Decadal changes in summertime reactive oxidized nitrogen and surface ozone over the Southeast United States

Widespread efforts to abate ozone (O3) smog have significantly reduced emissions of nitrogen oxides (NOx) over the past 2 decades in the Southeast US, a place heavily influenced by both anthropogenic and biogenic emissions. How reactive nitrogen speciation responds to the reduction in NOx emissions in this region remains to be elucidated. Here we exploit aircraft measurements from ICARTT (July– August 2004), SENEX (June–July 2013), and SEAC4RS (August–September 2013) and long-term ground measurement networks alongside a global chemistry–climate model to examine decadal changes in summertime reactive oxidized nitrogen (RON) and ozone over the Southeast US. We show that our model can reproduce the mean vertical profiles of major RON species and the total (NOy) in both 2004 and 2013. Among the major RON species, nitric acid (HNO3) is dominant (∼ 42–45 %), followed by NOx (31 %), total peroxy nitrates (6PNs; 14 %), and total alkyl nitrates (6ANs; 9–12 %) on a regional scale. We find that most RON species, including NOx , 6PNs, and HNO3, decline proportionally with decreasing NOx emissions in this region, leading to a similar decline in NOy . This linear response might be in part due to the nearly constant summertime supply of biogenic VOC emissions in this region. Our model captures the observed relative change in RON and surface ozone from 2004 to 2013. Model sensitivity tests indicate that further reductions of NOx emissions will lead to a continued decline in surface ozone and less frequent high-ozone events.


Introduction
Since the 1990s, the US Environmental Protection Agency (US EPA) has targeted emissions of nitrogen oxides (NO x ) to improve air quality by lowering regional photochemical smog (the 1990 Clean Air Amendment).Satellite-and ground-based observations imply significant declines in US NO x emissions, with a decreasing rate of roughly −4 % yr −1 after 2005 (Krotkov et al., 2016;Russell et al., 2012;Tong et al., 2015;Miyazaki et al., 2017;Lu et al., 2015;Lamsal et al., 2015).This has proven effective at lowering near-surface ozone (O 3 ) in the past few decades (Cooper et al., 2012;Simon et al., 2015;Hidy and Blanchard, 2015;Stoeckenius et al., 2015;Xing et al., 2015;Yahya et al., 2016;Astitha et al., 2017).The average of the annual fourth highest daily maximum 8 h average (MDA8) ozone over 206 sites has decreased by 31 % from 101 ppb in 1980 to 70 ppb in 2016 across the continental US, with more significant reductions in rural areas of the eastern US in summer (Simon et al., 2015;Cooper et al., 2012).Here we use both aircraft-and ground-based datasets, combined with a high-resolution chemistry-climate model, to evaluate responses of reactive oxidized nitrogen (RON) and surface ozone to the NO x emission reductions in the Southeast US.
In the troposphere, ozone is produced through photochemical reactions involving NO x and volatile organic compounds (VOCs) in the presence of sunlight.During photooxidation, a large fraction of NO x is transformed into its reservoirs, including nitric acid (HNO 3 ), peroxy nitrates (RO 2 NO 2 ; dominated by peroxyacetyl nitrate, or PAN), and alkyl nitrates (RONO 2 ).These species, together with NO x , are known as total reactive oxidized nitrogen (NO y = NO x + HNO 3 + HONO + 2 × N 2 O 5 + total peroxy nitrates, or PNs + total alkyl nitrates, or ANs).Some of these reservoir species, particularly those with an organic component, tend to be less soluble and longer lived.They may carry reactive nitrogen far from the NO x source region (Stohl et al., 2002;Parrish et al., 2004;Li et al., 2004) and thereby affect NO x concentrations and ozone formation on a regional to global scale (Liang et al., 1998;Horowitz et al., 1998;Perring et al., 2013;Paulot et al., 2016;Hudman et al., 2004).
Extensive datasets in the Southeast US offer a great opportunity to study the decadal changes in RON and surface ozone resulting from NO x emission decline.Aircraft campaigns during the summers of 2004 and 2013, including the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT; Fehsenfeld et al., 2006;Singh et al., 2006), the Southeast Nexus (SENEX; Warneke et al., 2016), and the Studies of Emissions and Atmospheric Composition, Clouds and Climate Coupling by Regional Surveys (SEAC 4 RS; Toon et al., 2016), provide detailed characterization of tropospheric composition in this region separated by nearly a decade.These data have been widely used to evaluate model estimates of RON and ozone (Singh et al., 2007;Pierce et al., 2007;Perring et al., 2009;Fischer et al., 2014;Hudman et al., 2007;Henderson et al., 2011;Hudman et al., 2009;Edwards et al., 2017;Baker and Woody, 2017;Travis et al., 2016;Mao et al., 2013b;Fisher et al., 2016;Yu et al., 2016;Liu et al., 2016).Together with measurements from surface networks, including the National Atmospheric Deposition Program (NADP) and EPA Air Quality System (AQS), these datasets enable a close examination of responses of RON and surface ozone to NO x emission reductions in this region.
Here we use a high-resolution global 3-D chemistryclimate model, the Geophysical Fluid Dynamics Laboratory (GFDL) AM3 model with updated isoprene and organic nitrate chemistry, to investigate decadal changes in summertime RON and surface ozone between 2004 and 2013 over the Southeast US.We first evaluate the model with comprehensive measurements from three aircraft campaigns in the summers of 2004 (ICARTT) and 2013 (SENEX and SEAC 4 RS).Model estimates of nitrate wet deposition flux are also evaluated against measurements from NADP; model estimates of NO y are compared with measurements from EPA AQS to provide an additional constraint on the fate of RON in the model.We then investigate the repartitioning of RON in response to NO x emission reductions from 2004 to 2013 on a regional scale.From there, we examine model estimates of decadal changes in summertime surface ozone at 157 EPA AQS monitoring sites over the Southeast US.We also demonstrate the sensitivity of RON and MDA8 ozone to a hypothetical NO x emission reduction over the next decade.

