4.1 Isotopic Impacts of NO3− Processing
 As discussed above (section 1), different photolysis pathways will induce different isotopic effects. For NO3− deposited to the snow that is then photolyzed, a theoretical fractionation factor, ϵ18, that assumes Rayleigh fractionation, can be used to quantify the change in δ18O with the degree of photolytic loss of NO3− as follows:
 For conditions at Dome C, Frey et al.  calculated ϵ18 as −34‰ serving to increase the residual δ18O-NO3− in the snow (δ18Ofinal) and recalculated for average Summit radiation conditions, ϵ18 = −32‰. As a mass dependent process, the loss of NO3− from the snow should have no impact on the Δ17O-NO3−. If we assume that some portion of the data presented in Figure 3a reflects direct deposition, photolysis would serve to move the snow composition away from the observed Δ17O-NO3− versus δ18O-NO3− relationship along lines of constant Δ17O by differing amounts depending on the amount of loss. There is, however, considerable uncertainty associated with the calculated oxygen enrichment factor for photolysis of NO3−: the calculations are for the gas phase only (i.e., there is no consideration of matrix effects), the quantum yield has no wavelength dependence, and it is assumed that photolyzed NO3− is lost directly to the gas phase only as NO2.
 The enrichment in snow δ18O-NO3− due to photolytic loss fits neither with the data from Summit nor with other measurements of photolytic loss made in the laboratory or field [Frey et al., 2009; McCabe et al., 2005]. In laboratory photolysis experiments and in situ snow measurements, depletion in 18O has been observed and assumed to be the result of competing factors of enrichment due to photolysis and mixing of the residual NO3− with a source depleted in 18O. The most likely source of this low δ18O, due to its abundance, is water or an isotopically similar oxidant. Indeed, laboratory studies of photolysis of nitrate have shown, when beginning with a single NO3− source, that photolysis of USGS35 NaNO3 results in a single line for δ17O-NO3− versus δ18O-NO3− that has markedly different slopes when in waters of differing isotopic composition [McCabe et al., 2005].
 If NO3− were to simply exchange oxygen atoms with water, with no loss at all, the expected result would be for the δ18O-NO3− and Δ17O-NO3− to be pulled toward that of water. (For our samples, δ18O-H2O = −38‰ to −20‰ and Δ17O-H2O = 0‰.) This would serve to decrease both δ18O and Δ17O of the NO3− in the snow, though not in the ratio that fits with our observations. For example, starting at the mean values for 2011 (δ18O-NO3− = 70.1‰ and Δ17O-NO3− = 25.3‰), if 10% of the oxygen atoms were to exchange with water of δ18O = −38‰, the new isotopic composition would be δ18O-NO3− = 59.3‰ and Δ17O-NO3− = 22.8‰, using equation (2).
 If the water δ18O = −20‰, δ18O-NO3− = 61.1‰, and Δ17O-NO3− would remain at 22.8‰. With increasing exchange, the data point would move toward the isotopic composition of water, i.e., a line set by the water composition end point (or range) and the starting composition of the NO3− (e.g., Figure 3b, green dashed line). The slope of increasing exchange would vary from 0.07 to 0.33 depending upon the initial NO3− composition and the composition of water but would never be equal to the observed value of 0.46.
 The competing processes of enrichment due to photolysis and mixing with a depleted oxygen source could result in apparent fractionation along the relationship we observe between δ18O-NO3− and Δ17O-NO3−, if they happen in a specific ratio. With a higher degree of loss, a larger amount of mixing with the depleted source would be required to maintain this relationship. This seems plausible, as the proposed mechanisms for mixing with water involve branching photolysis, with some fraction of the NO3− becoming NOx and another portion of NO3− following a path that remains in the condensed phase. The condensed phase NO2−/NO2 can exchange oxygen atoms with the solvent water and then reform NO3− (Figure 1, pathway d). As more NO3− is photolyzed to gas-phase NOx, more NO3− may also be photolyzed to the condensed-phase substance (NO2−), thus increasing the oxygen exchange with water. As long as these reactions occur in the necessary ratios, the linear relationship between δ18O and Δ17O of the residual nitrate can be maintained. For instance, using the range of NO3− loss from snow concentration studies [Burkhart et al., 2004; Dibb et al., 2007], if a 10% loss of NO3− were occurring, 7% of the remaining NO3− oxygen atoms must exchange with water in order to maintain the observed relationship between δ18O-NO3− and Δ17O-NO3−, (i.e., applying equation (1) and then equation (2)). At a 25% loss, a 16% exchange is needed, assuming that the water has a δ18O of −30‰ and Δ17O of 0‰.
 If the water had a constant δ18O, competing photolytic enrichment and exchange with water would be a logical explanation for the relationship observed between δ18O-NO3− and Δ17O-NO3−. The water observed over the May–June 2010 and May–June 2011 seasons, however, varies in δ18O from −38‰ to −20‰. With a 25% photolytic loss and water with δ18O of −20‰, a 22% exchange of remaining oxygen atoms is required to maintain the relationship between δ18O-NO3− and Δ17O-NO3−, while with water δ18O of −38‰, a 15% exchange is required to maintain the relationship. The differences in exchange required with varying water isotopic composition change with differing degrees of NO3− loss. If the isotopic composition of the water were to vary in concert with the photolysis of NO3−, we would expect to find a relationship between δ18O-H2O and δ18O-NO3−, but there is none.
