Interactive comment on “ Understanding the impact of recent advances in isoprene photooxidation on simulations of regional air quality ”

The present manuscript describes the update of isoprene oxidation in the SAPRC-07 mechanism whithin the CMAQ model. The performance of the model relative to 3 field campaingns over the US is shown to lead to significant improvements. The results are thorougly analized. Finally, a series of sensitivity simulations aimed at evaluating the uncertainty expecially with respect to alky nitrates chemistry are shown. The manuscript is of great interest and falls well within the scope of ACP. It is clear and well structured. I recommend publication after the major comments below are taken into


Introduction
Isoprene emissions are the dominant source of non-methane volatile organic compounds to the atmosphere (Guenther et al., 2006). Being highly reactive (lifetime ∼ 3 h at OH concentration of 1 × 10 6 mol cm −3 ), isoprene plays a central role in tropospheric chemistry. When isoprene reacts with OH, six isomeric hydroperoxy radicals (ISOPO 2 ) are formed. When NO concentrations are greater than ∼ 150 ppt, ISOPO 2 primarily reacts with NO to yield mainly alkoxy radicals and NO 2 . Subsequently, O 3 is formed through NO 2 photolysis, while OH and NO are regenerated in autocatalytic cycles. The minor channel forms hydroxynitrates (ISOPN , Table 1), which sequester NO x and therefore regulate O 3 formation locally. There have been a number of laboratory studies reporting the ISOPN yield, with the value 8440 Y. Xie et al.: Understanding the impact of recent advances in isoprene photooxidation ranging from 4.4 % to 15 % (Tuazon and Atkinson, 1990;Chen et al., 1998;Chuong and Stevens, 2002;Sprengnether et al., 2002;Patchen et al., 2007;Paulot et al., 2009a;Lockwood et al., 2010).
With the presence of a double bond, ISOPNs are highly reactive with respect to OH and O 3 (Giacopelli et al., 2005;Paulot et al., 2009a;Lockwood et al., 2010). At OH concentration of 1 × 10 6 mol cm −3 and 40 ppbv O 3 , the lifetime of ISOPN with respect to OH is about 4 h (Paulot et al., 2009a) and to O 3 is about 3 h (Lockwood et al., 2010), giving a combined lifetime slightly less than 2 h. The products from ISOPN oxidation might either retain the nitrate functionality (Grossenbacher et al., 2001;Giacopelli et al., 2005;Perring et al., 2009a) or release NO x (Paulson and Seinfeld, 1992). Paulot et al. (2009a) estimate about 50 % recycling efficiency of NO x (α) from ISOPN + OH/NO reactions and identified some of the organic nitrates produced by this reaction, including methyl vinyl ketone nitrate, methacrolein nitrate, and propanone nitrate. A recent field study confirmed the existence of these secondary organic nitrates compounds . Some of these organic nitrates products, e.g., propanone nitrate, are found to be considerably longerlived than ISOPN and can serve as temporary reservoirs for NO x , thereby impacting the NO y budget as well as O 3 levels far from isoprene source region (Paulot et al., 2009a;. The products and NO x recycling efficiency from ISOPN oxidation by O 3 are poorly understood (Giacopelli et al., 2005;Lockwood et al., 2010).
Furthermore, there are also large uncertainties in isoprene+NO 3 oxidation chemistry. Although only accounting for about 6-7 % of isoprene oxidation (Horowitz et al., 2007;Ng et al., 2008), studies suggest that isoprene can be an important sink for NO 3 radical (Starn et al., 1998;, and the reaction contributes substantially (30-60 %) to the formation of isoprene nitrates (ING) (von Kuhlmann et al., 2004;Horowitz et al., 2007;Paulot et al., 2012). The isoprene+NO 3 reaction is initiated by NO 3 addition to a double bond, followed by reaction with O 2 to produce a nitroxy peroxy radical (NISOPO 2 ). Subsequent reaction of NISOPO 2 with NO, NO 3 , RO 2 , and HO 2 can either result in functionalization, conserving the nitrate group, or release NO 2 through dissociation. Large nitrates yields (∼ 80 %) have been reported in a number of experimental studies (Barnes et al., 1990;Berndt and Boge, 1997;Rollins et al., 2009;Perring et al., 2009b;Kwan et al., 2012), with the primary products identified as C5-nitrooxycarbonyl (NIT1), C5-nitrooxyhydroperoxide (NISOPOOH), and C5hydroxynitrate (ISOPN). Few studies have investigated the fate of these organic nitrates (Rollins et al., 2009); therefore, their oxidation products and NO x recycling efficiency are not well known.
In addition to photochemical loss, isoprene nitrates are also subject to removal by deposition. Measurements have found these multi-functional nitrates are soluble in water with large Henry's law constants (Shepson et al., 1996;Treves et al., 2000). Their dry deposition velocities are less constrained, with measured values ranging from that of PAN (0.4 cm s −1 , Shepson et al., 1996) to HNO 3 (2.7 cm s −1 , . Studies also disagree regarding the relative importance of removal by deposition (Horowitz et al., 2007) and removal by photooxidation (Shepson et al., 1996;Ito et al., 2007;Perring et al., 2009a).
Simulations of tropospheric ozone production are known to be highly sensitive to ISOPN yield (von Kuhlmann et al., 2004;Wu et al., 2007;Ito et al., 2007;Weaver et al., 2009;Paulot et al., 2012). For example, Wu et al. (2007) found a 10 % reduction in global tropospheric ozone production while increasing the ISOPN yield from 4 % to 12 %. Similar to the effects of ISOPN yield, recent work by  also highlighted the importance of NO x recycling efficiency as well as the representation of secondary nitrate photochemistry for the simulation of tropical ozone.
In additional to laboratory studies, field observations have been used to probe isoprene nitrate chemistry. For example, Horowitz et al. (2007) found that a 40 % recycling efficiency together with a 4 % ISOPN yield and a fast dry deposition rate best captured the alkyl and multifunctional nitrates measurements collected during the INTEX-NA/ICARTT (Intercontinental Chemical Transport Experiment-North America/International Consortium for Atmospheric Research on Transport and Transformation) field study during summer 2004 (Singh et al., 2006;Fehsenfeld et al., 2006). Using the same datasets, Perring et al. (2009a) inferred a NO x recycling efficiency between 3 % and 33 % depending on the assumed ISOPN yields (4 -12 %).
