Limited effect of anthropogenic nitrogen oxides on secondary organic aerosol formation

Globally, secondary organic aerosol (SOA) is mostly formed from emissions of biogenic volatile organic compounds (VOCs) by vegetation, but it can be modified by human activities as demonstrated in recent research. Specifically, nitrogen oxides (NOx =NO+NO2) have been shown to play a critical role in the chemical formation of low volatility compounds. We have updated the SOA scheme in the global NCAR (National Center for Atmospheric Research) Community Atmospheric Model version 4 with chemistry (CAM4-chem) by implementing a 4-product volatility basis set (VBS) scheme, including NOx-dependent SOA yields and aging parameterizations. Small differences are found for the no-aging VBS and 2-product schemes; large increases in SOA production and the SOA-to-OA ratio are found for the aging scheme. The predicted organic aerosol amounts capture both the magnitude and distribution of US surface annual mean measurements from the Interagency Monitoring of Protected Visual Environments (IMPROVE) network by 50 %, and the simulated vertical profiles are within a factor of 2 compared to aerosol mass spectrometer (AMS) measurements from 13 aircraft-based field campaigns across different regions and seasons. We then perform sensitivity experiments to examine how the SOA loading responds to a 50 % reduction in anthropogenic nitric oxide (NO) emissions in different regions. We find limited SOA reductions of 0.9– 5.6, 6.4–12.0 and 0.9–2.8 % for global, southeast US and Amazon NOx perturbations, respectively. The fact that SOA formation is almost unaffected by changes in NOx can be largely attributed to a limited shift in chemical regime, to buffering in chemical pathways (lowand high-NOx pathways, O3 versus NO3-initiated oxidation) and to offsetting tendencies in the biogenic versus anthropogenic SOA responses.


Introduction
Organic aerosols (OAs) account for a substantial fraction of atmospheric fine particulate matter and can have significant impacts on both air quality (Huang et al., 2014;Zhang et al., 2007) and climate (Carslaw et al., 2010).Previous research suggests that organic compounds make up between 10 and 90 % of the total aerosol mass at continental midlatitudes and in tropical forests (Andreae and Crutzen, 1997;Kanakidou et al., 2005;Putaud et al., 2010;Seinfeld and Pankow, 2003).Aside from primary organic aerosols (POAs) that are directly emitted into the atmosphere, another major fraction of OA is composed of secondary organic aerosols (SOAs), which are formed through chemical transformation of anthropogenic and biogenic volatile organic compounds (AVOCs and BVOCs).AVOCs include aromatics, alkanes and alkenes of about 25, 44 and 38 TgC year −1 , respec-Y.Zheng et al.: Limited effect of anthropogenic nitrogen oxides tively, from industrial processes, fossil fuel use, biomass burning and road vehicles (Williams and Koppmann, 2007).Isoprene and monoterpenes are the dominant BVOC emissions with estimated global source strengths of about 500 and 150 TgC year −1 , respectively (Guenther et al., 2012).POA can also re-evaporate upon dilution and participate in the chemical oxidation processes leading to the formation of SOA (Robinson et al., 2007).
Biogenic SOA (BSOA) is usually regarded as natural aerosol and as such cannot be addressed by emission control legislation.Recent research implied that anthropogenic compounds facilitate BSOA formation, thus providing the possibility to control BSOA by regulating the emission of other precursor pollutants like AVOCs, POA and nitrogen oxides (Carlton et al., 2010;Emanuelsson et al., 2013;Hoyle et al., 2011;Lin et al., 2013;Rollins et al., 2012;Volkamer et al., 2006).For example, Carlton et al. (2010) have shown that, in the southeast US, up to 50 % of the total BSOA surface atmospheric loading is attributed to controllable pollution emissions.Spracklen et al. (2011) found that at the global scale the model with a large human-interfered SOA source was the most consistent with observations, which includes a maximum of 10 % SOA (10 Tg year −1 ) from fossil sources, and the extra is mostly likely due to an anthropogenic pollution enhancement of BSOA.The potential impacts of human activities are visible in every step of BSOA formation: the amount of naturally emitted BVOCs through land use and land cover change, the oxidative transformation of BVOCs to semivolatiles through altering atmospheric oxidant concentrations, and the partitioning behavior to the aerosol phase through modifying the load and miscibility of pre-existing organic aerosol (Hoyle et al., 2011).
Among the multiple human-induced influences, nitrogen oxides (NO x = NO + NO 2 , emitted from many fossil-fuel driven activity sectors) play a critical role in SOA formation through several aspects.First, through the competitive chemistry of organo-peroxy radicals (RO 2 ) formed from oxidation of AVOC and BVOC precursors, which can react mainly with NO at high NO x or hydroperoxyl (HO 2 ) and peroxy radicals (RO 2 ) at low NO x conditions (Kroll and Seinfeld, 2008;Ziemann and Atkinson, 2012).Calculating the SOA yield dependence on NO x is challenging because the OH / O 3 ratio depends on the VOC / NO x ratio (Presto et al., 2005).Lane et al. (2008) suggested that SOA yields could be calculated by a linear combination of the "pure" mass yields scaled by the strength of each branch.In many SOA models (e.g., Heald et al., 2008;Lane et al., 2008;Pye et al., 2010), the representative reactions for each branch are low − NO x condition : RO 2 + HO 2 → ROOH, high − NO x condition : RO 2 + NO → RONO 2 .
For AVOCs like light aromatics (Ng et al., 2007) and BVOCs like isoprene (Kroll et al., 2006) and monoterpenes (Presto et al., 2005), both the ROOH groups and the RONO 2 groups can be low in volatility thus facilitating SOA formation, but RONO 2 is not the dominant product of the RO 2 +NO channel; therefore, the high-NO x pathway usually has lower yields of SOA.Second, NO x can influence SOA formation through nighttime nitrate radical (NO 3 ) chemistry.This pathway has a distinctive chemical signature due to the high yields of organic nitrate (RONO 2 ), which also forms during daytime photooxidation in the presence of NO but with a lower yield.The importance of NO 3 -initiated SOA formation has been confirmed by chamber experiments (Griffin et al., 1999;Ng et al., 2008) and field studies, e.g., in Bakersfield, California, NO 3 chemistry contributed approximately a third of the nighttime increase in total OA (Rollins et al., 2012).Finally, NO x levels can impact the atmospheric oxidation capacity.In the NO x -limited regime (in terms of O 3 formation), the OH-initiated oxidation of CO, methane (CH 4 ) and other VOCs in the presence of NO x produces O 3 .Thus, in such conditions, increasing NO x by human activities should, in principle, lead to the increase in atmospheric oxidation capacity (OH and O 3 ) (Seinfeld and Pandis, 2006), and result in higher SOA yields.For example, using a chemical transport model, PMCAMx, Lane et al. (2008) suggested that a 50 % reduction in NO x emissions could decrease predicted ground-level BSOA by an average of 0.5 µg m −3 in the eastern US by lessening the atmospheric oxidant levels.
Due to the multiple impacts of NO x on SOA formation, it is important to understand how NO x emission controls alter the particulate matter atmospheric loading.The goal of this study is to improve the SOA scheme in a global climatechemistry model by incorporating a 4-product volatility basis set (VBS) framework (Pye et al., 2010), which has four representative volatility bins to better represent the volatility distribution of all semivolatiles in the atmosphere than the default 2-product scheme (Heald et al., 2008;Odum et al., 1996).The model is then used to investigate the impacts of anthropogenic NO x emission reduction on SOA formation.Section 2 describes the observational data sets used in this study.In Sect.3, we describe the default and updated SOA parameterizations embedded within the global chemistryclimate model framework.We perform control simulations using six different model configurations, including the default 2-product scheme and the updated SOA scheme with and without NO x -dependent yields for monoterpene, and with and without simplified SOA aging parameterizations.Section 4 shows the results.The control simulations are evaluated and assessed against several observational data sets.Then, we perform sensitivity simulations to probe the impacts of a global 50 % anthropogenic NO emission reduction on SOA production.We conduct this experiment as a simplified potential future scenario based on the 50 % NO x emission reduction from power plants in the southeast US by pollution control programs in the past decade (Frost et al., 2006;Kim et al., 2006).Section 5 summarizes the findings of this study.

