Interactive comment on “ Investigation of post-depositional processing of nitrate in East Antarctic snow : isotopic constraints on photolytic loss , re-oxidation , and source inputs ”

Abstract. Snowpits along a traverse from coastal East Antarctica to the summit of the ice sheet (Dome Argus) are used to investigate the post-depositional processing of nitrate (NO3−) in snow. Seven snowpits from sites with accumulation rates between 24 and 172 kg m−2 a−1 were sampled to depths of 150 to 300 cm. At sites from the continental interior (low accumulation, 200 ng g−1 in surface snow and decreases quickly with depth to 91 kg m−2 a−1), there are no obvious trends detected with depth and instead seasonality in nitrate mass fraction and isotopic composition is found. In comparison to the interior sites, the coastal pits are lower in δ15N (−15–71 ‰ vs. air N2) and higher in δ18O of nitrate (53–111 ‰ vs. VSMOW). The relationships found amongst mass fraction, δ15N, δ18O and Δ17O (Δ17O = δ17O–0.52 × δ18O) of nitrate cannot be explained by local post-depositional processes alone, and are instead interpreted in the context of a primary atmospheric signal. Consistent with other Antarctic observational and modeling studies, the isotopic results are suggestive of an important influence of stratospheric ozone chemistry on nitrate formation during the cold season and a mix of tropospheric sources and chemistry during the warm season. Overall, the findings in this study speak to the sensitivity of nitrate isotopic composition to post-depositional processing and highlight the strength of combined use of the nitrogen and oxygen isotopes for a mechanistic understanding of this processing.


G. Shi et al.: Isotopes of nitrate in East Antarctic snow
In the presence of sunlight, NO and NO 2 recycle rapidly with ozone (O 3 ), peroxy radical (HO 2 ), an organic radical (RO 2 , where R = organics), or halogen radicals (XO, where X = Br, Cl or I) according to the following reactions: During the day, i.e., when sunlight is present, oxidation of NO 2 by the hydroxyl radical (OH) produces nitric acid (HNO 3 ): At night and in colder environments, oxidation of NO 2 by O 3 is promoted and HNO 3 can be formed from hydrolysis of dinitrogen pentoxide (N 2 O 5 ), ) or by abstraction of a hydrogen atom by the nitrate radical (NO 3 ) from dimethyl sulfide (DMS) or a hydrocarbon (HC), NO 3 + DMS or HC → HNO 3 + products.(R7) Important NO x inputs to the troposphere include fossil fuel combustion, biomass burning, soil microbial activity, lightning, and injection from the stratosphere (Delmas et al., 1997;Lee et al., 1997).There has been interest in using ice core NO − 3 records to reconstruct past atmospheric NO x sources, and atmospheric loading and variability in concentration over time.Increasing NO − 3 concentrations in Greenland ice core records have been linked to increasing anthropogenic emissions (fossil fuel and/or agricultural) since the Industrial Revolution (Mayewski and Legrand, 1990;Hastings et al., 2009).In contrast, such increases in NO − 3 have not been observed in Antarctica (Wolff, 1995;Wolff et al., 2012), suggesting that concentrations in snow are mainly controlled by natural sources.
The partitioning of NO x inputs using ice core NO − 3 concentrations is difficult, however, since concentration alone cannot identify specific NO x sources, and NO − 3 can be lost from snow by post-depositional processes such as photolysis and possibly volatilization as HNO 3 (Wolff, 1995;Röthlisberger et al., 2000;Frey et al., 2009).Measurements of nitrogen and oxygen stable isotope ratios in NO − 3 provide further constraints for past NO x sources and oxidation chemistry (Alexander et al., 2004;Hastings et al., 2009;Hastings, 2010).In the atmosphere, the oxygen isotopes in NO − 3 reflect the oxidants involved in the production of NO − 3 (e.g., Reactions (R1)-(R7) above; Hastings et al., 2003;Michalski et al., 2003;Alexander et al., 2009), and the nitrogen isotopes can reflect NO x sources and possible imprints of transport and chemistry (Hastings et al., 2003;Elliott et al., 2007;Savarino et al., 2007;Morin et al., 2008;Altieri et al., 2013).However, post-depositional processing in snow can modify the isotopic composition of NO − 3 .At Dome C in East Antarctica (where the snow accumulation rate is < 35 kg m −2 a −1 , i.e., typically < 10 cm snow a −1 ), NO − 3 mass fractions decrease from hundreds of ng g −1 in surface snow to tens of ng g −1 at a depth of 10 cm, and this decrease corresponds to large changes in isotopic composition (Röthlisberger et al., 2000;Blunier et al., 2005;Frey et al., 2009;Erbland et al., 2013) such that this processing should be identifiable where it occurs.The influence of post-depositional alteration on NO − 3 , however, appears closely related to annual snow accumulation and at sites with higher accumulation rates, such as Summit, Greenland (200 kg m −2 a −1 , i.e., 60 cm snow a −1 ; Dibb and Fahnestock, 2004), the post-depositional effects are rather minor, and the atmospheric signal appears to be preserved (Hastings et al., 2004;Fibiger et al., 2013, and references therein).
In recent studies, the spatial variability of photolytic and volatile NO − 3 loss in East Antarctic upper snow has been investigated (Frey et al., 2009;Erbland et al., 2013), and represents important progress in understanding air-snow transfer of NO − 3 .However, there are still a number of questions regarding the interpretation of NO − 3 isotopes due to the complicated post-depositional behavior of NO − 3 .Distinguishing the form, extent and relative importance of the different possible isotope effects associated with post-depositional processes is critical for understanding what NO − 3 in an ice core represents.
In this study, samples from 150 to 300 cm deep snowpits, have been collected at seven sites along a traverse from the East Antarctic coast to Dome Argus (Dome A: the summit of the Antarctic ice sheet), and mass fraction and isotopic composition of NO − 3 were determined.The key objectives of this study are the following: (1) to investigate the effects of post-depositional processes on isotopic composition of NO − 3 at different depths in the snowpack; and (2) to understand the variation of NO − 3 isotopes in different environments across the East Antarctic Ice Sheet (EAIS).The results of this study are of significance to a further understanding of postdepositional processing of snow NO − 3 and the interpretation of NO − 3 isotopic composition archived in ice cores.