Recently, there have also been many studies devoted to isoprene oxidation under low NO x conditions Paulot et al., 2009b;Peeters et al., 2009;Peeters and Müller, 2010;Crounse et al., 2011). These studies address some of the important challenges with isoprene chemistry including (i) the apparent discrepancy between observed and modeled HO x abundance under isoprene rich conditions Ren et al., 2008;Butler et al., 2008;Lelieveld et al., 2008); and (ii) the link between isoprene gas phase oxidation and the formation of organic aerosols (Claeys et al., 2004;Edney et al., 2005;Kleindienst et al., 2007;Paulot et al., 2009b;Surratt et al., 2010).
The tropical forest has been a focus area for many recent studies Butler et al., 2008;Karl et al., 2009;Stone et al., 2011;Whalley et al., 2011) that aim to improve the understanding of isoprene oxidation chemistry. These locations are mostly characterized by very large biogenic emissions with pristine conditions, where isoprene peroxy radicals are dominated by reactions with HO 2 and isomerization. Nevertheless, anthropogenic emissions are known to substantially affect the relative importance of isoprene oxidation pathways and products (e.g., alkyl and multifunctional nitrates, epoxides). The southeastern United States is a region known for its abundant isoprene emissions and anthropogenic pollutants, providing a good opportunity to understand the impact of isoprene nitrates in a region influenced by both biogenic emissions and a range of NO x concentrations.
In this study, we incorporate recent advances in isoprene oxidation chemistry into the SAPRC-07 chemical mechanism (Carter, 2010a) within the US EPA Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006). Data from the INTEX-NA/ICARTT campaign during the summer of 2004 (Singh et al., 2006;Fehsenfeld et al., 2006) are used to evaluate model performance and constrain the uncertainties in the isoprene nitrate chemistry. The observations include measurements of speciated NO y , organic compounds, and aerosols, comprising a comprehensive dataset for characterizing the summer continental boundary layer of the eastern United States. In contrast to previous modeling studies, we use a highly detailed representation of isoprene chemistry and examine the effect on areas where both biogenic and anthropogenic emissions are important. In Sect. 2, we describe the modeling system and the updated isoprene oxidation scheme. In Sect. 3.1, we evaluate the CMAQ model results with INTEX-NA/ICARTT observations. We further investigate the sources and sinks of the isoprene peroxy radicals and isoprene nitrates (Sects. 3.2 and 3.3). We then examine the sensitivity of model results to the most important uncertainties in our current understanding of isoprene nitrate chemistry (Sect. 3.4), with an emphasis on O 3 and OH concentrations (Sect. 3.5).

Methods and data
We start this section by describing the updates to isoprene oxidation chemistry, followed by discussion of the photochem-ical modeling system and the observational datasets used to constrain the model results.

Updated isoprene oxidation scheme
We base our core chemistry on a modified version of SAPRC-07, including changes for air toxics (i.e., SAPRC-07T (Hutzell et al., 2012), referred to as the BASE scheme). A new version of the mechanism is developed in this study (referred to as the standard isoprene scheme (IS)), which incorporates recent advances in isoprene oxidation chemistry (Paulot et al., 2009a, b;Lockwood et al., 2010;Peeters and Müller, 2010;Crounse et al., 2011;Rollins et al., 2009). The changes include a more explicit representation of isoprene nitrate formation from OH/NO and NO 3 pathways as well as modification to the isoprene chemistry under low-NO x conditions. The updated chemistry is summarized in Fig. 1; see Tables S1 and S2 (Supplement) for a complete listing of the compounds and reactions. Each of the speciated isoprene nitrates is described in Table 1.
In the standard SAPRC-07 mechanism (Carter, 2010a, b), isoprene nitrates are represented, along with all other non-PAN analogue organic nitrates, as a lumped organic nitrate species (RNO 3 ), with k OH of 7.2 × 10 −12 cm 3 mol −1 s −1 and NO x recycling efficiency of about 30 %. In the OH/NO reactions of isoprene, the RNO 3 yield is 9.3 %. This RNO 3 yield is a derived value. It is treated as a parameter and adjusted to improve the comparison of SAPRC-07 simulated O 3 measurements from laboratory chamber experiments. The O 3 comparison is very sensitive to the RNO 3 yield parameter. In the NO 3 oxidation pathway, a large fraction (75 %) of organic nitrates is assumed to be long-lived and is treated as lost nitrogen (XN), and the RNO 3 yield is only 6.4 %. The IS mechanism treats isoprene nitrates explicitly. Isoprene oxidation reactions from OH are shown in Fig. 1a. In the presence of NO, isoprene nitrates can form as described by Pathway III in Fig. 1a. The formation and the subsequent oxidation of organic nitrates are largely based on the work of Paulot et al. (2009a). The δ-and β-hydroxy isoprene nitrates (ISOPN) are represented separately, as their fate varies substantially from each other (Paulot et al., 2009a). Formation and photooxidation of organic nitrates from MVK and MACR, i.e., methyl vinyl ketone nitrate (MVKN) and methacrolein nitrate (MACRN), are also treated following Paulot et al. (2009a). ISOPN reacts rapidly with OH (weighted average k OH = 6.2 × 10 −11 cm 3 mol −1 s −1 for δand β-branch), releasing about 60 % of NO x , with the rest forming methyl vinyl ketone nitrate (MVKN), methacrolein nitrate (MACRN), propanone nitrate (PROPNN), and ethanal nitrate (ETHLN). PROPNN reaction with OH (Zhu et al., 1991;Paulot et al., 2012) and photolysis (Roberts and Fajer, 1989;Barnes et al., 1993;Saunders et al., 2003) are added, and PROPNN is substantially longer-lived (k OH = 4.0 × 10 −13 cm 3 mol −1 s −1 ) than other second-generation isoprene nitrates. Photolysis rates of MVKN and ETHLN are assumed to be the same as that of PROPNN. For ISOPN+O 3 reactions, we use the reaction rate coefficients (weighted average k O 3 = 8.4 × 10 −17 cm 3 mol −1 s −1 ) from Lockwood et al. (2010). As little is known about the products and NO x recycling efficiency from this pathway, we estimate the likely products largely based on the SAPRC mechanism generation system (Carter, 2000(Carter, , 2010b. We assume a lower NO x recycling (30 %) from ISOPN + O 3 than from ISOPN+OH/NO reactions. In addition to the species listed above, a general isoprene nitrate species (RNO 3 I) is added to represent additional second-generation organic nitrates formation. RNO 3 I is assumed to react readily with OH (k OH = 8 × 10 −12 cm 3 mol −1 s −1 ).