Terminology and data sets
Table 1 summarizes major abbreviations used in this study.The term OA refers to the total particle-phase organic matter including carbon, hydrogen, oxygen and other possible elements.The term OC refers to only the mass of carbon in these organic compounds.Both OA and OC are used based on different measurement techniques.Similarly, primary organic carbon (POC) is the carbon mass in POA; secondary organic carbon (SOC) is the carbon mass in SOA.In this study the term SOA (secondary organic aerosol) and SOG (secondary organic gas) refer to particle phase and gas phase, respectively.

IMPROVE OC measurements
The US total OC data set is from the Interagency Monitoring of Protected Visual Environments (IMPROVE, Hand et al., 2011).IMPROVE OC is collected using quartz fiber filters for 24 h every third day, analyzed offline by thermal optical reflectance (TOR) (Chow et al., 1993), and corrected for an approximate positive artifact (Dillner et al., 2009).Assumptions made in this correction may not always be appropriate (Watson et al., 2009), and the potential negative artifacts due to the volatilization of particulate organics are not accounted for.We choose 120 surface sites from the IMPROVE network that are within the bottom layer in corresponding model grids.The original 3-day data from 2005 to 2009 have been averaged to seasonal and annual mean values.OC concentrations from sites within the same model grid cell (1.9 • × 2.5 • latitude by longitude) are averaged for comparison to modeled OC concentrations in corresponding model grid cells.

Aircraft-based OA measurements from an aerosol mass spectrometer (AMS)
The OA data sets come from 13 aircraft field campaigns that took place between 2005 and 2009 (Heald et al., 2011).In these campaigns, total OA density was measured using an AMS in standard temperature and pressure conditions (STP: 298 K, 1 atm) and provides fast online submicron aerosol composition (Canagaratna et al., 2007).For each field campaign, the 1 min raw data are averaged temporally and horizontally along the flight track for comparison to the simulated monthly mean OA vertical profile in the corresponding month and location in the model.Each observed OA profile is further averaged vertically to a single value for comparison to the simulated OA concentration averaged over the same range of altitudes.

Surface OA/OOA/HOA measurements from AMS
We select 42 surface AMS measurements in 2000-2008 from previous studies (Spracklen et al., 2011;Zhang et al., 2007) that differentiate between hydrocarbon-like OA (HOA, a surrogate for POA from combustion and biomass burning) and oxygenated OA (OOA, a surrogate for SOA from all sources).The HOA and OOA are determined by a multiple component analysis (MCA; Zhang et al., 2007).The averaged OOA, HOA and OA data for each campaign have been compared to the simulated monthly mean SOA, POA and total OA in the corresponding model grid.Most of these measurements were taken before 2005.We did not perform simulations in this period due to the lack of GEOS-5 meteorological data (described in Sect.(Guenther et al., 2012).The anthropogenic, biomass burning and other (except biogenic) emissions in CESM are as described in Lamarque et al. (2012).These consist of anthropogenic emissions from the Precursors of Ozone and their Effects in the Troposphere (POET) inventory for 2000 (Granier et al., 2005), with Asia replaced by the Regional Emission inventory for ASia (REAS v1) for each year (Ohara et al., 2007).The biomass burning emissions are from GFED-v2 (van der Werf et al., 2006) for 2005-2008 and from the Fire INventory of NCAR (FINN-v1) for 2009 (Wiedinmyer et al., 2011).All the SOA schemes discussed in this study consider that SOAs are only generated from oxidization of gas-phase VOCs.The SOA formation from organic compounds emitted originally in the condensed phase is not considered.Simulations are performed with a 30 min time step, a horizontal resolution of 1.9 • × 2.5 • and 56 levels from the surface to approximately 40 km.

