Quantifying atmospheric nitrate formation pathways based on a global model of the oxygen isotopic composition ( 1 17 O ) of atmospheric nitrate

The oxygen isotopic composition ( 117O) of atmospheric nitrate is a function of the relative abundance of atmospheric oxidants (O 3, ROx=OH+HO2+RO2) and the formation pathway of nitrate from its precursor NO x (=NO+NO2). Coupled observations and modeling of nitrate117O can be used to quantify the relative importance of chemical formation pathways leading to nitrate formation and reduce uncertainties in the budget of reactive nitrogen chemistry in the atmosphere. We present the first global model of atmospheric nitrate 117O and compare with available observations. The largest uncertainty for calculations of nitrate117O is the unconstrained variability in the 117O value of tropospheric ozone. The model shows the best agreement with a global compilation of observations when assuming a117O value of tropospheric ozone equal to 35‰ and preferential oxidation of NO x by the terminal oxygen atoms of ozone. Calculated values of annual-mean nitrate 117O in the lowest model layer (0–200 m above the surface) vary from 7‰ in the tropics to 41‰ in the polar-regions. The global, annual-mean tropospheric inorganic nitrate burden is dominated by nitrate formation via NO 2+OH (76%), followed by N2O5 hydrolysis (18%) and NO3+DMS/HC (4%). Calculated nitrate117O is sensitive to the relative importance of each nitrate formation pathway, suggesting that observations of nitrate117O can be used to quantify the importance of individual reactions (e.g. N 2O5 hydrolysis) leading to nitrate formation if the117O value of ozone is known. Correspondence to: B. Alexander (beckya@u.washington.edu)


Introduction
The formation and cycling of reactive nitrogen in the atmosphere has important implications for air quality, the oxidation capacity of the atmosphere, and atmospheric nitrate (nutrient) deposition.Combustion of fossil fuels, biofuels, and biomass and lightning converts or "fixes" inert nitrogen gas (N 2 ) into a highly reactive form (NO x =NO+NO 2 ).Other sources of NO x to the atmosphere include microbial processes in soils and transport from the stratosphere (Logan, 1983).Anthropogenic activities currently dominate NO x sources to the troposphere (Jaeglé et al., 2005).The influence of human activities on the atmospheric nitrogen budget is evident in the record of increasing nitrate concentrations over the past ∼100 years in Greenland ice cores (Mayewski et al., 1990).
The photochemical cycling of NO x leads to the formation of tropospheric ozone (O 3 ), a major air pollutant.Tropospheric ozone and its byproduct, the hydroxyl radical (OH), largely determine the oxidizing capacity of the atmosphere and the lifetime of most reduced trace gases (Thompson, 1992).The formation of nitrate, defined herein as gas-phase HNO 3 plus particulate NO − 3 , is the main sink of NO x in the atmosphere.Nitrate is soluble and is lost from the atmosphere through wet and dry deposition to the Earth's surface, providing a nutrient source to many ecosystems (Galloway et al., 2008).
Figure 1 summarizes the chemistry of NO x cycling and nitrate formation in the atmosphere.Upon emission, NO x cycles rapidly during the daytime between NO and NO 2 via oxidation (Reactions R1-R3) and photolysis (Reaction R4).
Published by Copernicus Publications on behalf of the European Geosciences Union.NO is oxidized to NO 2 by ozone (O 3 ) (Reaction R1) and peroxy radicals (HO 2 and RO 2 ) (Reactions R2-R3).NO 2 is also produced through decomposition or oxidation of peroxyacyl nitrates (PANs), reactions of peroxy radicals (RO 2 ) with peroxynitrates (RO 2 NO 2 ) and alkyl nitrates (RONO 2 ), and oxidation or decomposition of nitrous acid (HONO), pernitric acid (HNO 4 ), nitrate radicals (NO 3 ) and dinitrogen pentoxide (N 2 O 5 ).NO 2 is lost from the atmosphere through oxidation and dry deposition to the surface.NO 2 is oxidized by OH (daytime) to form HNO 3 (Reaction R5) or by O 3 to form NO 3 (Reaction R6).NO 3 is rapidly photolyzed during the daytime, so that NO 3 concentrations are only significant at night.At night, NO 3 reacts with dimethylsulfide (DMS) or hydrocarbons (HC) (Reaction R7) or with NO 2 to form N 2 O 5 followed by hydrolysis on the surface of aerosols (Reaction R8) to form HNO 3 .The lifetime of NO x against conversion to HNO 3 in the boundary layer varies from ∼1 day (tropics and in summer in the mid-to-high latitudes) to ∼3 days (high latitudes in winter) (Levy et al., 1999).Nitrate is lost from the atmosphere mainly through wet and dry deposition.