The oxidation of isoprene by NO 3 (Fig. 1b) is also updated in the IS mechanism. A nitrooxy-carbonyl (NIT1) yield of 70 % is used in the RO 2 + NO/NO 3 reactions (Rollins et al., 2009) (Pathway III in Fig. 1b). The RO 2 + HO 2 reaction (Pathway I in Fig. 1b) yields a nitrooxy-hydroperoxide (NISOPOOH). For RO 2 + RO 2 reactions (Pathway II in Fig. 1b), a branching ratio of 50 % is assumed for the radical propagating pathway versus carbonyl (NIT1) and alcohol (ISOPN) formation reactions. The subsequent oxidation of NIT1 by NO 3 largely follows Rollins et al. (2009), using a reaction rate coefficient of 7.0 × 10 −14 cm 3 mol −1 s −1 , which is about an order of magnitude slower than that of isoprene + NO 3 reactions. Reaction of NIT1 with O 3 is also included, and the reaction rate coefficient (2.5 × 10 −17 cm 3 mol −1 s −1 ) is estimated by scaling the NIT1 + NO 3 reaction rate coefficient using methacrolein as a reference compound. Products from NIT1+O 3 reactions and the NIT1+OH oxidation scheme are estimated, largely based on the SAPRC mechanism generation system. The IS mechanism assumes 70 % NO x recycling efficiency from NIT1 + O 3 oxidation. NIT1 is estimated to react rapidly with OH (k OH = 3.0 × 10 −11 cm 3 mol −1 s −1 ), with the reaction having a large yield of propanone nitrate (PROPNNB) and releasing little NO x . NISOPOOH is treated following Paulot et al. (2012), which assumes its fate is similar to that of hydroxyhydroperoxide formed from isoprene+OH reactions.
In addition to the changes in isoprene nitrate chemistry, isoprene oxidation chemistry under low-NO x conditions is updated. In the standard SAPRC-07 mechanism, the low-NO x oxidation of isoprene is represented by the formation of lumped organic hydroperoxides (R6OOH) from the RO 2 +HO 2 pathway.
Additional changes in the mechanism include revised reaction rates for RO 2 +HO 2 . The reaction rate coefficient in the base SAPRC-07 mechanism is independent of the size of the peroxy radical: k= 3.80×10 −13 exp(900/T). (1) In fact, k increases with the size of the peroxy radical (Rowley et al., 1992;Jenkin and Hayman, 1995;Boyd et al., 2003), and we use the expression derived by Saunders et al. (2003): where n is the number of atoms of carbon in the peroxy radical. For isoprene peroxy radicals, this increases the reaction rate by a factor of 2 at 298 K.

CMAQ model
CMAQ model version 4.7 is used for photochemical air quality modeling. The inputs to CMAQ are the anthropogenic emissions and meteorological conditions; outputs are concentrations and deposition fluxes for each hour of the simulation. CMAQ employs an Eulerian grid structure to explicitly simulate biogenic emissions, gas-phase, aqueous, and mixedphase chemistry, advection and dispersion, aerosol thermodynamics and physics, and wet and dry deposition. A more complete description and evaluation of the CMAQ processes and inputs are available in Foley et al. (2010), Carlton et al. (2010), and Appel et al. (2011). CMAQ with the base SAPRC-07T mechanism has been shown to reliably simulate atmospheric concentrations of ozone and aerosols (Hutzell et al., 2012). The spatial domain covers the entire continental US and a portion of Canada and Mexico. The model includes 24 vertical layers at 36 km horizontal resolution. Chemical boundary conditions are based on the default vertical profiles of gaseous species and aerosols in CMAQ that represent clean air conditions. The model simulation period is 00:00 UTC 1 July to 00:00 UTC 16 August 2004, with the first 96 hours discarded as spin-up to remove the impact of initial conditions.
We use 3-D meteorology fields developed by the 5th Mesoscale Meteorological model (MM5; Grell et al., 1994) (Warneke et al., 2010;Carlton and Baker, 2011). Model predictions compared with measurements during a July 1998 intensive field campaign in the Ozarks find that MEGAN tends to overpredict isoprene emissions, while BEIS shows underpredictions (Carlton and Baker, 2011). To account for the potential underestimation from BEIS, we increase the isoprene emissions calculated by BEIS version 3 (BEIS3) by 50 %.
The treatment of biogenic secondary organic aerosol (SOA) formation follows that in CMAQv4.7 (Carlton et al., 2010). Biogenic SOA formation from gas-phase oxidation of isoprene, monoterpenes, and sesquiterpenes is represented. Isoprene SOA is formed exclusively by OH oxidation. The yield from isoprene semivolatile products and their partitioning parameters are based on the low-NO x experimental results of Kroll et al. (2006). Monoterpene and sesquiterpene SOA is formed by OH, NO 3 , and O 3 oxidation, with the partitioning parameters based on the experimental data of Griffin et al. (1999). Nonvolatile SOA formation including acid enhanced isoprene SOA (Surratt et al., 2007) is also accounted for in the model. Evaluation of CMAQ using observations between the surface and 2 km (10-18 LST, except NO 3 is 21-04 LST). * indicates data from WP-3D. HCHO is based on the average of the University of Rhode Island and NCAR measurements due to the systematic difference (∼ 35 %) between the two datasets. The CMAQ model results are sampled at the GPS location of the 1 min measurements. Error is reported the median of the quantity CMAQ minus observations, divided by the median of the observations.