Default SOA parameterization
In CAM4-chem, the default SOA formation follows the 2product approach (Odum et al., 1996).Each parent VOC is oxidized to generate 2 semivolatile surrogates, which can partition into pre-existing organic particles including both POA and SOA.The partitioning of the semivolatile products is described by absorptive partitioning theory into carbonaceous aerosol material (Donahue et al., 2006;Pankow, 1994).CAM-chem tracks POC in its emission, transport and deposition module.Later, in Sect.4.2, we assume a POA-to-POC ratio of 1.4 (Aiken et al., 2008;White and Roberts, 1977) to calculate POA and OA to compare with observations.The model simulates anthropogenic SOA (ASOA) from NO xdependent OH-initiated oxidation of anthropogenic aromatics (benzene, toluene and xylene), BSOA from the OHinitiated oxidation of isoprene, and the ozonolysis, OH-and NO 3 -initiated oxidation of monoterpene (Table 2).The surrogate SOA products are assumed to be C 10 H 16 O 4 for SOA from monoterpenes (SOAM), C 5 H 12 O 4 for SOA from isoprene (SOAI), and C 6 H 7 O 3 , C 7 H 9 O 3 , and C 8 H 11 O 3 for SOA from benzene, toluene and xylenes; therefore, the O : C ratio is constant for each SOA species.The overall O : C ratio in total OA depends on the split between POA and SOA, and the fraction of each SOA species.Fossil content is regarded as POA including both hydrophobic and hydrophilic compounds and is not included in SOA in CAM4-chem.The default 2-product model in CAM4-chem only applies low-NO x yields parameterization for all OH-and O 3 -initiated BSOA formation.The SOA mass yields (summarized in Table S1) are from Heald et al. (2008) and references therein.

Updated SOA scheme
We update the SOA model to include a 4-product VBS scheme, which has 4 semivolatile surrogates for each parent VOC species.The saturation concentrations (C*) at 295 K for the 4 product groups are 0.1, 1, 10 and 100 µg m −3 , respectively.This VBS has a wider range of volatilities than the default 2-product parameterization that can better represent the volatility distribution of atmospheric semivolatiles.Another goal of implementing this VBS framework is to facilitate implementation of advanced processing including the aging effect.Changing the enthalpies of vaporization (see Tables S1 and S2 in the Supplement) has no significant effect on simulated SOA burden (difference smaller than 2 %).In addition to the current reactions used in the 2-product model, we have added the NO x -dependent pathway for SOA formation from monoterpenes and the NO 3 -initiated oxidation of isoprene into the VBS (see Table 2).SOA formed from OH-initiated photooxidation of isoprene still only has one set of yields following the low-NO x parameterizations.We do not change this isoprene-SOA parameterization to remain consistent with the VBS framework from Pye et al. (2010).Additional simulations that include the high-NO x pathway of isoprene chemistry are discussed in the Supplement.The SOA mass yields (summarized in Table S2) are from Pye et al. (2010) and references therein.
In the 4-product VBS model, the partitioning between the high-NO x (RO 2 +NO) and low-NO x (RO 2 +HO 2 ) pathway VBS_agLow Same as VBS Same as VBS, with multi-generational aging applied to all species; k OH = 5.2 × 10 −12 cm 3 molec −1 s −1 .
VBS_agAVOC Same as VBS Same as VBS, with multi-generational aging applied to ASOA only; is determined by the branching ratio β (Pye et al., 2010): .
Thus, 100 × β % of the parent hydrocarbon channels through the high-NO x pathway, and 100 × (1 − β) % of the parent hydrocarbon channels through the low-NO x pathway.This format of β is a simplification of , where k RO 2 +NO and k RO 2 +HO 2 represent the reaction rate coefficients of RO 2 +NO and RO 2 +HO 2 , respectively.Field studies that quantified the elemental composition of OA indicate the importance of aged oxygenated OA (Aiken et al., 2008;Chen et al., 2011;Heald et al., 2010).Several regional modeling studies have found the "aging" process necessary to produce reasonable OA mass (Athanasopoulou et al., 2013;Hodzic and Jimenez, 2011;Knote et al., 2015;Lane et al., 2008;Tsimpidi et al., 2010).In this study we implement a simplified aging parameterization into the global model to provide a rough assessment of the SOA sensitivity in VBS to the effect of aerosol aging.At every model time step, each gas-phase SOA product except for the lowest volatility product (C* = 0.1 µg m −3 ) is assumed to be further oxidized by OH with a reaction rate constant k OH of 4×10 −11 cm 3 molec −1 s −1 (Atkinson and Arey, 2003;Tsimpidi et al., 2010), which reduces its volatility by an order of magnitude.The oxygen-to-carbon ratio (O : C) is assumed to be constant for each surrogate SOA product; thus, an increase in SOA mass due to the addition of oxygen is not considered in the aging process.Considering the complexity of various SOA species and the large uncertainties in aging process, the assumption of a fixed O : C ratio for each SOA product surrogate is acceptable for global model parameterizations.The aging rate k OH = 4 × 10 −11 cm 3 molec −1 s −1 is at the high end of previously suggested parameters (Lane et al., 2008).We do simulations with and without this aging parameterization to quantify the possible range of global SOA strengths and additional simulations (see Sect. 3.4) to examine the effect of different aging parameters.
Particle-phase SOA as well as gas-phase SOG are removed from the atmosphere by wet and dry deposition.Dry deposition follows a resistance-in-series formulation (Heald et al., 2008;Wesely, 1998).SOA and other soluble aerosols are removed by both in-cloud scavenging and below-cloud washout (Barth et al., 2000;Lamarque et al., 2012).