Sample collection
The Chinese National Antarctic Research Expedition (CHINARE) team conducts an annual inland traverse from the coastal Zhongshan Station (Indian Ocean sector) to Dome A in East Antarctica (Fig. 1).This traverse covers a distance of about 1250 km.On the traverse route from Zhongshan to Dome A, seven snowpits were excavated during the 2012-2013 austral summer season (Fig. 1).Full information about each pit, including location, snow depth, sampling resolu- tion, collection date, mean annual snow accumulation, etc., is summarized in Table 1.Snowpits were excavated manually and one snow wall was scraped clean and flat with a high-density polyethylene (HDPE) scraper.Snow samples were collected using 250 mL narrow-mouth HDPE vials pushed horizontally into the snow wall beginning at the bottom of the pit and moving upwards.The scraper and vials were pre-cleaned with Milli-Q ultrapure water (> 18.2 M ), dried in a class 100 super clean hood at 20 • C and then sealed in the clean polyethylene (PE) bags that were not opened until the field sampling started.Field blanks consisting of sampling bottles filled with Milli-Q water were analyzed for ion concentrations.All personnel wore PE gloves and face masks, and the pit sites were generally 1 km away from the traverse route to avoid possible contamination from expedition team activities.After collection, the vials were again sealed in clean PE bags and preserved in a clean insulated cabinet.All together, 530 snow samples were collected.All samples were transported to China in a freezer at −25 • C and then shipped frozen to Brown University in Providence, RI, USA.

Snow NO −
3 mass fractions (denoted as w(NO − 3 ) in the following context) were determined using a Westco Scientific SmartChem 200 discrete chemistry analyzer.The standard deviation (SD) of w(NO − 3 ) of 55 field blanks run within sets of samples was 0.8 ng g −1 , which is comparable to blank Milli-Q water run on the same system.The pooled standard deviation of samples run in replicate (n = 50) in different sample sets is 1.5 ng g −1 .
Snow NO − 3 isotopic compositions were measured according to the denitrifier method by using denitrifying bacteria to convert NO − 3 to N 2 O gas, which is collected and injected into a stable isotope ratio mass spectrometer (IRMS; Thermo Scientific DELTA V Plus; Sigman et al., 2001;Casciotti et al., 2002;Kaiser et al., 2007).At Brown, a minimum of 5 nmol of NO − 3 is required for an accurate isotopic determination of 15 N / 14 N and 18 O / 16 O ratios in snow samples with w(NO − 3 ) as low as 10.0 ng g −1 , which can be analyzed directly without a pre-concentration step (i.e., for a 5 nmol NO − 3 run, a sample of 10.0 ng g −1 requires a 31 mL injection).
NO − 3 isotopic ratios (δ 15 N(NO − 3 ), δ 17 O(NO − 3 ), δ 18 O(NO − 3 )) are defined as where R is the ratio of n( 15 3 ) isotopic data were calibrated using the international reference materials IAEA-NO-3, USGS34 and USGS32, and δ 18 O(NO − 3 ) were calibrated using IAEA-NO-3, USGS34, and USGS35 (Michalski et al., 2002;Böhlke et al., 2003).Determining the isotopic composition in large volume samples has been extensively tested in the laboratory, and it is critical to run reference materials very close to the same concentrations (i.e., same volume injections) as samples, to eliminate any potential volume effects.Included in the supplementary materials are data from internal working standards that show excellent reproducibility over a variety of injection volumes in different runs (Tables S1 and S2 in the Supplement).Precision of the isotopic analyses is calculated in two ways (Buffen et al., 2014).First, the pooled standard deviation (1σ p ) of all standards run within individual sample sets was calculated.For δ 15 N(NO − 3 ), the 1σ p of standards is 0.3 ‰ (IAEA-NO-3, n = 80), 0.3 ‰ (USGS34, n = 80), 1.1 ‰ (USGS32, n = 53) and 0.4 ‰ (USGS35, n = 80); and for δ 18 O(NO − 3 ) this is 0.6 ‰ (IAEA-NO-3, n = 80), 0.6 ‰ (USGS34, n = 80), 0.7 ‰ (USGS32, n = 53) and 0.7 ‰ (USGS35, n = 80).Second, the pooled standard deviation of all replicate samples run in at least two different sets was examined (n = 38 pairs of samples) and yielded 0.8 ‰ for δ 15 N(NO − 3 ) and 0.5 ‰ for δ 18 O(NO − 3 ).The pooled standard deviation of the replicate samples is probably the most representative measure of precision as it accounts for the total variation within the denitrifier method (i.e., from sample preparation to isotopic determination), and the variance is not diluted compared to the much higher number of standards that are pooled across sample sets (compared to individual samples that are only run once or twice).
During the NO − 3 reduction by bacteria, a small number of oxygen atoms may be exchanged between water and the intermediates of denitrification (e.g., NO − 2 ) and must be corrected for the isotopic determination.In general, this exchange is < 10 %, and typically < 3 %, of the total O atoms  (Ding et al., 2010;Xiao et al., 2013).
in the produced N 2 O and is corrected for using the measured oxygen isotope compositions of snow (δ 18 O(H 2 O)) and water in the bacteria/media (see Casciotti et al., 2002;Kaiser et al., 2007 for correction schemes).δ 18 O(H 2 O) was determined using the standard CO 2 equilibration method (Johnsen et al., 1997).The standard deviation of reference (VSMOW) measurements (n = 20) was 0.10 ‰ .Full snowpit profiles of δ 18 O(H 2 O) were only completed for P1 and P7; only surface snow samples (3 cm) were measured for P2-P6.
A correction is also needed for δ 15 N(NO − 3 ) to account for the contribution of the 14 N 14 N 17 O isotopologue to the m/z 45 signal measured by the IRMS (Kaiser et al., 2007).Because atmospheric NO − 3 contains a non-zero 17 O (i.e., 17 O = δ 17 O-0.52 × δ 18 O > 0 ‰), simply assuming δ 17 O = 0.52 × δ 18 O can yield an overestimate of the true δ 15 N(NO − 3 ) by as much as 1-2 ‰ (Sigman et al., 2001;Hastings et al., 2003).To account for this contribution, a measured or estimated 17 O(NO − 3 ) is used to correct the δ 15 N(NO − 3 ) values.Previous East Antarctic investigations have shown that 17 O(NO − 3 ) mainly ranges from 25 to 35 ‰ in snow NO − 3 (Erbland et al., 2013) and we find a similar range for P1 in our study (see below).For P1, the measured 17 O(NO − 3 ) values reported below were used to correct δ 15 N(NO − 3 ), while a mid-range value of 17 O(NO − 3 ) = 30 ‰ was used for P2-P7.Using this mid-range value of 17 O(NO − 3 ) = 30 ‰ leads to an average δ 15 N(NO − 3 ) difference of 1.6 ‰ compared to using 17 3 ) used to correct the data (i.e., 17 O(NO − 3 ) = 25 or 35 ‰) results in a δ 15 N(NO − 3 ) difference of 0.3 ‰, which is comparable to our reported analytical precision and is negligible when compared to the range of sample δ 15 N(NO − 3 ) values.
For determination of 17 O(NO − 3 ), the sample N 2 O produced by the denitrifier method was thermally decomposed to N 2 and O 2 in a heated gold tube, and the O 2 was then measured at m/z 32 and 33 signals on the IRMS (Kaiser et al., 2007).A minimum of 35 nmol NO − 3 is needed for the analysis, but the low w(NO − 3 ) and low sample volumes available in this study limited the measurement of both 17 O(NO − 3 ) and δ 15 N and δ 18 O on the same sample.A preconcentration procedure is therefore needed for the measurement of 17 O(NO − 3 ) (e.g., Morin et al., 2008;Frey et al., 2009;Erbland et al., 2013).Briefly, NO − 3 was trapped in an anion exchange resin, and then eluted by a 1M NaCl solution.A variety of NaCl salts (including the products of Fisher Scientific, Extrapure, and Macron) were tested and found to contain NO − 3 , and thus procedural blanks were determined for each batch of NaCl used.(Note that NO − 3 was found in every batch of NaCl tested, with values as high as 547 ng g −1 in the salt, and was different even for bottles with the same lot number.)For the samples from P1, 28 ng g −1 was measured for the 1M solution of NaCl (Fisher Scientific) used for elution.During the concentrating procedure, one Milli-Q water blank and two sets of standards (USGS34 and USGS35) with similar w(NO − 3 ) to the snow samples were processed simultaneously.The measured 17 O(NO − 3 ) was then corrected by two steps: (1) 17 O(NO − 3 ) in concentrated samples was linearly corrected using the standards USGS34 and USGS35 run within individual sample sets; and (2) the output of step (1) was further corrected by the standards used during the concentration procedure to account for the impact of procedural influence (e.g., the NaCl blank).A mean difference of 2.5 ‰ for 17 O(NO − 3 ) was obtained between the results corrected by step (2) and data without the step (2) correction.Precision for repeated measurement of 17 O(NO − 3 ) is only 0.44 ‰ (see also Table S2 in the Supplement), but without correcting for the blank associated with the eluent NaCl, we find that the pre-concentration method can result in an underestimation at least on the order of 2.5 ‰ for 17 O(NO − 3 ).