The oxygen isotopic composition of nitrate reflects the relative importance of different oxidants in NO x cycling and nitrate formation (Michalski et al., 2003;Hastings et al., 2003).Atmospheric nitrate exhibits an anomalous ("massindependent") oxygen isotopic composition.The isotopic composition of atmospheric nitrate is considered anomalous due to enrichment in 17 O relative to 18 O over the expected relationship (δ 17 O≈0.5×δ 18O) that results from purely massdependent fractionation processes (Matsuhisa et al., 1978).Isotope ratios are expressed in "delta notation", where x=17 or 18 and the standard used for oxygen isotopic analysis is Standard Mean Ocean Water (SMOW).
Generally, δ values are expressed in per mil (‰).The oxygen isotopic anomaly in nitrate, quantified here as 17 O=δ 17 O−0.52×δ 18O, results mainly from the transfer of an isotopic anomaly in atmospheric ozone during oxidation of NO and NO 2 (Reactions R1 and R6 in Fig. 1).
Reactive bromine (BrO) can also play a role in both NO x cycling and nitrate formation in polar regions (Evans et al., 2003;Saiz-Lopez et al., 2008).BrO participates in NO x cycling and nitrate formation through the following reactions: BrO is expected to have large 17 O values due to the involvement of O 3 in BrO formation (Morin et al., 2007).The formation of nitrate through BrO oxidation and BrONO 2 hydrolysis is thus expected to lead to large 17 O (nitrate) values on the upper end of the range of observations.The global importance of reactive bromine in NO x cycling and nitrate formation remains to be quantified, but is thought to have significant regional impacts on both tropospheric NO x concentrations and ozone production (Yang et al., 2005).
Understanding the importance of different oxidants for NO x cycling and the pathways of nitrate formation is critical for understanding the budget of reactive nitrogen in the atmosphere.Here we present the first global chemical transport model of the oxygen isotopic composition of atmospheric nitrate.Comparison with available observations sheds light on previous assumptions used in box model studies regarding the isotopic composition of ozone and the isotopic transfer function during NO x oxidation reactions, and provides a means to test and validate the model's representation of reactive nitrogen chemistry.

Model description
We utilize the GEOS-Chem global 3-D model of coupled aerosol-oxidant chemistry (Park et al., 2004)  The tropospheric O 3 -NO x -hydrocarbon simulation was first described by Bey et al. (2001) with updates by Fiore et al. (2002) and Martin et al. (2002).Emissions of NO x total 39.3 TgN yr −1 including sources from fossil fuel burning (including aircraft) (25.2 TgN yr −1 ), biofuel (2.2 TgN yr −1 ), biomass burning (5.3 TgN yr −1 ), lightning (0.1 TgN yr −1 ) and soil (including fertilizer) (6.6 TgN yr −1 ).Anthropogenic NO x emissions were taken from the Global Emission Inventory Activity (GEIA) (Benkovitz et al., 1996), scaled by country on the basis of energy statistics to the year 1995 as described by Bey et al. (2001).The monthly inventory of emissions from biomass burning are from the Global Fire Emissions Database (GFEDv2.1)for the year 2005 (Randerson et al., 2007;vanderWerf et al., 2006).Soil NO x emissions are computed using a modified version of the algorithm proposed by Yienger and Levy (1995) with the canopy reduction factors described by Wang et al. (1998).Emissions of NO x from lightning are linked to deep convection following the parameterization of Price and Rind (1992) with vertical profiles taken from Pickering et al. (1998).Stratospheric NO y (=NO x +HNO 3 ) concentrations are calculated using NO y production rates and its partitioning into NO x and HNO 3 provided by a global 2-D stratospheric chemistry model (Schneider et al., 2000).The model represents transport of stratospheric HNO 3 and NO x into the troposphere; however, there is no explicit representation of polar stratospheric cloud (PSC) sedimentation and resulting stratospheric denitrification.