AM3 Model
We apply a high-resolution (50 × 50 km 2 ) version of the GFDL AM3 global chemistry-climate model to study decadal changes in RON and ozone over the Southeast US.Chemistry-climate models provide a unique capability to both evaluate model representation of these observed changes and use that to improve future projections of air quality in the same region.The model configuration is to a large extent similar to that used in another paper (Li et al., 2016) and a short summary is provided below.The dynamical core, physical parameterizations, cloud and precipitation processes, and cloud-aerosol interactions mainly follow Donner et al. (2011), except that convective plumes are computed on a vertical grid with finer resolution (Paulot et al., 2016).Dry deposition in the model has been updated to use dry deposition velocities calculated in the GEOS-Chem model (Paulot et al., 2016) to reflect the rapid deposition of organic nitrates and oxidized volatile organic compounds (OVOCs; Nguyen et al., 2015).The current time step for chemistry and transport in our model is 20 min.We show below in Sect.3.1 that with the current setting our model can reproduce the vertical profiles of RON.The sensitivity of RON to operator duration can be found in Philip et al. (2016).
Isoprene emissions are computed in the model using the Model of Emissions of Gases and Aerosols from Nature (MEGAN).In 2004, isoprene emissions over the continental US (25-50 • N, 130-70 • W) are computed to be 8.0 Tg C in July and August together, with a previous model estimate of 7.5 Tg C by Mao et al. (2013b).In 2013, model estimates of isoprene emissions were scaled down by 20 % following Li et al. (2016).The resulting isoprene emissions are 7.7 Tg C in July-August in this region, with little difference compared to 2004.Monoterpene emissions follow Naik et al. (2013) and do not vary interannually, with a total of 4.0 Tg C in July and August.
Anthropogenic emissions follow the Representative Concentration Pathway 8.5 (RCP8.5)projection (Lamarque et al., 2011) for both 2004 and 2013 to compare the model to observations in a consistent fashion and also enable the future projection of air quality in this region.As shown in Table 1, anthropogenic NO x emissions over the continental US during July-August 2004 amount to 0.42 Tg N month −1 , which is consistent with Hudman et al. (2007) but 11 % lower than the EPA estimate of 0.47 Tg N month −1 (Granier et al., 2011).For the year 2013, we apply a 25 % reduction to anthropogenic NO x emissions from the RCP8.5 projection (from base year of 2010) to best reproduce the vertical profiles of RON during SENEX as shown below in Sect.3.1.This adjustment is also consistent with recent estimates of NO x emissions over the Southeast US (Anderson et al., 2014).The resulting anthropogenic NO x emissions (0.25 Tg N month −1 ) are 14 % lower than the 2011 EPA national emission inventory (NEI11v1) estimate of 0.29 Tg N month −1 (0.28 Tg N month −1 from the updated NEI11v2 emission inventory), although both inventories have a similar spatial distribution (Fig. S1 in the Supplement).We also apply a diurnal variation to anthropogenic NO x emissions following Mao et al. (2013b).Soil NO x emissions in our model, 3.6 Tg N yr −1 globally (Naik et al., 2013), are considerably lower than other model estimates, including 5.5 Tg N yr −1 in Yienger and Levy (1995) and 9.0 Tg N yr −1 in Hudman et al. (2012).As a result, anthropogenic NO x emissions over the continental US are 0.84 Tg N for July-August 2004, and 0.50 Tg N in July-August 2013, with a 40 % reduction from 2004 to 2013 (Table 1).This relative change in anthropogenic NO x emissions is consistent with EPA estimates (https://www.epa.gov/air-emissions-inventories/air-pollutant-emissions-trends-data) and satellite observations (Krotkov et al., 2016;Lu et al., 2015).Compared to the NEI11v1 inventory, RCP8.5 used in our model shows similar relative differences in the Southeast US and nationally.

Gas-phase chemistry
We apply the same isoprene mechanism as described by Li et al. (2016).A full list of the reactions can be found in Table S1.This mechanism is based on Mao et al. (2013b), but has been significantly revised to incorporate recent laboratory updates on isoprene oxidation by OH, ozone, and NO 3 (Schwantes et al., 2015;Bates et al., 2014Bates et al., , 2016;;Peeters et al., 2014;St. Clair et al., 2016;Praske et al., 2015;Müller et al., 2014;Lee et al., 2014;Crounse et al., 2011).One major feature is the suppression of the δ-isoprene hydroxy peroxy radical (δ-ISOPO 2 ) and subsequent reaction pathways in the model, as these channels are considered to be of minor importance under ambient conditions (Peeters et al., 2014;Bates et al., 2014).The fraction of ISOPO 2 undergoing isomerization is calculated using bulk isomerization estimates (Crounse et al., 2011).As a result, the firstgeneration isoprene alkyl nitrate is assumed to be β-hydroxy nitrate (ISOPNB) in the model with a yield of 10 % from the ISOPO 2 + NO pathway.This differs from a recent GEOS-Chem study of organic nitrates over the Southeast US that assumed a 9 % yield of the first-generation isoprene alkyl nitrate comprised of 90 % ISOPNB and 10 % δ-hydroxy nitrate (ISOPND; Fisher et al., 2016).The treatment of βand δ-ISOPO 2 will not only affect the speciation of organic nitrates but also the production of ozone due to different NO x recycling efficiency in their secondary products.We also include updated chemistry for methylvinyl ketone (MVK; Praske et al., 2015), an updated yield of hydroxy hydroperoxides (ISOPOOH; Bates et al., 2016;St. Clair et al., 2016), fast photolysis of carbonyl organic nitrates (Müller et al., 2014), and an updated ozonolysis rate of ISOPNB (Lee et al., 2014).
In addition, we reduce the yield of organic nitrates (MACRN) from methacrolein (MACR) oxidation from 15 to 3 %, which is estimated from the measured yield of nitrates from MVK oxidation (Praske et al., 2015).
Another major model revision involves the treatment of the nighttime oxidation of isoprene.Instead of following Mao et al. (2013b), we revise the nighttime oxidation of isoprene largely based on the Leeds Master Chemical Mechanism v3.2 (MCM v3.2), allowing for a more complete description of isoprene oxidation by NO 3 .In particular, MCM v3.2 suggests significant production of propanone nitrate (PROPNN) from the photooxidation of the C 5 carbonyl nitrate, which is consistent with recent laboratory experiments (Schwantes et al., 2015).We also update the products of the reaction of the nitrooxy alkylperoxy radical (INO 2 ), the peroxy radical from isoprene oxidation by NO 3 , with HO 2 to reflect a lower molar yield (0.77) of C 5 nitrooxy hydroperoxide (INPN; Schwantes et al., 2015).The differences between MCM v3.2 and the most updated version, MCM v3.3.1, in isoprene nighttime chemistry appears to be small (Jenkin et al., 2015).We therefore use MCM v3.2 as the reference in this work.
We include a highly simplified chemistry for the oxidation of monoterpenes in this work, mainly to quantify their contribution to organic nitrates.Monoterpenes are lumped into one chemical species (C 10 H 16 ) in our model.The organic nitrate yield is set to 26 % from OH-initiated oxidation (Rindelaub et al., 2015) and to 10 % from NO 3 -initiated oxidation (Browne et al., 2014).Details of the monoterpene chemistry can be found in Table S2 in the Supplement.
In the condensed phase, organic nitrates can undergo hydrolysis reactions producing HNO 3 (Darer et al., 2011;Hu et al., 2011;Rindelaub et al., 2015;Boyd et al., 2015;Szmigielski et al., 2010;Sato, 2008;Jacobs et al., 2014;Bean and Hildebrandt Ruiz, 2016).However, the hydrolysis rate varies greatly with structures of organic nitrates (Bean and Hildebrandt Ruiz, 2016;Darer et al., 2011;Hu et al., 2011;Boyd et al., 2015;Rindelaub et al., 2016).Here we assume a firstorder irreversible reactive uptake for the heterogeneous loss of organic nitrates onto aerosols (Reaction R1), followed by their hydrolysis reaction returning HNO 3 and alcohols (Reaction R2; Fisher et al., 2016): where RONO 2 , AONJ, and ROH represent gas-and particlephase organic nitrates and alcohols, respectively.We allow the heterogeneous loss of organic nitrates to sulfate, black carbon, primary organic carbon, sea salt, mineral dust, and SOA following Mao et al. (2013a).Besides the base case that only includes ISOPNB for heterogeneous loss (Jacobs et al., 2014), we include two additional sensitivity tests to evaluate the potential impact of organic nitrate hydrolysis.One is the "hydro_full" case including the heterogeneous loss of a C 5 dihydroxy dinitrate (DHDN) and monoterpene nitrates only from OH oxidation during daytime (TERPN1; nighttime monoterpene nitrates are excluded), and the other is the "no_hydro" case assuming no heterogeneous loss for any organic nitrate.We adopt an effective uptake coefficient of 0.005 for ISOPNB and DHDN and 0.01 for TERPN1, following Fisher et al. (2016), with a 3 h bulk lifetime in the particle phase (Pye et al., 2015;Lee et al., 2016; Table S3 in the Supplement).The details of each case are listed in Table 2.