Snowpit w(NO −
3 ) A summary for all measurements of w(NO − 3 ), δ 15 N(NO − 3 ) and δ 18 O(NO − 3 ) in each snowpit is given in Fig. 2 and the detailed profiles with depth are illustrated in Fig. 3.In general, w(NO − 3 ) is lower than 200 ng g −1 in P1 and P2, which are characterized with higher annual snow accumulation (see 3 for each snowpit (P1-P7), plotted as a function of snow accumulation rate.Box and whisker plots represent maximum (top end dash symbol for each box), minimum (bottom end dash symbol for each box), the range 1-99 % (top and bottom X symbol for each box), percentiles (5th, 25th, 75th, and 95th), and median (50th, solid line) and mean (open square near center of each box).Note that the data of 17 O(NO − 3 ) are only available for P1 snowpit and refer to the right-hand axis.
present in both pits.In contrast, pits P4-P7 from the lower snow accumulation sites show the highest w(NO − 3 ) in surface snow, which falls sharply from > 200 ng g −1 near the surface to below 50 ng g −1 within the top meter, and do not contain regular fluctuations.The markedly decreasing trend of w(NO − 3 ) with depth seems to fit an exponential model as has been seen previously (Traversi et al., 2009).

Isotopic compositions of NO
3 ) and δ 18 O(NO − 3 ), the coastal and inland pits differ greatly in terms of the average values and the variability with depth.For the coastal sites P1-P3, δ 15 N(NO − 3 ) is generally lower than in the inland snowpits P4-P7, varying between −14.8 and 70.8 ‰, while δ 15 N(NO − 3 ) in the inland pits ranges from 15.5 to 460.8 ‰ (Figs. 2 and 3).This high value of 460.8 ‰ in pit P7 (which is at Dome A) is the highest natural δ 15 N(NO − 3 ) on Earth so far reported to our knowledge.In the inland pits (P4-P7), δ 15 N(NO − 3 ) is lower in the uppermost layers and strongly increases deeper in the snowpack, with most of the increase occurring in the top 20-30 cm.
In contrast to δ 15 N(NO − 3 ), δ 18 O(NO − 3 ) is higher on average in the coastal pits (P1-P3), ranging between 52.5 and 111.2 ‰, compared to the inland sites (P4-P7), where δ 18 O(NO − 3 ) varies between 16.8 and 84.0 ‰ (Figs. 2 and 3).It is noted that the averages of δ 18 O(NO − 3 ) for P4-P7 are comparable, while δ 15 N(NO − 3 ) means vary significantly, from 133.6 to 335.2 ‰.There is no obvious trend in the δ 18 O(NO − 3 ) profiles with depth in P1-P3, but this is not the case for the inland sites.δ 18 O(NO − 3 ) decreases over the top 20-30 cm, but gradual and consistent increases are observed below 30 cm in P4, P5 and P7 which continue to the pit base (200-300 cm; Fig. 3).A similar decrease in δ 18 O(NO − 3 ) is observed in the top of P6, but it is not clear if an increasing trend exists in the profile below.
The difference between the coastal and inland pits observed here is similar to that observed in the Erbland et al. (2013) study.A comparison between the two studies is presented in the Supplement as Fig. S1.

