify the main features and uncertainties of photochemistry in Beijing.By focusing on the chemical pathways controlling O 3 production and loss, and their responses to varied precursors, we then quantify the rates of O 3 formation and diagnose its chemical regime.

Model simulated concentrations of OH, HO 2 and RO 2
Figure 1 shows the average diurnal profiles of OH, HO 2 and RO 2 concentrations simulated in the standard model (S0).The 20-day average diurnal maximum concentrations of OH, HO 2 , and RO 2 are 9×10 6 , 6.8×10 8 , and 4.5×10 8 molecules cm −3 , respectively.RO x radical measurements over mainland China are still sparse.To put our simulated concentrations in context, the simulated maximum OH concentration (9×10 6 molecules cm −3 ) in this study for Beijing is ∼50 % higher than that simulated over Mountain Tai in June of 2006 (6×10 6 molecules cm −3 ) (Kanaya et al., 2009), and lies between the observed (13×10 6 molecules cm −3 ) and simulated (7×10 6 molecules cm −3 ) values at a site in PRD (Hofzumahaus et al., 2009).The maximum HO 2 concentration simulated for Beijing (6.8×10 8 molecules cm −3 ) is close to that over Mountain Tai (Kanaya et al., 2009), and is only half of that in PRD (Hofzumahaus et al., 2009).We note that HO 2 concentrations simulated over China (Kanaya et al., 2009 and this work) are quite sensitive to γ (HO 2 ), due to the large abundance of aerosols.Interestingly, removing the aerosol HO 2 sink in this study would lead to higher simulated HO 2 and OH concentrations that are in good agreement with the observed values at the PRD site by Hofzumahaus et al. (2009), although the locations and time of the two studies are different.Compared to urban areas outside China, the simulated OH and HO 2 concentrations for Beijing are similar to those observed in Mexico City (Shirley et al., 2006;Dusanter et al., 2009), and yet higher than those in New York City (Ren et al., 2003) and Birmingham of the UK (Emmerson et al., 2005a).RO 2 radicals include all organic peroxy radicals derived from VOC oxidation, and are categorized into 7 groups (Fig. 1), i.e. methyl peroxy radicals (CH 3 O 2 ), first generation peroxy radicals from alkanes (ALKA p ), alkenes except isoprene (ALKE p ), isoprene (ISO p ), aromatics (ARO p ), acyl peroxy radicals (RCO 3 ), and peroxy radicals from OVOCs (OVOC p ).The three most abundant groups of RO 2

