Large Contributions from Biogenic Monoterpenes and Sesquiterpenes to Organic Aerosol 1 in the Southeastern United States 2

Abstract. Atmospheric organic aerosol (OA) has important impacts on climate and human health but its sources remain poorly understood. Biogenic monoterpenes and sesquiterpenes are critical precursors of OA. The OA generation from these precursors predicted by models has considerable uncertainty owing to a lack of appropriate observations as constraints. Here, we perform novel lab-in-the-field experiments, which allow us to study OA formation under realistic atmospheric conditions and offer a connection between laboratory and field studies. Based on the lab-in-the-field experimental approach and positive matrix factorization analysis on aerosol mass spectrometry data, we provide a measure of OA from monoterpenes and sesquiterpenes in the southeastern U.S. Further, we use an upgraded atmospheric model and reproduce the measured OA concentration from monoterpenes and sesquiterpenes at multiple sites in the southeastern U.S., building confidence in the observed attribution of monoterpene SOA. We show that the annual average concentration of OA from monoterpenes and sesquiterpenes in the southeastern U.S. is ~ 2.1 µg m −3 . This amount is substantially higher than represented in current regional models and accounts for 21 % of World Health Organization PM 2.5 standard, indicating a significant contributor of environmental risk to the 77 million habitants in the southeastern U.S.


Introduction
Organic aerosol (OA) constitutes a substantial fraction of ambient fine particulate matter (PM) and has large impacts on air quality, climate change, and human health (Carslaw et al., 2013;Lelieveld et al., 2015).OA can be directly emitted from sources (primary OA, POA) or formed by the oxidation of volatile organic compounds (VOCs) (secondary OA, SOA).Global measurements revealed the dominance of SOA over POA in various atmospheric environments (Jimenez et al., 2009;Ng et al., 2010).The VOCs can be emitted from natural sources (i.e., biogenic) or human activities (i.e., anthropogenic).However, the relative contribution of biogenic and anthropogenic sources to SOA formation in the atmosphere is poorly constrained.This knowledge is critical for formulating effective pollution control strategies that aim at reducing ambient PM concentrations and accurately assessing the climate effects of OA (Hallquist et al., 2009).Biogenic VOCs such as monoterpenes (MT, C10H16) and sesquiterpenes (SQT, C15H24) are recognized as critical precursors of SOA (Tsigaridis et al., 2014;Hodzic et al., 2016;Pye et al., 2010).The predicted global SOA production from MT and SQT varies from 14 to 246 Tg yr -1 (Spracklen et al., 2011;Pye et al., 2010).This large variation in model estimates arises from a number of factors (including uncertainty in SOA yield) and introduces significant uncertainties in estimating OA concentrations and its subsequent influences on climate and human exposure.
The large model uncertainties call for ambient observations to constrain model results.
Isolating and measuring SOA production from specific sources are challenging because SOA is a complex mixture consisting of thousands of compounds and SOA evolves dynamically in the atmosphere.A widely used method to apportion OA into different characteristic sources is positive matrix factorization (PMF) analysis on the organic mass spectra measured by aerosol mass spectrometer (AMS) (Ulbrich et al., 2009;Jimenez et al., 2009;Ng et al., 2010).PMF-AMS analysis groups OA constituents with similar mass spectra and temporal variations into characteristic OA subtypes (i.e., factors).This analysis has revealed two ubiquitous OA subtypes in ambient environments, less-oxidized oxygenated OA (LO-OOA) and more-oxidized oxygenated OA (MO-OOA), which are differentiated by their degree of carbon oxidation.LO-OOA and MO-OOA together account for more than half of total submicron OA (Crippa et al., 2014;Xu et al., 2015a;Jimenez et al., 2009).Primarily based on comparison of their mass spectra with those of laboratory-generated SOA, previous studies proposed that LO-OOA is freshly formed SOA from various sources and evolves into MO-OOA with photochemical aging in the atmosphere (Jimenez et al., 2009;Ng et al., 2010).These studies have significantly advanced our knowledge of the composition and evolution of ambient OA; however, there are still uncertainties associated with the sources of these OA factors.Firstly, the current understanding on LO-OOA and MO-OOA offers little mechanistic information regarding the specific sources of these factors at a measurement site.Atmospheric models typically use the lumped LO-OOA and MO-OOA concentration to constrain simulated total SOA concentration (Spracklen et al., 2011;Tsigaridis et al., 2014), which hinders our ability to diagnose the cause for the discrepancies between modeled and observed aerosol concentrations (Spracklen et al., 2011).Secondly, the assumption that LO-OOA represents fresh SOA has yet to be directly verified.Also, it is not known whether fresh SOA is exclusively apportioned into LO-OOA.For example, rather than being produced from continued photochemical aging, recent studies hypothesize that the rapidly produced HOMs (highly oxygenated molecules) from the oxidation of VOCs may contribute to MO-OOA (Ehn et al., 2014).
Thus, considering the large abundance of these two OA subtypes and that they are surrogates for ambient SOA, understanding the sources of compounds composing these two OA subtypes is critical to constrain atmospheric models and SOA budget.
In this study, we integrate lab-in-the-field experiments, extensive ambient ground measurements, and state-of-the-art modeling to constrain the OA budget from MT and SQT.We provide direct evidence that newly formed SOA from α-pinene (representative monoterpene, which accounts for about half of monoterpenes emissions (Guenther et al., 2012)) and βcaryophyllene (representative sesquiterpene) dominantly contributes to LO-OOA in the southeastern U.S. The modeled SOA from the oxidation of MT and SQT (denoted as SOAMT+SQT) accurately reproduces the magnitude and diurnal variability of LO-OOA measured at multiple sites in the southeastern U.S. The agreement between model and measurements supports the hypothesis that LO-OOA can be used as a measure of SOAMT+SQT in the southeastern U.S. The lab-in-thefield approach allows for the study of SOA formation under realistic atmospheric conditions, which bridges laboratory studies and field measurements and provides a direct way to evaluate the atmospheric relevancy of laboratory studies.