Observational datasets
We use data from the INTEX-NA/ICARTT campaigns in the summer of 2004 (1 July to 15 August) (Singh et al., 2006;Fehsenfeld et al., 2006) to evaluate model performance and constrain the uncertainties in the isoprene nitrate chemistry. Observations are available from NASA DC-8 and NOAA WP-3D (18 flights on each aircraft). The two aircrafts have different sampling emphasis, with DC-8 targeted at regional airmasses over North America and WP-3D aimed at local flows downwind of urban centers and point sources in the northeastern United States (Fig. 3). During INTEX-NA/ICARTT, the total alkyl and multifunctional nitrates ( ANs) are measured by thermal dissociation laser induced fluorescence (TD-LIF, Day et al., 2002). The dataset represents the most spatially extensive measurements of ANs over the eastern United States to date. While ANs includes isoprene nitrates (ING), the ANs measurement also includes other organic nitrates as well. As mentioned, these same data have been used in previous studies investigating the formation and fate of isoprene nitrates (Horowitz et al., 2007;Perring et al., 2009a).

Evaluation of IS mechanism with INTEX-NA/ICARTT observations
With the updated isoprene nitrate chemistry, the IS simulation shows improved performance over a range of species measured by aircraft during the INTEX-NA/ICARTT 2004 field study. Model performance of selected species within the boundary layer is summarized in Fig. 2. Isoprene nitrates (ING) and ANs are substantially shorter-lived in the IS scheme ( Table 2). As a result, ANs concentrations decrease by 20 %, reducing the bias from 60 % to 30 %. NO x concentrations are slightly underestimated in the base case (−7 %), and have little bias (1 %) in the IS case. The increase in NO x is due to shorter photochemical lifetime of isoprene nitrates and reduced net removal of NO x by isoprene nitrate chemistry. HNO 3 (not shown) is biased low in both cases (−44 % for the base, −33 % for IS scheme). For PAN and NO y (NO y = NO x + PNs + ANs + HNO 3 + minor species), both model runs Isoprene concentrations from the DC-8 measurements are slightly overestimated by the base case (bias of 26 %), whereas the IS case shows very little bias (4 %). For measurements from the WP-3D, slight underestimations are found in both runs (−12 % from the base, −26 % from IS). The decreases in isoprene concentrations in the IS case can be related to the increase in OH (∼ 30 %, as shown below). Oxidation products from isoprene, methyl-vinyl ketone (MVK) and methacrolein (MACR) are also measured by the WP-3D flights. Modeled MVK and MACR are well correlated with observations (R 2 = 0.66), however both simulations underpredict the concentrations of these species by 40 %. Underestimation of MVK and MACR could be partly attributed to the bias in isoprene, while errors in modeled OH concentrations might cause additional underpredictions of these isoprene oxidation products. Formaldehyde (HCHO) increases slightly (4 %) in the IS case, and both simulations show some underpredictions (bias from base case = −12 %, and bias from IS = −9 %).
For radical species, NO 3 increases by 35 % as a result of increased NO x and O 3 , and the simulation shows good agreement with observations (bias of 4 %). Concentrations of OH increase by 30 %, with the bias improved from −30 % in the base simulation to about −20 % in the IS case. Performance of HO 2 also improves slightly with bias reduced from about −30 % to −25 %. A recent study suggests the traditional OH measurements made by laser-induced fluorescence (LIF) using the Penn State Ground-based Tropospheric Hydrogen Oxides Sensor (used in 2004 INTEX-NA/ICARTT study) tend to overestimate the concentrations due to internally generated OH, likely from oxidation from biogenic VOCs (Mao et al., 2012). It is also reported that HO 2 measured by LIF is likely subject to interference by RO 2 from isoprene, alkenes, or aromatics (Fuchs et al. 2011). Therefore, caution should be taken when interpreting OH and HO 2 performance, especially when biogenic VOCs are abundant. We find the increase in CMAQ simulated HO x is partly driven by their increased production. The change is related to a number of sources: rapid photolysis of HPALD, increased photolysis from the increase of O 3 , and increased photolysis from the increased abundance of isoprene oxidation products (e.g., glyoxal and methylglyoxal). Meanwhile, the changes in NO x due to updates in isoprene nitrate chemistry further increase OH concentrations by shifting the partitioning between HO 2 and OH through HO 2 +NO reaction.
Above the planetary boundary layer in the free troposphere, the impact of the more detailed isoprene chemistry is smaller (Fig. S1). Furthermore, the model underestimates ozone and NO y in the free troposphere, potentially due to errors in vertical mixing or boundary conditions. Because of these biases and because the short lifetime of isoprene means that relatively little is transported to the free troposphere, we restrict our analysis to concentrations within the planetary boundary layer, where the impact of these biases are reduced.
Changes in modeled surface concentrations of selected species from the IS simulation compared to the base case are shown in Fig. 3. NO x concentrations increase over extended regions of the continent, with a maximum increase of 40-80 pptv (4 %) over the southeastern US and California where there are abundant isoprene emissions. In contrast to the increase of NO x in isoprene rich regions, decreases are found in urban areas with very high NO x but relatively low isoprene concentrations (e.g., Los Angeles). In these urban locations, OH concentrations increase as a result of increased O 3 levels, causing more rapid formation of HNO 3 and therefore a decrease in NO x .
Corresponding to the increase in NO x concentrations, O 3 levels increase by up to 3 ppbv with the spatial pattern correlated well with the changes in NO x . Concentrations of OH, HO 2 (not shown), and NO 3 radical also increase by ∼ 30 %, ∼ 6 %, and ∼ 20 % respectively at these locations. With the increases in oxidants levels (OH, O 3 , NO 3 ), biogenic SOA is increased by ∼ 0.1 µg/m 3 (i.e., 10-20 %) in the southeastern United States. We find greater changes when averaging within the boundary layer (<2 km) for the southeastern US: NO x , 9 %; HO 2 , 11 %; and NO 3 , 67 %. We further evaluate model performance using the correlation of O 3 versus ANs (Fig. 4) as measured during INTEX-NA/ICARTT. Both are produced from RO 2 reactions with NO, and therefore might provide useful information regarding the ANs formation and fate. We restrict our analysis to samples with high isoprene concentrations (> 500 pptv) to focus on isoprene rich air masses. Using linear regression, the observations are fit to a line with slope of 35.1±6.4 ppbv( AN)/ppbv(O 3 ). We find observed ANs and O 3 are correlated with R 2 of 0.24. The slope is in good agreement with reported morning values at Granite Bay, CA (Cleary et al., 2005), and between the morning and afternoon values found at La Porte, TX (Rosen et al., 2004). Horowitz et al. (2007) use the entire INTEX-NA/ICARTT summer 2004 dataset and find that observations of O 3 versus Fig. 5. Simulated fate of isoprene peroxy radicals in the IS and E (slow HO 2 +RO 2 ) cases. Results are averaged over the entire modeling period for the southeastern US within 2 km of the surface. ISOPO 2 refers to isoprene peroxy radicals formed from isoprene+OH reactions (Fig. 1a), and NISOPO 2 refers to those formed from isoprene+NO 3 reactions (Fig. 1b).