 If the degree of NO3− photolysis and the δ18O-H2O were to vary synchronously, that would require them to both be controlled by the same factors. If the only control on sublimation of water, and therefore δ18O-H2O increase, was actinic flux, then it would be possible to relate it to the degree of photolysis of NO3−. The δ18O-H2O should, however, be primarily controlled by relative humidity, which should have no effect on NO3− photolysis. In addition, concurrent changes in NO3− photolysis and δ18O-H2O would require δ18O-H2O to reset to the same values each evening before NO3− photolysis restarts in the morning. This is improbable, as the water deposition can come from a variety of sources with different δ18O-H2O, e.g., riming, fog deposition, or fresh snowfall. In addition, if sublimation were driving the change in δ18O-H2O, there should be a change in deuterium excess in the snow [Stichler et al., 2001], but all the samples fall along a line with a slope of 8 (δ18O = 8.0 * δD + 6.0, R2 = 0.99). This indicates that all isotopic changes in water are derived at equilibrium; therefore, sublimation cannot be the source of variation in δ18O-H2O. The most likely source of δ18O-H2O variation is deposition of new water.
 Additionally, stratospheric O3 concentration, and therefore UV penetration to the surface at Summit, is an important control on the photolysis of NO3−. It is notable that despite significant depletion in stratospheric O3 during spring 2011 compared to spring 2010 [Manney et al., 2011], the observed relationship between Δ17O and δ18O of NO3− is the same in both years (Figure 3a).
 In summary, the observed relationship between δ18O-NO3− and Δ17O-NO3− cannot be explained by postdepositional processing of NO3− in the snow, considering our current understanding of the isotopic imprints of the processes discussed above. The oxygen isotopic signals observed in NO3− at Summit are more plausibly explained as representing atmospheric NO3− deposition to Summit.
4.2 Atmospheric Production of NO3−
 Most linear relationships of the type found between δ18O-NO3− and Δ17O-NO3− at Summit are interpreted as the result of mixing of different oxidants that react with NOx to produce atmospheric NO3− [e.g., Michalski et al., 2004]. The linear relationship between δ18O-NO3− and Δ17O-NO3− suggests isotopic mixing between a high end-member with δ18O = 100‰ and Δ17O = 39‰ and a low end-member with δ18O = 18‰ and Δ17O = 0‰ (Figure 3a). The high end-member likely results from stratospheric O3. The lower end-member is more difficult to identify. The atmospheric oxidant with the closest isotopic composition is molecular oxygen (O2) (δ18O-O2 = 23.9‰, Δ17O-O2 =0‰ versus Vienna standard mean ocean water [Barkan and Luz, 2005]). A mixing line between these two oxidants (O2 and stratospheric O3) (Figure 3b, red solid line) is the best fit to the surface snow data, compared with other oxidants of Δ17O = 0‰. For instance, H2O vapor (or OH in isotopic equilibrium) would have δ18O between −30‰ and −10‰ (Figure 3b, green dashed line shows −20‰), which does not fit the data. If OH maintains some of its O3 character from O(1D) [e.g., Kunasek et al., 2008], the mixing line would remain the same, but the lower endpoint would be moved toward O3. Assuming an equilibrium fractionation of 44‰ between OH and H2O [Michalski et al., 2012] results in an OH-O3 mixing line that is an even worse fit for the data (Figure 3b, orange dotted line). Thus, it would appear that oxygen atoms from stratospheric O3 and atmospheric O2 are the main controls on the isotopic composition of NO3− that is ultimately deposited to Summit. Furthermore, the influence of the stratosphere on NO3− (e.g., NO3− formed in the troposphere via reaction of NOx and stratospheric O3) may account for the higher than expected summertime Δ17O-NO3− at Summit based on models [Alexander et al., 2009; Kunasek et al., 2008].
4.3 NOx Production From Snow NO3−
 The conclusion, based upon the isotopic data, that the NO3− seen in Summit snow is a direct atmospheric signal that reflects little to no postdepositional loss contrasts with the estimates of NO3− loss based upon snow concentration measurements made in the past [Burkhart et al., 2004; Dibb et al., 2007]. Gas-phase observations and recent modeling work, however, suggest that very small fractions of NO3− are involved in the postdepositional processing at Summit [Honrath et al., 1999; Thomas et al., 2011].
 Honrath et al.  calculated that only a tiny amount of the NO3− must photolyze in order to give a 1000 pptv NOx concentration in the interstitial snow (firn) air (i.e., 6 × 10−11% of the NO3− in a 5 µmol L−1 snow sample must be converted to NOx). These numbers, however, were never translated into boundary layer concentrations.
 In a 1-D model that matches well with observed NOx concentrations, Thomas et al.  showed that only 0.10% of the NO3− in the top 10 cm of snow is required to be lost over a 3 day period in order to explain the NOx concentrations measured in the boundary layer at Summit. Assuming a summer accumulation rate of 5.1 cm mo−1 [Dibb and Fahnestock, 2004], the top 10 cm of snow will be entirely replaced by fresh snow in less than 60 days. In that case, a loss of 2.1% of the NO3− in the snow is required to account for the measured NOx concentrations, backing our interpretation that the postdepositional processing of NO3− is small in magnitude and has little to no effect on the isotopes observed.
 The contrast between prior snow concentration measurements and the isotopic measurements, as well as the modeled NO3− loss, is difficult to reconcile. We have demonstrated that large photolytic losses of NO3− are not driving these measurements. Evaporative loss or volatilization of HNO3 will also not account for the discrepancy, as HNO3 concentrations in the atmosphere would have to be 4–10 times larger than the NOx concentrations, which is inconsistent with measurements at Summit [Dibb et al., 1998; Honrath et al., 2002]. The lower amount of NO3− loss predicted from NOx concentrations in air fits with the isotope data, while loss predicted from snow concentration measurements does not. It is possible that the calculations based on snow concentrations are confounded by the spatial heterogeneity of NO3− or by fluctuations in water content (e.g., evaporation). The isotopes of NO3−, therefore, present a more sensitive record of NO3− chemistry than concentration alone in the snow at Summit.