Discussion
After deposition, NO − 3 can be lost from snow by photolysis and volatilization as HNO 3 (sometimes referred to as evaporation or physical release in other studies), and the extent of loss via these post-depositional processes is expected to be accumulation dependent (Röthlisberger et al., 2002;Grannas et al., 2007).At lower accumulation sites, NO − 3 loss is relatively high, synchronous with a large degree of isotopic fractionation (Blunier et al., 2005;Frey et al., 2009;Erbland et al., 2013).In contrast, post-depositional alteration of snow NO − 3 in high accumulation regions can be minor, and seasonal and interannual cycles can be preserved in the snowpack (e.g., Wagenbach et al., 1994;Hastings et al., 2004).
Based on the site differences in annual snow accumulation rate and the profile trends of w(NO − 3 ), δ 15 N(NO − 3 ) and δ 18 O(NO − 3 ), the seven pits are divided into two groups within the following discussion: group I includes the coastal, medium-high accumulation sites P1-P3 (> 91 kg m −2 a −1 ) and group II are the low-accumulation and further inland sites P4-P7 (< 55 kg m −2 a −1 ).Below we consider what processes (and fractionation constants) can explain observations from the group I and group II snowpits, and whether it is possible to predict values at depth based on the loss processes near the surface.3 ) in P3 and P7 was broken so the trend in the deeper snow can be seen.

NO − 3 loss in inland upper snowpack
If it is assumed that post-depositional loss of snow NO − 3 is accompanied by a Rayleigh-type fractionation, the observed changes in δ 15 N and δ 18 O in a snowpit profile can be described as a function of w(NO where δ snow,0 and δ snow denote isotopic ratios in the initial and remaining NO − 3 , respectively, and w snow,0 and w snow are the initial and remaining NO − 3 mass fractions, respectively (e.g., Blunier et al., 2005).ε can be obtained from the slope of the linear regression for ln(w snow ) vs. ln(δ snow + 1), while [ln(δ snow,0 + 1) − ε × ln(w snow,0 )] would be the intercept.It is noted that ε is related to the fractionation factor α by ε = α − 1 (Criss, 1999).
Solar radiation decreases exponentially in the snowpack, with attenuation described in terms of an e-folding depth (z e ) where the actinic flux is reduced to 1/e (37 %) of the surface value.Accordingly, roughly 95 % of snowpack photochemistry should occur above the depth of three times z e (Warren et al., 2006).For the individual pits here, we calculate apparent ε values (ε app ) from data in the upper 20, 40 and 60 cm, to evaluate the impacts of post-depositional processes on snow NO − 3 (Table 2).Zatko et al. ( 2013) calculated a z e of about 20 cm for remote Antarctic sites.For group II, relatively strong relationships are observed between w(NO − 3 ) and δ 15 N(NO − 3 ) or δ 18 O(NO − 3 ) in the top 60 cm (as indicated by the statistically significant R 2 values for Eq.(2); Table 2).These pits are characterized by negative 15 ε app , with values of −77.8 ‰ (P4), −93.1 ‰ (P5), −50.2 ‰ (P6) and −61.3 ‰ (P7) for the upper 20 cm snow layer (i.e., the 0-20 cm interval; Table 2); whereas 18 ε app values are positive, indicating a depletion of 18 O(NO − 3 ) with decreasing w(NO − 3 ).The observed fractionation constants ( 15 ε app and 18 ε app ) for group II (P4-P7) are comparable to those from other snowpits on the East Antarctic plateau (Frey et al., 2009;Erbland et al., 2013; see also Fig. S1 in the Supplement).
In the upper 20 cm of the snowpack, a significant NO − 3 loss with increasing depth is seen in the group II pits and corresponds to a large enrichment of 15 N(NO − 3 ).A large loss of NO − 3 leading to such high δ 15 N(NO − 3 ) values in the surface snow is consistent with the calculated low 15 ε app in the upper snowpack and our expectations based on other findings in East Antarctica (e.g., Savarino et al., 2007;Erbland et al., 2013).Such strongly positive δ 15 N values (> 100 ‰) have not been observed in atmospheric NO − 3 .