The aerosol and oxidant chemistry are coupled through the formation of sulfate and nitrate, heterogeneous chemistry, and aerosol effects on photolysis rates.Optical prop-erties are calculated for each aerosol component as a function of local relative humidity as described by Martin et al. (2003).Reactions involving aerosols are described by Jacob (2000) with updates for N 2 O 5 hydrolysis as described in Evans and Jacob (2005).Photolysis frequencies are computed using the Fast-J radiative transfer algorithm (Wild et al., 2000) which allows for Raleigh scattering as well as for Mie scattering by clouds and aerosols.Total inorganic nitrate is partitioned between the gas and particulate (≤1 µm diameter) phases for the K + -Ca 2+ -Mg 2+ -NH where k R1 , k R2 , and k R3 represent rate constants for Reactions (R1), (R2) and (R3) respectively. 17O(O * 3 ) represents the isotopic anomaly that is transferred from O 3 to NO 2 during oxidation of NO. 17 O(O * where 17 O(O 3 ) is the bulk isotopic composition of ozone and 17 O(O * 3 ) is used in Eqs. ( 2) and (4).The above calculation Eq. (2a) assumes that 17 O(HO 2 )=0‰, in contrast to observations of 17 O(H 2 O 2 ) generally between 1-2‰ (H 2 O 2 forms mainly through the self reaction of HO 2 ) (Savarino and Thiemens, 1999b).However, the assumption that 17 O(HO 2 )=0‰ versus 17 O(HO 2 )=2‰ changes our calculated 17 O(nitrate) by less than 1‰ over the global range of calculated A values (Fig. 3 and Sect.4.2) and simplifies the calculations.The assumption that 17 O(HO 2 )=0‰ is also consistent with other studies (Michalski et al., 2003;Morin et al., 2008;Kunasek et al., 2008).NO x will achieve isotopic equilibrium locally with O 3 , HO 2 , and RO 2 during the daytime prior to conversion to nitrate.For the daytime (Reaction R5), the local A day value is calculated in the model based on the relative NO 2 production rates (Reactions R1-R3) between 10:00-14:00 hours local time.For the nighttime (Reactions R6, R6 and R6, R8), calculation of the local A value is more complicated because photochemical steadystate does not exist at night.Since peroxy radical concentrations (whose production rates are dependent upon UV radiation) will drop off before NO x cycling (which occurs at visible wavelengths) ceases, NO x will continue achieve isotopic equilibrium in the late afternoon resulting in A values near unity.Thus, for the nighttime (Reactions R6, R7 and R6, R8), the local A night value is calculated based on the relative NO 2 production rates between midnight to 2am local time.This assumes that the relative importance of O 3 and peroxy radicals in the oxidation of NO to NO 2 at night is similar to that during dusk when NO x photochemical steady-state exists.This assumption may result in an overestimate of 17 O(nitrate) R6, R7 and 17 O(nitrate) R6, R8 due to a broadening of the peroxy radical diurnal cycle from photolysis of species other than ozone (e.g.HCHO, HONO, H 2 O 2 , CH 3 O 2 H) (Fleming et al., 2006;Carpenter et al., 1997), but is likely to be small on the global scale.
Nitrate in the model is transported as four separate tracers depending on its production pathway (Reactions R5-R8 in Fig. 1 plus stratospheric nitrate).Each nitrate tracer is assigned a 17 O value according to its involvement with O 3 during formation, similar to that proposed by (Michalski et al., 2003): We assume O 3 will dominate NO x cycling in the stratosphere leading to A=1.We use a factor of 5/6 to calculate the 17 O value of stratospheric nitrate ( 17 O(nitrate) strat ) to be consistent with other studies (McCabe et al., 2007;Savarino et al., 2007) that assume stratospheric nitrate forms via N 2 O 5 hydrolysis.This neglects nitrate formation via ClONO 2 decomposition and possible non-zero values of 17 O(H 2 O) strat (Franz et al., 2005;Zahn et al., 2006), leading to a possible underestimate of 17 O(nitrate) strat .However, our results are insensitive to this assumption as stratospheric nitrate is a negligible source of nitrate in the model (Sect.4.2). 17O(nitrate) is calculated according to the local importance of O 3 in NO x cycling (A) and the relative abundance of each nitrate tracer (from transport and local production) according to the following,  (nitrate).This error will be most significant in remote regions with no local sources of NO x (e.g.polar winter) -the magnitude of this error (4-5‰) is explored in more detail in Sect.4.3.
