Deposition, recycling and archival of nitrate stable isotopes between the air-snow interface: comparison between Dronning Maud Land and Dome C, Antarctica

The nitrate (NO 3- ) isotopic composition δ 15 N-NO 3- of polar ice cores has the potential to provide constraints on past ultraviolet 10 (UV) radiation and thereby total column ozone (TCO), in addition to the oxidising capacity of the ancient atmosphere. However, understanding the transfer of reactive nitrogen at the air-snow interface in Polar Regions is paramount for the interpretation of ice core records of δ 15 N-NO 3- and NO 3- mass concentrations. As NO 3- undergoes a number of post-depositional processes before it is archived in ice cores, site-specific observations of δ 15 N-NO 3- and air-snow transfer modelling are necessary in order to understand and quantify the complex photochemical processes at play. As part of the Isotopic Constraints 15 on Past Ozone Layer Thickness in Polar Ice (ISOL-ICE) project, we report new measurements of NO 3- concentration and δ 15 N-NO 3- in the atmosphere, skin layer (operationally defined as the top 5 mm of the snow pack), and snow pit depth profiles at Kohnen Station, Dronning Maud Land (DML), Antarctica. We compare the results to previous studies and new data, presented here, from Dome C, East Antarctic Plateau. Additionally, we apply the conceptual one-dimensional model of TRansfer of Atmospheric Nitrate Stable Isotopes To the Snow (TRANSITS) to assess the impact of photochemical processes that drive the 20 archival of δ 15 N-NO 3- and NO 3- in the snow pack. We find clear evidence of NO 3- photolysis at DML, and confirmation of our hypothesis that UV-photolysis is driving NO 3- enough to preserve the seasonal cycle of NO 3- concentration and δ 15 N-NO 3- , in contrast to Dome C where the profiles are smoothed due to stronger photochemistry. TRANSITS sensitivity analysis of δ 15 N-NO 3- at DML highlights that the dominant factors controlling the archived δ 15 N-NO 3- signature are the snow accumulation rate and e-folding depth, with a smaller role from changes in the snowfall timing and TOC. Here we set the framework for the interpretation of a 1000-year ice core record of δ 15 N-NO 3- from DML. Ice core δ 15 N-NO 3- records at DML will be less sensitive to changes in UV than at Dome C, however 35 the higher snow accumulation rate and more accurate dating at DML allows for higher resolution δ 15 N-NO 3- records. 3- δ 15 N-NO 3- snow pit Fig. Both the simulated depth profile of NO 3- mass concentration and δ 15 N-NO 3- for an accumulation rate of 6 cm yr -1 (w.e.) show seasonal variability in the first year with a range of of 380 ng g -1 and 20 ‰, which decreases with depth to a range of 95 ng g -1 and 10 ‰ in the fourth year. Also plotted are the simulated NO 3- and δ 15 N-NO 3- depth profiles for accumulation rates of 2.5 cm yr -1 (w.e.) and 100 cm yr -1 (w.e.). As the accumulation rate increases, the annual layers of δ 15 N-NO 3- become thicker, the seasonal amplitude increases, the mean annual δ 15 N-NO 3- value decreases, and the δ 15 N-NO 3- values in the top 10 cm decrease. At very low snow accumulation rates, the seasonal cycle is smoothed, as in the case of Dome C. A similar pattern is observed for the simulated NO 3- mass concentrations with depth: seasonal cycles of NO 3- mass concentrations are more pronounced at higher snow 445 accumulation rates, while inter-annual variability is smoothed at Dome C. The simulated archived (i.e., annual average of the first year below 1 m) NO 3- mass concentration, δ 15 N-NO 3- , and NO 3- mass flux values are 120 ng g -1 , 130 ‰, and 210 pg m -2 yr -1 , respectively. The simulated annual average 15 ε app is -19 ‰ for the top 30 cm (i.e., active photic zone with an e-folding depth of 10 cm).