RO x budgets
Figure 2 illustrates schematically the RO x daytime (06:00-18:00) budgets simulated in the model.Minor RO x radical sources (<0.1 ppbv h −1 ), e.g.ozonelysis of alkenes, are not shown.The reactions of NO + HO 2 (19.8 ppbv h −1 ) and NO + RO 2 (12.2 ppbv h −1 ) are the two largest pathways of radical cycling in the system, mainly due to the abundance of NO (e.g.∼ 5ppbv around noontime).As a result, the fast RO x cycling is mainly driven by these two NO + RO x reactions, as typically seen in NO x -rich environments (Emmerson et al., 2005b;Shirley et al., 2006;Dusanter et al., 2009;Elshorbany et al., 2009).
Besides the efficient radical cycling, the primary sources and sinks of radicals are also very large in Beijing.Photolysis of OVOCs turns out to be the predominant primary RO x source (4.0 ppbv h −1 ), and also the largest sources of HO 2 (2.4 ppbv h −1 ) and RO 2 (1.6 ppbv h −1 ), consistent with previous urban studies (Jenkin et al., 2000;Emmerson et al., 2005b;Dusanter et al., 2009).Photolysis of excess HONO is the second largest RO x source (3.0−0.8 = 2.2 (ppbv h −1 )), as well as the largest source of OH.At noontime the excess HONO produces OH at ∼5 ppbv h −1 (Fig. 3), a rate that is comparable to that found by Su et al. (2008) at Xinken in PRD, and slightly higher than that at another site, Backgarden, in the same region (Hofzumahaus et al. 2009).By contrast, the reaction of O 1 D + H 2 O only contributes 0.4 ppbv h −1 of the primary OH production, less than 1/5 of the excess HONO source, which is consistent with the finding in PRD by Hofzumahaus et al. (2009).It should be noted that these aforementioned excess daytime HONO sources inferred over China are significantly larger than most urban areas outside China (e.g.Acker et al., 2006;Kleffmann, 2007 and references therein;Dusanter et al., 2009;Elshorbany et al., 2009;Costabile et al., 2010).Summing up all these aforementioned sources gives a total primary RO x production rate at 6.6 ppbv h −1 (2.6 ppbv h −1 for OH, 2.4 ppbv h −1 for HO 2 , and 1.6 ppbv h −1 for RO 2 ), which is comparable to that in Santiago, Chile (7.0 ppbv h −1 ) (Elshorbany et al., 2009), but ∼50 % higher than those in Mexico City in 2006 (4.75 ppbv h −1 ) (Dusanter et al., 2009), and Birmingham of the UK (4.5 ppbv h −1 ) (Emmerson et al., 2005b).
RO x radicals are ultimately removed from the atmosphere via deposition of radical reservoir species, e.g.HNO 3 , H 2 O 2 , ROOH.The net radical losses via NO x -radical reactions are 3.6 ppbv h −1 , including NO 2 + OH producing HNO 3 (1.7 ppbv h −1 ), RO 2 + NO 2 producing organic peroxy nitrates (RO 2 NO 2 , mostly PANs) (1.1 ppbv −1 ), and RO 2 + NO producing organic nitrates (RONO 2 ) (0.8 ppbv h −1 ).By contrast, the radical loss rates via radical-radical reactions producing peroxides such as H 2 O 2 and ROOH are much lower (0.6 ppbv h −1 in total).Such a contrast has been typically seen as a feature of chemistry in NO x -rich urban environments.Another important and still uncertain RO x sink in Figs. 2 and 3 is the aerosol uptake of HO 2 (1.1ppbv h −1 ), mainly owing to the abundant aerosols in Beijing.The magnitude of this radical sink varies significantly with values of γ (HO 2 ) used in the model (Sect. 3.3.3).It is also noteworthy that the two RO 2 + NO x reaction pathways collectively contribute a net RO x loss at 1.9 ppbv h −1 , larger than that of NO 2 +OH (1.6 ppbv h −1 ), which is different from most urban environments outside China (Emmerson et al., 2005b;Dusanter et al., 2009;Elshorbany et al., 2009).Figure 4 shows the diurnal transition of RO 2 NO 2 production and loss.During most of the daytime, RO 2 NO 2 production dominates over its loss processes (mainly via thermal decomposition) leading to the net formation of RO 2 NO 2 and thus sequestering of NO 2 and R(O)O 2 radicals.RO 2 NO 2 loss starts to dominate over production from late afternoon into evening.Such a diurnal transition of RO 2 NO 2 production and loss differs from the often used steady-state assumption of RO 2 NO 2 , and has considerable impacts on O 3 production (Sect.3.2).RONO 2 is formed from minor channels in NO + RO 2 reactions, and the importance of these channels is known to be a function of the size of RO 2 .RONO 2 has longer lifetimes (at least 2 days) than RO 2 NO 2 , and its loss by transport and deposition is a net loss of RO x radicals.Another feature of the chemical system in Fig. 2 is the coupling of NO x and VOCs chemistry and the comparable importance of their roles as RO x sources and sinks.Both NO x and VOCs affect major RO x primary sources, i.e. the source OH from excess HONO (2.2 ppbv h −1 ) and photolysis of OVOCs (4.2 ppbv h −1 ).Both of them affect RO x sinks through organic nitrates.This feature of chemistry could have implications for O 3 sensitivities to NO x and VOCs (Farmer et al., 2011).In Section 3.2 -3.3, we examine the formation of O 3 , and its sensitivities to various factors, including excess HONO, aromatics, γ (HO 2 ), as well as NO x and VOCs.