We perform sensitivity studies to cover the range of observed tropospheric 17 O(O 3 ) values (25-35‰).We assume stratospheric 17 O(O 3 )=40‰in all simulations, a value towards the upper end of observations (10.4-45.7‰)(Mauersberger et al., 2001) and consistent with model studies (Liang et al., 2006).We note that calculations of 17 O(nitrate) formed within the troposphere use tropospheric values for 17 O(O 3 ).We do not distinguish between tropospheric and stratospheric-derived 17 O(O 3 ) within the troposphere.Globally, production within the troposphere dominates tropospheric ozone (Prather and Ehhalt, 2001); however, this distinction may be important locally, particularly at high altitudes in the mid-to high-latitudes, and its importance will vary seasonally.For example, (Terao et al., 2008) estimated that stratospheric-derived ozone peaked at 40% (30%) of total ozone in spring in high (mid) latitudes at 500 hPa; however, the contribution of stratospheric ozone at the surface is generally small (Fiore et al., 2003).The magnitude of the error introduced upon our tropospheric 17 O(nitrate) calculations will depend upon the difference between the value of 17 O(O 3 ) produced in the troposphere and stratosphere, as well as the influence of stratospheric-derived ozone at any particular location.Since nitrate production is dominated in the boundary layer, we expect this error in general to be small at the surface at low altitudes where most of the observations of 17 O(nitrate) are located, but could be of seasonal importance in the mid-to high-latitudes.
Previous box-model studies have assumed either an equal likelihood of all three O-atoms of O 3 (Michalski et al., 2003;Kunasek et al., 2008) or favored oxidation by the terminal Oatom of O 3 (Morin et al., 2008) during oxidation of NO and NO 2 .We perform additional sensitivity studies to account for differing assumptions regarding the transfer mechanism of the isotopic anomaly from O 3 to NO 2 and NO 3 upon oxidation (Reactions R1 and R6).Whether or not the terminal O-atom of ozone is favored as the reacting atom influences the resulting isotopic composition of nitrate because the asymmetric isotopologues of ozone are enriched in heavy oxygen isotopes with respect to bulk ozone (e.g. 18O 16 O 16 O vs. 16 O 18 O 16 O) (Janssen, 2005;Bhattacharya et al., 2008;Michalski and Bhattacharya, 2009).
Observations of 17 O(nitrate) were made using two types of analysis: the "silver salt pyrolysis" method (Michalski et al., 2002) (observations from La Jolla, South Pole, Summit, Chile, Ecuador and COCA) and the "denitrifier" method (Kaiser et al., 2007) (observations from Alert, Dumont D'Urville, Princeton, Bermuda, and the Atlantic cruise).The silver salt pyrolysis method relies on isolation of dissolved, inorganic nitrate (NO − 3 ) prior to analysis.Nitrate anion separation ensures that only inorganic nitrate is measured, assuming that soluble organic nitrate does not dissociate in water.Observations of C 1 -C 5 alkyl nitrates in wet deposition (rain, snow, frost) (Hauff et al., 1998) suggest that they do not readily dissociate.The denitrifier method does not require isolation of NO − 3 and uses denitrifying bacteria (Pseudomonas aureofaciens) to convert nitrate (and nitrite) into N 2 O, which is then quantitatively thermally decomposed into O 2 and N 2 .It is generally assumed that the denitrifier method selectively measures inorganic nitrate; however, this has yet to be specifically demonstrated.Hawari et al. (2000) showed that biological degradation of RDX (hexahydro-1,3,5-trinitro-1,3,5triazine) produced N 2 O as a byproduct, suggesting that certain types of microorganisms can convert soluble organic nitrates into N 2 O.It is not known whether or not Pseudomonas aureofaciens will do the same.As will be shown, whether or not we include organic nitrates in our isotope calculations is significant for calculated 17 O(nitrate) in regions where organic nitrates are abundant.NO 2 +O 3 reactions.Although the comparison in Fig. 2 suggests that 17 O(O 3 )=35‰ is the appropriate value to use in calculations of 17 O(nitrate), it is no substitute for direct observations of 17 O(O 3 ), as model biases will impact our results.Model