3
Photolysis of snow NO − 3 is thought to primarily occur within a disordered interface, sometimes referred to as a quasiliquid layer, at the surface of the ice crystal via the reactions with Reaction (R8) exceeding Reaction (R9) by a factor of 8-9 (Warneck and Wurzinger, 1988;Dubowski et al., 2001;Chu and Anastasio, 2003).Snow NO − 3 photolysis products are mainly NO 2 , greatly exceeding NO under most conditions (Dibb et al., 2002).Under acidic conditions (pH < 5), HONO (g) formation from NO − 2 protonation is also important (Grannas et al., 2007 and references therein).Only NO 2 produced near the ice crystal surface-air interface can be released to the firn air and subsequently escape from the snowpack to the overlying atmosphere (Boxe et al., 2005).
In order to identify the relative importance of photolysis and volatilization (Sect.4.1.3)on NO − 3 loss, the fractionation constant of each process should first be quantified.The photolysis rate constant j NO − 3 (s −1 ) is expressed as where σ NO − 3 (cm 2 ) is the spectral absorption cross section; φ NO − 3 is the quantum yield (0-1), and I is the spectral actinic flux (photons cm −2 s −1 nm −1 ).Frey et al. (2009) proposed a theoretical model for estimating nitrate photolytic isotopic fractionation constants, which is based on a framework originally developed by Yung and Miller (1997) for stratospheric N 2 O.The framework exploits mass-dependent differences in the vibrational frequencies and ground-state energies for a given set of isotopologues.These differences result in a modeled spectral absorption cross section for the heavier isotopologue which is shifted to longer wavelengths, thus influencing the rate constant.The isotopic fractionation constant can then be calculated by where j corresponds to the heavy isotopologue (e.g., 15 N 16 O − 3 ); j corresponds to the light isotopologue (e.g., 14 N 16 O − 3 ), and it is assumed that the different isotopologues retain similar spectral absorption curves and equal quantum yields.
We calculate the photolytic 15 ε and 18 ε at sites P1 and P7 for peak summer radiation conditions (solstice solar noon on 21 December 2012) using actinic fluxes derived from the Tropospheric Ultraviolet and Visible (TUV 5.0) radiation transfer model (Madronich and Flocke, 1998) assuming clear sky conditions and a total overhead ozone column of 300 DU for both sites.We use the nitrate absorption cross section from Chu and Anastasio (2003) and calculate quantum yields using the equation given in this same work for −10 • C at P1 (φ = 0.0040) and −30 • C at P7 (φ = 0.0019).
The resulting fractionation constants are 15 ε = −45.3‰ (P1) to −48.0 ‰ (P7) and 18 ε = −32.5 ‰ (P1) to −34.4 ‰ (P7), and the relatively small variability between the two sites indicates that the calculated values are representative for the two site groupings.Berhanu et al. (2014) have recently proposed absorption cross sections derived from the measurements of Chu and Anastasio (2003) but modeled directly for the 14 N 16 O − 3 and 15 N 16 O − 3 isotopologues at −30 • C. When using these cross sections, 15 ε is calculated to be −48.9‰ at P1 and −52.8 ‰ at P7.
The negative ε values suggest that photolysis will lead to a strong enrichment of both 15 N and 18 O in NO − 3 remaining in the snow.For 15 ε app calculated from observations in the upper 60 cm of the group II pits (Table 2), the higher R 2 values imply that photolysis can largely explain enrichment of 15 N in NO − 3 with the decrease of w(NO − 3 ).At Dome C, where snow accumulation is typically less than 35 kg m −2 a −1 , close to the values of P4-P7 (Fig. 1), photolysis has also been reported as responsible for large increases in δ 15 N(NO − 3 ) with depth in the snow (Frey et al., 2009;Erbland et al., 2013).The negative calculated 18 ε (−32.5 to −34.4 ‰), however, does not agree with the highly positive 18 ε app values based on the observations (Table 2).

Aqueous phase "secondary" NO − 3 formation
If the post-depositional loss of NO − 3 in the group II pits was driven solely by photolysis, 18 O should also be enriched in the remaining NO − 3 according to the modeled photolytic 18 ε values (−32.5 to −34.4 ‰).However, δ 18 O decreases over the top 20 cm (Fig. 3) and the apparent 18 ε values ( 18 ε app ) calculated from the observed data in the upper 20 cm are instead positive, varying from 16.7 to 30.2 ‰ (Table 2).Furthermore, simple photolysis will lead to a linear relationship of δ 18 O(NO − 3 ) vs. δ 15 N(NO − 3 ) with a slope of roughly 0.7, i.e., equal to the ratio of the fractionation constants.How- 3 ) in the top 20 cm (Fig. 4), with slopes varying from −0.4 to −0.2.
Similar negative relationships have been observed in other East Antarctic snowpits (Frey et al., 2009;Erbland et al., 2013) and, following from experimental and theoretical work (McCabe et al., 2005;Jacobi and Hilker, 2007), were attributed to the aqueous-phase re-oxidation of the products thought to be important at NO − 3 concentrations that are orders of magnitude higher than those found in snow (Bunton et al., 1952).) For the top 2.5 cm of snow in P7, w(NO − 3 ) and δ 18 O(NO − 3 ) is 374 ng g −1 and 74.9 ‰ respectively, while w(NO − 3 ) decreases to 37.6 ng g −1 at a depth of 20 cm.Based on the modeled effect of photolysis, δ 18 O in the remaining NO − 3 at 20 cm could be predicted by where δ 18 O remaining represents δ 18 O in the remaining NO − 3 ; δ 18 O 0 is that of initial NO − 3 ; f is the fraction of NO − 3 remaining in the snow; and 18 ε is the photolysis fractionation constant (−34.4 ‰).If photolysis alone is responsible for the NO − 3 loss, δ 18 O(NO − 3 ) is expected to be 163 ‰ at 20 cm.Instead, at a depth of 20 cm the observed δ 18 O(NO − 3 ) is 42.6 ‰.This then would require a cumulative 54 % of the O atoms in the remaining NO − 3 to reflect that of water, assuming a δ 18 O(H 2 O) of −60 ‰.This back-of-theenvelope calculation would produce varying degrees of exchange amongst the inland snowpits (group II) due to the difference in mass fraction and δ 15 N(NO − 3 ) and δ 18 O(H 2 O) in the upper snowpack.For this calculation to be more accurate the initial deposition of primary NO − 3 from the atmosphere should be known (vs. that remaining in the surface snow sample collected for study) and it should be considered that the photolytic loss of NO − 3 and formation of secondary NO − 3 effectively happen simultaneously.Still, this simplified mass balance approach, overall, indicates that re-oxidation plays a significant role in determining how the δ 18 O of NO − 3 evolves in the snow column during burial.