Lab-in-the-field perturbation experiments
The perturbation experiments were performed in July-August 2016 on the rooftop of the Environmental Science and Technology building on the Georgia Institute of Technology campus.This measurement site is a representative urban site in Atlanta.Multiple ambient field studies have been performed at this site previously (Xu et al., 2015b;Hennigan et al., 2009;Verma et al., 2014).
A 2m 3 Teflon chamber (cubic shape) (Fig. 1) was placed outdoor on the rooftop of the building.
The eight corners of the chamber were open (~2"×2") to the atmosphere to allow for continuous exchange of air with the atmosphere.The perturbation procedure is briefly described below and illustrated in Fig. A1.Firstly, we continuously flushed the chamber with ambient air using two fans, which were placed at two corners of the chamber.During this flushing period, all instruments sampled ambient air and were not connected to the chamber.The flushing period lasted at least 3 hours to ensure that the air composition in the chamber is the same as ambient composition.
Secondly, we stopped both fans and connected all instruments to chamber.Because of the continued sampling by the instruments (~20 liter per minute) and the open corners of the chamber, ambient air continuously entered the chamber, even though the two fans were turned off.Thirdly, after sampling the chamber for about 30min, we injected a known amount of VOC (liquid) into the chamber with a needle, where the liquid vaporized upon injection.We continuously monitored the chamber composition for ~40 min after VOC injection.Lastly, we disconnected all instruments from the chamber, sampled ambient air, and turned on two fans to flush the chamber to prepare for the next perturbation experiment.
We perturbed the chamber content by injecting one of the following VOCs: isoprene, αpinene, β-caryophyllene, m-xylene, or naphthalene, which are major biogenic or anthropogenic emissions, respectively.The injected VOC amounts were carefully selected.If the injection amount is too large, it is not atmospherically relevant, produces too much SOA, and will bias subsequent analysis.If the injection amount is too small, the produced SOA would fall below the detection limit of the experimental approach.The OA concentration in the chamber after perturbation ranges from 4 to 16 µg m -3 , which is within the range of typical ambient OA concentration.The VOC oxidation occurred in ambient air (inside the chamber) and lasted ~40 min.Several previous studies have used ambient air as background (Palm et al., 2017;Leungsakul et al., 2005).An important distinction between our study and pervious work is that we perturbed the ambient air only by injection of VOCs and no extra oxidant precursors (i.e., O3, photolysis of H2O and O2, or photolysis of NOx) were added to the chamber.Our approach allows for study of SOA formation from the specific VOCs injected and evaluate into which PMF factor the SOA is apportioned.Each perturbation experiment can be divided into the following four periods: Amb_Bf (30min ambient measurement period before sampling chamber), Chamber_Bf (from sampling chamber to VOC injection, a period ~30min), Chamber_Af (from VOC injection to stop sampling chamber, a period ~40min), and Amb_Af (30min ambient measurement period after sampling chamber).We calculate the changes in the mass concentration of OA factors after perturbation based on the difference between Chamber_Bf and Chamber_Af, after taking ambient variation into account.The detailed procedure is presented in Appendix A. We develop a comprehensive set of criteria to determine if the changes are statistically significant and if the changes are simply due to ambient variations.The details of these criteria are also discussed in Appendix A.

Analytical instruments
A suite of analytical instruments was deployed to characterize both the gas-phase and particlephase compositions.The particle-phase composition was monitored by a scanning mobility particle sizer (SMPS, TSI) and a high resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS, Aerodyne), which shared the same sampling line.A diaphragm pump (flow rate ~8 liter per minute) was connected to this sampling line, which increased the sampling flow rate and suppressed particle loss in the sampling line by reducing the residence time in the tubing.The HR-ToF-AMS measures the chemical composition and size distribution of submicron non-refractory species (NR-PM1) with high temporal resolution.The instrument details about HR-ToF-AMS have been extensively discussed in the literature (Canagaratna et al., 2007;DeCarlo et al., 2006) and the operation of HR-ToF-AMS in this study is described in the section S2 of Supplement.
The gas-phase composition and oxidation products was monitored by an O3 analyzer (Teledyne T400, lower detectable limit 0.6ppb), an ultrasensitive chemiluminescence NOx monitor (Teledyne 200EU, lower detectable limit 50ppt), and a high-resolution time-of-flight chemical ionization mass spectrometer (HR-ToF-CIMS).The HR-ToF-CIMS with I -as regent ion can measure a suite of oxygenated volatile organic compounds (oVOCs) at high frequency (1Hz).
Detailed working principles and sampling protocol can be found in Lee et al. (2014).The concentrations of VOCs were not measured in this study.All gas-phase measurement instruments shared the same sampling line.Similar to the particle sampling line, a diaphragm pump (flow rate ~8 liter per minute) was connected to the gas sampling line to reduce the residence time in the tubing.