the opposing effects of higher concentrations of both photolabile NO3and light absorbing impurities (e.g. dust and black carbon) in Antarctica and Greenland respectively. At Concordia Station on Dome C in East Antarctica, the light penetration depth (e-folding depth) is ~10 cm for wind pack layers and ~20 cm for hoar layers (France et al., 2011). Based on the 65 propagation of light into the snow pack, the snow pack can be divided into three layers. The first layer is known as the skin layer (a few mm thick) where direct solar radiation is converted into diffuse radiation. The second layer is called the active photic zone (below the skin layer layer), where solar radiation is effectively diffuse and the intensity of the radiation decays exponentially (Warren, 1982). The third layer is called the archived zone (below the active photic zone), where no photochemistry occurs. 70 Previous research has focused predominantly on the high elevation polar plateau (Dome C). Here, the exponential decay of NO3mass concentrations in the snow pack and thus post-depositional processing of NO3were attributed to either evaporation or ultra-violet (UV)-photolysis Röthlisberger et al., 2002). The open debate of which postdepositional process controlled NO3mass concentrations in the snow pack led to the use of a new isotopic tool, the stable isotopic composition of nitrate (δ 15 N-NO3 -) (Blunier et al., 2005). More recently, theoretical , laboratory 75 Erbland et al., 2013;Erbland et al., 2015;Shi et al., 2019;, and field Frey et al., 2009) evidence show that NO3mass loss from the surface snow to the overlying atmosphere and its associated isotopic fractionation is driven by photolysis. The physical release or evaporation of NO3is negligible Shi et al., 2019).
At Dome C, the large redistribution and net mass loss of NO3below the skin layer and the simultaneous isotopic fractionation 80 of δ 15 N-NO3in the snow pack indicate that post-depositional processes significantly modify the original NO3concentration and δ 15 N-NO3composition . Skin layer observations of δ 15 N-NO3in the surface snow at Dome C show strong enrichment compared to the atmospheric δ 15 N-NO3signature. Furthermore, snow pit profiles show an exponential decrease of NO3concentration and an enrichment in the δ 15 N-NO3composition with depth . Once NOx is produced by NO3photolysis, it is expected to have a lifetime in the polar troposphere of <1 day before it is oxidised to nitric 85 acid (HNO3) at Dome C and South Pole (Davis et al., 2004b), and can then be redeposited to the skin layer (e.g. Mulvaney et al., 1998).
This research at Dome C laid the foundation for Erbland et al. (2015) to derive a conceptual model of UV-photolysis induced post-depositional processes of NO3at the air-snow interface. "Nitrate recycling" is the combination of NOx production from NO3photolysis in snow, the subsequent atmospheric processing and oxidation of NOx to form atmospheric HNO3, the 90 deposition (dry and/or wet) of a fraction of the HNO3, and the export of another fraction. In NO3recycling, the skin layer is an active component of the atmosphere. This recycling can occur multiple times before NO3is eventually archived below the active photic zone in ice cores (Davis et al., 2008;Erbland et al., 2015;Zatko et al., 2016;Sofen et al., 2014). We refer to atmospheric NO3as the combination (i.e., total) of HNO3 (gas phase) and particulate NO3 -.
Year round measurements of NO3mass concentrations and δ 15 N-NO3in the skin layer and atmosphere at Dome C have 95 provided insights into the annual NO3cycle in Antarctica ( Fig. 1)  . In the early winter, the stratosphere https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License.
undergoes denitrification via formation of PSC. As PSC sediment slowly, there is a delay between the maximum stratospheric NO3concentration and the maximum NO3concentration deposited in the skin layer in late winter (Mulvaney and Wolff, 1993;Savarino et al., 2007). In spring, surface UV increases and initiates photolysis-driven post-depositional processes, which redistribute NO3between the snow pack and overlying air throughout the sunlit summer season. This results in the δ 15 N-NO3 -100 isotopic enrichment of the NO3skin layer reservoir, and maximum atmospheric NO3mass concentrations in October-November. In summer, NO3resembles a strongly asymmetric distribution within the atmosphere-snow column with the bulk residing in the skin layer and only a small fraction in the atmospheric column above.
Over longer time scales, UV-driven post-depositional processing of NO3is also driven by changes in the degree of postdepositional loss of NO3with greater NO3loss during the glacial period relative to the Holocene. The observed glacial-105 interglacial difference in post-depositional processing of NO3is dominated by variations in snow accumulation rate (Geng et al., 2015).
Nitrate is not preserved in the snow pack at sites with very low snow accumulation rates (i.e., Dome C: 2.5-3 cm yr -1 ) because snow layers remain close to the surface and in contact with the overlaying atmosphere for a relatively long time enhancing the effect of post-depositional processes. At sites with low snow accumulation rates, the source signature of δ 15 N-NO3is erased 110 by post-depositional process. Therefore, photolysis induced NO3loss and δ 15 N-NO3fractionation is dependent on snow accumulation. Three distinct transects from coastal Antarctica to the East Antarctic Plateau show that NO3fractionation is strongest with decreasing snow accumulation (Shi et al., 2018;Erbland et al., 2013;Noro et al., 2018). Skin layer NO3mass concentrations are significantly higher at low snow accumulation sites, for example ~160 ng g -1 (winter) to 1400 ng g -1 (summer) at Dome C compared to 50 ng g -1 (winter) to 300 ng g -1 (summer) at Dumont d'Urville (DDU) on the Antarctic 115 coast. Furthermore, the strong inverse linear relationship between NO3concentration and accumulation rate was revealed in a composite of seven ice cores across Dronning Maud Land (DML) (Pasteris et al., 2014).
Yet, NO3photolysis leaves its own process-specific imprint in the snow pack (Shi et al., 2019;Erbland et al., 2015;, which opens up the possibility to use δ 15 N-NO3to infer past surface-UV variability ). However, NO3photolysis rates in snow depend on a number of site-specific factors as does the degree of photolytic isotopic fractionation 120 of NO3eventually preserved in ice cores . These factors need to be quantitatively understood at a given ice core site to enable quantitative interpretation of ice core records. Here, we carry out a comprehensive study of the air-snow transfer of NO3at Kohnen Station in DML, East Antarctica through δ 15 N-NO3measurements in the atmosphere, skin layer and snow pits, and compare the observations to Dome C. Due to the previous research outlined above, we assume that the photolysis is the dominant driver of NO3post-depositional processes, and later assess the validity of this 125 this assumption (section 4.3). We apply the Transfer of Atmospheric Nitrate Stable Isotopes To the Snow (TRANSITS) model  to i) understand how NO3mass concentrations and δ 15 N-NO3are archived in deeper snow and ice layers, and ii) investigate the sensitivity of changes in the past snow accumulation rate, snowfall timing, e-folding depth, and TCO on the δ 15 N-NO3signature. In order to interpret this novel UV proxy, it is paramount to understand the air-snow transfer processes specific to an ice core site, and how δ 15 N-NO3is archived in the deeper snow and ice layers (Geng et al., 2015;Morin et al., 130 https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License. 2009;Erbland et al., 2015). Within the framework of the Isotopic Constraints on Past Ozone Layer Thickness in Polar Ice (ISOL-ICE) project, this study provides a basis for the interpretation of δ 15 N-NO3from a 1000-year ice core recovered in 2016/17 at Kohnen Station.