3
Volatilization, or physical release, of HNO 3 may also be a pathway for post-depositional loss of NO − 3 in snow (Röthlisberger et al., 2000;Erbland et al., 2013).The importance of this process is unclear, however, as loss proceeds only as HNO 3 , HNO 3(aq) HNO 3(g) , (R12) and thus requires highly acidic conditions given the very high dissociation constant for HNO 3 (Sato et al., 2008).Volatilization may also be inhibited at low temperatures as suggested by laboratory and field observations (Erbland et al., 2013;Berhanu et al., 2014).
The current understanding of the isotopic impact of volatilization is somewhat limited.An experiment conducted by Erbland et al. (2013) at Dome C suggested that the 15 ε (mean ± 1σ ) for volatilization varied from 0.9 ± 3.5 ‰ (−30 • C) to −3.6 ± 1.1 ‰ (−10 • C) (i.e., close to nonfractionating).However, the loss of NO − 3 in these experiments may have been driven by the large losses of snow via sublimation.No observational or experimental data for 18 ε are available.Theoretical model estimates of volatile fractionation, assuming that the aqueous-phase equilibrium in Reaction (R12) is the controlling step in the overall fractionation (Frey et al., 2009), predict values of 15 ε and 18 ε to be from 12.6 ‰ (0 • C) to 16.8 ‰ (−73 • C) and between 1.1 ‰ (0 • C) and 0.6 ‰ (−73 • C), respectively (Table S3 in the Supplement).
For the summertime temperatures at the P4-P7 sites (< −30 is negatively correlated with δ 18 O(NO − 3 ) (Fig. 4), disagreeing with both expectations (i.e., field experiment observations and modeled results; Table S3 in the Supplement).The current understanding of volatile fractionation, however, is very limited and experimental data for 18 ε are not available to date.This makes it hard to attribute the relationships between w(NO − 3 ) and δ 15 N(NO − 3 ) and δ 18 O(NO − 3 ) to volatilization at our sites given the existing state of knowledge.

Isotopic variability with depth at inland sites (group II snowpits)
As shown in Table 2, the logarithmic relationships between δ 15 N or δ 18 O and mass fraction of NO − 3 are strongest in the upper 60 cm of group II snowpits, as seen in the R 2 values.While generally weaker in the layers below, the R 2 values remain significant even when the entire snowpit is considered.For instance, for 15 ε app , P4, P5, and P6 all have R 2 above 0.5 (with a p < 0.05) when observations from the entire snowpit are considered.Interestingly, while δ 15 N(NO − 3 ) maintains a negative relationship with w(NO − 3 ) at all depths, the sign of apparent fractionation constant for δ 18 O(NO − 3 ) changed (Table 2; Fig. 5); i.e., the relationships between δ 18 O(NO − 3 ) and w(NO − 3 ) shift from being positive in the upper 20 cm to generally negative in the deeper snow (Fig. S2 in the Supplement).This leads to there generally being no association between w(NO − 3 ) and δ 18 O(NO − 3 ) when the entire snowpit depth interval is considered (Table 2).
It is also useful to discuss this variability with depth in terms of the isotopic fractionation constants.If there was only a single isotopically fractionating process driving the observed changes in the snow, such as photolysis, the apparent fractionation constants for δ 15 N(NO − 3 ) and δ 18 O(NO − 3 ) would be similar throughout the snow column.This expectation relies on two important assumptions.First, the boundary conditions that influence the fractionation constants remain similar over time.This is to say that factors such as the contribution of different NO − 3 sources, accumulation rate, the influence of total overhead ozone on the spectral actinic flux and photolysis rate, and the influence of snow chemistry on the photolability of NO − 3 (e.g., Davis et al., 2008;Meusinger et al., 2014), remain similar over time.Second, the isotopic imprint of photolysis is set in the upper snowpack and then preserved below.This requires the assumption that once NO − 3 is moved below the photic zone, no additional in situ modifications take place.Stemming from these expectations, the isotopic composition of buried NO − 3 could be back-calculated to that originally at the surface if the isotopic imprint of alteration at the surface could be quantified in terms of a fractionation constant.
Based on the observations with depth, it is clear that the ε app values are dependent on the depth range chosen (Table 2).Both 15 ε app and 18 ε app vary but with distinct differences.The 15 ε app tends to become more positive with depth, while 18 ε app decreases and even switches from positive to negative values.When taken together with variability in the strength of the isotopic relationships with w(NO − 3 ) and the observation that isotopic composition continues to change below the expected photic zone depths, especially for δ 18 O(NO − 3 ) (Fig. 3), it would seem that the assumptions above do not all apply.Either the fractionation constants change over time because of a change in boundary conditions and/or the isotopic imprint of photolysis is not preserved below the photic zone.The former hypothesis is much more likely as, thus far, there is little other evidence of processes well below the photic zone modifying buried NO − 3 .

Predicting w(NO −
3 ), δ 15 N(NO − 3 ) and δ 18 O(NO − 3 ) values in buried snow Erbland et al. (2013) proposed that snow w(NO − 3 ) and isotopic compositions may approach constant values, called "asymptotic" values, below the photic zone (or zone of active NO − 3 loss).By means of an exponential regression, asymptotic values are calculated by where M (x) is the w(NO − 3 ), δ 15 N(NO − 3 ) or δ 18 O(NO − 3 ) at depth x (cm); M (as.) is the asymptotic value for these parameters; M (0) is the value at the surface of the snowpit; and c is a constant.Asymptotic values for each snowpit are calculated from the best fit (minimizing the sum of squared residuals) of M (x) vs. depth.
Based on observations from only the top 20 cm of snow in different snowpits on the EAIS, the Erbland et al. (2013) study predicted values below the photic zone.Because our snowpits extend deeper along the entire traverse, and as seen in Fig. 3 the snowpits did not typically follow a simple exponential decrease, we explored whether the asymptotic values change when derived from different depth ranges in the snowpit.In order to compare the asymptotic values derived from different snow depth ranges, observations from four intervals (0-20, 0-40, 0-60, and 0-100 cm) were selected to make this calculation, and the results are listed in Table 3.Several interesting results emerge.For asymptotic calculations of w(NO − 3 ), δ 15 N(NO − 3 ) and δ 18 O(NO − 3 ), all show important variations depending on the depth interval over which they are calculated.The variance, described in Table 3 by the standard error, is relatively large for the asymptotic values, but generally decreases the more observations that are included.In concert with this, the greater amount of observations included in the calculation the better the fit of the predicted values with the observed values as evidenced by the changing R 2 with calculations over different depth intervals.The sensitivity of the calculation of asymptotic values for different depth intervals in each of the group II snowpits is shown in Fig. 6.
What is the depth interval over which it is necessary to calibrate the asymptotic calculation?In other words, how much 3 ) and δ 18 O and δ 15 N of NO − 3 in the near surface (< 20 cm) and at depth (100-200 cm for group II snowpits) snow ranges.Least squares regressions are noted with lines and are significant at p < 0.05.It is noted that the linear regression slope for individual snowpits is the apparent fractionation constant ( 15 ε app and 18 ε app ; Table 2).The red dashed lines in the panels of P3 (ln(δ 15 N + 1) vs. ln(w(NO − information must we know about surface conditions to make the asymptotic relationship useful for predicting isotope values at depth? Figure 6 clearly shows that the more observations that are included the better the fit to the data will be, which is logical.But the range of profiles predicted by the asymptotic regressions also make it clear that much more information exists in the observations than can be explained by the simple assumption that photolytic loss, particularly in the top 20 cm, is the overwhelming signal captured at these low-accumulation sites.Overall, the δ 18 O(NO − 3 ) in deeper snow is more difficult to predict compared to w(NO − 3 ) and δ 15 N(NO − 3 ).This conclusion is consistent with the changes in δ 18 O(NO − 3 ) at deeper depths (Fig. 3) and the changes in ε app with depth (Table 2).
In cases where there is significant post-depositional loss and/or processing of NO − 3 , the δ (as.) , in theory, could help account for the impact of post-depositional processing compared to preservation in reconstructing a primary  3 ), δ 15 N and δ 18 O of NO − 3 calculated based on four different snow depth intervals (0-20, 0-40, 0-60 and 0-100 cm) of each snowpit.p is the significance level of observed data fitted using the exponential regression Eq. ( 6), and R 2 denotes the squared correlation coefficient of observed data compared to the regression model predicted values.Also given is the standard error (1σ ) of asymptotic values.