Experiment setup
In this study, we apply six different treatments of SOA formation, as summarized in Table 2. 2-product is the default SOA model; VBS and VBS_agHigh are the updated 4-product VBS schemes with and without the aging effect.VBS_agHigh applies the high aging rate k OH = 4 × 10 −11 cm 3 molec −1 s −1 to all species, thus, presumably providing the higher bound of simulated SOA loadings.These three schemes (2-product, VBS and VBS_agHigh) are the www.atmos-chem-phys.net/15/13487/2015/main SOA schemes that we use to compare with observations (Sect.4.2) and study the sensitivity to NO x perturbations.For each of the three schemes, we perform one control run and one sensitivity run in which anthropogenic NO emissions are reduced by 50 % (Sect.4.3).We perform additional simulations to explore the impact of different aging and NO xdependency parameterizations: VBS_agLow applies a lower k OH of 5.2 × 10 −12 cm 3 molec −1 s −1 (Hu et al., 2013) to all species, which is close to the lower limit suggested by other studies (Hodzic and Jimenez, 2011;Spracklen et al., 2011); VBS_agAVOC applies k OH = 4 × 10 −11 cm 3 molec −1 s −1 to AVOCs only, as suggested by some studies that ASOA ages longer than BSOA (Lee-Taylor et al., 2015).VBS_lowNO x is the same as VBS except that all SOAM is assumed to be formed through the low-NO x (RO 2 +HO 2 ) pathway.This VBS_lowNO x scheme is done to isolate the influence of the NO x -dependent pathway for SOAM formation, which is not considered in the default 2-product settings.All simulations are conducted for the years 2004-2009 with offline meteorology from GEOS-5 reanalysis and specific monthly anthropogenic emissions.The year 2004 result is discarded as spin-up.

Comparison of various SOA schemes
The annual mean zonally averaged SOA concentration is shown in Fig. 1.The tropical maximum in the lower troposphere is due to large year-round Amazonian BVOC emissions coupled with extensive seasonal biomass burning that provides ample pre-existing POA onto which the semivolatiles can condense.The second surface maximum in the Northern Hemisphere mid-latitudes, 30-60 • , is mostly attributed to (1) summertime BVOC emissions from broadleaf deciduous forest in the temperate and boreal zones, especially the southeast US which has very high BVOC emissions in the summer (Guenther et al., 2006); (2) plentiful supply of anthropogenic and biomass burning emitted POA; and (3) large amounts of AVOC emissions from human activities.In most simulations, indicated by the white contour lines in Fig. 1, the BSOA from isoprene and monoterpene oxidation accounts for more than 70 % of the total SOA in most latitudes and altitudes, which actually includes both "naturally formed" and "anthropogenically influenced" BSOA.The rest is ASOA from the oxidation of AVOCs.In the VBS_agAVOC run, ASOA accounts for a larger fraction in Northern Hemisphere mid-latitudes than other simulations ranging from 30 to 50 % because in this scheme the aging process is only applied to ASOA.
Table 3 details the annual global SOA budget in each control experiment.Compared to the default 2-product approach, the VBS scheme predicts a smaller global annual burden of SOA (19 % lower than the 2-product), although the surface concentration is 11 % higher with compensating lower concentrations at higher elevations.Due to the higher yields in the VBS (see Tables S1 and S2), more parent hydrocarbon is consumed near the source location and less is transported to the upper troposphere relative to the 2-product scheme.Their different volatility also contributes to the difference in SOA concentrations.Table 3 suggests a shorter SOA lifetime of 8.9 days in the VBS than the lifetime of 11.4 days in the 2-product scheme due to the larger wetdeposition flux, which is consistent with the higher surface concentration.The SOA global burden in the VBS_lowNO x run is 14 % higher than the VBS.This is consistent with the fact that the high-NO x (RO 2 +NO) channel is less SOAproducing compared to the low-NO x (RO 2 +HO 2 ) channel.The yields of SOAM at 10 µg m −3 under high-and low-NO x are 0.09 and 0.19, respectively (Table S2).
For present climate, the differences in annual burden between the two VBS models without aging effect (VBS and VBS_lowNO x ) and the 2-product model are relatively small (< 20 %), because for most parent hydrocarbon species they are fitted into the same chamber data (see Heald et al., 2008;Pye et al., 2010, and references therein).In contrast, in VBS_agHigh, adding the aging effect accelerates the shift of volatile mass towards lower volatility bins and hence more mass in the particle phase, and results in an overall doubling of the net SOA (particle phase) production, which is important for SOA environmental impacts.We find that the SOA production is sensitive to the assumed OH oxidation rate constant (k OH ) for aging of the semivolatile intermediates.For example, using VBS_agLow scheme with a lower k OH of 5.2 × 10 −12 cm 3 molec −1 s −1 (Hu et al., 2013), the annual mean SOA production rate would be 44.6 ± 2.0 Tg[C] year −1 , in comparison to a production rate of 58.6 ± 2.4 Tg[C] year −1 in the VBS_agHigh scheme with k OH = 4 × 10 −11 cm 3 molec −1 s −1 , and a production rate of 28.6 ± 1.6 Tg[C] year −1 in the VBS scheme without aging parameterization.This single aging parameter represents the multi-generational aging of hundreds of thousands of oxidation intermediate species that are involved in the SOA formation (Lee-Taylor et al., 2015) and is currently not well characterized for individual precursors and chemical environments.In the rest of this study, we will use the three schemes, 2-product, VBS and VBS_agHigh, to compare with observations and explore the NO x -dependent effects.