Positive Matrix Factorization (PMF) analysis
PMF analysis has been widely used for aerosol source apportionment in the atmospheric chemistry community (Jimenez et al., 2009;Crippa et al., 2014;Xu et al., 2015a;Ng et al., 2010;Ulbrich et al., 2009;Beddows et al., 2015;Visser et al., 2015).PMF solves bilinear unmixing factor model (Paatero and Tapper, 1994;Ulbrich et al., 2009) by minimizing the summed least squares errors of the fit weighted with the error estimates of each measurement.We utilized the PMF2 solver, which does not require a priori information and reduces subjectivity.In this study, we performed PMF analysis on the high-resolution mass spectra of organic aerosol (inorganic species are excluded) of combined ambient and perturbation data in the one-month measurements.
Considering that (1) the perturbation data only account for ~10% of total data and (2) the OA concentration is similar in the perturbation experiments and typical ambient measurements, the perturbation experiments do not create a new factor that does not already exist in the ambient data.This is desirable because it allows PMF analysis to apportion the newly formed OA in the perturbation experiments into pre-existing OA factors in the atmosphere.
We resolved five OA factors, including hydrocarbon-like OA (HOA), cooking OA (COA), isoprene-derived OA (isoprene-OA), less-oxidized oxygenated OA (LO-OOA), and moreoxidized oxygenated OA (MO-OOA).The time series and mass spectra of OA factors are shown in Fig. 2. The same 5 factors have been identified at the same measurement site and extensively discussed in the literature (Xu et al., 2015a;Xu et al., 2015b;Xu et al., 2017).Below, we only provide a brief description on these OA factors and more details are discussed in section S3 of Supplement.The mass spectrum of HOA is dominated by hydrocarbon-like ions (CxHy + ions) and HOA is a surrogate of primary OA from vehicle emissions (Zhang et al., 2011).For COA, its concentration is higher at meal times and its mass spectrum is characterized by prominent signal at ions C3H5 + (m/z 41) and C4H7 + (m/z 55), which likely arise from unsaturated fatty acids (Huang et al., 2010;Mohr et al., 2009).The mass spectrum of isoprene-OA is characterized by prominent signal at ions C4H5 + (m/z 53) and C5H6O + (m/z 82) and it is related the reactive uptake of isoprene oxidation products, isoprene epoxydiols (IEPOX) (Budisulistiorini et al., 2013;Hu et al., 2015;Robinson et al., 2011;Xu et al., 2015a).LO-OOA and MO-OOA are named based on their differing carbon oxidation state.

Details of multiple ambient sampling sites
Measurements at multiple sites in the southeastern U.S. were performed as part of Southeastern Center for Air Pollution and Epidemiology study (SCAPE) and Southern Oxidant and Aerosol Study (SOAS).Detailed descriptions about these field studies have been discussed in the literature (Xu et al., 2015a;Xu et al., 2015b) and section S4 of Supplement.The sampling periods are shown in Table S1 and the sampling sites are briefly discussed below.
• Georgia Tech site (GT): This site is located on the rooftop of the Environmental Science and Technology building on the Georgia Institute of Technology (GT) campus, which is about 30-40m above the ground and 840m away from interstate I75/85.This is a representative urban site in Atlanta.This is also where the perturbation experiments in this study were conducted.
• Jefferson Street site (JST): This is a central SEARCH (SouthEastern Aerosol Research and Characterization) site, which is in Atlanta's urban area with a mixed commercial and residential neighborhood.It is about 2 km west of the GT site.The JST and GT sites are in the same grid cell in CMAQ.
• Yorkville site (YRK): This is a central SEARCH site located in a rural area in Georgia.This site is surrounded by agricultural land and forests and is at about 80 km northwest of JST site.
• Centreville site (CTR): This is a central SEARCH site in rural Alabama.The sampling site is surrounded by forests and away from large urban areas (55km SE and 84 km SW of Tuscaloosa and Birmingham, AL, respectively).The is the main ground site for the SOAS campaign.

Laboratory chamber study on SOA formation from α-pinene
To compare with results from the lab-in-the-field perturbation experiments, we performed laboratory experiments to study the SOA formation from α-pinene photooxidation under different NOx conditions in the Georgia Tech Environmental Chamber (GTEC) facility.The facility consists of two 12 m 3 indoor Teflon chambers, which are suspended inside a temperature-controlled enclosure and surrounded by black lights.The detailed description about chamber facility can be found in Boyd et al. (2015).The experimental procedures have been discussed in Tuet et al. (2017).
In brief, the chambers were flushed with clean air prior to each experiment.Then, α-pinene and oxidant sources (i.e., H2O2, NO2, or HONO) were injected into chamber.Once the concentrations of species stabilize, the black lights were turned on to initiate photooxidation.The experimental conditions are summarized in Table S2.Considering that the OA mass concentration affects the partitioning of semi-volatile organic compounds (Odum et al., 1996) and hence affects the organic mass spectra measured by AMS, we calculated the average mass spectra in these laboratory studies by only using the data when the OA mass concentration is below 10 µg m -3 , which is similar to that in our ambient perturbation experiments.