Methods
The ISOL-ICE project aims to understand natural causes of past TCO variability by i) an air-snow exchange study to enable 135 the interpretation of ice core records of NO3and δ 15 N-NO3 -, ii) reconstructing a 1000-year record of UV using a new ice core proxy based on δ 15 N-NO3 - (Ming et al., in prep;Winton et al., 2019a), and iii) numerical modelling of the natural causes of TCO variability. In the air snow-transfer study presented here, we report new atmospheric, skin layer and snow pit NO3and δ 15 N-NO3observations from DML, and compare them to new and published Erbland et al., 2013; observations from Dome C. Published data from Dome C comprises year round atmospheric and skin layer 140 measurements from 2009-2010 , and multiple snow pit profiles Frey et al., 2009).
We present a new extended time series at Dome C of year round atmospheric and skin layer NO3and δ 15 N-NO3from 2011-2015.

Study sites
The ISOL-ICE campaign was carried out at the summer only, continental Kohnen Station where the deep European Project 145 for Ice Coring in Antarctica (EPICA) Dronning Maud Land (EDML; 75°00' S, 0°04' E; 2982 m a.s.l.; https://www.awi.de/en/expedition/stations/kohnen-station.html) ice core was recovered in 2001-2006 to a depth of ~2800 m (Wilhelms et al., 2017). As part of the ISOL-ICE campaign, a new ice core (ISOL-ICE; (Winton et al., 2019a)) was drilled 1 km from the EDML borehole (Fig. 2). In addition, the ISOL-ICE air-snow transfer study site was located ~200 m from the EDML ice core site (Fig. 2). Here we compare two ice core drilling sites in Antarctica: Kohnen Station (referred to as DML 150 henceforth) and EPICA Dome C (75°05'59'' S, 123°19'56'' E) (Fig. 2). Both sites are similar in terms of the latitude and therefore in terms of radiative forcing at the top of the atmosphere (Table 1). Satellite images of TCO over Antarctica show that the lowest annual TCO values are centred over the South Pole region encompassing DML and usually Dome C although the spatial variability is significant from year to year (https://ozonewatch.gsfc.nasa.gov/). The sites are different in terms of their location with respect to moisture source, elevation and precipitation regime. The DML site is situated ~550 km from the 155 ice shelf edge, is subject to cyclonic activity and receives ~80 % of its precipitation from frontal clouds (Reijmer and Oerlemans, 2002). While Dome C is more remote (~1100 km from the coast) and diamond dust is the dominant form of precipitation. The annual snow accumulation rate also differs between the sites; while both sites have exceptionally low accumulation compared to the coast, DML (annual mean: 6 cm yr -1 (water equivalent; w.e.)) receives more than double that of Dome C (annual mean: 2.5 cm yr -1 (w.e.)) (Le Meur et al., 2018;Hofstede et al., 2004;Sommer et al., 2000). Throughout the 160 study we refer to our sampling site as "DML". https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License.