Comparison with the IMPROVE network OC observations
The IMPROVE surface observations and the model outputs are averaged from 2005 to 2009.Modeled OC concentrations are calculated as the sum of primary carbon (directly emitted and transported in the model) and the carbon contained in each SOA species that is calculated assuming the surro-   2a and  3 show the model-IMPROVE comparison of annual mean surface total OC concentrations using the models 2-product, VBS and VBS_agHigh.The total OC in the 2-product and the VBS models are similar to each other and are close to the IMPROVE OC magnitude.They capture the observed spatial distribution to within 50 % (r 2 = 0.45 and 0.47, respectively).They capture the low OC values in the middle and west inland areas, and high OC values in the southeast US where considerable BVOC is emitted from forest as well as POC and AVOC emitted from economic sectors.In the northeast US and some coastal polluted regions in California, OC is greatly overestimated by the models.Figure 2c indicates large simulated POC concentrations in these regions while IMPROVE total OC (= POC + SOC) is not even as large as the simulated POC concentrations.Therefore, the positive bias of the two no-aging simulations in the northeast US is likely due to an overestimate of POC emissions in the inven-tory or due to the assumption that all POAs are non-volatile once emitted and stay in the particle phase until deposition.The fact that IMPROVE sites are predominantly located in remote clean regions might also contribute to this discrepancy.In Fig. 2b the white contour lines illustrate the annual mean fraction of SOC in total OC.Table 4 summarizes the fractions in each season.In the two no-aging models, the annual mean SOC-to-OC ratio ranges from 20 to 30 % in the northeast US and from 40 to 60 % in the southeast US.Even in summer, the ratio does not exceed 50 % in the northeast US and 70 % in the southeast US, which is lower than the suggested values from Ahmadov et al. (2012) and Shrivastava et al. (2008).The aging experiment VBS_agHigh increases the SOC-to-OC ratio greatly (68 and 81 % in summertime northeast and southeast US) and overestimates OC across the entire US due to large SOA formation from aging, which is consistent with previous studies in that the aging coefficient we apply here (k OH = 4 × 10 −11 cm 3 molec −1 s −1 ) is at the  high end of suggested aging rates.The VBS_agHigh scheme slightly improves the replication of spatial distribution of annual mean OC concentrations (r 2 = 0.53 as compared to 0.45 and 0.47 in the 2-product and VBS schemes) but not in summer (r 2 = 0.13, Table 5).Assuming only ASOA ages, the VBS_agAVOC scheme does not improve the simulated spa-tial distribution (r 2 = 0.48 in annual average and r 2 = 0.18 in summer; Table 5).

Vertical profiles of OA from aircraft-based AMS measurements
To assess the simulated OA vertical profile in these models, we select 13 aircraft campaigns that had available AMS measurements between 2005 and 2009.The comparison of vertical profiles is shown in Fig. 4. The VBS_agHigh scheme provides a higher OA concentration than the other two no-aging simulations.Overall, the intermodel differences are smaller than the model-observation differences.In biomass-burninginfluenced regions, the observed OA profile is usually associated with large variations at elevated altitude, indicating sporadic fire plumes.For example, for the AMMA campaign (west Africa), the aircraft tracked biomass-burning plumes, thus giving several maxima of observed mean OA at multiple altitudes.In this case, the observed median value at each layer is a more reliable value for evaluation of the simulations (Heald et al., 2011)

OA, SOA and POA from surface AMS measurements
The observed OOA is a surrogate for SOA, and HOA is a surrogate for POA in AMS measurements (Aiken et al., 2009;Lanz et al., 2007;Zhang et al., 2005).We use 42 shortterm surface AMS measurements (Spracklen et al., 2011;Zhang et al., 2007) and classify their locations into four groups: North America (17 sites), Europe (12 sites), East Asia (12 sites) and Amazon ( 1 mean model results have been averaged into a climatology to compare to the observations, which may lead to large modelobservation differences.Figure 6 compares the measured and simulated OA, OOA(SOA) and HOA(POA).The comparisons between observations and simulations show large discrepancies (in opposite directions) for primary and secondary species.POA is identical in the three simulations.Consistent with the comparison to the IMPROVE network in Sect.4.2.1, the models overestimate POA in most regions and especially in North America, which will promote condensation of semivolatiles onto pre-existing organic matter, thus forming more SOA.The SOA concentrations in the two no-aging models, 2-product and VBS, are close to observed OOA in North America and lower in other regions.By including the aging, the VBS_agHigh simulation increases SOA concentration, leading to an overestimation in North America and still an underestimation in most other regions.The total OA concentrations in all models exceed observed OA in North America.In Fig. 7 we plot the comparison of SOA(OOA)-to-OA ratios from the observations and simulations.The 2-product and VBS models significantly underestimate the observed OOA-to-OA ratios that range from 0.4 to 1.The VBS_agHigh model makes an improvement but is still lower than the observations due to the large amount of simulated POA.Overall, the intermodel differences are smaller than the model-observation differences.These conclusions are consistent with Sect.4.2.1 and 4.2.2.