Observational datasets
We use measurements from a series of field campaigns (2004  ICARTT, 2013 SENEX, and 2013 SEAC 4 RS) to evaluate model performance on ozone, NO x , HNO 3 , PAN, ANs, and NO y over the Southeast US in summer.
The ICARTT aircraft campaign provided a detailed characterization of tropospheric chemistry over the eastern US in the summer of 2004 (1 July-15 August 2004).Two aircraft, the NASA DC-8 and the NOAA WP-3D, were deployed to collect measurements of ozone, RON, isoprene, and its oxidation products.Here we focus on data including ozone, NO x , HCHO (tunable diode laser (TDL) absorption spectrometry), HNO 3 (mist chamber-ion chromatograph by the University of New Hampshire and chemical ionization mass spectrometry (CIMS) by the California Institute of Technology), PAN, and ANs (including gas and aerosol RONO 2 ) collected on the NASA DC-8 aircraft over the Southeast US.Details of the instrument operation and accuracy are summarized in Singh et al. (2006) and the references therein.
Two aircraft campaigns were conducted in the summer of 2013 over the Southeast US.The first is the NOAA SENEX campaign using NOAA WP-3D aircraft to investigate the interaction between biogenic and anthropogenic emissions and the formation of secondary pollutants (27 May-10 July 2013).We focus on daytime measurements of ozone, NO x , HNO 3 , PAN, speciated RONO 2 , and NO y in this work.Details of the instrument operation and accuracy are summarized in Warneke et al. (2016) and the references therein.The second is the NASA SEAC 4 RS campaign, which took place in August-September 2013, with a focus on vertical transport of atmospheric pollutants from the surface to the stratosphere.Here we focus on observations of ozone, NO 2 , HCHO (laser-induced fluorescence, or LIF), ANs (including gas and aerosol RONO 2 ), and speciated RONO 2 collected on the NASA DC-8 aircraft to evaluate model representation of ANs and RONO 2 originating from isoprene oxidation.Details of the instrument operation and accuracy are summarized in Toon et al. (2016) and the references therein.
Besides these aircraft campaigns, we also use surface observations for model evaluation, including nitrate (NO − 3 ) wet deposition flux and concentration from the National Trends Network (NTN) of NADP (accessible at http://nadp.sws.uiuc.edu/data/)and surface ozone and NO y from EPA AQS (accessible at https://aqs.epa.gov/aqsweb/documents/data_mart_welcome.html).We focus on NO − 3 wet deposition fluxes at 53 NADP sites and MDA8 ozone at 157 EPA AQS sites (Fig. S2 in the Supplement) in the Southeast US during July-August 2004 and 2013.NO y measurements at 10 out of the 157 AQS sites in the same episodes are compared with model estimates as an additional constraint on the decadal changes in NO y .We choose July-August as our "summer" since this is the common period of all the measurements used in the model evaluation.

Model evaluation
We evaluate our model against observations from aircraft campaigns in 2004 and 2013.For each of the three field campaigns, all measurements are averaged to a 1 min time resolution.Data from biomass burning (CH 3 CN ≥ 225 ppt or HCN ≥ 500 ppt), urban plumes (NO 2 ≥ 4 ppb or NO x / NO y ≥ 0.4 if NO y is available), and stratospheric air (O 3 / CO > 1.25 mol mol −1 ) are excluded (Hudman et al., 2007) in all the analyses, as these subgrid processes may not be well represented in our model.We focus on the Southeast US region using data within the domain of 25-40 • N latitude and 100-75 • W longitude for our analyses.A map of all the flight tracks of each campaign is shown in Fig. S3 in the Supplement.All model results are sampled along the flight track with 1 min time resolution.