ANs are only weakly correlated (R 2 = 0.12) with slope of 81.7. Compared to the correlation and slope reported by Horowitz et al. (2007), we find slightly stronger correlation and reduced slope (a factor of 2) using observations coincident with high isoprene concentrations. Following the methods of Rosen et al. (2004) and Cleary et al. (2005), the slope of 35.1 implies a ANs yield of 5.7 %. As the high isoprene observations are mostly found in the afternoon hours, it is expected that this net yield is lower than the actual yield, as some ANs are lost over the course of a day to deposition and photochemistry. Similar findings have been reported by Horowitz et al. (2007), where the inferred yield was about a factor of 2 lower than the isoprene nitrate yield from OH initiated chemistry.
The CMAQ simulated correlations of O 3 versus ANs are compared with the observed relationship in Fig. 4. In comparison to the base simulation, the IS scheme greatly reduces the bias relative to the observed slope (−19 % vs.
−46 %). Slightly stronger correlations are found in the two model simulations (squared correlation coefficient of 0.33) compared to those from the observations. In our analysis of O 3 versus ANs correlations, the model predicts about 40 % ANs from sources other than isoprene (not shown). Prior measurements have inferred a substantial contribution of isoprene to ANs (Perring et al., 2009a;. Additional measurements of speciated organic nitrates are needed to better constrain the contribution of isoprene to the composition of ANs. The IS mechanism with comprehensive isoprene chemistry more accurately simulates observed AN, O 3 , and NO x . Nevertheless, substantial uncertainty remains in isoprene nitrate chemistry. In the following sections, we use observations to investigate and constrain these uncertainties.  Fig. 6. Simulated production and loss processes for isoprene nitrates from the IS simulation and its sensitivity cases (listed in Table 2). Results are summarized over the entire modeling period for the southeastern US within 2 km of the surface.
Our examination of the isoprene peroxy radical fate and the sources and sinks of isoprene nitrates uses the IS simulation as a reference calculation to obtain a better understanding of the key processes.

Fate of isoprene peroxy radicals
To quantify the fate of isoprene peroxy radicals, we attribute the sinks to each chemical pathway using the CMAQ model results aggregated between the surface and 2 km in altitude and over the southeastern US. The results are shown in Fig. 5. For the peroxy radicals produced from OH reactions (ISOPO 2 ), the IS simulation suggests primary reactions with Pathway III, NO (44 %) and Pathway I, HO 2 (40 %). Isomerization plays a less important role (9 %) than reported in some previous studies (Peeters et al., 2009;Peeters and Müller, 2010;Archibald et al., 2010;Stavrakou et al., 2010), as we use the rate measured by Crounse et al. (2011). The RO 2 + HO 2 reaction pathway has important implications as the epoxides formed from isoprene hydroxyhydroperoxide oxidation have been shown to be an important gas-phase precursor of SOA (Paulot et al., 2009b;Surratt et al., 2010;Froyd et al., 2010;Lin et al., 2012).
For peroxy radicals from NO 3 reactions (NISOPO 2 ), the model predicts their fate is dominated by reactions with Pathway I, HO 2 (43 %) and Pathway II, RO 2 (35 %, including NISOPO 2 self-reactions (6 %), and their reactions with methyl peroxy radical (9 %), acetyl peroxy radical (8 %), and a peroxy radical operator (RO2C) (12 %)). The remaining portion is composed of reactions with Pathway III: NO (16 %) and NO 3 (7 %). The dominance of reaction of NISOPO 2 with HO 2 is consistent with the laboratory and modeling results from Rollins et al. (2009) and can have significant impacts on the nighttime oxidants levels, as a re-cent study (Kwan et al., 2012) suggests a large yield (50 %) of OH formation from this pathway (not included in this study). Furthermore, recent laboratory work suggests that NISOPO2 self-reactions may form lower-volatility dimers that contribute to SOA (Ng et al., 2008;Kwan et al., 2012). In our simulation of the southeastern US, six percent of the NISOPO2 radicals undergo self reactions, which suggests this may be a relevant source of SOA but it is not a large sink of NISOPO2. It should be noted, however, that there are large uncertainties in the RO 2 + RO 2 , RO 2 + HO 2 , RO 2 + NO, and RO 2 + NO 3 reaction rates (Orlando and Tyndall, 2012), and better constraints from laboratory and field studies are needed to confirm the relative importance of these pathways.