The impact of anthropogenic NO x pollution on surface SOA
For each of the 2-product, VBS and VBS_agHigh schemes, we perform a control run and a sensitivity run in which the anthropogenic NO emissions are reduced by 50 % to explore the impact of NO x pollution on surface SOA concentrations.Other NO sources including biomass burning and soil emissions are not changed.The 50 % reduction in anthropogenic NO emissions leads to a 36 % decrease in annual mean total NO emissions and a 38 % decrease in surface NO x concentrations at global scale (Fig. S1 in the Supplement).The global surface level of oxidants OH, O 3 and NO 3 decrease by 13, 8 and 29 %, respectively (Fig. S2).The surface NO : HO 2 ratio has been greatly reduced by 67 %, while the change in branching ratio (β = NO NO+HO 2 ) is small (−3.4 %), indicating the NO concentration in the model is too high for HO 2 to compete.The spatial distribution and probability density function of β are plotted in Figs.S3 and  Figure 5.Comparison between averaged OA concentration (µg m −3 ) from 13 AMS field campaigns and the three model simulations: 2-product, VBS and VBS_agHigh.The campaign information is summarized in Heald et al. (2011) (Fig. 1 and Table 1 therein).All data in each campaign are temporally, horizontally and vertically averaged to a single value and compared to the model outputs averaged over the same period and location.S4 in the Supplement.We choose a polluted and a clean region as examples: the southeast US (32-40 • N, 95-77 • W) and the Amazon (17 • S-5 • N, 77-55 • W), both of which are mostly in the NO x -limited regime in terms of ozone formation due to their large BVOC emissions (Lane et al., 2008), i.e., the concentration of O 3 and OH are positively related to concentration of NO x .We examine the dependence of annual mean surface SOA concentrations on β and oxidants level at the global scale and over the southeast US and Amazon regions in Fig. 8.The comparison of the sensitivity (red) and the control runs (green) indicates that the 50 % reduction in anthropogenic NO emissions leads to a small decrease in β, oxidation level and SOA concentrations.In Fig. 8, the small SOA concentrations associated with small β values (β < 0.6) mostly happen over the ocean (not shown) or polar regions where VOC precursors hardly exist and NO x concentrations are low.In the range 0.6 < β < 1.0, the common regime over land, the highest SOA concentrations occur at relatively lower β values, which are mostly located in tropical rain forests with large BVOC emissions and high SOA production efficiency through the low-NO x pathway.The influences of β and oxidant level are tightly related because high β indicating high NO x is usually associated with high concentrations of oxidants.The dependence of SOA on oxidant concentration indicates a maximum at medium oxidant level of approximately 0.8 × 10 12 molec cm −3 .The low SOA concentrations at high oxidant level mostly occur in polluted regions where SOA production is overwhelmingly dominated by the high-NO x (low-yields) pathway.
The SOA production in response to NO x perturbations is complex, as described in Sect. 1.For example, in the VBS_agHigh scheme, we consider monoterpene SOA (SOAM) coming from NO 3 -initiated oxidation and the lowand high-NO x pathway for both OH-and O 3 -initiated oxidation.As shown in Fig. 9, with a 50 % reduction in anthropogenic NO emissions, total surface SOAM concentration decreases, dominated by the decrease in NO 3 -oxidation branch.This decrease in total SOAM mass is a result of addition or cancellation of various changes in each branch, and the relative importance of different branches may change with different regions.One interesting phenomenon is that when NO x emissions are reduced, the low-NO x OH-and O 3 -initiated oxidation branches form less SOAM mass in the Amazon but more SOAM in human-influenced regions like mid-latitude broadleaf forest in the southeast US, coastal Asia and boreal forest in northern Europe.To further understand and quantitatively evaluate the complex NO x influence on SOA formation, we examine the predicted change in surface SOA concentrations in different pathways in response to the decrease in anthropogenic NO emissions, as illustrated by Fig. 10.Table 6 details the relative contribution of each pathway to the total SOA change.The results for various SOA types, i.e., aromatic SOA, isoprene SOA and monoterpene SOA, are discussed below.

Anthropogenic SOAs from benzene, toluene and xylenes (ASOA)
ASOA in the three models are assumed to form from OHinitiated oxidation, including both low-NO x and high-NO x pathways, i.e., AVOCs+OH(HO 2 ) and AVOCs+OH(NO).
In the southeast US as shown in Fig. 10, all models predict an increase in the low-NO x pathway and a decrease in the high-NO x pathway.This is because the model assumes linear interpolation between low-and high-NO x pathways based on the branching ratio (Sect.3.3).When NO x is reduced, more AVOCs are oxidized under the low-NO x pathway, which has higher yields (see Tables S1 and S2).Due to the limited change in β, the effect of shifting to a highyield HO 2 pathway is very small.The total ASOA formation The observed HOA is a surrogate for POA from combustion and biomass burning.Simulated POA is identical in the three simulations.In the POA-HOA comparison, data points with observed HOA smaller than 0.01 µg m −3 have been set to 0.01 µg m −3 to be shown in the plots.
Table 6.The relative contributions (%) of each SOA formation pathway to the total SOA concentration change in the southeast US and the Amazon, defined as depends on both the low-/high-pathway partitioning and the oxidation capacity, so it can either increase (e.g., 2-product, VBS) or decrease (VBS_agHigh).In the Amazon, the ASOA changes follow the same pattern as in the southeast US, but their relative contributions are very small due to the low concentrations of AVOCs and anthropogenic NO x .The contributions of ASOA changes to the total SOA change (defined as change in ASOA |total SOA change| ) range from −3.7 to 9.2 % in the southeast US and from −0.6 to 1.0 % in the Amazon.