Mean vertical profiles of ozone and RON
Figure 1 shows the observed and modeled mean vertical profiles of ozone, NO x , HNO 3 , PAN, ANs, and NO y during ICARTT and SENEX.We use ANs measurements from SEAC 4 RS to evaluate model performance during summer 2013 due to the lack of ANs measurements from SENEX.Our model results include both gas and aerosol RONO 2 in ANs, although aerosol RONO 2 accounts for 7-11 % of ANs in the planetary boundary layer (PBL, < 1.5 km).We do not consider inorganic nitrates in the particle phase in this analysis due to the lack of a thermodynamic model for inorganic aerosols in the current version of AM3.This simplification is expected to have minimal effects, as they only account for a small fraction of aerosol nitrates in the Southeast US (Ng et al., 2017).To investigate the impact of RONO 2 hydrolysis, we include two model simulations: the base case with heterogeneous loss of ISOPNB and a sensitivity run, "no_hydro", without heterogeneous loss of organic nitrates.
Mean observed ozone in the surface layer decreased from 50 ppb during ICARTT to 35 ppb during SENEX, which is consistent with the declining trend in surface MDA8 ozone at AQS monitoring sites (Sect.4.2).As we show in Sect.4.2, this decline in ozone is mainly driven by NO x emission reduction, with little influence by meteorology in the two years.Our model can reproduce the vertical gradient and the relative change in ozone from 2004 to 2013, except for a positive absolute bias of 6-12 ppb in the boundary layer.The performance statistics of ozone in the boundary layer listed in Table S4 also indicate positive biases in the model, with a fractional bias (FB) of 9.4-17 %, fractional error (FE) of 16-19 %, normalized mean bias (NMB) of 9.4-16 %, and normalized mean error (NME) of 16-19 %.This overestimate of ozone is higher than that reported (3-5 ppb) by Mao et al. (2013b) for their simulation of the ICARTT dataset, likely due to faster photolysis of carbonyl nitrates that increases the NO x recycling efficiency from isoprene oxidation.
We further examine mean vertical profiles of NO x and its reservoirs in 2004 and 2013 (Fig. 1).In the boundary layer along the flight tracks, HNO 3 is the most abundant RON, accounting for 40-46 % of NO y , followed by NO x (18-23 %), PAN (20 %), and ANs (11-21 %).Between 2004 and 2013, mean observed NO y in the boundary layer decreased by 20 %, from 2.0 to 1.6 ppb, a weaker change than the 35 % reduction of total NO x emissions (Table 1).The responses of major RON species are mostly proportional to the change in NO x emissions, with the notable exception of ANs.We find significant decreases in NO x (−35 %) and HNO 3 (−29 %) as well as a slight decrease in PAN (−13 %) from observations.The relative trends of HNO 3 and PAN are opposite to those found in the Los Angeles (LA) basin, where PAN decreased much faster than HNO 3 (Pollack et al., 2013).This difference results mainly from the rapid decrease in anthropogenic VOC emissions in the LA basin that also serve as major precursors of PAN.In contrast, isoprene is the major precursor of PAN over the Southeast US.Its emissions show a constant supply (within 5 % differences over the two summers) in this region.Showing a different trend from the above compounds, ANs increase from 0.23 to 0.27 ppb (+17 %) near the surface.As we show below in Sect.4.1, these changes (except for ANs) are mostly consistent with model estimates on regional average.The discrepancy in their trends of vertical profiles and regional average might be due to representative errors from the three aircraft campaigns on spatial (Fig. S3 in the Supplement) and temporal (different episodes; refer to the observation data description in Sect.2.4) scales.
The model can reproduce RON species in the boundary layer but tends to underestimate them in the free troposphere.This is likely due to insufficient production of NO x from lightning in the free troposphere in our model, which is 0.048 Tg N in total over North America during July-August of 2004, lower by almost a factor of 5 compared to the value (0.27 Tg N from 1 July-15 August 2004) reported by Hudman et al. (2007).This underestimate can be improved by scaling up lightning emission by a factor of 5-10 (Fang et al., 2010).We do not adjust the lightning NO x emission in this work due to its high uncertainty (Ott et al., 2010;Pickering et al., 1998).
The hydrolysis of organic nitrates affects RONO 2 significantly in the boundary layer.By introducing the hydrolysis of ISOPNB, we find that the model relative bias of ANs is reduced from +20 to +2 % during ICARTT (Fig. 1).The performance metrics in Table S4 in the Supplement also indicate better agreement of the model with observations if hydrolysis of ISOPNB assumed.However, the relative bias is increased in magnitude from −9 to −24 % during SEAC 4 RS.This low bias can be partially due to neglecting small alkyl nitrates, which could contribute 20-30 ppt to ANs (less than 10 % near the surface) during SEAC 4 RS (Fisher et al., 2016).Including small alkyl nitrates will increase modeled ANs a bit in ICARTT as well.The hydrolysis of ISOPNB also leads to a slight increase in HNO 3 (Table S4 in the Supplement).The impact of the hydrolysis of ISOPNB on boundary layer ozone appears to be small.This is mainly because without hydrolysis, the dominant loss of ISOPNB is oxidation by OH, which then leads to the formation of secondary organic nitrates including MVKN, MACRN, and DHDN.The majority of these organic nitrates (MVKN and DHDN) return NO x slowly due to their long lifetimes (Table S5 in the Supplement), resulting in a similar effect on ozone production as the hydrolysis of ISOPNB.In addition to the good agreement of ANs, our model shows good agreement with speciated RONO 2 measured during SENEX and SEAC 4 RS, including ISOPN and the sum of MVKN and MACRN (Fig. 2).We find that the large discrepancy between ANs and speciated alkyl nitrates (Fig. S4 in the Supplement) can be explained  by a combination of monoterpene nitrates and DHDN and nighttime NO 3 oxidation products from isoprene, accounting for 20-24, 14-17, and 23-29 % of ANs, respectively, in the boundary layer.
Given the good agreement between observed and modeled RON in both 2004 and 2013, we find that the ozone bias shown in Fig. 1 cannot be completely explained by an overestimate of anthropogenic NO x emissions.A recent GEOS-Chem study (Travis et al., 2016) shows that the ozone bias can be largely reduced by scaling down anthropogenic NO x emissions.We find that a similar reduction of anthropogenic NO x emissions in 2013, from 0.25 Tg N month −1 to 0.15 Tg N month −1 , would lead to an underestimate of NO y , HNO 3 , and PAN by 30, 33, and 30 %, respectively.Such a reduction would also be inconsistent with the relative changes in EPA estimates of NO x emissions shown above.Indeed, other processes, such as ozone dry deposition, may also contribute to this bias and warrant further investigation.