Snowpit Depth
w(NO − 3 ) (as.) , ng g −1 δ 15 N (as.) , ‰ δ 18 O (as.) , ‰ 22.3 ± 6.9 0.00 0.97 490.0 ± 23.0 0.00 0.90 23.2 ± 7.0 0.00 0.78 0-60 cm 23.3 ± 3.9 0.00 0.97 448.7 ± 33.1 0.00 0.86 29.7 ± 2.5 0.00 0.75 0-100 cm 17.9 ± 2.3 0.00 0.97 383.8 ± 9.4 0.00 0.75 33.9 ± 1.4 0.00 0.64 atmospheric signal.Our results show that δ (as.) (and ε app ) is sensitive to the depth interval over which exponential change is assumed.As suggested above, changes in factors such as the contribution of different NO − 3 sources, accumulation rate, total overhead ozone, and the influence of snow chemistry on photolability of NO − 3 may not remain similar over the time period covered by the snowpits.There do not appear to be significant trends in the annual accumulation rates based on data from Dome A. Wang et al. (2013) have compiled existing stake and snowpit accumulation measurements from Dome A and show (1) little spatial variability (surrounding 50 km) and ( 2 ).Automatic weather station measurements presented in the same work show somewhat higher accumulation in the spring and summer (roughly 6-7 mm per month) vs. fall and winter (roughly 3-6 mm per month) with fairly stable values in the warmer months.Based on the Dome A studies, it is unlikely that significant changes in accumulation have occurred in the area of the group II snowpits.Therefore, this cannot explain the difference between what is predicted based upon the δ (as.) and what is observed (Fig. 6).Nor does it seem likely that changes in ε app can be explained by accumulation rate.If it is assumed that snow accumulation has been constant for the group II snowpits, then the snowpits can be roughly dated based on measured accumulation and snow density (Fig. S3).This approximate dating suggests that the bottom of the P4 snowpit is from about the year 2000; P5 dates from ∼ 1994; P6 from 1985; and P7 from 1970.All four snowpits show a change in the relationship between δ 18 O(NO − 3 ) with w(NO − 3 ) and with δ 15 N(NO − 3 ) between near surface snow (< 20 cm) and deeper snow (Fig. 5).Based on the approximate dating, the timing of this change is very different in the different snowpits.For example, for a depth of 100 cm, snow in P4 is dated to ∼ 2007, P5 to ∼ 2004, and P6 and P7 to ∼ 2000.(Even given the imprecision of the dating method it is unlikely that more accurate dating would conclude that changes in the snowpits occur exactly together.) Can changes in stratospheric ozone concentration help to explain this change?Both the photolytic rate constant and fractionation constants would be sensitive to significant changes in overhead ozone concentration (i.e., less stratospheric ozone leads to more penetration of light at wavelengths that can photolyze NO − 3 ).Based on the approximate dating of the pits, P7 overlaps with the preozone hole era (generally considered prior to 1980), but there is no obvious change in the isotope observations (1980 occurs at a depth of about 250 cm; Figs. 3 and S3).Moreover, both ground-based observations at the South Pole and satellite-based observations (TOMS/OMI) do not show any significant trend in total overhead ozone during spring and early summer over the time period 2000-2010 (http://www.antarctica.ac.uk/met/jds/ozone/graphs.html; http://www.cpc.noaa.gov/products/stratosphere/winter_bulletins/sh_09/; http://ozonewatch.gsfc.nasa.gov/statistics/annual_data.html).This does not prove that there is not a link between the observed changes in the isotopic composition of snow NO − 3 and overhead ozone concentration, but this link over time is not obvious in this region.Finally, it is notable that the DC07 and DC04 pits (about 70 cm) from Frey et al. (2009) cover 8-10 years and thus overlap some with data from P4-P7.Neither of the DC pits show increasing δ 18 O(NO − 3 ) with depth (or decreasing δ 15 N(NO − 3 ) such as that in P7).Given the large spatial influence of stratospheric ozone on surface irradiance in Antarctica, it seems unlikely that Dome A and its surrounding region would be affected by this process and not Dome C.This suggests that something more localized, such as a change in the photolability of NO − 3 due to changes in snow chemistry, may have an important influence in recent decades.