O 3 production and loss rates
The formation of O 3 in the troposphere is via the reactions of NO and peroxy radicals.On the other hand, due to the fast cycling of both O 3 and NO 2 under urban conditions, O 3 loss is due to a number of reactions leading to the destruction of O 3 and NO 2 .The daytime average P(O 3 ) is the sum of HO 2 + NO (19.8 ppbv h −1 ) and RO 2 + NO (12.2 ppbv h −1 ) at 32 ppbv h −1 (Fig. 2), comparable to previous calculations for Beijing during CAREBeijing-2006(Lu et al., 2012), and is near the top of the existing reported values for urban environments (e.g.Ren et al., 2003;Shirley et al., 2006;Kanaya et al., 2008;Wood et al., 2009).The reaction of HO 2 + NO accounts for roughly 2/3 of P(O 3 ).RCO 3 + NO, CH 3 O 2 + NO and AROp + NO are the predominant RO 2 + NO reactions, due to the relative abundance of RO 2 radicals (Fig. 1) and the large reaction rate constant of RCO 3 + NO.The mean daytime peak of P(O 3 ) is ∼60 ppbv h −1 , occurring around 11:00 (Fig. 5), earlier than the peaks of both HO 2 and RO 2 around 13:00 because of the decreased NO concentrations from morning to early afternoon.P(O 3 ) is also found to peak around 10:00-11:00 local time in Mexico City (Shirley et al., 2006).The corresponding PBL averaged O 3 production and net formation rates are shown in Fig. S1 of the Supplement.The rates are about a factor of 4 lower reflecting the simulated decrease of theese rates with altitude (Fig. S2 in the Supplement).
The daytime mean and maximum L(O 3 ) rates are 6.2 ppbv h −1 and 12 ppbv h −1 , respectively, roughly 1/5 of P(O 3 ).L(O 3 ) consists of NO 2 → HONO (2.2 ppbv h −1 ), NO 2 + OH (1.7 ppbv h −1 ), RO 2 + NO 2 (1.1 ppbv h −1 ), O 1 D + H 2 O (0.4 ppbv h −1 ), and other minor reactions.Given the noontime O 3 and NO 2 concentrations (∼55 ppbv and ∼10 ppbv) and the loss rate of O 3 , the chemical lifetime of O 3 is ∼5 h.It is interesting that the unknown source of HONO (assuming NO 2 to be the precursor) also serves as a L(O 3 ) term, and is in fact the largest L(O 3 ) reaction (∼40 %), directly affecting O 3 formation and the lifetime of O 3 .Such a potential O 3 loss process associated with HONO is worth further investigation.The average daytime net formation rate of O 3 , i.e.P(O 3 ) -L(O 3 ), is ∼26 ppbv h −1 .
Rapid surface O 3 formation at a similar rate in Beijing and its vicinity were also reported previously (Lou et al., 2010).However, it should be noted that the O 3 formation rate near the surface is one of many factors affecting surface O 3 concentrations (e.g., Shirley et al., 2006).Inspection of the model results shows a decrease of O 3 production rates with altitude (Fig. S2 in the Supplement) due largely to a corresponding decrease in NO concentrations.When averaged over the whole PBL, the daytime average gross and net O 3 production rates of 6.7 ppbv h −1 and 5.0 ppbv h −1 , respectively, are much slower than those near the surface (Figs.S1 and S3 in the Supplement).Our results point to the importance of obtaining the vertical profiles of O 3 precursors in the PBL in order to better constrain model simulations and improve the understanding of factors controlling O 3 concentrations near the surface.While the vertical turbulent transport and surface dry deposition are taken into account in the 1-D model, horizontal advection also plays an important role.Although the 1-D model cannot be used in predicting surface O 3 concentrations, the increase of O 3 concentrations in daytime due to the net production in the PBL is reflected in the surface O 3 observations (Liu et al., 2010).