Community Multiscale Air Quality (CMAQ) Model
We used the Community Multiscale Air Quality (CMAQ) atmospheric chemical transport model to simulate the pollutant concentrations across the southeastern U.S. CMAQ v5.2gamma was run over the continental U.S. for time periods between May 2012 to July 2013 with 12km × 12km horizontal resolution.We focus our analysis on the southeastern U.S., which comprises 11 states (Arkansas, Alabama, Florida, Georgia, Kentucky, Louisiana, Mississippi, North Carolina, South Carolina, Tennessee, and Virginia).The meteorological inputs were generated with version 3.8 of the Weather Research and Forecasting model (WRF), Advanced Research WRF (ARW) core.We also applied lightning assimilation to improve convective rainfall (Heath et al., 2016).Anthropogenic emissions were based on the EPA (Environmental Protection Agency) NEI (National Emission Inventory) 2011 v2.Biogenic emissions were predicted by the BEIS (Biogenic Emission Inventory System) v3.6.1.The gas-phase chemistry was based on CB6r3 (Carbon Bond v6.3).
We performed two simulations with different organic aerosol treatment.The "default simulation" generally follows the scheme of Carlton et al. (2010), with IEPOX SOA following Pye et al. (2013) and documented in Appel et al. (2017) (Fig. S1a).The traditional two-product absorptive partitioning scheme (Odum et al., 1996) is used in "default simulation" to describe SOA formation from monoterpenes using data from laboratory experiments by Griffin et al. (1999).In the "updated simulation", we incorporate two recent findings.Firstly, we implemented MT+NO3 chemistry to explicitly account for the organic nitrate compounds that have recently been shown to be a ubiquitous and important component of OA (Pye et al., 2015;Kiendler-Scharr et al., 2016;Lee et al., 2016;Ng et al., 2017).We follow the scheme described in Pye et al. (2015) to represent the formation and partition of organic nitrates from monoterpenes via multiple reaction pathways (i.e., oxidation by NO3 and oxidation by OH/O3 followed by RO2+NO).Secondly, we improved the parameterization of SOA formation from MT+O3/OH based on a recent study by Saha and Grieshop (2016), who applied a dual-thermodenuder system to study the α-pinene ozonolysis SOA.
The authors extracted SOA yield parameters by using an evaporation-kinetics model and volatility basis set (VBS).The SOA yields in Saha and Grieshop (2016) are consistent with recent findings on the formation of HOMs (Ehn et al., 2014;Zhang et al., 2015) and help to explain the observed slow evaporation of α-pinene SOA (Vaden et al., 2011).In the updated simulation, we use the VBS framework with parameters derived from Saha and Grieshop (2016).The properties of 7 volatility bins are listed in Table S3.A schematic of SOA treatment in "updated simulation" is shown in Fig. S1b.Additional details of the CMAQ simulations are given in the section S5 of Supplement.

α-pinene perturbation experiments
A total of 19 α-pinene perturbation experiments were performed at different times of the day (i.e., from 9am to 9pm) to probe a wide range of reaction conditions.The concentrations of O3 and NOx during α-pinene perturbation experiments are summarized in Table S4.Initially ~14 ppb α-pinene is injected into chamber, but only a small fraction of α-pinene is reacted in the chamber, with most of α-pinene being carried out of the chamber due to dilution with ambient air (section S6 of Supplement).Fig. 3 shows the time series of OA factors in a typical α-pinene perturbation experiment.
The most striking feature is a burst increase of LO-OOA after α-pinene injection.This is the most direct and compelling evidence that freshly formed α-pinene SOA contributes to LO-OOA.About 15 min after α-pinene injection, LO-OOA concentration starts to decrease, as ambient air continuously flows into the chamber and dilutes the concentration of LO-OOA (section S6 of Supplement).As shown in Fig. S2, the major known gas-phase oxidation products of α-pinene measured by HR-ToF-CIMS (Eddingsaas et al., 2012;Yu et al.;Lee et al., 2016) show an immediate increase after α-pinene injection.This verifies the rapid oxidation of α-pinene in the chamber.conditions vary between experiments, we note that both LO-OOA enhancement amount and LO-OOA formation rate (i.e., slope of LO-OOA increase) correlate positively with ozone concentration (Fig. 5).This correlation suggests that the concentration of oxidants, both ozone and hydroxy radical (OH, which is not measured in this study but is known to positively correlate with ozone in the atmosphere), is a controlling variable for OA formation in α-pinene experiments.This is likely because higher oxidant concentrations lead to more α-pinene consumption and hence more OA production with the same reaction time.
MO-OOA only increases in 1 out of 19 α-pinene experiments.The lack of enhancement in MO-OOA suggests that the HOMs, which are rapidly produced from the α-pinene oxidation (Ehn et al., 2014), are unlikely contributors to MO-OOA, though more future studies on the apportion of HOMs by PMF analysis are warranted.Isoprene-derived OA (isoprene-OA) increases in 7 out of 19 α-pinene experiments.This increase is surprising because the isoprene-OA factor is typically interpreted as SOA from the reactive uptake of IEPOX.Our results demonstrate that the isoprene-OA factor (also referred to as "IEPOX-OA" in some studies) could have interferences from αpinene SOA.This conclusion could be applicable to isoprene-OA factor resolved at other monoterpenes-influenced sites.Primary OA factors, i.e., HOA and COA, only show slight increases in 1 or 2 α-pinene experiments, indicating a lack of interference from α-pinene SOA in these factors.

β-caryophyllene perturbation experiments
A total of 6 β-caryophyllene perturbation experiments were performed.Initially ~10 ppb β-caryophyllene is injected into chamber The concentrations of O3 and NOx during βcaryophyllene perturbation experiments are summarized in Table S4.In all β-caryophyllene perturbation experiments, LO-OOA also shows a significant enhancement (Fig. 4b).This clearly demonstrates that the freshly formed SOA from β-caryophyllene oxidation can be another source of LO-OOA.In addition to LO-OOA, COA shows an unexpected increase in 5 out of 6 βcaryophyllene experiments.We have ample evidence that the COA factor at the measurement site has contributions from cooking activities.Firstly, the diurnal variation of COA peaks during meal times (Fig. S3a).Secondly, the COA concentration shows clear increase on football days, consistent with barbecue activities on campus and close to the measurement site.Thirdly, the COA concentration is enhanced on the days right before the start of a new semester when there are many fraternity/sorority rush events (i.e., barbecue activities) on campus (Fig. S3b and S3c).However, the COA enhancement in β-caryophyllene experiments underscores the fact that COA may not be purely from cooking activities in areas with large biogenic emissions.

Perturbation experiments with other VOCs
In addition to α-pinene and β-caryophyllene, we also performed perturbation experiments by injecting isoprene, m-xylene, or naphthalene, which are important biogenic and anthropogenic emissions, respectively.However, the SOA formation from these VOCs is not detectable.This is mainly due to either lower SOA yields (of isoprene) or slower oxidation rates (of m-xylene and naphthalene) compared to α-pinene and β-caryophyllene (section S6 of Supplement).The perturbation experiments with other VOCs confirm the stronger ability of α-pinene and βcaryophyllene to produce SOA (Kroll et al., 2006;Ng et al., 2007;Griffin et al., 1999).
We have also performed perturbation experiments by injecting acidic sulfate particles to probe reactive uptake of IEPOX.We observed enhancement in isoprene-OA concentration after the injection of sulfate particles.The detailed results are included in Appendix B.