Production and loss of isoprene nitrates
Using the same averaging domain over the southeastern US described in Sect. 3.2, the production and loss of isoprene nitrates are shown in Fig. 6 and Table 2. Modeling results from the IS case suggest a large fraction (∼ 40 %) of isoprene nitrates production occurs from isoprene oxidation by NO 3 , and subsequent reactions with NO, NO 3 , HO 2 , and RO 2 . This branching ratio of sources from NO 3 oxidation is generally consistent with the findings (∼ 50 %) from Horowitz et al. (2007); however, direct evidence for enhanced formation from this pathway is rare. Ambient observations have been largely limited to mono-functional alkyl nitrates and total alkyl nitrates (Buhr et al., 1990;Flocke et al., 1998;Rosen et al., 2004;Cleary et al., 2005;Day et al., 2009). Only in a few cases the diurnal profiles of specific individual multifunctional organic nitrates have been quantified (Grossenbacher et al., 2001(Grossenbacher et al., , 2004. More ambient measurements are needed to verify model predictions. For the loss of total isoprene nitrates (ING), we count photochemical reactions as a loss process only if isoprene nitrates are converted to species that no longer retain the nitrate functional group. The model results show that ING loss is dominated by photooxidation by OH (38 %) and O 3 (14 %). O 3 plays a less important role than OH despite reacting rapidly with isoprene nitrates (Lockwood et al., 2010), reflecting the lower NO x recycling efficiency we assumed in O 3 +ISOPN reactions (∼ 30 % in O 3 +ISOPN versus ∼ 60 % in OH + ISOPN reactions). The additional loss consists of dry deposition (8 %) and cloud processes (18 %, i.e., wet deposition and convective mixing), photolysis (4 %), and advection/diffusion out of the modeling subdomain (17 %). The large contribution from photooxidation is consistent with findings by Shepson et al. (1996), Ito et al. (2007), Perring et al. (2009a), andPaulot et al. (2012). In contrast, Horowitz et al. (2007) concludes that dry deposition is the main sink. The discrepancy is partly due to a faster dry deposition velocity (as of HNO 3 ) used by Horowitz et al. (2007) and the fact that second-generation isoprene nitrates are assumed to be removed only by deposition processes in that study. The IS simulation also suggests there are large differences in the daytime versus nighttime lifetime of isoprene nitrates; the isoprene nitrates have a longer lifetime at night. ING 0 , ING 1 , and ING 2 attribute 16 %, 30 %, and 11 %, respectively, to ANs concentrations, with the remaining (∼ 40 %) from sources other than isoprene (Fig. S2, Supplement). Comparing the relative role played by these different generations of isoprene nitrate species (not shown), ING 0 and ING 1 account for about 40 % and 50 % of the total loss of ING (excluding advection/diffusion), respectively. The large contribution from ING 1 is due to its large fraction in the ING composition and also the fact that even though ING 1 is longer-lived than ING 0 (daytime lifetime ∼ 6.2 h versus ∼ 1.2 h), it is assumed to release NO x entirely during photooxidation, whereas a significant portion of ING 0 (∼ 70 %) is assumed to maintain the nitrate functional group in its reactions with OH, O 3 , and NO 3 . Being much longerlived (daytime lifetime about 18 h) and accounting for a small fraction of ING, ING 2 contributes only 7 % of the total loss, and may play an important role in the transport of NO x to remote regions and the free troposphere.

Sensitivity simulations and constraints with INTEX-NA/ICARTT data
In the previous sections, we have shown the peroxy radical fate as well as production and loss of isoprene nitrates from the IS simulation. Here we examine the sensitivity of model results to the most important uncertainties in our current understanding of isoprene nitrate chemistry, including the yield from the OH/NO reaction pathway, NO x recycling efficiency from ING 0 , dry deposition rate, and RO 2 +HO 2 reaction rate (Table 2 and Figs. 1, 5, and 6).

Sensitivity simulations
A. Daytime NO x recycling: when ING 0 reacts with OH, O 3 , and NO 3 , it is uncertain whether they will maintain the nitrate functional group or release NO x (Paulson and Seinfeld, 1992;Giacopelli et al., 2005;Paulot et al., 2009a). In the IS simulation, we use the recycling efficiency (∼ 60 %) from Paulot et al. (2009a) for the ISOPN+OH pathway. Since little is known about the recycling efficiency from ISOPN+O 3 reactions and from nitrates formed from isoprene+NO 3 reactions, we derive the likely products largely based on the SAPRC mechanism generation system. We investigate the uncertainty due to recycling efficiency separately for ING 0 formed from the OH and NO 3 pathways. When recycling is completely turned off, we assume an even split in production of ING 1 (represented by RNO3I) and ING 2 (represented by PROPNN/PROPNNB for OH/NO 3 pathways). When 100 % recycling efficiency is assumed for ING 0 formed from isoprene oxidation by OH (run A1), the overall recycling efficiency increases to ∼ 70 %. Isoprene nitrates production increases slightly (4 %) as a result of in-creased NO x , and therefore increases in the fraction of the peroxy radicals reacting with NO. The fraction of isoprene nitrates (ING) lost via photooxidation increases from ∼ 55 % in the IS simulation to ∼ 70 % in the A1 case. Loss through ING 0 increases by about a factor of 2, causing a substantially reduced lifetime of ING (∼ 6 hours to ∼ 4 hours during the day). In contrast, when recycling of NO x is completely turned off for these hydroxy isoprene nitrates (run A2), the overall recycling efficiency is reduced to ∼ 10 %. Production of isoprene nitrates decreases slightly (−3 %) as a result of decreased NO x and thus a decrease in the fraction of the peroxy radicals reacting with NO. Loss via dry deposition and clouds becomes comparable to that via photooxidation, with each process accounting for 35-40 % of the total loss of ING. Loss through ING 0 decreases by about a factor of 2, and consequently the lifetime of ING increases to ∼ 10 h during the day.
B. Nighttime NO x recycling: when 100 % recycling of NO x is assumed for the ING 0 formed from the isoprene oxidation by NO 3 (run B1), the change is generally similar (though smaller) to trends in the simulation examining the daytime recycling efficiency (run A1). When the recycling is completely turned off (run B2), the overall effects on the production and loss are relatively small, as the IS scheme already assumes the nitrates formed from isoprene + NO 3 reactions release little NO x while reacting with OH and NO 3 . The most significant change is found in the lifetime of ING at night, which increases by a factor of 2. Note that we only examine the recycling of NO x from isoprene nitrates formed exclusively from NO 3 oxidation (C5-nitrooxycarbonyl and C5-nitrooxyhydroperoxide), which accounts for ∼ 90 % of isoprene nitrates production at night.
C. Dry deposition: in the IS run, ING 0 and ING 1 use Henry's law constants of H 298 = 1.7 × 10 4 M atm −1 (Ito et al., 2007) and ING 2 use a value of H 298 = 10 3 M atm −1 (Sander, 1999) to estimate their loss by wet deposition. We assume the dry deposition velocity is equal to that of PAN (∼ 0.6 cm s −1 at daytime), which is on the slow side of measured values (Shepson et al., 1996;. A sensitivity simulation (run C) is conducted to examine the effect of this uncertainty in dry deposition velocity in which we increase the rate to that of HNO 3 (∼ 3 cm s −1 ), considered to be an upper limit. In this simulation, we find that loss of ING via dry deposition increases from 8 % to 26 % of the total. Combined with removal by clouds, the deposition loss (40 %) becomes comparable to that of photooxidation (47 %). The lifetime of ING during the day also drops considerably (to ∼ 4 h).