Snow and aerosol sampling
Daily skin layer samples (which we operationally define as the top 5 mm of the snow pack following Erbland et al. (2013)) were collected from the DML site (Fig. 2) in January 2017 during the ISOL-ICE ice core drilling and atmospheric monitoring campaign. To prevent contamination from the nearby Kohnen Station, snow samples were collected from the "flux site" within 165 the station's designated clean air sector (defined as 45° from both ends of the station building) located ~1 km from the station (Fig. 2). The skin layer samples were was collected in polyethylene bags (Whirl-pak®) using a stainless steel trowel. A total of 45 skin layer samples were collected between 31 December 2016 and 29 January 2017 from a designated sampling site each day during the campaign (75°00.184' S, 000°04.527' E; Fig. 2). To determine the spatial variability of NO3in the skin layer at the flux site, an additional five skin layer samples were collected in a ~2500 m 2 area of the flux site (75°00.161' S -170 000°04.441' E, 75°00.175' S -000°04.518' E; Fig. 2c).
Adjacent to the skin layer samples, snow was sampled from a 1.6 m snow pit at the flux site (snow pit B; Fig. 2c) and a 2 m snow pit at the "ice core" site (snow pit A; Fig. 2b). Two parallel profiles were sampled, i) for major ion mass concentrations (including NO3 -) collected in pre-washed 50 mL Corning® centrifuge tubes at 3 cm resolution by inserting the tube directly into the snow face, and ii) for stable NO3isotope analysis collected in Whirl-pak® bags at 2 cm resolution using a custom-175 made stainless steel tool. Exposure blanks (following the same method as the samples by opening the tube/ Whirl-pak® bag at the field site but not filling the sample container with snow) were also collected for both types of samples. Snow density and temperature were measured every 3 cm, and a visual log of snow pit stratigraphy was recorded.
Daily aerosol filters were collected using high-volume aerosol samplers custom-built at the Institute of Environmental Geosciences (IGE), University of Grenoble-Alpes, France described previously Erbland et al., 2013). The 180 high-volume sampler collected atmospheric aerosol on glass fibre filters (Whatman GF/A filter sheets; 20.3 × 25.4 cm) at an average flow rate of 1.2 m 3 min −1 at standard temperature and pressure (STP; temperature: 273.15 K; pressure: 1 bar) to determine the concentration and isotopic composition of atmospheric NO3 -. It is assumed that the atmospheric NO3collected on glass fibre filters represents the sum of atmospheric particulate NO3and HNO3 (gas phase). The bulk of HNO3 present in the gas phase is most likely adsorbed to aerosols on the filter, as described previously . 185 The high-volume sampler was located 1 m above the snow surface at the flux site at the DML site (Fig. 2c), where a total of 35 aerosol filters were sampled daily between 3 and 27 January 2017. In addition, we coordinated an intensive 4-hour sampling campaign in phase with Dome C, East Antarctica (Fig. 2) between 21 and 23 January 2017. At Dome C, high-volume sampler is located on the roof of the atmospheric shelter (6 m above the snow surface), where a total of 12 samples were collected. At DML, loading and changing of aerosol collection substrates was carried out in a designated clean area. Aerosol laden filters 190 were transferred into individual double zip-lock plastic bags immediately after collection and stored frozen until analysis at the British Antarctic Survey (BAS; major ions) and IGE (NO3isotopic composition). For the atmospheric NO3work, three types of filter blanks were carried out; i) laboratory filter blanks (n = 3; Whatman GF/A filters that underwent the laboratory procedures without going into the field), ii) procedural filter blanks (DML: n = 4; Dome C: n = 1; filters that had been treated https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License. as for normal samples but which were not otherwise used; once a week, during daily filter change-over, a procedural blank 195 filter was mounted in the aerosol collector for 5 min without the collector pump in operation -this type of filter provides an indication of the operational blank associated with the sampling procedure), and iii) 24 h exposure filter blanks sampled at the beginning and end of the field campaign (DML: n = 2; Dome C: n = 1; filters treated like a procedural blank but left in the collector for 24 h without switching the collector on). All samples were kept frozen below -20 ºC during storage and transport prior to analysis. 200 In addition, skin layer and aerosol samples have been sampled continuously at Dome C over the period 2009-2015 following Erbland et al. (2013);Frey et al. (2009). The sampling resolution for skin layer is every 2-4 days, and weekly for aerosol samples. Data from 2009-2010 have previously been published by Erbland et al. (2013), and we report the 2011-2015 data here ( Fig. 1).

Major ion mass concentrations in snow and aerosol 205
Aerosol NO3and other major ions were extracted in 40 mL of ultra-pure water (resistivity of 18.2 MΩ; Milli-Q water) by centrifugation using Millipore Centricon® Plus-70 Filter Units (10 kD filters) in a class-100 clean room at the BAS. Major ion mass concentrations in DML snow samples were determined in an aliquot of melted snow from skin layer and snow pit samples, and aerosol extracts by suppressed ion chromatography (IC) using a Dionex™ ICS-4000 Integrated Capillary HPIC™ System ion chromatograph. A suite of anions, including NO3 -, chloride (Cl -), methanesulfonic acid (MSA) and sulphate (SO4 2-210 ), were determined using an AS11-HC column and a CES 500 suppressor. Cations, including sodium (Na + ), were determined using a CS12A column and a CES 500 suppressor. During the course of the sample sequence, instrumental blank solutions and certified reference materials (CRM; ERM-CA616 groundwater standard and ERM-CA408 simulated rainwater standard; Sigma-Aldrich) were measured regularly for quality control and yielded an accuracy of 97 % for NO3 -. Nitrate mass concentrations in Dome C samples were determined by colorimetry at IGE following the procedure described in Frey et al. 215 (2009). Blank concentrations for exposure blank, procedural blank and laboratory blank and detection limits are reported in Table S1. The non-sea-salt sulphate (nss-SO4 2-) fraction of SO4 2was obtained by subtracting the contribution of sea-saltderived SO4 2from the measured SO4 2mass concentrations (nss-SO4 2-= SO4 2--0.252 × Na + , where Na + and SO4 2are the measured concentrations in snow pit samples and 0.252 is the SO4 2-/ Na + ratio in bulk seawater.