Perturbation experiments vs. mass spectra comparison
The perturbation experiments provide more insights into the sources of OA factors than traditional mass spectra comparison.Previous studies concluded that LO-OOA (also denoted as semi-volatile oxygenated organic aerosol, SV-OOA) represents freshly formed SOA, mainly based on the observation that the mass spectra of laboratory-generated fresh SOA from various sources better resemble the mass spectrum of LO-OOA than other factors (Jimenez et al., 2009;Ng et al., 2010).While this mass spectra comparison approach sheds light on the potential sources of LO-OOA, it does not allow for evaluating whether freshly formed SOA in the atmosphere is exclusively apportioned into LO-OOA.The perturbation experiments, on the other hand, provide a way to evaluate this explicitly.Here, we directly produce SOA from a specific known VOC in ambient air matrix and determine where it is apportioned into.For example, we show that while fresh SOA from α-pinene and β-caryophyllene oxidations are mainly apportioned into LO-OOA, they could also be possibly apportioned into isoprene-OA factor and COA, respectively.
The perturbation experiments have the potential to utilize subtle differences across the entire the mass spectrum to evaluate the sources of OA factors.Based on previous laboratory studies, the mass spectra of fresh SOA from α-pinene oxidation and β-caryophyllene oxidation share much similarity, but there are subtle differences in the mass spectra (Bahreini et al., 2005; Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1109Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.Tasoglou and Pandis, 2015).For example, in the perturbation experiments a fraction of the fresh β-caryophyllene SOA is apportioned into COA factor, but we do not observe similar behavior for α-pinene SOA.This is likely because f55 (i.e., the ratio of m/z 55 to total signal in the mass spectrum) is typically higher in β-caryophyllene SOA than α-pinene SOA and the mass spectrum of COA is characterized by prominent signal at m/z 55 (Fig. 2).

Connection between laboratory and field studies
Due to the difficulties associated with accurately measuring complex chemical processes in the atmosphere, laboratory studies have been an integral part in our understanding of atmospheric chemistry (Burkholder et al., 2017).However, the representativeness of laboratory studies under simplified conditions with respect to the complex atmosphere is difficult to evaluate.One unique feature of our lab-in-the-field approach is that the VOC oxidation and SOA formation proceed under realistic atmospheric conditions.Taking advantage of this, we provide a direct link between laboratory studies and ambient observations.Previous laboratory studies have shown that NO can affect SOA composition by influencing the fate of organic peroxy radical (RO2, a critical radical intermediate formed from VOC oxidation) (Kroll and Seinfeld, 2008;Sarrafzadeh et al., 2016;Presto et al., 2005).To evaluate the representativeness of laboratory studies and directly investigate the effects of NO on SOA composition, in Fig. 6, we compare the chemical composition of α-pinene SOA formed in laboratory studies under different NO conditions (denoted as SOAlab) with those in α-pinene ambient perturbation experiments (denoted as SOAambient).The degree of similarity in OA mass spectra (i.e., evaluated by the correlation coefficient) between laboratory αpinene SOA generated under NO-free condition (i.e., denoted as SOAlab,NO-free, using H2O2 photolysis as oxidant source) and SOAambient shows a strong dependence on ambient NO concentration, under which the SOAambient is formed.The degree of similarity in mass spectra decreases rapidly when ambient NO increases from 0.1 to 0.2ppb, and then reaches a plateau at ~0.3ppb NO.The opposite trend is observed when laboratory α-pinene SOA generated in the presence of high NO concentrations (i.e., denoted as SOAlab,high-NO, using the photolysis of NO2 or nitrous acid as oxidant source) are compared with SOAambient.These observations directly demonstrate the transition of RO2 fate as a function of NO under ambient conditions.For the perturbation experiments performed when ambient NO is below ~0.1ppb, the mass spectra of SOAambient are similar to SOAlab,NO-free, consistent with that RO2 mainly reacts with hydroperoxyl (HO2) or isomerizes.In contrast, for the perturbation experiments performed when ambient NO is above ~0.3ppb, the mass spectra of SOAambient are similar to SOAlab,high-NO, consistent with that the RO2 fate is dominated by NO.This NO level (~0.3ppb) is consistent with the NO level required to dominate the fate of RO2 in the atmosphere, as calculated by using previously measured HO2 and kinetic rate constants (section S8 of Supplement).These observations also directly illustrate that the SOA composition from laboratory studies can be representative of atmosphere.We note that the mass spectra of SOAambient are generally more similar with that of laboratory SOA generated using NO2 photolysis as oxidant source than using nitrous acid photolysis.This suggests that laboratory experiments using NO2 photolysis as oxidant source better represent ambient high NO oxidation conditions in the southeastern U.S. than experiments using nitrous acid do.Possible explanations are discussed in section S7 of Supplement.This finding provides new insights into designing future laboratory experiments to better mimic the oxidations in ambient environments.