O 3 production efficiency
Alongside O 3 formation, NO x is transformed into oxidized nitrogen compounds NO z (NO z ≡ NO y -NO x ), e.g.RONO 2 , RO 2 NO 2 , and HNO 3 , and then eventually removed from the atmosphere by deposition.NO z compounds at daytime account for 20-50 % of total NO y (Fig. 6).The O 3 production efficiency (OPE) of NO x is defined as the amount of O 3  produced during the lifetime of NO x (Liu et al., 1987).Based on our model calculated P(O 3 ) and P(NOz) (Fig. 2), we estimate a daytime average OPE to be 9.7, much larger than that estimated by Wang et al. (2010) for the summer of 2008, and yet within the estimates by Chou et al. (2011) for the summer of 2006.It is also within the estimated range for Mexico City (4-12) (Lei et al., 2007;Wood et al., 2009).Considering the moderate concentrations of HO 2 and RO 2 compared to other urban environments, the relatively high OPE from our calculation is mainly due to the high daytime NO concentration (∼5 ppbv at noontime).When averaged over the whole PBL, the OPE value of 11.7 is higher than estimated near the surface.

Sensitivity studies -assessing the impacts of HONO, aromatics and aerosol uptake of HO 2
Based on the results from the standard model (S0) results discussed above, we found that excess HONO, reactive aromatic VOCs, and aerosol uptake of HO 2 are important factors in the photochemical system in Beijing.In the next section, in order to further address the uncertainties in these factors, we extend our analyses of these individual factors by comparing results from a series of sensitivity simulations listed in Table 1.

Impacts of excess HONO on RO x budgets and O 3 formation
The large net OH source from the photolysis of excess HONO relative to other primary OH sources has been shown in Fig. 2. Figure 7 shows the sensitivity simulation results without excess HONO (S1).The standard model (S0) has ∼60 % higher daytime average OH concentration than S1 due to the excess HONO.Increased OH leads to more active photochemistry and thus ∼50 % increases of HO 2 and RO 2 concentrations, as well as P(O 3 ).A similar amplifying effect by HONO on P(O 3 ) was also noted by Lou et al. (2010).A second consequence is the large sink (∼40 %) of O 3 due to excess HONO production (Section 3.2).The impact of excessive HONO is even larger (∼130 %) when aromatics are not included (comparing Figs.S3 to S2 in Fig. 7).This excess HONO term is usually not considered in the budget of O 3 in previous studies.For locations like Beijing, it appears necessary to take into account the production and loss of O 3 due to excess HONO.More importantly, the nature of excess HONO is currently unknown and needs to be considered as a major source of uncertainty for understanding O 3 .We note that the daytime HONO concentrations measured in this and other studies over China, i.e. roughly 1 ppbv on average during daytime, and the inferred excess HONO formation rates (e.g.Su et al., 2008Su et al., , 2011;;Hofzumahaus et al., 2009;and this work), are substantially higher than those found elsewhere.Although known interference was tested and removed (the Supplement), unknown interference was possible (e.g.Pinto et al., 2010).A well-designed intercomparison with other instruments (such as a long-path differential optical absorption spectroscopy (DOAS) instrument) in Beijing would be necessary.The detailed photochemical analysis presented here by comparing the relative importance of the primary HO x sources (e.g., Fig. 2) would provide a quantitative assessment on the impact of a potential HONO interference on the O 3 and radical photochemistry if the interference were quantified.We also note that the excess HONO source was only included in the surface layer of the model since the nature of the heterogeneous reaction is not well characterized.Extending this source into the PBL would further increase its impact.