Isoprene SOA (SOAI)
In our current models, the isoprene only has one set of yields for OH-initiated daytime oxidation following low-NO x parameterization and is oxidized by NO 3 during the night (the latter is not considered in the 2-product model).The OH oxidation is the dominant branch to form SOAI in both regions.When anthropogenic NO emissions are halved, both OH-and NO 3 -initiated branches decrease in the southeast US and the Amazon due to reduced atmospheric OH and

Monoterpene SOA (SOAM)
Monoterpenes are oxidized by OH, O 3 and NO 3 in all models but the 2-product model only considers the low-NO x pathway (Table 2).In the southeast US with large human influence, the surface SOAM formation is largely attributed to NO 3 -initiated oxidation as indicated by most models, which dominates the reduction in response to reduced NO x .This branch itself contributes to the total SOA change ( change in SOAM from NO 3 −oxidation |total SOA change| ) by −48.9 to −65.3 %.This reduction in NO 3 branch compared to its normal value is relatively small because the decrease in NO 3 concentration is only 24 %.In the VBS and VBS_agHigh models, the partitioning between high-vs.low-NO x pathways determines the tendency of increasing yields from the low-NO x pathway and decreasing yields through the high-NO x pathway.The OH oxidation in the southeast US follows such tendency.However, the SOA formed from both high-and low-NO x pathways of ozonolysis increases.One possible explanation is the "buffering" between O 3 -and NO 3 -initiated oxidation, both of which mostly happen at night.Compared to the control run, NO 3 is significantly lower in the sensitivity run; thus, more monoterpenes would be oxidized by O 3 under both low-and high-NO x conditions.Adding up the changes in all branches, the SOAM change contributions to total SOA change are about −36.1 to −60.7 %.In the Amazon's pristine environment, most branches demonstrate a slight reduction in SOAM in all models.Since the absolute magnitude of anthropogenic NO x is small, the major influence of the NO x might be the decline in level of atmospheric oxidants: OH, O 3 and NO 3 decrease by 14, 6 and 16 %, respectively.Despite the minor lessening of oxidation capacity, the SOAM reduction and total SOA reduction are negligible.

Summary of surface SOA concentration change
The changes in total SOA concentration at the surface in different regions are summarized in Table 7.In both humaninfluenced and clean regions, the 50 % reduction in anthropogenic NO emissions leads to a decline of BSOA, which dominates the overall SOA decrease.The ASOA could either increase (in models without aging parameterization) or decrease (in models with aging considered).Among the multiple effects of NO x , BSOA is mostly influenced by changes in NO 3 -initiated oxidation.Both BSOA and ASOA are also influenced by the change in atmospheric oxidation capacity and the partitioning between high-vs.low-NO x pathways.
The annual mean total surface SOA reductions in the southeast US, the Amazon and global average range from 119 to 518, 30 to 153, 3.6 to 43 ng m −3 , respectively.The corresponding percentage reductions are 6.4-12.0,0.9-2.8 and 0.9-5.6 %.These changes are comparable with previous estimates (Carlton et al., 2010;Lane et al., 2008) but all are smaller than the magnitude of 1 standard deviation, indicating that such changes are not statistically significant compared to interannual variations caused by climate and emission variations.The column concentrations of tropospheric SOA are also examined (results not shown here), and the conclusion still holds -no significant change of SOA column concentration when anthropogenic NO emissions are reduced by 50 %.One major reason is the small reduction in branching ratio β, which limits the shift between high-vs.low-NO x chemical regimes.The fact that SOA is stable in response to anthropogenic NO x changes is also attributed to the buffering of various branches (e.g., increased ozonolysis and decreased NO 3 oxidation), the partitioning between low-and high-NO x pathways and the offset from opposite tendencies of BSOA and ASOA responses (in the no-aging models).The complexity includes the competition between NO and HO 2 to react with RO 2 , its substantial influence on atmospheric oxidation capacity, and the nighttime NO 3 direct oxidation of isoprene and monoterpenes.In this study, we have updated the SOA scheme in the global chemistry-climate model CAM4-chem to include a 4-product VBS scheme that has a broader representation of volatility distribution and quantitatively evaluated and explained the multiple impacts of anthropogenic NO x on SOA at global scale.
We updated the SOA scheme in CAM4-chem to a 4product VBS scheme.Compared to the default 2-product model, the VBS scheme has 11 % higher surface SOA concentration.While the total annual mean SOA burden is 19 % smaller (0.69 ± 0.03 Tg[C] as compared to 0.85 ± 0.04 Tg[C]) and has a lifetime that is shorter (8.9 ± 0.2 days as compared to 11.4 ± 0.4 days).Due to the different volatility and higher yields of SOA in the VBS, more VOC is oxidized near the surface and less is transported to higher levels, and more SOA is washed out near the surface.We explored an aging parameterization with a constant reaction rate with OH (k OH = 4 × 10 −11 cm 3 molec −1 s −1 , the higher-limit in previous studies), which almost doubles the net annual SOA production and significantly increases the SOA concentration both at the surface and in the lower free troposphere.The global SOA burden with aging considered (i.e., VBS_agHigh scheme) increases to 1.08 ± 0.06 Tg[C] and the corresponding lifetime is 6.7 ± 0.1 days.By applying a lower aging reaction rate (k OH = 5.2 × 10 −12 cm 3 molec −1 s −1 , the lowerlimit in previous studies), we found that the simulation of SOA is quite sensitive to the assumed k OH .Despite the significance to SOA formation and properties, the aging effect is still poorly understood at the global scale.Further laboratory and process-modeling constraints at different conditions are needed.
The simulated total OC concentrations in the 2-product and the VBS models without aging are similar, and they capture the magnitude and distribution of annual mean surface OC concentrations in the US from the IMPROVE network, by 45-47 %, but overestimate OC in the polluted northeast US and western coastal regions.The model with an implementation of aging (VBS_agHigh) slightly improves the replication of annual mean spatial distribution (r 2 = 53 %) but overestimates the magnitude.All three models perform poorly in summertime.Compared to AMS measurements from 13 aircraft-based field campaigns, the simulations of OA vertical profiles are within a factor of 2 across different regions and seasons.The VBS_agHigh scheme performs better than the two no-aging models to reproduce these observed OA concentrations (r 2 = 56 %, rmsd = 1.45).Further climatological comparisons with surface AMS observations indicate reasonable simulated total OA concentrations but overestimation of POA in some polluted regions, which is consistent with the comparison to the IMPROVE network.This overestimation of POA may come from a higher biased POC from the emissions inventory in certain regions (e.g., the northeast US).If so, it would partially conceal the fact that the current parameterized SOA yields and overlooking of ag- M1a" M1b" M2a" M2b"M3" I1a" I3" A1a"A1b"
Finally, we performed sensitivity experiments to examine how the SOA loading responds to a 50 % reduction in anthropogenic NO emissions in different regions.The BSOA generally decreases due to the reduction in NO 3 -initiated reaction and the reduced atmospheric oxidation capacity, while the ASOA increases in the two no-aging models mainly because of the increased partitioning to the low NO x pathway; more AVOCs are oxidized through the low-NO x pathway that has higher yields.In the aging model, ASOA decreases due to the more important effect of reduced oxidation capacity.Decreases in the total surface SOA concentra-tions are 6.4-12.0,0.9-2.8 and 0.9-5.6 % for the southeast US, the Amazon and global NO x perturbations, respectively, which, however, are not significant.The fact that SOA formation is stable to changes in NO x can be largely attributed to limited shift in low-and high-NO x regimes, to buffering in chemical pathways (e.g., O 3 versus NO 3 -initiated oxidation), and to offsetting tendencies in the biogenic versus anthropogenic SOA responses.Our results, based on the global chemistry-climate model CAM4-chem with simplified SOA schemes, indicate that air quality control on anthropogenic NO x may not have substantial impacts on organic aerosol loadings at large regional scales.Further modeling studies including both process-based and parameterized schemes need to be done to carefully examine the NO x impact on SOA formation.
The Supplement related to this article is available online at doi:10.5194/acp-15-13487-2015-supplement.