Nitrate isotopic composition in snow and aerosol 220
Samples were shipped frozen to IGE where the NO3isotope analysis was performed. The denitrifier method was used to determine the stable NO3isotopic composition in samples at IGE following Morin et al. (2008). Briefly, samples were preconcentrated due to the low NO3mass concentrations found in the atmosphere and snow over Antarctica. To obtain 100 nmol of NO3required for NO3isotope analysis, the meltwater of snow samples and aerosol extracts were sorbed onto 0.3 mL of anion exchange resin (AG1-X8 chloride form; Bio-Rad) and eluted with 5 x 2 mL of 1 M NaCl (high purity grade 99.0 %; 225 American Chemical Society (ACS grade); AppliChem Panreac) following Silva et al. (2000). Recovery tests yielded 100 %  Erbland et al., 2013). Once pre-concentrated, NO3is converted to N2O gas by denitrifying bacteria, Pseudomonas aureofaciens. The N2O is split into O2 and N2 on a gold furnace heated to 900 °C followed by gas chromatographic separation and injection into the isotope ratio mass spectrometer (IRMS) for duel O and N analysis using a Thermo Finnigan™ MAT 253 IRMS equipped with a GasBench II™ and coupled to an in-house-built NO3interface (Morin 230 et al., 2009).
Certified reference materials (IAEA USGS-32, USGS-34 and USGS-35;Böhlke et al., 1993;Böhlke et al., 2003) were prepared (matrix match 1 M NaCl in identical water isotopic composition as samples; ACS grade) and subject to the same analytical procedures as snow and aerosol samples. The nitrogen isotopic ratio was referenced against N2-Air (Mariotti, 1983).
We report 15 N/ 14 N of NO3 -(δ 15 N-NO3 -) as δ-values following Eq. (1). 235 δ 15 N-NO3 -= ( 15 N/ 14 Nsample / 15 N/ 14 Nstandard -1) (1) For each batch of 60 samples, the overall accuracy of the method is estimated as the reduced standard deviation of the residuals from the linear regression between the measured reference materials (n = 16) and their expected values. For the snow (n = 118) and aerosol samples (n = 35), the average uncertainty values obtained for δ 15 N was 0.5 ‰ for both datasets.

Light attenuation through the snow pack (e-folding depth) 240
Measurements of light attenuation through the snow pack were made at the two snow pit sites during the ISOL-ICE campaign following a similar approach of previous studies (France and King, 2012;France et al., 2011). Vertical profiles of down-welling irradiance in the top 0.4 m of the snow pack were measured using a high-resolution spectrometer (HR4000; Ocean Optics) covering a spectral range of 280 to 710 nm. To do this, a fiber optic probe attached to the spectrometer and equipped with a cosine corrector with spectralon diffusing material (CC-3-UV-S; Ocean Optics) was inserted into the snow to make 245 measurements at approximately 0.03 m depth intervals. The fiber optic probe was either inserted horizontally into pre-cored holes, at least 0.5 m in length to prevent stray light, into the side wall of a previously dug snow pit or pushed gradually into the undisturbed snow pack starting at the surface at a 45º angle, which was maintained by a metal frame. Most measurements with integration time ranging between 30 and 200 ms were carried out at noon to minimise changing sky conditions, and each vertical snow profile was completed within 0.5 hr. The spectrometer was calibrated against a known reference spectrum from 250 a Mercury Argon calibration source (HG-1; Ocean Optics), dark spectra were recorded in the field by capping the fibre optic probe and spectral irradiance was then recorded at depth relative to that measured right above the snow surface.
The e-folding depth was then calculated according to the Beer-Bouguer Lambert law. Stratigraphy of the snow pack recorded at each site showed presence of several thin (10 mm) wind crust layers over the top 0.4 m of snow pack. However, calculating e-folding depths for each layer in between wind crusts yielded inconclusive results. Therefore, reported e-folding depths (Fig. 255 S1, Table S2) are based on complete profiles integrating potential effects from wind crust layers. Resulting e-folding depths relevant for the photolysis of NO3 - (Table S2) show significant standard deviations, and also considerable variability (0.9-4.0 cm) between profiles, which reflect both systematic experimental errors as well as spatial variability of snow optical properties.
They are lower than at Dome C but similar to previous model estimates for South Pole (France et al., 2011;Wolff et al., 2002). The origin of the reduced e-folding depth relative to Dome C is not known but is likely due to greater HUmic-LIke Substances 260 (HULIS) content or different snow morphology (Libois et al., 2013;Zatko et al., 2013). We use e-folding depths observed in this study at DML and those reported previously at Dome C as guidance for our model sensitivity study to quantify the impact of the variability of e-folding depth on archived δ 15 N-NO3in snow.

Nitrate photolysis rate coefficient
Hemispheric or 2π spectral actinic flux from 270 to 700 nm was measured at 2.1 m above the snow surface using an actinic 265 flux spectroradiometer (Meteorologieconsult GmbH; Hofzumahaus et al. (2004). 2π NO3photolysis rate coefficients J(NO3 -) were then computed using the NO3absorption cross section and quantum yield on ice estimated for -30 ºC from Chu and Anastasio (2003). The mean 2π J-NO3value at DML during January 2017 was 1.02 x 10 -8 s -1 , and 0.98 x 10 -8 s -1 during the 1 to 14 January 2017 period. The observed 2π J(NO3 -) at DML was a factor of three lower than Dome C (2.97 x 10 -8 s -1 ; 1 to 14 January 2012) which was previously measured using the same instrument make and model, and at the same latitude (Kukui et 270 al., 2013). Only ~5 % of the apparent inter-site difference can be attributed to TCO being ~25 DU larger at DML (306 DU) than at Dome C (287 DU) during the comparison period. The remainder was possibly due to greater cloudiness at DML and differences in calibration. In this study, the observed 2π J(NO3 -) is used to estimate the snow emission flux of NO2.