NO −
3 wet deposition flux and concentration  Since errors in modeled precipitation could strongly affect the modeled NO − 3 wet deposition flux (Appel et al., 2011;Grimm and Lynch, 2005;Metcalfe et al., 2005;Paulot et al., 2014;Tost et al., 2007), we also evaluate the modeled NO − 3 concentration (C p,i ), which is calculated by using the modeled NO − 3 wet deposition flux (F p,i ) and observed precipitation (P o,i ; C p,i = F p,i /P o,i ), as a separate constraint.The model shows a similar declining trend from the observations with a relative bias of −23 and −41 % in NO − 3 concentration for 2004 and 2013, respectively.Our results are consistent with the base case of Paulot et al. (2016), which showed that convective removal is likely insufficient in AM3, leading to underestimates of both NO − 3 wet deposition flux and concentrations.Our results are somewhat different from a recent GEOS-Chem study (Travis et al., 2016).The authors found that reducing anthropogenic NO x emissions from NEI11v1 by 53 % can significantly improve the overestimate of 71 % for NO − 3 wet deposition flux in their model during August-September of 2013.A further reduction of anthropogenic NO x emissions in our model (to 0.15 Tg N month −1 ), as suggested by Travis et al. (2016), would lead to an even greater negative bias compared to observations.

RONO 2 and related species
We further evaluate RONO 2 and related species in this section, with a large focus on measurements from ICARTT and SEAC 4 RS.The major pathway for the production of daytime RONO 2 is the reaction of NO with RO 2 originating from VOC oxidation by OH: where α is the branching ratio for alkyl nitrate formation.NO 2 subsequently undergoes photolysis to produce ozone: For isoprene, α is 9 ± 4 % (for ISOPN) according to a recent study (Xiong et al., 2015).For monoterpenes, specifically αpinene, α ranges from 1 to 26 % (Rindelaub et al., 2015;Nozière et al., 1999;Aschmann et al., 2002).Here, we use 10 % for isoprene and 26 % for monoterpenes.As RONO 2 and ozone are both produced from Reaction R4, a correlation between them is expected.We show that the model can roughly reproduce the correlation of O x (= O 3 + NO 2 ) vs. ANs during both ICARTT and SEAC 4 RS (Fig. 4), although the slope has a positive relative bias of about 21 and 33 %, respectively, largely due to an overestimate of ozone in the model.The good agreement between observed and modeled O x vs. daytime RONO 2 provides additional support for our treatment of the yields and fate of these daytime isoprene nitrates.
Another metric to evaluate RONO 2 chemistry is the correlation between ANs and HCHO, as the latter is a coproduct from Reaction R4.We show in Fig. 4 that the model can roughly capture the observed ANs-HCHO slope, with an underestimate by 25 and 13 % during ICARTT and SEAC 4 RS, respectively.The underestimate is in part due to small alkyl nitrates that are neglected in the model, as mentioned in Sect.3.1.During ICARTT, the slope estimated by AM3 is 0.12, similar to the value (0.15) from a previous GEOS-Chem study using a different isoprene oxidation mechanism that assumed a higher α (of 4.7 % from ISOPNB and 7.0 % from ISOPND vs. 10 % of ISOPNB and zero ISOPND in AM3) and a lower yield of HCHO (66 % vs. 90 % in AM3; Mao et al., 2013b).The reason for such similarity between the two models might be twofold: (a) the additional contribution of monoterpene nitrates to ANs in AM3 compensates for the decrease in α from isoprene nitrates compared to GEOS-Chem and (b) the coarse grid resolution of GEOS-Chem simulations (2 • × 2.5 • ) may lead to a higher estimate of HCHO compared to the result from a finer grid resolution (Yu et al., 2016).
Since HCHO can be produced from pathways of isoprene hydroxy peroxy radicals (ISOPO 2 ) other than Reaction R4 (such as isomerization of ISOPO 2 and ISOPO 2 + HO 2 ), changes in the slope of ANs vs. HCHO may help to quantify decadal changes in isoprene oxidation pathways.We find in Fig. 4 that the observed slope of ANs-HCHO shows very little change from 2004 to 2013.This is in part due to substantial HCHO production from isoprene oxidation un-  der low NO x conditions (Li et al., 2016) and in part due to the buffering of ANs in response to decreasing NO x , as shown below in Sect.4.1.Our model is able to reproduce such behavior.We also find that the branching ratios for the reactions of ISOPO 2 change marginally from 2004 to 2013 over the Southeast US (Fig. S5 in the Supplement).The frac-tion of ISOPO 2 + NO decreases from 81 % in 2004 to 66 % in 2013.The fraction of ISOPO 2 + HO 2 increases from 15 to 28 %, and the fraction of ISOPO 2 isomerization increases from 4 to 6 %.Our results are slightly different from the results of GEOS-Chem, which found a lower contribution from the NO pathway (54 %) and a higher contribution from isomerization (15 %) during August-September of 2013 (Travis et al., 2016).
We also compare the correlation between major daytime isoprene nitrates and HCHO during 2013, which provides a constraint on the yield of these nitrates.Our model shows a slight overestimate on the slope (Fig. 4d), which is consistent with the comparison of mean vertical profiles shown in Fig. 2. The computed slope (5 %) in this study is different from that (2.5 %) of a recent GEOS-Chem simulation by Fisher et al. (2016).This is partially due to the different treatment of βand δ-ISOPO 2 between GEOS-Chem and AM3.Another factor is that MVKN and MACRN are not allowed to hydrolyze in AM3, leading to a higher abundance of these two nitrates.
Figure 5 shows the mean vertical profiles of modeled monoterpene nitrates (MNs) and isoprene nitrates (INs) during ICARTT and SEAC 4 RS.INs are the most abundant RONO 2 , accounting for 76-80 % below 3 km over the Southeast US.In the measurements, ISOPN + MVKN + MACRN only contributes one-third of ANs (Fig. S4 in the Supplement).We show below that the discrepancy between ANs and speciated RONO 2 can be explained by other daytime and nighttime INs and MNs in the model.More than 60 % of modeled INs originate from isoprene oxidation during daytime.The first-generation nitrate ISOPN contributes slightly more (31 %) than the second-generation nitrates MVKN + MACRN (28 %) to the total daytime INs during ICARTT.This is different from Mao et al. (2013b), who showed a higher contribution of MVKN + MACRN than the firstgeneration INs due to the different treatment of βand δ-ISOPO 2 .We see more ISOPN (32 %) than MVKN + MACRN (26 %) from the daytime INs during SEAC 4 RS, which is consistent with Fisher et al. (2016).A large uncertainty in our model is attributed to DHDN, which contributes 32 % to the daytime INs.Fisher et al. (2016) showed less DHDN during SEAC 4 RS since it was removed rapidly by hydrolysis (1 h lifetime) in their model.Our sensitivity test (hydro_full; Fig. S6 in the Supplement) indicates that AM3 would significantly underestimate ANs if we assume a similar heterogeneous loss of DHDN as ISOPN.In fact, DHDN was hypothesized originally in Lee et al. (2014) for the imbalance of nitrogen in their lab experiments, and may serve as a proxy for a large number of unidentified daytime INs.It remains unclear what the dominant loss of DHDN is.Daytime nitrates from monoterpene oxidation are another important source of ANs in this region, accounting for 17-20 % (24-26 ppt) of the total.Fisher et al. (2016) estimate a smaller burden of MNs of about 10-20 ppt due to a lower molar yield (18 % vs. 26 % in AM3) and faster hydrolysis of MNs in their model.
Nighttime chemistry contributes about 30-36 % of ANs, which is dominated by isoprene oxidation as well (Fig. 5).About 33-41 % of the INs are produced during night, similar to the value (44 %) reported by Mao et al. (2013b) but with different speciation due to the different treatment of chemistry.PROPNN contributes about 29-38 % of the total INs.PROPNN in this work is mainly produced from the oxidation of C 5 nitrooxy hydroperoxide (INPN) and C 5 carbonyl nitrate (ISN1; dominantly by photolysis) that are generated from isoprene oxidation by NO 3 during the nighttime.This is different from Fisher et al. (2016), who showed that PROPNN is partially from the δ-ISOPO 2 + NO pathway and partially from the oxidation of ISN1 by NO 3 .In our model, we see a rapid increase in PROPNN after sunrise in the boundary layer (Fig. S7 in the Supplement), which is consistent with observations at the Southern Oxidants and Aerosols Study (SOAS) ground site CTL (Schwantes et al., 2015).Our model overestimates the mean vertical profile of PROPNN by a factor of 3 (not shown).As our model may largely underrepresent the chemical complexity of nighttime isoprene oxidation as shown by Schwantes et al. (2015), we consider PROPNN as a proxy for other unspecified isoprene nighttime nitrates.Overall, PROPNN contributes a significant fraction of ANs in the model, 23-29 % in the boundary layer as shown in Sect.3.1.With monoterpene nitrates and isoprenederived DHDN and nighttime NO 3 oxidation products taken into account, we find that our model can reproduce both ob-served ANs and speciated alkyl nitrates (Fig. S4 in the Supplement).