D. Isoprene nitrate yield: with a yield of 12 % from OH/NO reactions and 70 % from NO 3 reactions, the IS simulation finds that about 40 % of isoprene nitrates are produced from the NO 3 pathways. There have been a number of laboratory studies reporting the yield of the OH/NO pathway, as mentioned previously. Holding the yield from the NO 3 pathway constant, we conduct a sensitivity simulation (run D) with a yield of 6 % from the OH/NO pathway, and this reduces the total production of isoprene nitrates by 20 %. The change also results in increased relative importance of NO 3 pathways, which accounts for half of isoprene nitrates.
E. RO 2 + HO 2 reaction rate: in the IS case, the RO 2 +HO 2 reaction rates increase with the size of the molecule, consistent with the laboratory measurements (Rowley et al., 1992;Jenkin and Hayman,1995;Boyd et al., 2003). To assess the effects of using a uniform reaction rate for this class of reactions (as has been assumed in numerous previous studies), a sensitivity simulation is conducted where we change the rates to the value used by the base run (about half of the rate used in the IS scheme for isoprene peroxy radicals at 298 K). This change has a significant impact on the relative importance of NO and HO 2 pathways for the peroxy radicals formed by isoprene reacting with OH (ISOPO 2 ). With the reduced rate of HO 2 +ISOPO 2 , NO becomes the dominant channel (∼ 50 %) and HO 2 accounts for only ∼ 25 %. As a result, isoprene nitrates production increases by ∼ 10 %. For the peroxy radicals (NISOPO 2 ) formed by isoprene reacting with NO 3 , the slower RO 2 +HO 2 rates switch the relative importance of HO 2 and RO 2 pathways, but the role of NO and NO 3 channels remains generally the same.

Observational constraints
From previous analysis (Sect. 3.1), we have shown that while the IS simulation has improved model performance for a range of species, there is still notable bias in terms of ANs and its correlation with O 3 . We also find uncertainties in the isoprene nitrate chemistry have a significant impact on their production and loss processes. Here we use observed ANs and O 3 concentrations to further constrain these uncertainties.
In environments with high isoprene concentrations (> 500 pptv), the slope of the correlation between ANs and O 3 (Fig. 7) is best mimicked by the simulation with reduced isoprene nitrate yield (run D), while the two 100 % recycling NO x cases (run A1 and B1) and the fast dry deposition case (run C) also show reasonably good performance and fall within the uncertainty range. On the other hand, the simulation with 0 % NO x recycling from ING 0 formed from the OH/NO pathway (run A2) substantially degrades the performance, underpredicting the slope by about 40 %.
The simulations that have a slope of ANs versus O 3 similar to the observations also show better agreement in terms of ANs concentrations. In environments with high isoprene concentrations (> 500 pptv), the IS simulation shows ∼ 45 % overestimation of the median ANs concentrations, while the base run substantially degrades the model performance with a bias of ∼ 90 %. Best agreement is achieved in the runs assuming 100 % recycling of NO x from ING 0 formed from isoprene oxidation by OH (run A1), using the isoprene nitrate yield of 6 % from the OH/NO pathway (run D), or using a fast dry deposition rate as HNO 3 (run C). When a recycling of 0 % from ING 0 oxidation is assumed (run A2), the agreement substantially worsens with predictions close to the base case. Note that a recent study by Crounse et al. (2012) suggests that methacrolein hydroxy peroxy radicals mostly undergo isomerization and decomposition without forming MACRN in the atmosphere. This, however, should only have minor effects on our conclusions as MACR and MVK together contribute 13-18 % of isoprene nitrates production in the model simulations.
In summary, the base case shows substantial overestimation of ANs concentrations and underestimation of the O 3 / ANs slope. The IS simulation improves both, though it still overpredicts ANs by ∼ 45 % and underpredicts the O 3 / ANs slope by 20 %. The NO x recycling efficiency from ING 0 formed from the OH/NO pathway appears to have the largest impact on the model performance, followed by isoprene nitrate yield, dry deposition rate, and the NO x recycling efficiency from ING 0 formed from the NO 3 pathway. Improved agreement with the observations can be produced by reducing isoprene nitrate concentrations via any one of the following: (1) using a 6 % yield of isoprene nitrates from the OH/NO pathway; (2) allowing isoprene nitrates to dry deposit as rapidly as HNO 3 ; (3) assuming a NO x recycling efficiency of 100 % from ING 0 formed from OH oxidation; or (4) assuming a NO x recycling efficiency of 100 % from ING 0 formed from NO 3 oxidation. On the other hand, assuming 0 % NO x recycling from ING 0 formed from OH oxidation shows substantially degraded performance, similar to that of the base run. Note that we evaluate the uncertainties in isoprene nitrate chemistry by varying one parameter at a time, and therefore combinations of changes might further improve the comparison with observations. For example, even though the simulation with 0 % NO x recycling efficiency for ING 0 formed from NO 3 oxidation shows some underestimation of the O 3 / ANs slope, it might result in acceptable performance if the isoprene nitrate yield is reduced or a fast dry deposition rate is used. Nevertheless, we find similar daytime lifetimes of isoprene nitrates (∼ 4 h for the 12 % yield cases and ∼ 6 h for the 6 % yield case) in both simulations that show good agreement with observations. However, our CMAQ simulation suggests that non-isoprene organic nitrates are 40 % of the ANs. Errors in the nonisoprene ANs may influence these results.

Impacts on OH and Ox
The IS scheme has significant impact on NO x , NO 3 , and biogenic SOA concentrations as compared to the base case. We also find uncertainties in the IS scheme can have large effects on the sources and sinks of isoprene nitrates. Here we further examine the sensitivity of O 3 and OH to these uncertainties in the isoprene nitrate chemistry. We start our investigation by quantifying the variations in fate of NO x due to the uncertainties in isoprene nitrate chemistry. Using the methodology of Paulot et al. (2012), the local net removal of NO x by isoprene nitrates, D ING , is defined as where P ING0 represents photochemical production of firstgeneration isoprene nitrates, P ING1 new represents production of isoprene nitrates from MVK and MACR, α × L ING0 represents NO x recycled from ING 0 , and L ING1 and L ING2 represent photochemical loss of ING 1 and ING 2 . Figure 8 shows that changes in O x (=O 3 + NO 2 ) production are well correlated with changes in the amount of NO x removed by isoprene nitrate chemistry (D ING ) in the southeastern United States, consistent with the findings of Paulot et al. (2012).