Air-snow transfer modelling
In order to evaluate the driving parameters of isotope air-snow transfer at DML we used the TRANSITS model (Erbland et 275 al., 2015) to simulate snow profiles of NO3concentration and δ 15 N-NO3and compare them to our observations. TRANSITS is a conceptual multi-layer 1D model which aims to represent NO3recycling at the air-snow interface including processes relevant for NO3snow photochemistry (UV-photolysis of NO3 -, emission of NOx, local oxidation, deposition of HNO3) and explicitly calculates NO3mass concentrations and δ 15 N-NO3in snow. Due to the reproducible depth profile of δ 15 N-NO3within 1 km (section 3.3), we assume δ 15 N-NO3composition is spatially uniform at DML and thus a 1D model is appropriate 280 for our site. The atmospheric boundary layer in the model is represented by a single box above the snow pack. The 1 m snow pack is divided into 1000 layers of 1 mm thickness. Below the photic zone of the snow pack, the NO3mass concentrations and δ 15 N-NO3values are assumed to be constant and thus archived during the model run. The model is run for 25 years, which is sufficient to reach steady state. The input data is provided in Table S3.
Photolysis rate coefficients of NO3 -(J(NO3 -)) above and within the snowpack are used by the TRANSITS model runs as input 285 for this study, and are modelled off-line using the tropospheric ultraviolet and visible (TUV)-snow radiative transfer model (Lee-Taylor and Madronich, 2002). The following assumptions were made: i) a clear aerosol-free sky, ii) extra-terrestrial irradiance from Chance and Kurucz (2010), and iii) a constant Earth-Sun distance as that on 27 December 2010 (Erbland et al., 2015). The TUV-snow radiative transfer model was constrained by optical properties of the Dome C snow pack (France et al., 2011), notably an e-folding depth of i) 10 cm in the top 0.3 m, and ii) 20 cm below 0.3 m , to compute 290 J( 14/15 NO3 -) profiles as a function of solar zenith angle (SZA) and TCO   The set up used in this paper is similar to Erbland et al. (2015) except for the following modifications. We use the TCO from the NIWA Bodeker combined dataset version 3.3, at the location of the snow pit site, averaged from 2000 to 2016 (http://www.bodekerscientific.com/data/total-column-ozone). The year round atmospheric NO3concentration is taken from 295 Weller and Wagenbach (2007), and the meteorology data is taken from Utrecht University automatic weather Station (AWS) at DML05/Kohnen (AWS9; https://www.projects.science.uu.nl/iceclimate/aws/files_oper/oper_20632). The snow accumulation rate is set to 6 cm yr -1 (w.e.) . We carried out a sensitivity analysis to evaluate the impact of variable accumulation rate, timing of snowfall, and e-folding depth on the snow profile of NO3and δ 15 N-NO3 -. The sensitivity tests were as followed: the snow accumulation rate was varied between the bounds seen in the last 1000-years at 300 DML; the snow accumulation rate was varied from year to year according to our snow pit profile; the timing of the snow accumulation was varied throughout the year; and the e-folding depth was varied within the range of observations from this study and previously at Dome C. To evaluate the sensitivity of archived δ 15 N-NO3to e-folding depth, changes the J( 14/15 NO3 -) profiles for Dome C  were recalculated and used as TRANSITS input by scaling the surface value of J( 14/15 NO3 -) with a new e-folding depth (2, 5, 10, 20 cm). An example is shown in Fig. S2a for SZA = 70º, TCO = 300 DU and 305 an e-folding depth of 5 cm. The top 2 mm are retained from the Dome C base case to account for non-linearities in snow radiative transfer in snow, which are strongest in the non-diffuse zone right below the snow surface (Fig. S2b). It is noted that TUV-snow model estimates of down-welling or 2π J(NO3 -) above the snow surface at the latitude of Dome C or DML (= 75º S) compare well to observations at Dome C in January 2012, whereas they are a factor three higher than measurements at DML in January 2017 (Table S4 and section 2.6). This should not affect the results of the sensitivity study, which aims to 310 explore relative changes of archived δ 15 -NO3due to a prescribed change in e-folding depth.

Snow pit dating
Dating of the snow pits was based on the measured concentrations of Na + , MSA, and nss-SO4 2following previous aerosol and ice core studies at DML (Göktas et al., 2002;Weller et al., 2018). Here, Na + mass concentrations have a sharp, well-defined 315 peak in the austral spring/late winter, while MSA and nss-SO4 2-, primarily derived from the biogenic production of dimethylsulfide (DMS), record maximum concentrations in the austral autumn. Non-sea salt SO4 2-(nss-SO4 2-) often displays a second peak corresponding to late austral spring/summer sometimes linked to MSA. Spring seasons were defined as 1 September and positioned at the Na + peak, while autumn seasons were defined as 1 April and positioned where a MSA and nss-SO4 2peak aligned (Fig. S3). Annual layer counting of Na + layers shows snow pit A spans 8 years from autumn 2009 to 320 summer 2017 and snow pit B spans 9 years from summer 2008 to summer 2017 with an age uncertainty of ± 1 year at the base of the snow pit. The mean snow accumulation rate for the snow pits is estimated to be 6.3 ± 1.4 cm yr -1 (w.e.), consistent with https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License. accumulation rates of 6.0 -7.1 cm yr -1 (w.e.) from snow pits and ice cores from DML Hofstede et al., 2004;Oerter et al., 2000).