Decadal changes in PBL RON and surface ozone over the Southeast US
As RON and related species from aircraft and surface measurements are well reproduced in our model in the summers of 2004 and 2013, we assume that the model is representative of this chemical environment and then use the model to derive monthly mean changes between the summers of 2004 and 2013.We also investigate the impacts of further decreases in NO x emissions by applying a hypothetical 40 % reduction of anthropogenic NO x emissions in 2013 but keeping other emissions and meteorology the same (the "hypo" case in Table 2).

PBL RON
We first examine the simulated decadal changes in RON in the boundary layer over the Southeast US as shown in Fig. 6.In the summer of 2004, the model suggests that NO y is mainly comprised of HNO 3 (45 %), NO x (31 %), PNs (14 %), and ANs (9 %).In response to a 40 % reduction in anthropogenic NO x emissions (35 % reduction in total NO x emissions; Table 1) from 2004 to 2013, NO y declines by 34 %.This modeled response is comparable to long-term NO y measurements from the AQS surface network, which shows on average a 45 % decrease from 2004 to 2013 over the Southeast US.Based on model estimates in Fig. 6, most RON species are reduced proportionally, with decreases of 38 % for HNO 3 , 32 % for NO x , and 34 % for PNs.The different changes in PNs and PAN (the majority of PNs) in Fig. 1 might be due to the difference in sampling regions.The only exception is ANs, with a smaller decline of 19 %.
As an important source of organic aerosols, ANs may contribute to the decrease in organic aerosols over the Southeast US in the past decade (Blanchard et al., 2016).
We then conduct a sensitivity test with an additional 40 % reduction of anthropogenic NO x emissions from 2013.We find that NO y decreases by 29 %, with a proportional decrease in HNO 3 , NO x , and PNs (Fig. 6).The slower decrease in NO y is likely due to ANs, which decrease at a slower rate and become a larger fraction of NO y .The buffering of ANs is consistent with previous studies (Browne and Cohen, 2012;Fisher et al., 2016), mainly due to lower OH resulting from decreased NO x (Fig. S8) and thus prolonged lifetimes of NO x and ANs (Browne and Cohen, 2012).As shown in Fig. S8 in the Supplement, averaged noontime OH decreases by 11 % from 2004 to 2013 and by 29 % after we impose an additional 40 % reduction of anthropogenic NO x emissions from 2013 levels.
The historical NO x emission reduction also affects reactive nitrogen export out of the boundary layer.Here we de-  fine exported nitrogen as the difference of the sources (chemical production and emissions) and sinks (chemical loss, wet and dry deposition).As shown in Table 3, total summertime NO y export from the Southeast US boundary layer decreases proportionally, from 24.1 Gg N in 2004 to 16.6 Gg N in 2013.The NO y export efficiency, calculated as net exported nitrogen divided by total NO x emissions, remains roughly the same (12 %) for 2004 and 2013, which is comparable to previous studies (Fang et al., 2010;Li et al., 2004;Parrish et al., 2004;Mao et al., 2013b;Sanderson et al., 2008;Hudman et al., 2007).Among all exported species, NO x contributes most to the net export from the PBL (6 % of total NO x emissions), followed by PAN (4 %) and ANs (2 %).We emphasize in Table 3 that a major fraction of NO x is exported through the top of the boundary layer (convection).From a budget calculation throughout the tropospheric column over the same region, we find that despite having the same NO y export efficiency (12 %), HNO 3 becomes the major exporter, accounting for half of NO y export efficiency from the total column (6 %).The contributions from PAN and ANs are roughly the same as their export from the boundary layer (4 and 2 %).This suggests that surface NO x ventilated through the boundary layer, converted to HNO 3 in the free troposphere, and exported as HNO 3 is likely the major NO y export mechanism over the Southeast US in our model, which is in agreement with previous observations (Parrish et al., 2004;Neuman et al., 2006).PAN and ANs together account for another half of NO y export efficiency.As PAN and ANs are of biogenic origin and longer lived than HNO 3 , they may play a key role in influencing RON and ozone in downwind regions (Moxim et al., 1996;Fischer et al., 2014).