Direct and secondary impacts from aromatics
Aromatics are the most reactive and abundant VOC group measured in Beijing (Liu et al., 2010).The direct impact of aromatics on radical budgets and O 3 formation is via contributing first generation RO 2 (AROp) upon oxidation by OH; and we refer the effect due to subsequent oxidation products as secondary.Comparing S0 with S2 (Fig. 7), adding aromatics leads to a factor of 2 increase of HO 2 , RO 2 , and P(O 3 ).These changes obviously could not be explained solely by the addition of AROp radicals (Fig. 5).Inspection of the model results shows that OVOCs concentrations increase drastically after adding aromatics (e.g.∼100 % increase of formaldehyde; ∼65 % increase of acetaldehyde, a factor of 5 increase of methylglyoxal and a factor of 10 increase of glyoxal (Table S3 in the supplement)), and their photolysis further produces substantial amounts of primary RO 2 and HO 2 .More significantly, the presence of aromatics in S0 increases OH by ∼30 % compared to S2 despite of the loss of OH by reacting with aromatics.Therefore, the overall increase of primary RO x production from the secondary impact by aromatics is large enough to compensate for the shift from OH to peroxy radicals in the RO x family.If the reactions of excess HONO are not included, the impacts by aromatics (from Figs.S3 to S1) are even more drastic, leading to more than 100 % increase of HO 2 , RO 2 and P(O 3 ), and 50 % increase of OH.We note that the finding on the significance of aromatic VOCs on photochemistry is qualitatively robust.However, the quantitative results presented here depend on the chemical mechanism used for aromatic VOC oxidation, for example, the yields of dicarbonyls, which are uncertain (e.g.Carter, 2009).In situ measurements of OVOC species, especially those dicarbonyls, such as methylglyoxal and glyoxal, will be needed to further constrain the model.When integrated in the PBL, the effect of aromatics is smaller (∼25 % by comparing case S0 to S2 in Fig. S3 of the Supplement).

Aerosol uptake of HO 2
Figure 8 shows the variations of daytime average HO 2 , OH concentrations and P(O 3 ) rates as a function of γ (HO 2 ) value.HO 2 concentration drops by www.atmos-chem-phys.net/12/7737/2012/Atmos.Chem.Phys., 12, 7737-7752, 2012 60 % from 4.05×10 8 molecules cm −3 at γ (HO 2 ) = 0 to 1.65×10 8 molecules cm −3 at γ (HO 2 ) = 0.2.Correspondingly, P(O 3 ) decreases by ∼50 % from 35.4 ppbv h −1 to 23.3 ppbv h −1 , and OH drops by 30 % from 5.26×10 6 molecules cm −3 to 3.65×10 6 molecules cm −3 .P(O 3 ) and OH changes are not as large as HO 2 in part because the impact of γ (HO 2 ) on RO 2 radicals is indirect and not as large.Figure 8 suggests that γ (HO 2 ) is a large source of uncertainty in current HO x simulation studies over polluted regions of China, where aerosol loading is high (Kanaya et al., 2009).Recently, Taketani et al. (2012) measured ambient γ (HO 2 ) in the range of 0.09 ∼ 0.4 over Mt.Mang (40 km north of Beijing), and in the range of 0.13 ∼ 0.34 at a remote site over Mt.Tai in China.These γ (HO 2 ) values for ambient aerosols are considerably larger than those estimated by Thornton et al. (2008) or the experimental values for non-metal aerosols compiled by Mao et al. (2012).Additional independent measurements are needed to confirm these high γ (HO 2 ) values over China.

Chemical regimes of O 3 production
We diagnose the P(O 3 ) chemical regimes in Beijing using two approaches.First, we examine the sensitivity of P(O 3 ) to perturbations of NO x and VOC concentrations.We also try to use previously proposed diagnostic equations (e.g.Sillman et al., 1990;Kleinman et al., 1997;Daum et al., 2000) for NO x -limited and VOC-limited regimes, as has been done in previous studies (Lei et al., 2007).The chemical environment in Beijing is strongly affected by various factors, e.g.excess HONO, aromatics, and aerosol uptake of HO 2 .These factors are not present in the US where those previous theoretical studies (e.g.Sillman et al., 1990;Kleinman et al., 1997) were conducted.We analyze each scenario listed in Table 1 and then discuss the possible impacts from those factors on P(O 3 ) chemical regimes in Beijing.