Abundance of SOAMT+SQT in the Southeastern U.S.
The ambient perturbation experiments provide direct evidence that the majority of freshly formed SOA from the oxidation of MT and SQT contributes to LO-OOA.Previous studies suggest that the oxidation of β-pinene (another important monoterpene) by nitrate radicals (NO3) contributes to LO-OOA in the southeastern U.S. (Boyd et al., 2015;Xu et al., 2015a), though this reaction alone cannot replicate the magnitude of LO-OOA, particularly during the daytime (Pye et al., 2015).Considering the large biogenic emissions in the southeastern U.S. (Guenther et al., 2012) and the new results from our perturbation experiments, we propose that the major source of LO-OOA in this region is the oxidation of MT and SQT by various oxidants (O3, OH, and NO3).To test this hypothesis, we use CMAQ to simulate pollutant concentrations across the southeastern U.S.
The SOAMT+SQT concentration in the default simulation (i.e., no explicit organic nitrate partitioning, Griffin et al. (1999) photooxidation parameterization) is significantly lower than LO-OOA by 55-84% (Fig. 7).In contrast, SOAMT+SQT in the updated simulation (explicit organic nitrates and Saha and Grieshop (2016) VBS for MT+O3/OH) accurately reproduces the magnitude and diurnal variability of LO-OOA for each site (Fig. 8a).The model bias is reduced to within ~20% for most sites, except for Centreville, Alabama (i.e., 43% for CTR_June dataset).The consistency between modeled SOAMT+SQT and measured LO-OOA at multiple sites and in different seasons supports our hypothesis that LO-OOA largely arises from the oxidation of MT and SQT in the southeastern U.S. Fig. 8b present maps of ground-level SOAMT+SQT concentration corresponding to the time periods of observational data.The SOAMT+SQT concentration is substantially higher in the southeast than other U.S. regions.The SOAMT+SQT is present throughout the year and reaches the largest concentration in summer.The spatial and seasonal variation of SOAMT+SQT concentration is consistent with MT and SQT emissions (Guenther et al., 2012).The annual concentration of SOAMT+SQT in PM2.5 in the southeastern U.S. is ~2.1 µg m -3 (i.e., average concentration over the six sampling periods and over the southeastern U.S. in the updated simulation).This accounts for 21% of World Health Organization PM2.5 guideline (i.e., 10 µg m -3 annual mean) and indicates a significant contributor of environmental risk to the 77 million habitants in the southeastern U.S. Also, the estimated concentration of SOAMT+SQT is substantially higher than represented in current models (Lane et al., 2008;Zheng et al., 2015).The oxidation of MT and SQT is likely an under-estimated contributor to natural PM in pre-industrial period, which determines the baseline state of atmosphere and the estimate of climate forcing by anthropogenic emissions (Carslaw et al., 2013).Models need to improve the description of the MT and SQT oxidation to reduce the uncertainties in estimated OA budget and subsequent climate forcing.
We note that we do not conclude that LO-OOA arises exclusively from MT and SQT, SOA from anthropogenic VOCs may also contribute to LO-OOA.However, the SOA contribution from anthropogenic VOCs is expected to be much smaller than that from biogenic monoterpenes and sesquiterpenes in the southeastern U.S. Firstly, as shown in the perturbation experiments, α-pinene and β-caryophyllene produce more SOA than m-xylene and naphthalene using the same experimental approach in ambient air matrix.Together with weaker emissions of anthropogenic VOCs than biogenic VOCs in the southeastern U.S. (Goldstein et al., 2009), the small contribution to SOA from anthropogenic VOCs is expected.Secondly, as indicated in Fig. S5, the modeled concentration of SOA from anthropogenic VOCs is on the order of 0.1 µg m -3 .Even if we double the SOA yields of anthropogenic VOCs to account for the potential vapor wall loss in laboratory studies (Zhang et al., 2014), the concentration of SOA from anthropogenic VOCs oxidation is still negligible compared to SOAMT+SQT.SOA from anthropogenic VOCs oxidation could be abundant in urban areas of the western U.S.There is evidence that LO-OOA in California is related to the oxidation of anthropogenic VOCs, as radiocarbon analysis suggests 68-75% of carbon in LO-OOA in California stems from fossil sources (Hayes et al., 2013;Zotter et al., 2014).The contribution from anthropogenic VOCs to LO-OOA awaits exploration through ambient perturbation experiments in various locations around the world.