Nitrate mass concentrations 325
Atmospheric NO3mass concentrations (Caerosol) were estimated from high-volume aerosol filters by the ratio of total NO3mass loading to the total volume of air pumped through the filter at STP conditions following Eq. (2), and assuming a uniform loading of the aerosol filter. Caerosol = NO3mass loading / air volume (STP) ( Aerosol mass concentrations range from 0.5 to 19 ng m -3 and show a downward trend throughout January 2017 (R 2 =0.55; p= 330 <0.001; Fig. 3). In contrast, NO3mass concentrations in the skin layer increase during the month from 136 to 290 ng g -1 .
Nitrate mass concentrations in both snow pits, which range from 23 to 142 ng g -1 , are substantially lower than those in the skin layer. Compared to Dome C, average annual atmospheric, skin layer and snow pit mass concentrations are lower at DML (Table 2), in agreement with higher NO3mass concentrations found at lower snow accumulation sites .
The NO3mass concentration profile in the upper 50 cm of the snow pack at Dome C shows an exponential decrease with 335 depth and becomes relatively constant at 35 ng g -1 at 20 cm compared to 160-1400 ng g -1 in the skin layer (Figs. 1 and 4; Frey et al., 2009). While the highest NO3mass concentrations in the snow pack at DML are also found in the skin layer, the concentration profile exhibits a different pattern. The sharp decrease in NO3mass concentration occurs in the top ~5 mm at which point the snow pit records inter-annual variability in the NO3mass concentration. Nitrate mass concentrations at DML exhibit a maximum in summer and winter minimum. 340 Although the Dome C depth profiles of NO3mass concentration do not record seasonal variability, year-round measurements of skin layer and atmospheric NO3mass concentrations exhibit sharp maximum during sunlit conditions in spring and summer and low mass concentrations in winter. This annual cycle is consistent both i) spatially across Antarctica (McCabe et al., 2007;Erbland et al., 2013;Frey et al., 2009), and ii) temporally over last 7 years ( Fig. 1) Erbland et al., 2013;Frey et al., 2009). 345 While the precision of the IC measurement of NO3is better than 2 %, the spatial variability at DML of NO3in the skin layer exceeds this. During the sampling campaign, five skin layer samples were taken from an area of ~2500 m 2 at the flux site (snow surface had sastrugi up to 10 cm) to understand how representative the snow pit mass concentrations are of the greater study area. We found that the spatial variability of NO3mass concentrations and δ 15 N-NO3at DML was 10 % and 17 % respectively. At Dome C, the spatial variability of NO3mass concentrations is between 15 and 20 %. We note that this 350 variability includes the natural spatial variability and the operator sampling technique.

Isotopic composition of nitrate
Atmospheric δ 15 N-NO3ranges from -49 to -20 ‰ at DML and -9 to 8 ‰ at Dome C during the January campaign, and is depleted with respect to the skin layer, which ranges from -22 to 3 ‰ at DML (Fig. 3). Similar to the NO3mass concentrations, in spring and summer with respect to winter (Fig. 4). The δ 15 N-NO3in both snow pits at DML show extremely good reproducibility with depth indicating there is little spatial variability within 1 km at the site (Fig. 4). The δ 15 N-NO3in snow pits at Dome C do not preserve a seasonal cycle. However, in parallel with the exponential decay of NO3mass concentrations with depth at Dome C, there is a strong increase in the δ 15 N-NO3with depth. At Dome C, δ 15 N-NO3increases up to 250 ‰ in the top 50 cm, this increase is weaker at DML (up to 80 ‰ in the top 10 cm at which point seasonal cycles are evident). At 360 Dome C, although no annual cycle is preserved in the snow pack, the year-round measurements of δ 15 N-NO3in the atmosphere decrease during sunlit conditions in spring and summer (Fig. 1). While the δ 15 N-NO3in the skin layer has a spring minimum that increases to a maximum at the end of summer ( Fig. 1). Skin layer δ 15 N-NO3is about 25 ‰ higher than atmospheric δ 15 N-NO3 -. Nitrate mass concentration and δ 15 N-NO3composition data for aerosol, skin layer and snow pit samples are available in Winton et al. (2019b). 365

Archived nitrate mass concentration and isotopic composition
We calculate archived values of NO3mass concentration and δ 15 N-NO3which represent the archived mass fraction and isotopic composition reached below the photic zone. Archived values were calculated by averaging the NO3and δ 15 N-NO3values below the photic zone, i.e., 30 cm (section 4.4). The archived NO3mass concentration and δ 15 N-NO3values for snow pit A were 60 ng g -1 and 50 ‰, the and archived NO3mass concentration for snow pit B was 50 ng g -1 . Note that δ 15 N-NO3 -370 values were measured below 30 cm in snow pit B. These measured values are half of those expected for a site with a snow accumulation rate of 6 cm yr -1 (w.e.) in the spatial survey from Erbland et al. (2013) (Table 2).