Sensitivity simulation results
In the sensitivity analyses (Table 1), we vary NO x and VOCs concentrations (110 %, 90 %, 70 %, and 50 % of the observed values) and examine the change of P(O 3 ), i.e.P(O 3 ).While it is desirable to study the direct change of O 3 due to precursor changes, a box or 1-D model cannot adequately simulate O 3 concentrations due to the large impact of advection as noted previously.We therefore diagnose the change of P(O 3 ).The sensitivity results are numerically accurate only when the change of the model state is small.We included large changes (50 % and 70 % reductions) in order to infer the impact of precursor changes on O 3 photochemical production regime.These results represent a linear extrapolation of the sensitivities in the photochemical system.Figure 9 shows the sensitivity experiment results.P(O 3 ) consistently shows positive responses to NO x , i.e. an increase of NO x leads to an increase of P(O 3 ), although the former is always larger than the latter, i.e., the P(O 3 ) − NO x lines are to the left of the 1:1 line.P(O 3 ) is largely determined by the product of NO and HO 2 (RO 2 ) concentrations (Fig. 9a).The non-linear dependence of P(O 3 ) on NO x is a reflection of the dependence of HO 2 and RO 2 on NO x .For example, increasing NO x leads to decreased peroxy radicals due in part to the conversion of peroxy radicals to OH and RO by reacting with NO.The degree of peroxy radical decrease is also a function of the change in primary RO x sources and sinks.Comparing the scenarios without aromatics and HONO (S3 and S3a) with those with both or either one of them (S0, S1 or S2), P(O 3 ) in the former scenarios (S3 and S3a) is much less sensitive to NO x (e.g., the flat shapes of the orange lines in Fig. 9a).The larger sensitivity of peroxy radicals to NO x change in S3 and S3a is because of a much smaller primary RO x source without excess HONO, aromatics or both (Sect.3.3).Similarly, inspection of the difference between S0 and S0a or among S3, S3a and S3b shows that a larger HO 2 aerosol sink leads to a lesser sensitivity of peroxy radicals to NO x and hence a higher sensitivity of P(O 3 ) to NO x .
The complexity of the P(O 3 )-NO x sensitivity also results in part from the change over the course of a day (Fig. 10).Generally, P(O 3 ) shows a larger sensitivity to NO x in the afternoon than in the morning.Under different scenarios, such as S0 and S3a in Fig. 10, P(O 3 )-NO x sensitivities show different transition patterns over the course of daytime.This daytime transition of P(O 3 )-NO x sensitivity is due to the fast decreasing NO x from morning towards afternoon and thus reducing the importance of NO x in sequestering radicals, while the primary RO x source increases into the afternoon.
In contrast to the largely varying degrees of sensitivities of P(O 3 ) to NO x , the sensitivity of PO 3 to VOCs is much more uniform and closer to the 1:1 linear response (Fig. 9b).The largest deviation from the 1:1 response line is the simulation without aromatics (S2) due to the large impact of aromatics to primary RO x sources (Sect.3.3.2). Figure 9a and 9b show that the P(O 3 )-VOC response resembles the VOClimited chemical regime (Sillman et al., 1990), although the P(O 3 )-NO x response does not, suggesting that photochemical O 3 production in Beijing is neither NO x -limited nor VOC-limited, but lies in a transition regime where reduction of either can reduce P(O 3 ).
Concurrent reduction of NO x and VOC concentrations leads to greater P(O 3 ) reduction than reducing either (Fig. 9c), although the additional reduction from VOCreduction only scenarios (Fig. 9d) varies.In agreement with the P(O 3 )-NO x sensitivity (Fig. 9a), the least change from the VOC-only scenarios is from the simulations S3 and S3a in which neither excess HONO nor aromatics is included.
, Q, and NO during the daytime (06:00-18:00) and afteroon (12:00-18:00).PBL shows similar dependence to the change of NO x and VOCs as in Fig. 9, although the column P(O 3 )-VOC sensitivity is weaker than near the surface and is similar to column P(O 3 )-NO x sensitivity.Inspection of model results suggests that the O 3 production chemical regime increasingly shifts toward being NO x -limited away from the surface (Fig. S4 in the Supplement).The co-benefit of concurrent reduction of both NO x and VOCs in reducing O 3 production is larger when integrated in the PBL column than near the surface (by comparing Fig. S4 of the Supplement with Figure 9).