Surface ozone
Since the mid-1990s, NO x emission controls have led to significant improvement in ozone air quality over the eastern US (Simon et al., 2015;Cooper et al., 2012).As NO x emissions continue to decrease, ozone production efficiency (OPE) may increase due to the lower NO x removal rate by OH and to some extent may compensate for the ozone reduction (Sillman, 2000).Meanwhile, surface ozone production may be further complicated by the increasing importance of RO 2 isomerization and RO 2 + HO 2 .Here we first evaluate our model against surface ozone observations in 2004 and 2013 and then project the future response of surface ozone to even lower NO x emissions to examine the efficacy of near-term NO x emission controls at lowering near-surface ozone levels.
We first examine the modeled surface ozone against observations at 157 EPA AQS monitoring sites over the Southeast US in July-August of 2004 and 2013 (Fig. S9 in the Supplement).In general, AM3 overestimates surface MDA8 ozone in both years by about 16 ppb on average, with NMB of 33-45 % and NME of 35-46 %.This positive bias of sum-mertime surface ozone has been a common issue in a number of modeling studies of this region (Fiore et al., 2009;Canty et al., 2015;Brown-Steiner et al., 2015;Strode et al., 2015;Travis et al., 2016).This might be partially attributed to overestimated anthropogenic NO x emissions from nonpower plant sectors, excessive vertical mixing in the boundary layer (Travis et al., 2016), or underestimates of ozone dry deposition (Hardacre et al., 2015;Val Martin et al., 2014).Further studies are warranted to investigate the cause of this bias in AM3.
Surface ozone concentrations over the Southeast US decline substantially from 2004 to 2013 in response to the large NO x emission reduction (Simon et al., 2015).MDA8 ozone averaged across all the monitoring sites is observed to decrease by 11 ppb (23 % of observed mean MDA8 ozone in July-August of 2004) resulting from an approximate 40 % reduction of anthropogenic NO x emissions (35 % reduction in total NO x emissions).This strong sensitivity of surface ozone to NO x emissions reflects the linear relationship between the ozone production rate and NO x concentrations when NO x is low (Trainer et al., 2000).Our model is able to capture this strong NO x -O 3 sensitivity, with the mean MDA8 ozone reduced by 10 ppb from 2004 to 2013.We find that a further 40 % reduction of anthropogenic NO x emissions with identical meteorological conditions could lead to an additional 9 ppb decrease in MDA8 ozone, a similar magnitude to the change from 2004 to 2013.
We further investigate the impact of temperature and moisture on surface ozone from 2004 to 2013.While several studies suggest that surface ozone increases with ambient temperature (Jacob and Winner, 2009;Bloomer et al., 2010;Wu et al., 2008;Steiner et al., 2010), Cooper et al. (2012) showed that this temperature-related impact is weak during the period 1990-2010 across the USA.Recent studies suggest that relative humidity (RH) or vapor pressure deficit (VPD) may play an important role in ozone variability through soilatmosphere or biosphere-atmosphere coupling (Kavassalis and Murphy, 2017;Camalier et al., 2007;Tawfik and Steiner, 2013).Our model shows marginal differences in RH (less than 1 %) and temperature (+2.4 K) within the PBL over the Southeast US between the summers of 2004 and 2013, which is consistent with observed changes in RH (+2.7 %) and temperature (+2.6 K) during ICARTT and SENEX.This small variation in the model is also consistent with climatology data (Hidy et al., 2014).Camalier et al. (2007) showed that RH has a much bigger impact on summertime ozone than temperature over the Southeast US, suggesting little influence of meteorology on ozone trends during this period.Using the same model but with the standard AM3 chemical mechanism, Lin et al. (2017) found that meteorology changes would have caused high surface ozone over the eastern US to increase by 0.2-0.4ppb yr −1 in the absence of emission controls from 1988 to 2014.Therefore, we conclude that the impact of climate variability and change on surface ozone is relatively small compared to NO x emission reductions over  the Southeast which is consistent with previous studies (Lam et al., 2011;Hidy et al., 2014;Lin et al., 2017;Rieder et al., 2015).Decreasing NO x emissions also reduce the frequency of high-ozone pollution events.Figure 7 shows the probability density function of observed and modeled MDA8 ozone at each monitoring site during July-August 2004 and 2013 and the probability density function of modeled MDA8 ozone under a hypothetical scenario with a 40 % reduction in anthropogenic NO x emissions compared to 2013.We show that the lowest ozone, about 20 ppb in current model simulations, remains invariant with NO x emission changes over the Southeast US, which is consistent with observations (Fig. 7a).Meanwhile, the high tail of MDA8 ozone events shifted from more than 100 ppb in 2004 to about 85 ppb after the 40 % reduction of anthropogenic NO x emissions from 2013.A similar shift is found in observations.The narrowing of the range of ozone with decreasing NO x is consistent with the observed trends reported by Simon et al. (2015).We also find that further reductions of NO x emissions will reduce both median ozone values and the high tail, suggesting that fewer high-ozone events will occur under continued NO x emission controls in the future.