The uncertainties in the isoprene nitrate chemistry can impact the production of O x (P(O x )) by about 10 %, while the sensitivity of P(O x ) to changes in D ING ( P(O x )/ D ING ) is about −15 %. The simulation using slower RO 2 + HO 2 rates behaves slightly differently from the other models runs, as an increased fraction of RO 2 reacts with NO, and consequently increased O x production also plays a role.  studied the effects of these uncertainties in tropical regions (South America, Southeast Asia, and Africa) with a range of NO x emissions. Compared to their results, we find P(O x )/ D ING in our study region is similar to tropical locations which have the highest NO x emissions, i.e., Southeast Asia or Africa during the biomass burning season. Much greater impact is reported by Paulot et al. (2012) for South America, which is characterized by very low NO and therefore larger ozone production efficiency. Furthermore, the NO x recycling efficiency from ING 0 appears to have the largest impact on ozone production in our study region, whereas more equally shared importance between deposition velocity, NO x recycling efficiency, and isoprene nitrate yield is indicated by the results of Paulot et al. (2012) for South America. The differences are likely a result of higher OH and O 3 concentrations, and therefore increased importance of photochemical loss in our study area.
Changes in OH concentrations are also well correlated with changes in D ING , and the uncertainties in isoprene nitrate chemistry can impact OH concentrations by ∼ 6 %. The results show OH concentrations generally increase as D ING decreases. However, further reduction of D ING also increases the formation of HNO 3 by reaction of NO 2 +OH, which removes NO x and OH simultaneously (case A1). Compared to the changes from the base to the IS scheme (∼ 30 %), we find less of an impact on OH due to the uncertainties in isoprene nitrate chemistry. This is because the increase in OH in the IS scheme is primarily driven by increased HO x production rather than changes in NO x .

Conclusions
Air quality in the southeastern United States is influenced by the interplay between abundant isoprene emissions and anthropogenic pollutants. We incorporate recent advances in our understanding of isoprene oxidation chemistry into the SAPRC-07 chemical mechanism within the CMAQ model, including isoprene nitrates formation and oxidation, epoxide formation, and isoprene peroxy radical isomerization. We use observations from the 2004 INTEX-NA/ICARTT field campaign to evaluate model performance and constrain the uncertainties in the isoprene nitrate chemistry.
We find the new model improves the simulation of the aircraft observations for a range of species. We find that most isoprene nitrates are relatively short-lived and prone to returning NO x , and thus produce increased NO x (4-9 %) and O 3 (2 ppbv) in the southeastern US. OH concentrations increase by ∼ 30 % primarily as a result of increased HO x production. Laboratory studies have also observed OH reformation from RO 2 + HO 2 reactions, with differing yields, depending on the source of peroxy radicals (Hasson et al., 2004;Jenkin et al., 2007;Taraborrelli et al., 2012). In this work we have considered OH reformation from isoprene peroxy radicals (Paulot et al., 2009b, Taraborrelli et al., 2012Liu et al., 2012), but not from acyl peroxy radicals and β-oxo peroxy radicals. Assuming OH yield of up to 50 %, recent modeling work has shown that these reactions increase OH concentrations by 5-40 % (Pugh et al., 2010;Kubistin et al., 2010;Stavrakou et al., 2010;. We would expect an additional increase in OH if this source of OH were included in the IS mechanism. The increase in oxidants (OH, O 3 , and NO 3 ) causes concentrations of biogenic SOA to increase by 15 %. Future work will integrate the detailed representation of isoprene oxidation described here with recent advances elucidating the chemical origins of isoprene SOA (Surratt et al., 2010;Lin et al., 2012).
HO 2 reactions with isoprene peroxy radical are estimated to account for ∼ 40 % of total reactions during both daytime and nighttime oxidation, highlighting the importance of a better understanding of the oxidation products from this pathway even for a region with elevated NO x concentrations. The isoprene peroxy radical isomerization accounts for ∼ 10 % of the total.
We examine the sensitivity of our model results to the key assumptions in the isoprene nitrate chemistry, including the yield from the OH/NO pathway, NO x recycling efficiency from first generation isoprene nitrates, dry deposition rate, and RO 2 +HO 2 reaction rates. Constrained by observed ANs and their correlation with O 3 , reasonably good agreement can be achieved using a range of isoprene nitrate yields and daytime lifetimes: between 6 %, ∼ 6 h and 12 %, ∼ 4 h.
Uncertainties in the isoprene nitrate chemistry can impact ozone production by ∼ 10 % and OH concentrations by ∼ 6 % in the southeastern United States. The effects are wellcorrelated with the changes in the net amount of NO x removed by isoprene nitrate chemistry, consistent with findings from Paulot et al. (2012). The uncertainties in NO x recycling efficiency appear to have a larger impact than uncertainties in isoprene nitrate yield and dry deposition velocity.
We find 40-50 % of the isoprene nitrates are formed from isoprene + NO 3 reactions, consistent with previous findings (Horowitz et al., 2007). Substantial uncertainty exists in the nighttime isoprene nitrate chemistry and peroxy radical fate, and further studies are required to elucidate the oxidation pathways and products. We find that photooxidation accounts for about half of the total loss of isoprene nitrates even when these compounds dry deposit as rapidly as HNO 3 . A significant fraction of the loss proceeds via production of secondary oxidation products whose fate remain very uncertain.
In regions at the confluence of biogenic emissions and anthropogenic pollution, accurate simulations of the ozone and secondary organic aerosol depend on representing the isoprene oxidation chemistry accurately. Further progress depends on improved understanding of isoprene oxidation pathways, rate of NO x recycling from isoprene nitrates, and the fate of the secondary, tertiary, and further oxidation products of isoprene.