Figure 1 .
Figure 1.Annual mean zonally averaged SOA concentration (µg m −3 ) (shown as colored shades) and the fraction of biogenic SOA (%) (shown as white contours) in CAM4-chem for different SOA treatments.

Figure 3 .
Figure 3.Comparison of averaged annual mean surface OC concentrations (µg[C] m −3 ) between IMPROVE measurements and the three simulations: 2-product, VBS and VBS_agHigh.Different colors indicate sites in different regions.In each subplot, the dashed line is the 1 : 1 line.The coefficients of determination (r 2 ), rootmean-square difference (rmsd) and the model-to-observation slope (k) are included.

Figure 4 .
Figure 4. Comparison between the observed vertical profiles of OA concentration (µg m −3 ) from 13 AMS field campaigns and the three model simulations: 2-product, VBS and VBS_agHigh.The campaign information is summarized in Heald et al. (2011) (Fig. 1 and Table1therein).The error bars are 1 standard deviation of the binned observations for each 0.5 km interval.The grey shades are simulated POA assuming a POA-to-POC ratio of 1.4.The model simulations are sampled for the corresponding months and locations for each campaign.The location and location type for each campaign is included in each subplot.

Figure 6 .
Figure6.Comparison of surface AMS measurements (units: µg m −3 ) and three simulations.First column: total OA; second column: SOA(OOA); third column: POA(HOA).The r 2 and rmsd are included in each subplot.The OOA is a surrogate for SOA from all sources.The observed HOA is a surrogate for POA from combustion and biomass burning.Simulated POA is identical in the three simulations.In the POA-HOA comparison, data points with observed HOA smaller than 0.01 µg m −3 have been set to 0.01 µg m −3 to be shown in the plots.
SOA change in each pathway|total SOA change|.The sums of all numbers in each simulation equal −100 % because the total SOA change in the sensitivity runs compared to the control runs are always negative.The reaction denotations are the same as defined in Fig.10.

Figure 7 .
Figure 7.Comparison of observed OOA-to-OA ratio from surface AMS measurements and simulated SOA-to-OA ratio from the 2product, VBS and VBS_agHigh schemes.r 2 and rmsd are included in each subplot.

Figure 8 .
Figure 8. Dependence of annual mean surface SOA concentration (µg m −3 ) on branching ratio and oxidants level at global scale, in the southeast US and the Amazon.The control runs and the sensitivity runs using VBS and VBS_agHigh schemes are shown.The 2-product results are similar to the VBS results (not shown).Data points over ocean are excluded.Note that the scales for the southeast US are different from the Amazon and the global subplots.

Table 1 .
Abbreviations used in this study.

Table 2 .
Summary of SOA treatments in CAM4-chem model runs.

Table 3 .
Summary of simulated annual mean global budget of SOA (particle-phase).

2015 13494 Y. Zheng et al.: Limited effect of anthropogenic nitrogen oxidesTable 4 .
Fraction of SOC in total OC (%) in the southeast US and the northeast US.

Table 5 .
Coefficients of determination (r 2 ) of IMPROVE measurements versus simulated total OC.

Table 7 .
Changes in surface SOA concentrations due to a 50 % reduction in anthropogenic NO emissions.Total SOA changes from each model are listed for global average, the southeast US and the Amazon.x plays a complex role in the chemical formation of SOA.