errors in transport may impact for example the vertical distribution of NO x , which will in turn influence 17 O(nitrate) due to changing A values (generally decreasing with height) or changes in the importance of different nitrate formation pathways with altitude.Global biases in oxidant (O 3 , NO 3 , OH, RO 2 , HO 2 ) concentrations or the formation rate of nitrate for specific pathways (e.g.N 2 O 5 hydrolysis) will also impact the isotopic calculations.For example, if the true 17 O(O 3 )=25‰, then our finding that 17 O(O 3 )=35‰ provides the best agreement between observed and modeled 17 O(nitrate) may indicate that the model globally and systematically underestimates (overestimates) O 3 and/or NO 3 (OH and/or peroxy radical) concentrations.The resulting magnitude of the bias in 17 O(nitrate) will be dependent upon the magnitude of the bias in O 3 , NO 3 , OH, or peroxy radical concentrations.Our calculated global mean OH concentration (10.7×10 5 molecules cm −3 ) compares well with (Wang et al., 2008) (10.6×10 5 molecules cm −3 ) who optimized the global abundance of OH by interpreting observations of methyl chloroform using an inverse technique and the GEOS-Chem model.This suggests no global, systematic bias in OH.Calculated O 3 concentrations in the GEOS-Chem model has been extensively evaluated against observations (Hudman et al., 2009;Zhang et al., 2008;Terao et al., 2008;Wu et al., 2007;Wang et al., 2009), suggesting no systematic global bias.Peroxy radical concentrations would have to be overestimated globally by a factor of ∼3 to account for a 10‰ difference in 17 O(NO x ).Although we may expect significant regional biases in these radical concentrations, a systematic global bias seems unlikely for species (or nitrate formation pathways) with a high degree of spatial variability.

Model comparison with observations
We note that Michalski et al. (2003) and Kunasek et al. (2008) achieved good agreement with observations of 17 O(nitrate) using a box models that assumed 17 O(O 3 )=35‰ with the statistical transfer function.In their analysis of observed 17 O(nitrate) in La Jolla, CA, Michalski et al. (2003) estimated a much larger fraction of nitrate formed via N 2 O 5 hydrolysis (30-90%) than we do (5-50%), revealing the sensitivity of 17 O(nitrate) to the fractional contribution of N 2 O 5 hyrolysis to total nitrate formation.We expect that Michalski et al. (2003) significantly overestimated the contribution of N 2 O 5 hydrolysis in their model due to the use of a reaction rate constant that did not account for variability of factors such as aerosol surface area, chemical composition, humidity and temperature.Also, it is generally thought that models overestimate the importance of nitrate formation via N 2 O 5 hydrolysis through overestimates of the reaction probability (Davis et al., 2008;Mc-Neill et al., 2006;Brown et al., 2009).Kunasek et al. (2008) also obtained good agreement with snowpit observations of 17 O(nitrate) in Summit, Greenland with a box model that assumed 17 O(O 3 )=35‰ with the statistical isotopic transfer function.Their box model used boundary conditions from the same global model used here (GEOS-Chem), but inherently did not account for transport of nitrate formed at lower latitudes (with lower values of 17 O(nitrate)).This demonstrates the importance of accounting for transport of nitrate from different regions with potentially different values of 17 O(nitrate), particularly for remote polar locations far from major sources of NO x .

Global variability of tropospheric nitrate formation pathways and 17 O(nitrate)
Figure 3 shows global plots of modeled June-July-August and December-January-February mean 17 O(nitrate) in the lowest model layer (0-200 m above the surface). 17O(nitrate) ranges from 7-41‰, with A day and A night generally ranging from 0.2-1.0 and 0.9-1.0,respectively.The spatial variability of 17 O(nitrate) is largely determined by the importance of O 3 in NO x cycling (A) due to the fact that the isotopic composition of two-thirds of the oxygen atoms of nitrate is determined during NO x cycling.The oxidation of NO x to nitrate plays a secondary but significant role in determining the final 17 O(nitrate) value.