Implications
In this study, we propose that LO-OOA can be used as a surrogate of fresh SOA from MT and SQT in the southeastern U.S., based on the weight of evidence provided by: ( 1 Using LO-OOA as a surrogate of SOAMT+SQT in the southeastern U.S., our ambient ground measurements suggest that at least 19-34% of OA in the southeastern U.S. is from the oxidation of biogenic monoterpenes and sesquiterpenes (Xu et al., 2015a).The fraction of biogenic OA in the southeastern U.S. is even larger if we consider that isoprene-OA could account for 21-36% of OA in summer (albeit potential interferences of SOA from monoterpenes oxidation) and that MO-OOA (24-49% of OA) likely contains SOA from long-term photochemical oxidation of biogenic VOCs.The dominant biogenic origin of SOA poses a challenge to control its burden in the southeastern U.S., if the roles of anthropogenic oxidants and other controlling factors are not recognized.Previous studies have shown that the SOA formation from biogenic VOCs can be mediated by anthropogenic emissions, such as nitrogen oxides and sulfur dioxide (Hoyle et al., 2011;Goldstein et al., 2009;Surratt et al., 2010;Rollins et al., 2012;Xu et al., 2015a).Thus, regulating anthropogenic emissions could help reduce SOA concentration (Lane et al., 2008;Pye et al., 2015;Zheng et al., 2015).For example, as observed in our ambient perturbation experiments, one controlling parameter of α-pinene SOA formation is the concentration of atmospheric oxidants (O3, OH, and NO3), which are known to strongly depend on NOx concentration.As it has been shown that anthropogenic emissions exert complex and non-linear influences on biogenic SOA formation (Zheng et al., 2015), the effectiveness of regulating anthropogenic emissions on biogenic SOA burden requires careful investigations.Importantly, the novel lab-in-the-field perturbation experiments substantially improve our understanding of ambient OA sources.This approach is easily applicable to other regions in the world.Future experiments conducted under various ambient environments and with diverse SOA precursors would facilitate accurate quantification of global OA sources as well as their climate and health impacts.Each perturbation experiment includes four periods: Amb_Bf (~30min), Chamber_Bf (~30min), Chamber_Af (~40min), and Amb_Af (~40min)."Amb" and "Chamber" represent that instruments are sampling ambient and chamber, respectively."Bf" and "Af" stand for before and after perturbation, respectively.The solid lines are measurement data.The dashed red lines are the linear fits of ambient data (i.e., combined Amb_Bf and Amb_Af).The slopes are used to extrapolate Chamber_Bf data to Chamber_Af period (i.e., dashed black lines).The validity of the linearity assumption is discussed in Appendix A. The difference between measurements (i.e., solid lines) and extrapolated Chamber_Bf (i.e., dashed black lines) represents the change caused by perturbation.S1.In panel (a), since the perturbation experiments show that 16% of SOA from α-pinene oxidation is apportioned into isoprene-OA (Fig. S7a), we only include 84% of modeled SOA from MT+O3/OH when comparing with LO-OOA for the sites with isoprene-OA (Fig. S7a).As mentioned above, one critical assumption is that the ambient variation is linear during both the Chamber_Bf and Chamber_Af periods (i.e., when instruments are connected to chamber and not sampling the ambient aerosol) and that the ambient variation can be represented by interpolating Amb_Bf and Amb_Af.We design the following pseudo-experiment to test the validity of this assumption.In brief, we perform the same analysis as we did for the perturbation experiments, but using ambient data only (i.e., no perturbation data).We firstly randomly select a data point, which defines the start point of one pseudo-test.Secondly, based on the start point, we obtain the concentration of OA factors during "Amb_Bf" period, (i.e., from start point to start point + 30min), "Chamber_Bf" period (i.e., from start point + 30min to start point + 60min), "Chamber_Af" period (i.e., from start point + 60min to start point + 100min), and "Amb_Af" period (from start point + 100 min to start point + 130min).This mimics the sampling periods in a real perturbation experiment.Thirdly, we calculate the slope of ambient period (i.e., combined "Amb_Bf" and "Amb_Af" periods) and the slope of chamber period (i.e., combined "Chamber_Bf" and "Chamber_Af" periods).Fourthly, we calculate if the slope of chamber period is significantly different from the slope of ambient period.We repeat this test 1000 times and then obtain the probability of whether the slopes of chamber period and ambient period are significantly different.
Fig. A2a shows the probability that the slopes of chamber period and ambient period are not significantly different for five factors.The larger this probability is, the more reliable the linearity assumption is.The average probability is ~50% for all factors, without discernible diurnal trends.This suggest that there is ~50% chance that the linear variation assumption is valid.Since the linearity assumption is not perfect, we develop another criterion to constrain the potential influence of ambient variation on the interpretation of perturbation results.
Criterion 4: From the above pseudo-experiment on ambient data only, we can calculate the relative change in slope between "chamber period" and "ambient period" by Chamber Amb Amb Slope -Slope relative change in slope = Slope Eqn 1 In each pseudo-experiment test, we calculate a relative change in slope between "chamber period" and "ambient period".By repeating the pseudo-experiment test 1000 times, we obtain a frequency distribution of the relative change in slope for each OA factor (Fig. A2b).This frequency distribution indicates the probability that certain relative change in slope occurs due to ambient variation.Take LO-OOA as an example, the probability that the relative change in slope varies by .Time series of OA in experiment ap_0801_1 to illustrate the analysis method.Each perturbation experiment includes four periods: Amb_Bf (~30min), Chamber_Bf (~30min), Chamber_Af (~40min), and Amb_Af (~40min)."Amb" and "Chamber" correspond to the periods when the instruments are sampling ambient and chamber, respectively."Bf" and "Af" stand for before and after perturbation, respectively.The solid lines are measurement data.The dashed red lines are the linear fit of ambient data (i.e., combined Amb_Bf and Amb_Af).The slope is used to extrapolate Chamber_Bf data to Chamber_Af period (i.e., black dashed line).The dense dashed purple line is the linear fit of the first 8 points during the Chamber_Af period.The sparse dashed purple line is the linear fit of all data points during the Chamber_Af period.During this period, the difference between measurements (i.e., solid green data points) and extrapolated Chamber_Bf (i.e., dashed black line) represents the change in organic concentration caused by perturbation.

Fig
Fig. 4a shows the perturbation-induced changes in the concentrations of OA factors for all α-pinene experiments.Out of 19 experiments, the LO-OOA concentration is enhanced in 14 experiments.Also, among all OA factors, LO-OOA shows the largest enhancement.This directly supports that freshly formed α-pinene SOA contributes to LO-OOA.The enhancement in LO-OOA concentration differs between experiments, mainly because the perturbations were performed at different times of day under different reaction conditions.Although the reaction Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1109Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.
) the large emissions of MT and SQT in this region; (2) the contribution from MT + NO3 to LO-OOA as shown in previous studies; (3) perturbation experiments providing direct evidence that the majority of fresh SOA from the oxidation of MT and SQT contributes to LO-OOA; (4) the consistency of modeled SOAMT+SQT with the magnitude and diurnal trend of LO-OOA at different sites and in different seasons.
Fig. 1.The instrument setup for ambient perturbation experiments.
Fig.3.The time series of OA factors in an -pinene perturbation experiment (Expt ID: ap_0801_1).Each perturbation experiment includes four periods: Amb_Bf (~30min), Chamber_Bf (~30min), Chamber_Af (~40min), and Amb_Af (~40min)."Amb" and "Chamber" represent that instruments are sampling ambient and chamber, respectively."Bf" and "Af" stand for before and after perturbation, respectively.The solid lines are measurement data.The dashed red lines are the linear fits of ambient data (i.e., combined Amb_Bf and Amb_Af).The slopes are used to extrapolate Chamber_Bf data to Chamber_Af period (i.e., dashed black lines).The validity of the linearity assumption is discussed in Appendix A. The difference between measurements (i.e., solid lines) and extrapolated Chamber_Bf (i.e., dashed black lines) represents the change caused by perturbation.