Conclusion and discussion
Near-surface ozone production over the Southeast US is heavily influenced by both anthropogenic and biogenic emissions.We investigate the response of NO y speciation to the significant NO x emission controls (about 40 % reduction) in this region over the past decade in light of the fast-evolving understanding of isoprene photooxidation.This knowledge is needed to predict nitrogen and ozone budgets in this region and elsewhere in the world with similar photochemical environments.Here we use extensive aircraft and ground observations combined with a global chemistryclimate model (GFDL AM3) to examine decadal changes in NO y abundance and speciation as well as in surface ozone mixing ratios over the Southeast US between the summers of 2004 and 2013.We then use the model to infer future NO y speciation and surface ozone abundances in response to further NO x emission controls in this region.
We first evaluate the model with aircraft and surface observations.When we apply the estimated 40 % reduction in anthropogenic NO x emissions from 2004 to 2013, our model reproduces the major features of the vertical profiles of NO x , HNO 3 , PAN, ANs, and NO y observed during aircraft campaigns over the Southeast US in the summers of 2004 and 2013.By including recent updates to isoprene oxidation, our model can largely reproduce the vertical profiles of ANs and several speciated alkyl nitrates, as as their correlations with O x and HCHO, lending support to the model representation of isoprene oxidation.On the other hand, we show that a discrepancy between measured ANs and speciated RONO 2 can be explained by a combination of monoterpene nitrates, dinitrates, and nighttime NO 3 oxidation products from isoprene.We also show that modeled ozone appears to be insensitive to the hydrolysis of ISOPNB because its photooxidation, mainly by OH, also returns little NO x .
Major RON species decline proportionally as a result of NO x emission reductions in the Southeast US, except for a slower rate in ANs.The slower decline of ANs reflects the prolonged lifetime of NO x when it is decreasing.Our model suggests that summertime monthly averaged NO x , HNO 3 , PAN, and NO y decline by 30-40 % in response to a 40 % reduction in anthropogenic NO x emissions from 2004 to 2013.This proportional decrease is likely driven by high concentrations of biogenic VOCs, the major precursor of PAN in this region that change little in magnitude from 2004 to 2013.In contrast, Pollack et al. (2013) find a faster PAN decrease than HNO 3 in the LA basin over the past several decades, partly due to the decrease in anthropogenic VOC emissions that are major PAN precursors.
Deposited and exported NO y decline with NO x emission reductions.The model also shows a decrease in NO − 3 wet deposition flux by 29 % from 2004 to 2013, which is consistent with observations from the NADP network (−24 %).We find from model calculations that the NO y export efficiency remains at 12 % in both 2004 and 2013, leading to a proportional decrease in exported NO y .The dominant NO y export terms include NO x and HNO 3 , each accounting for 6 % of the total exported NO y , followed by PNs (4 %) and ANs (2 %).
The response of surface ozone to NO x emission reductions reveals a strong NO x -O 3 sensitivity in summertime over the Southeast US.Observations from the EPA AQS surface network suggest that mean MDA8 ozone during July-August decreased by 23 %, from 48 ppb in 2004 to 37 ppb, in 2013.Despite a positive absolute bias of up to 12 ppb in boundary layer ozone and 16 ppb in surface MDA8 ozone, our model shows a 10 ppb decrease in surface MDA8 ozone from 2004 to 2013, which is very close to the observed 11 ppb decrease from the EPA data.The bias of ozone in our model is not entirely attributed to uncertainties in NO x emissions, as the overestimate suggested by earlier work would lead to an underestimate of NO y (Travis et al., 2016).Care should be exercised in applying the modeling results for surface ozone regulation purposes given the high-ozone bias shown in our model.We find from model calculations that modeled MDA8 ozone will continue to decrease by another 9 ppb, assuming that anthropogenic NO x emissions are reduced by 40 % from 2013 levels with meteorology and other emissions kept the same.In addition, further NO x emission reduction leads to less frequent high-ozone events.This continued strong sensitivity of surface ozone to NO x emissions can guide the development of effective emission control strategies for improving future air quality.

Figure 1 .
Figure 1.Mean vertical profiles of ozone and reactive oxidized nitrogen from observations during ICARTT (a) and SENEX (b) over the Southeast US (25-40 • N, 100-75 • W) during daytime and model estimates from AM3 with hydrolysis of ISOPNB (red) and AM3 without hydrolysis of alkyl nitrates (green).The solid and dashed black lines in the HNO 3 of ICARTT represent measurements collected using mist chamber-ion chromatograph by the University of New Hampshire (UNH) and chemical ionization mass spectrometry by the California Institute of Technology (CIT), respectively.NO y from ICARTT is calculated as the sum of NO x , HNO 3 (UNH in the solid line and CIT in the dashed line), PAN, and total alkyl nitrates ( ANs).Observed and modeled ANs in (b) are during SEAC 4 RS.

Figure 2 .
Figure 2. Mean vertical profiles of ISOPN and MVNK + MACRN during (a) SENEX and (b) SEAC 4 RS over the Southeast US (25-40 • N, 100-75 • W).Black lines are the mean of observations.Red and green lines are the mean of modeled results with hydrolysis of ISOPNB and without hydrolysis of alkyl nitrates, respectively.Gray shades are 1 standard deviation (±σ ) of averaged profiles of the measured tracers.

Figure 3
Figure 3 shows a comparison of NO − 3 wet deposition flux between observations and model results during the summers of 2004 and 2013.The observed NO − 3 wet deposition flux is calculated by multiplying the measured NO − 3 concentration and precipitation at each monitoring site as F o,i = C o,i P o,i , where F o,i is the monthly mean NO − 3 wet deposition flux, C o,i and P o,i are the monthly mean observed NO − 3 concentration and precipitation at monitoring site i.The modeled NO − 3 wet deposition flux includes HNO 3 and all the alkyl nitrates.Observations indicate a 24 % reduction of NO − 3 wet deposition flux in summer from 2004 to 2013 over the Southeast US, likely due to NO x emission reductions.This reduction in monthly averaged NO − 3 wet deposition flux is well captured by our model (−29 %), despite a low relative bias of 40 % in both years and NMB of −39 to −43 % (TableS4in the Supplement).

Figure 4 .
Figure 4. O x versus ANs correlation (a, c) and ANs versus formaldehyde correlation (b, d) within the boundary layer (< 1.5 km) during ICARTT (a, b) and SEAC 4 RS (c, d).Observations are in black diamonds; model estimates from AM3 with ISOPNB hydrolysis are represented by red symbols.Green symbols represent the correlation using modeled ISOPN + MVKN + MACRN.Blue symbols represent the correlation using observed ISOPN + MVKN + MACRN from SEAC 4 RS.Solid lines are the reduced major axis regression lines.

Figure 5 .
Figure 5. Mean vertical profiles of modeled alkyl nitrates from isoprene and monoterpene oxidation (a, b) and major isoprene nitrate species (c, d) during ICARTT (a, c) and SEAC 4 RS (b, d) from AM3 with hydrolysis of ISOPNB.
Figure 6.Modeled mean NO x , HNO 3 , total peroxy nitrates ( PNs), total alkyl nitrates ( ANs), and NO y averaged over the boundary layer (< 1.5 kmg) of the Southeast US during July-August of 2004 (a), 2013 (b), and a scenario assuming 40 % reduction of 2013 anthropogenic NO x emissions (c).Numbers in parentheses indicate mean concentrations over the plotted region.Note that different color scales represent the concentration of each species.

Figure 7 .
Figure 7. Observed (a) and simulated (b) probability density function of MDA8 ozone at AQS monitoring sites in Fig. S2 in the Supplement during the summers of 2004 and 2013 and a scenario with a 40 % reduction in the anthropogenic NO x emissions of 2013.

Table 3 .
Monthly NO y budget in the boundary layer (< 1.5 km) of the Southeast US during July-August of 2004, 2013, and a scenario with 40 % reduction of anthropogenic NO x emissions of 2013 a .We define the boundary of the Southeast US as 25-40 • N, 100-75 • W. All budget terms are in Gg N. b PNs includes PAN, peroxymethacryloyl nitrate (MPAN), and a C 5 hydroxy peroxyacyl nitrate (C5PAN1) produced by oxidation of ISN1.
a c Alkyl nitrates produced from oxidation of isoprene and monoterpenes by OH. d Alkyl nitrates produced from oxidation of isoprene and monoterpenes by NO 3 .