Figure 4 shows the fractional importance of each nitrate production pathway to annual mean nitrate concentrations in the lowest model layer (0-200 m above the surface).Nitrate production via NO 2 +OH (Reaction R5) dominates (up to 87%) in the tropics where OH concentrations are highest.Nitrate production via N 2 O 5 hydrolysis (Reaction R8) dominates (up to 74%) at high northern latitudes over the continents and the Arctic, consistent with observations and modeling by Tie et al. (2003) and Stroud et al. (2003).Nitrate production via reaction of NO 3 with (primarily) DMS (Reaction R7) is most important (up to 46%) in the high latitudes in the marine boundary layer.The stratospheric source of nitrate is negligible (annual average maximum of 2% in Antarctica).The annual-mean fractional contribution to the tropospheric inorganic nitrate burden for nitrate formed via NO 2 +OH (Reaction R5), N 2 O 5 hydrolysis (Reactions R6, R8), and NO 3 +DMS/HC (Reactions R6, R7) is 76%, 18%, and 4%, respectively.The remaining inorganic nitrate burden (2%) is from the stratosphere.It is important to note here that we are likely underestimating the stratospheric source in polar-regions due to the lack of an explicit stratospheric denitrification mechanism from the sedimentation of PSCs in the model.Stratospheric nitrate is thought to be an important (but unquantified) source of nitrate to Antarctica (Legrand and Kirchner, 1990;Mc-Morrow et al., 2004;Mulvaney and Wolff, 1993;Wagenbach et al., 1998;Weller et al., 2002;Savarino et al., 2007).Indeed, the lack of a stratospheric denitrification mechanism in the model may explain the model's inability to capture the seasonality of the observations at DDU, Antarctica (Savarino et al., 2007) (see Fig. 5).

Model discrepancies
Figure 5 compares monthly-mean 17 O(nitrate) model calculations with observations available over the course of a full year (Alert, Summit, Princeton, La Jolla, DDU, and the South Pole).The model reproduces the seasonality of the 17 O(nitrate) observations.The model shows good agreement (generally within 5‰) with the observations in lowto mid-latitudes (La Jolla, Bermuda, COCA, Atlantic cruise, Chile and Ecuador; Fig. 2), though additional observations in low-latitude forested regions are necessary to validate model predictions of 17 O(nitrate) less than 20‰.In general, the model tends to underestimate the observations during spring and summer in polar-regions by 0-5‰, and overestimate the observations the northern hemisphere mid-to highlatitudes by 1-7‰ with maximum overestimates in polar winter.
The model underestimate (5‰) of 17 O(nitrate) maximizes in polar regions in spring when nitrate formation from BrONO 2 hydrolysis is expected to be significant (Morin et al., 2007;Evans et al., 2003;Saiz-Lopez et al., 2008).BrO radical concentrations can be as high as ∼30 ppt during spring in both the Arctic and Antarctic close to seaice (Chance, 1998;Richter et al., 1997;Wagner and Platt, 1998;Kreher et al., 1997).Several indirect lines of evidence suggest a non-negligible contribution of bromine radical chemistry in the inland Arctic (Summit, Greenland) during summer months (Grannas et al., 2007;Peterson and Honrath, 2001;Sjostedt et al., 2007), and are supported by BrO observations of 1-3 ppt during early summer at Summit, Greenland (Huey et al., 2007).The model underestimate of 17 O(nitrate) in the polar-regions during spring is likely due to the lack of BrO chemistry in the model, which may also contribute to the underestimate in summer.Kunasek et al. (2008) estimated a 4‰ increase in summertime 17 O(nitrate) at Summit, Greenland after adding reactive bromine chemistry to their box model assuming BrO concentrations of 3 ppt.This suggests that BrO chemistry can fully account for the 3‰ underestimate in calculated 17 O(nitrate) at Summit in July (Fig. 5).In addition, photochemical processing of snowpack nitrate during spring and summer leads to a significant local source of NO x in the polar-regions (Jones et al., 2000;Honrath et al., 2002) that is not accounted for in the model.Local reprocessing of previously deposited nitrate during periods of active bromine radical chemistry will enhance the importance of bromine in NO x cycling and nitrate formation in the polar-regions.
Non-zero 17 O(OH) values due to incomplete exchange of OH with water vapor in the cold, dry polar atmosphere (Morin et al., 2007) may also contribute to the underestimate of 17 O(nitrate) in winter and spring.Kunasek et al. (2008) estimated that ∼10% of the original 17 O(OH) could be retained at Summit, Greenland in spring and summer.This would lead to a maximum underestimate of 1-2‰ in calculated 17 O(nitrate) and may partially explain the spring/summer discrepancy in polar regions.