Seasonal shifts in NO −
3 sources to coastal snow (group I snowpits) Pit profiles from the higher accumulation sites (group I) do not fit an exponential function but instead show periodic variability in mass fraction and isotopic composition of NO − 3 (Fig. 3).
As discussed above, sharp decreases in w(NO − 3 ) in the top few centimeters of inland East Antarctic snowpits are interpreted as evidence of severe photolytic NO − 3 loss.w(NO − 3 ) in the top 10 cm of the coastal P1 snowpit also decreases from the surface (Fig. 3) and, if viewed in isolation, could also be taken as evidence for post-depositional loss.However, annual average snow accumulation at P1 is approximately 50 cm snow a −1 , and the full profile clearly shows that similarly high w(NO − 3 ) values are observed below 10 cm, as would be expected from seasonal cycles.We expect that if the coastal sites studied by Erbland et al. (2013) had been continuously sampled below 20 cm, similar features would have been observed.This should serve as caution in interpreting the behavior of NO − 3 at high accumulation sites based on observations that do not cover a full annual cycle of snowfall.
Although a significant correlation between ln(1 +δ 15 N) and ln(w(NO − 3 )) was observed in the upper snow layers (upper 20, 40 and 60 cm; Table 2) in P3, no relationship was found between the two parameters when the top most sample was removed (ln(1 +δ 15 N) vs. ln(w(NO − 3 )) in P3; Fig. 5).This is different from the inland snowpits, where the top most sample is less influential on the linear fit (i.e., ln(1 +δ 15 N) vs. ln(w(NO − 3 )) in P4-P7; Fig. 5).In addition, the 15 ε app values found in P3, while statistically significant, are much higher than those calculated for P4-P7 (Table 2).The 18 ε app is almost entirely not statistically significant for P1, P2 and P3.In the case of the group I snowpits, the relationships amongst δ 15 N(NO − 3 ), δ 18 O(NO − 3 ), and w(NO − 3 ) are difficult to interpret as evidence of photolysis (Figs. 4 and 5).
Profiles of the group I pits (P1-P3) show large variations in w(NO − 3 ) and isotopic composition throughout the snowpack, with some correspondence to δ 18 O(H 2 O) which is a proxy for temperature (Fig. 7; see also Fig. 3).The seasonality is most apparent at site P1 due to the highest sampling resolution (3.0 cm per sample compared to 5.0 cm per sample for pits P2 and P3) and highest snow accumulation rate (172 kg m −2 a −1 ), though all group I sites feature high accumulation rates above 91 kg m −2 a −1 (Table 1).
It is difficult to assign samples to four distinct seasons based on δ 18 O(H 2 O) alone, so we choose a conservative classification of two periods: a warm season corresponding to higher δ 18 O(H 2 O) and a cold season characterized by lower δ 18 O(H 2 O) (Fig. 7).These assignments are also consistent with other established seasonal tracers measured in P1 in that the δ 18 O(H 2 O) peaks (warm season) correspond to spikes in methanesulfonic acid and low Na + mass fractions, while the opposite pattern is present during the identified cold seasons (C.-J.Li, personal communication, 2014).The snow ac-

Figure 1 .
Figure 1.Snowpit locations sampled during the 2012-2013 Chinese National Antarctic Research Expedition (CHINARE) inland traverse.The numbers in parentheses denote the annual snow accumulation rates (kg m −2 a −1 ) which are extended to 2013 from bamboo stick field measurements (Ding et al., 2011).

Figure 2 .
Figure 2. Statistics of mass fraction and isotopic compositions of NO −3 for each snowpit (P1-P7), plotted as a function of snow accumulation rate.Box and whisker plots represent maximum (top end dash symbol for each box), minimum (bottom end dash symbol for each box), the range 1-99 % (top and bottom X symbol for each box), percentiles (5th, 25th, 75th, and 95th), and median (50th, solid line) and mean (open square near center of each box).Note that the data of 17 O(NO − 3 ) are only available for P1 snowpit and refer to the right-hand axis.

Figure 3 .
Figure 3. Detailed profiles of w(NO − 3 ) (a) and isotopic composition δ 15 N (b) and δ 18 O of NO − 3 (c) in the P1-P7 snowpits.Note that the x axis of w(NO −3 ) in P3 and P7 was broken so the trend in the deeper snow can be seen.

Figure 4 .
Fig. 4. Relationships between δ 18 O and δ 15 N of NO 3 -in the topmost 20 cm of the snowpits.Least squares regressions are noted with lines and are significant at p<0.05.
• C), physical release should deplete both 15 N and 18 O in the remaining snow NO − 3 according to the modeled ε values, whereas the field experiment observations would suggest negligible change in δ 15 N with decreasing w(NO − 3 ) in the snow.The observations for P4-P7 show increasing δ 15 N(NO −3 ) with decreasing w(NO − 3 ), and that δ 15 N(NO − 3 )
Figure 5.The relationships between w(NO −3 ) and δ 18 O and δ 15 N of NO − 3 in the near surface (< 20 cm) and at depth (100-200 cm for group II snowpits) snow ranges.Least squares regressions are noted with lines and are significant at p < 0.05.It is noted that the linear regression slope for individual snowpits is the apparent fractionation constant ( 15 ε app and 18 ε app ; Table2).The red dashed lines in the panels of P3 (ln(δ 15 N + 1) vs. ln(w(NO − 3 ))) and P1 (ln(δ 18 O + 1) vs. ln(w(NO − 3 ))) are indicative of the change in the linear fit if the top most samples (open circle in P3, top row; open diamond in P1, second row) are removed.

Table 2 .
Observed fractionation constants for 15 N and 18 O of NO − 3 ( 15 ε app and 18 ε app ) calculated for different snow layer depths from the linear regression of ln(δ snow + 1) vs. ln(w snow ) in Eq. (2).Five different depth intervals were selected for calculating ε app : 0-20, 0-40, 0-60 cm, 100 cm-bottom and the entire pit.Also given are the standard error (1σ ), R 2 values and the significance level, p, where bolded values represent p < 0.05.

Table 3 .
Asymptotic values of w(NO −