Evaluation with diagnostic equations of O 3 production for different chemical regimes
Various studies have provided relatively simple diagnostics for O 3 production regimes (e.g.Sillman et al., 1990;Kleinman et al., 1997;Daum et al., 2000).Lei et al. (2007) summarized these studies into two equations: where k t is the weighted average rate constant for reaction of HO 2 and RO 2 with NO; k eff is the effective rate constant for peroxide (H 2 O 2 and ROOH) formation; Q is the total primary source of RO x radicals, in ppbv h −1 ; L N , L R and L ON are the radical loss rates due to the reactions of OH + HO 2 , RO 2 + R'O 2 , and radical-NO x reactions excluding OH + NO 2 , respectively; Y is the average yield of HO 2 and RO 2 for each OH + VOC reaction; L OH−VOC and L OH−NO 2 are the loss rates of OH due to reactions with VOCs and NO 2 , respectively; PER is the peroxide formation rate.We compare the correlations between model calculated hourly P(O 3 ) (ppbv h −1 ) and those from the diagnostic equations (Lei et al., 2007).The results for different model sensi-tivity simulations are shown in Table 2.We also show in Table 2 the correlations with NO and the primary RO x source Q.In the standard model (S0), P(O 3 ) shows better correlation with ) (afternoon R 2 = 0.50; daytime R 2 = 0.19), mainly due to the much better P(O 3 )-Q correlation (afternoon R 2 = 0.66; daytime R 2 = 0.74) than the P(O 3 )-NO correlation (afternoon R 2 = 0.22; daytime R 2 = 0.002).These suggest that P(O 3 ) during our observations in Beijing behave more like in the VOC-limited regime than the NO xlimited regime.This is particularly true when morning data are taken into account since O 3 production can clearly reside in the VOC-limited regime (Fig. 10).It is interesting to note that the VOC-limited regime in the morning disappears when the sensitivity is integrated in the PBL (Fig. S5 in the Supplement).In the afternoon when O 3 production is large, however, both diagnostics show reasonably good correlations with the P(O 3 ).An outlier is the scenario of S2 when aromatics are not included; the chemical regime clearly shifts into the VOC-limited regime given the much better correlation (R 2 = 0.79) with In the scenario of S0a (without HO 2 aerosol uptake) for the afternoon, P(O 3 ) correlate even better with the NO x -limited diagnostics (R 2 = 0.56) than with the VOC-limited diagnostics (R 2 = 0.49).Comparing the results of S0, S0a, and S0b, aerosol uptake of HO 2 tends to shift the O 3 production more towards the VOClimited regime.
In general, the diagnostic equations are consistent with our sensitivity simulations, suggesting that under the most realistic scenario (S0), O 3 production in Beijing is in the transition regime.Aromatics and excess HONO tend to shift O 3 production into NO x -limited regime, while aerosol HO 2 sink tends to shift it towards VOC-limited regime.