The model tends to overestimate 17 O(nitrate) (1-7‰) in the northern mid-to high latitudes in winter when N 2 O 5 hydrolysis dominates nitrate production (Fig. 4), suggesting that the model may be overestimating this nitrate production pathway.The reaction probability of N 2 O 5 (γ N 2 O 5 ) on the surface of aerosols is influenced by temperature, humidity, and aerosol composition (Kane et al., 2001;Hallquist et al., 2003;Thornton et al., 2003).Davis et al. (2008) suggested that γ N 2 O 5 may be overestimated by Evans and Jacob (2005) (used in GEOS-Chem), especially in regions of relatively low temperatures and high relative humidity.In addition, the presence of aerosol surface coating by organics has been shown to decrease γ N 2 O 5 (McNeill et al., 2006) and is not accounted for in the model.We examine the sensitivity of 17 O(nitrate) to N 2 O 5 hydrolysis by setting the reaction probability of N 2 O 5 equal to zero (γ N 2 O 5 =0), effectively shutting off this nitrate formation pathway.Removing N 2 O 5 hydrolysis results in relative increase in both NO 2 +OH (Reaction R5) and NO 3 +DMS/HC (Reactions R6, R7), with the former (latter) reaction becoming more important in the tropics and mid-latitudes in summer (mid-to high-latitudes in winter).This tends to increase surface 17 O(nitrate) by up to 7‰ (seasonal mean) in regions where nighttime reactions dominate (mid-to high-latitudes in winter), and decrease surface 17 O(nitrate) by up to 3‰ (seasonal mean) in regions where daytime reactions dominate (tropics, midlatitudes in summer).Removing N 2 O 5 hydrolysis results in a 30% increase in the global, annual-mean tropospheric NOX (=NO+NO 2 +NO 3 +HNO 2 +2N 2 O 5 +HNO 4 ) burden.This effect of N 2 O 5 hydrolysis on reactive nitrogen partitioning and loss is consistent with, but somewhat smaller than that found by an earlier analysis (50%) (Dentener and Crutzen, 1993), likely due to the different γ N 2 O 5 values used.
The model overestimates 17 O(nitrate) at Princeton, NJ year round by 4-7‰ and at Bermuda by 1-5‰.These data sets represent event-based rainwater samples averaged over each month of collection and thus may not adequately represent the monthly-mean state.These observations may also be influenced by the presence of soluble organic nitrates, which are not included in the model calculations.
Given that the denitrifier method may be measuring a combination of organic and inorganic nitrates, we examine the potential influence of soluble organic nitrates by including them in our isotope calculations.Soluble organic nitrates included in our isotope calculations form in the model mainly from reactions of NO+RO 2 to form isoprene nitrates (RONO 2 ), where RO 2 in this case originates from isoprene oxidation products.Soluble organic nitrates also form to a minor extent from reactions between isoprene nitrates (formed via isoprene+NO 3 ) and RO 2 (methylperoxy and peroxyacetyl radicals).Isoprene is a biogenic hydrocarbon whose main source is terrestrial vegetation, with emissions largely dependent upon vegetation type and density, light and temperature (Guenther et al., 1995).We use the Model of Emissions from Gases and Aerosols in Nature (MEGAN) to compute process-based biogenic emissions of isoprene and other trace gases (Guenther et al., 2006) as described in Palmer et al. (2006) andMillet et al. (2008).The model does not include soluble organic nitrates formed via oxidation of anthropogenic hydrocarbons in the presence of NO x , which would most impact the formation of organic nitrates during winter when emission rates for biogenic hydrocarbons are low.The model treats soluble organic nitrates in a similar manner to inorganic nitrate, partitioning the organic nitrates between the gas and aerosol-phase according to thermodynamic equilibrium of HNO 3 .Organic nitrates are then dry or wet deposited.The formation of organic nitrates is most important in regions where isoprene emissions are highest (forested regions in the low-to mid-latutides, maximum in summer).
The 17 O value of organic nitrates is calculated according to the following, for reactions with NO+RO 2 and isoprene nitrates+RO 2 , respectively: For the latter reaction, we assume that the organic nitrate retains two of the original oxygen atoms of NO 3 , which may be an underestimate.However, our results are not particularly sensitive to this assumption since organic nitrates formed via this reaction pathway are a minor source of total (inorganic plus organic) nitrate (<6% annual mean).Organic nitrates formed via the NO+RO 2 pathway dominate (80% annual mean) total nitrate in tropical forested regions, and are as high as 20% (annual mean) elsewhere.Model results that include organic plus inorganic nitrate in calculations of 17 O(nitrate) are shown in Fig. 5 for the observations that utilized the denitrifier method.Including organic nitrates derived from isoprene oxidation lowers the calculated 17 O(nitrate) values by as much as 3‰ at a midlatitude continental location (Princeton, NJ) during summer, and as much as 10‰ in tropical forested regions.