Fig. 6 .
Fig.6.The correlation coefficients between the mass spectra of OA formed in laboratory under different NO conditions ("SOAlab") and those of OA formed in ambient α-pinene perturbation experiments ("SOAambient").The subscripts "lab" and "ambient" indicate the SOA formed under laboratory conditions and ambient conditions, respectively.Three different oxidant sources (i.e., H2O2, HONO, and NO2) are used to create different NO concentrations in laboratory studies.The mass spectra of "SOAambient" are calculated by comparing the mass spectra of OA during Chamber_Af and those of extrapolated Chamber_Bf (section S7 of Supplement).To calculate reliable mass spectra of "SOAambient", only the experiments with significant OA enhancement are analyzed and shown here (Appendix A).The x-axis is the average NO concentration during each perturbation experiment.The data points on the same vertical line (i.e., the same NO concentration) are from the same perturbation experiment, but compared to three different laboratory experiments.The dashed lines are used to guide eyes.The bars on top of the figure represent the 10 th , 50 th , and 90 th percentiles of NO concentration for CTR (Centreville, AL), YRK (Yorkville, GA), and JST (Jefferson Street, GA) in 2013.The NO concentration is measured by the SouthEastern Aerosol Research and Characterization (SEARCH) network.The 90 th percentile of NO concentration in JST is 14.8 ppb, which is not shown in the figure.
Fig. 8. (a) top panel: the diurnal trends of LO-OOA and modeled SOA from monoterpenes and sesquiterpenes (SOAMT+SQT) at different sampling sites in the southeastern U.S. (b) bottom panel: maps of modeled ground-level SOAMT+SQT concentration.Model results shown here are from the updated simulation.Abbreviations correspond to Centreville (CTR), Jefferson Street (JST), Yorkville (YRK), Georgia Institute of Technology (GT).Detailed sampling periods are shown in TableS1.In panel (a), since the perturbation experiments show that 16% of SOA from α-pinene oxidation is apportioned into isoprene-OA (Fig.S7a), we only include 84% of modeled SOA from MT+O3/OH when comparing with LO-OOA for the sites with isoprene-OA (Fig.S7a).The mean bias (MB), mean error (ME), and normalized mean bias (NMB) for each site are shown in each panel.The slopes and correlation coefficients (R) are obtained by least square fit.The error bars indicate the standard error.In panel (b), average SOAMT+SQT concentration in PM2.5 during each sampling period is reported.
Fig. 8. (a) top panel: the diurnal trends of LO-OOA and modeled SOA from monoterpenes and sesquiterpenes (SOAMT+SQT) at different sampling sites in the southeastern U.S. (b) bottom panel: maps of modeled ground-level SOAMT+SQT concentration.Model results shown here are from the updated simulation.Abbreviations correspond to Centreville (CTR), Jefferson Street (JST), Yorkville (YRK), Georgia Institute of Technology (GT).Detailed sampling periods are shown in TableS1.In panel (a), since the perturbation experiments show that 16% of SOA from α-pinene oxidation is apportioned into isoprene-OA (Fig.S7a), we only include 84% of modeled SOA from MT+O3/OH when comparing with LO-OOA for the sites with isoprene-OA (Fig.S7a).The mean bias (MB), mean error (ME), and normalized mean bias (NMB) for each site are shown in each panel.The slopes and correlation coefficients (R) are obtained by least square fit.The error bars indicate the standard error.In panel (b), average SOAMT+SQT concentration in PM2.5 during each sampling period is reported.
Fig.A1.Time series of OA in experiment ap_0801_1 to illustrate the analysis method.Each perturbation experiment includes four periods: Amb_Bf (~30min), Chamber_Bf (~30min), Chamber_Af (~40min), and Amb_Af (~40min)."Amb" and "Chamber" correspond to the periods when the instruments are sampling ambient and chamber, respectively."Bf" and "Af" stand for before and after perturbation, respectively.The solid lines are measurement data.The dashed red lines are the linear fit of ambient data (i.e., combined Amb_Bf and Amb_Af).The slope is used to extrapolate Chamber_Bf data to Chamber_Af period (i.e., black dashed line).The dense dashed purple line is the linear fit of the first 8 points during the Chamber_Af period.The sparse dashed purple line is the linear fit of all data points during the Chamber_Af period.During this period, the difference between measurements (i.e., solid green data points) and extrapolated Chamber_Bf (i.e., dashed black line) represents the change in organic concentration caused by perturbation.
Fig. B1.The mass spectra and time series of OA factors in the 2015 acidic sulfate particle perturbation measurements.Note that the perturbation periods are included in the time series.
. Phys.Discuss., https://doi.org/10.5194/acp-2017-1109Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.1146 Fig. B2.The statistically significant changes in the concentrations of OA factors after perturbation 1147 by acidic sulfate particles.The experiments are sorted by perturbation time.The changes in 1148 concentration are the difference between measurements during the Chamber_Af period and mass 1149 concentration extrapolated from the Chamber_Bf period.A set of criteria are developed to evaluate 1150 if the changes are significant and if the changes are due to ambient variation (Appendix A).H2O-1151 Org factor in these sulfate perturbation experiments represents organic contaminations in 1152 .Phys.Discuss., https://doi.org/10.5194/acp-2017-1109Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.all pnts in Bag_Af Linear fit of first 8 pnts in Bag_Af Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2017-1109Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 2 January 2018 c Author(s) 2018.CC BY 4.0 License.
a b c Average concentration during the Chamber_Af period.Atmos