Conclusions
In this work, summertime photochemistry at Beijing is investigated with a 1-D photochemical model constrained by the observations near the surface.Through detailed chemical budget analysis, we find that VOC (especially aromatics) oxidation and heterogeneous chemistry (including HONO formation and HO 2 uptake by aerosols) play potentially critical roles in RO x budgets and O 3 formation.A series of modeling experiments are conducted to explore the model sensitivities to these processes, particularly in light of the uncertainties in the understanding of heterogeneous processes.Although the quantitative results presented are somewhat dependent on the parameters we chose for the heterogeneous reactions, the detailed diagnostics of radical sources and sinks and model sensitivity results can be used to estimate the significance of these processes when better information is available or alternative assumptions are made in future studies.Sensitivity experiments are also done for determining the response of P(O 3 ) to the change of NO x and VOCs in order to understand the O 3 photochemical regimes near the surface.
Through a detailed chemical budget analysis, we find that summertime photochemistry in Beijing is characterized by fast formation, recycling and removal of RO x radicals.The total RO x primary source (and sink) (6.6 ppbv h −1 ) near the surface in Beijing is close to the largest values reported for urban environments.Photolysis of OVOCs (4.2 ppbv h −1 ) and excess HONO (2.2 ppbv h −1 ) are the two largest RO x sources, much more important than that from O 1 D+H 2 O (0.4 ppbv h −1 ).Formation of RO 2 NO 2 (1.0 ppbv h −1 ) and RONO 2 (0.7 ppbv h −1 ) are as important as the typical major RO x sink via OH + NO 2 reaction (1.6 ppbv h −1 ).Aromatics are the major player in OVOC and organic nitrate formation.Aerosol uptake of HO 2 may also be a major RO x sink due to the large aerosol surface area in Beijing, and this sink is quite sensitive to the value of γ (HO 2 ).The importance of aromatics, heterogeneous HONO production, and possibly large aerosol update of HO 2 signifies the unique photochemical environments in Beijing.These characteristics are likely to be present over many regions of the polluted eastern China with large clusters of cities and industrial regions.Observation-based modeling studies of RO x radical chemistry over these regions that include high quality comprehensive measurements of RO x radicals, HONO, organic nitrates, and VOCs (aromatics in particular) and their oxidation products will be necessary to reduce the uncertainties of the factors discussed in this study and develop more accurate mechanistic and quantitative understanding of the photochemical system.
The chemical production of O 3 near the surface in Beijing is extremely fast, at 32 ppbv h −1 on average during daytime.The high concentrations of NO (∼5 ppbv at noontime), excess HONO and aromatic VOCs are the major driving factors.The O 3 production rate decreases rapidly away from the surface in the PBL, leading to a factor of 4 lower gross and net O 3 production rates in the PBL compared to the surface.The chemical loss of O 3 near the surface is also fast, about 6 ppbv h −1 , and the heterogeneous formation of excess HONO via NO 2 →HONO is actually the largest (∼40 %) O 3 loss term.Sensitivity simulations and analysis using diagnostic equations suggest that the O 3 production in Beijing does not lie in either the VOC-limited regime or in the NO xlimited regime, but in the transition regime, where reduction of either NO x or VOCs could lead to reduced O 3 production.The transition regime feature is even more pronounced than near the surface when the sensitivity to NO x or VOCs of column P(O 3 ) rate integrated from the surface to the top of the PBL is considered.It implies that there is flexibility in choosing either NO x or VOC reduction to achieve the most cost-effective O 3 reduction.The co-benefit of concurrent reduction of both NO x and VOCs is small based on modeling analysis of near-surface observations.However, it becomes significant when integrated in the PBL column.
In this study, we have focused on understanding the photochemistry using observation-constrained modeling.Our results point to chemical characteristics not yet well represented in current 3-D modeling studies.The large clustering of concentrated city and industrial regions in the eastern China, such as North China Plain (NCP), Yangtze River Delta (YRD), and PRD, suggests that fast photochemistry plays a critically important role in determining O 3 levels in these regions.3-D modeling analysis, ideally constrained by in situ or remote sensing observations, will be necessary to understand the interplay of chemistry and transport on the regional and global scales.

Fig. 1 .
Fig. 1.Average diurnal profiles of OH, HO 2 and RO 2 (black lines) in the standard model (S0).The vertical bars show the hourly standard deviations.The color lines in the rightmost panel show the major components of RO 2 , which are described in the text.

Fig. 2 .
Fig. 2. Daytime (06:00-18:00) average budgets of RO x radicals.Primary RO x sources and sinks are in red and blue, respectively.The production and loss rates are in ppbv h −1 .

Fig. 3 .
Fig. 3. Average diurnal profiles of major RO x primary sources and sinks.

Fig. 4 .
Fig. 4. Average diurnal profiles of net formation rates of RO 2 NO 2 and HNO 3 .Production and loss rates of RO 2 NO 2 are also shown.

Fig. 9 .
Fig. 9. Changes of O 3 production ( P(O 3 )) as a function of NO x , VOCs, and both under different scenarios in Table1.