The overestimate of 17 O(nitrate) in the polar winter (2-7‰) can be explained by the use of local A values in calculating 17 O(nitrate) where there is a negligible local source of NO x and most of the nitrate originates from lower latitudes.The importance of ozone in NO x cycling shows strong latitudinal dependence resulting in A values increasing with latitude (maximum in polar winter).The overestimate of 17 O(nitrate) due to the use of local A values is most pronounced in remote polar-regions in winter where nitrate originates from lower latitudes and local NO x production (e.g.snowpack photodenitrification) is absent.An overestimate of the A day value by 0.2 and A night by 0.1 (typical differences between A values at 90 • S and 45 • S), leads to an overestimate of 17 O(nitrate) by 4-5‰ (accounting for the 20-40% contribution of (Reaction R5) to total nitrate concentrations at the surface in polar winter).This is within the range of the overestimate in polar winter.

Conclusions
We use the global 3-D chemical transport model, GEOS-Chem, to simulate the oxygen isotopic composition ( 17 O) of atmospheric nitrate and compare with available observations.These comparisons allow for quantitative assessment of the relative importance of different oxidants in NO x cycling and nitrate formation pathways, and are critical for interpreting ice core observations of 17 O(nitrate) in terms of changes in past oxidant concentrations (Alexander et al., 2004).
The largest uncertainty for calculations of 17 O(nitrate) is the unconstrained variability in tropospheric 17 O(O 3 ).The best agreement with global observations occurs when assuming 17 O(O 3 )=35‰ and using the Savarino et al. (2008) isotopic transfer function during the reaction of O 3 with NO and NO 2 .The spatial variability of 17 O(nitrate) is largely determined by the importance of O 3 in NO x cycling due to the fact that the isotopic composition of two-thirds of the oxygen atoms of nitrate is determined during NO x cycling.The oxidation pathway of NO x to HNO 3 plays a secondary but significant role in determining 17 O(nitrate) values.The global, annual-mean tropospheric inorganic nitrate burden (0.38 Tg N) is dominated by nitrate formation via NO 2 +OH (76%), followed by N 2 O 5 hydrolysis (18%) and NO 3 +DMS/HC (4%).
Calculated values of annual-mean 17 O(nitrate) in the lowest model layer (0-200 m above the surface) vary from 7‰ in the tropics to 41‰ in polar regions.Modeled annualmean 17 O(nitrate) values as high as 40‰ exist in the polar regions with negligible contribution from stratospheric nitrate (<2%).Modeled annual-mean 17 O(nitrate) values as low as 7‰ are predicted in tropical forested regions due to the dominance of peroxy radicals and OH in NO x cycling and nitrate formation, respectively.Additional measurements of 17 O(nitrate) in the tropics are needed to validate these results.
The largest model discrepancies are in polar-regions during spring and summer (0-5‰ underestimate), and in midto high latitudes in winter (1-7‰ overestimate).The former is likely due to the lack of reactive bromine chemistry in the model, which should peak in spring in polar regions and lead to large (>40‰) 17 O(nitrate) (Morin et al., 2007).The high bias in polar winter is due to the use of the local fractional importance of O 3 (versus HO 2 +RO 2 ) in NO x cycling (A value) for calculations of 17 O(nitrate) in regions where nitrate originates from lower latitudes.This overestimate is particularly pronounced in winter when local NO x production (e.g., snowpack photodenitrification) is negligible.The role (or lack thereof) of organic nitrates in observations of 17 O(nitrate) that utilize the denitrifier method (Kaiser et al., 2007;Casciotti et al., 2002) needs to be assessed.Whether or not organic nitrates are included in the calculation of 17 O(nitrate) has a significant impact in lowto mid-latitude continental regions, and may partially account for the high bias in the mid-latitude 17 O(nitrate) values.

BFig. 1 .
Fig. 1.Simplified chemistry leading to inorganic nitrate formation in the model.

Fig. 4 .
Fig. 4. Annual-mean fractional importance of each nitrate production pathway leading to total inorganic nitrate at the surface in the model: NO 2 +OH (top left), N 2 O 5 hydrolysis (top right), NO 3 +DMS/HC (bottom left), and stratospheric nitrate (bottom right).