Nitrate mass flux estimates
The total deposition flux (F) of NO3is partitioned into wet and dry deposition fluxes (Fwet and Fdry respectively; Eq. (3)), and can be estimated using the measured mass concentration of NO3in the snow pack (Csnow) and the local snow accumulation 375 rate (A; Eq. (4)). Estimates of the dry deposition rate (Fdry) of NO3were calculated using Eq. (5) using the atmospheric mass concentrations of NO3 -(Caerosol) and a dry deposition velocity (Vdry deposition) of 0.8 cm s −1 , and are reported in Table S5. This deposition velocity is based on the dry deposition of HNO3 at South Pole (Huey et al., 2004) which has a similar snow accumulation rate (6.4 cm yr -1 (w.e.); Mosley- Thompson et al. (1999)) to DML. Other estimates of dry deposition velocities include 0.05-0.5 cm s -1 for HNO3 over snow (Hauglustaine et al., 1994;Seinfeld and Pandis, 1998), 1.0 cm s -1 for NO3over 380 the open ocean (Duce et al., 1991), and an apparent deposition velocity of 0.15 cm s -1 for summer HNO3 at Dome C . The estimated apparent NO3deposition velocity at Dome C is low because of the strong recycling of NO3on the polar plateau in summer, i.e., reactive nitrogen is re-emitted from the skin layer to the atmosphere. Thus, the dry deposition velocity at DML is likely to lie between 0.15 and 0.8 cm s -1 . We assume that a constant deposition velocity throughout the campaign is appropriate for DML. 385  (5) Note that Eq. (4) does not take into account post-depositional processes of non-conservative ions, such as NO3 -. We follow the 390 approach of Erbland et al. (2013) who use an archived NO3flux (Fa) to represent the downward NO3flux which escapes the photic zone towards deeper snow layers. Using simple mass balance, we can then estimate the flux of NO3 -(Fre-emit), which is re-emitted from the snow pack to the overlaying atmosphere (Eq. (6)).
Fre-emit = F -Fa (6) Taking a simple mass balance approach, a schematic of NO3mass fluxes are illustrated in Fig. 5a. As the atmospheric 395 campaign did not cover an entire annual cycle, we use estimates of atmospheric NO3fluxes at DML reported by Pasteris et al. (2014) and Weller and Wagenbach (2007) of 43 and 45 pg m -2 s -1 , respectively, as a year round dry deposition fluxes. Due to the linear relationship of ice core NO3mass concentrations with the inverse accumulation, the authors assume that the magnitude of the dry deposition flux is homogenous over the DML region. Mean annual mass concentrations of NO3in our snow pits suggest a total NO3deposition mass flux of 110 pg m -2 s -1 and therefore a wet deposition mass flux of 65 pg m -2 s -400 1 .
However, at relatively low snow accumulation sites where photolysis drives the fractionation of NO3from the surface snow to atmosphere , it is necessary to take into account the skin layer in the NO3flux budget as this air-snow interface is where air-snow transfer of NO3takes place. We use the available NO3mass concentrations measured in aerosol, skin layer, and snow pits from the ISOL-ICE campaign to estimate the mass flux budget for January 2017 (Fig. 5b). The dry 405 deposition mass flux of atmospheric NO3during January 2017 at DML averages 64 ± 38 pg m -2 s -1 (Table S5). The NO3mass flux to the skin layer is 360 pg m -2 s -1 , however only 110 pg m -2 s -1 of NO3is archived. Considering the active skin layer, only 30 % of deposited NO3is archived in the snow pack while 250 pg m -2 s -1 is re-emitted to the overlaying atmosphere.

Fractionation constants
Fractionation constants were calculated following the approach of Erbland et al. (2013) which assumes a Rayleigh single loss 410 and irreversible process of NO3removal from the snow. As this approach may oversimplify the processes occurring at the airsnow interface, Erbland et al. (2013) referred to the quantity as an "apparent" fractionation constant. Thus, the apparent fractionation constant represents the integrated isotopic effect of the processes involving NO3in the surface of the snow pack and in the lower atmosphere. The apparent fractionation constant is denoted as 15 εapp and calculated using Eq. (7). ln(δ 15 Nƒ + 1) = 15 ε x lnƒ + ln(δ 15 N0 + 1) (7) 415 where δ 15 Nƒ and δ 15 N0 are the δ-values in the initial and remaining NO3 -, and ƒ is the remaining NO3mass concentration. The ε values are related to the commonly used fractionation factor α by ε = α − 1. The 15 εapp derived for snow pits in the photic zone is 12 ‰.

Simulated nitrate mass concentrations and isotopic ratios from TRANSITS modelling
Simulated TRANSITS results for the air snow interface are illustrated in Fig. 6. In the atmosphere, the TRANSITS model is 420 forced with the smoothed profile of year-round atmospheric NO3measurements from the DML site (Weller and Wagenbach, 2007) which has the highest mass concentrations in spring and summer with a maximum of 80 ng m -3 in November and a winter minimum of 2 ng m -3 (Fig. 6b). Although we only have measurements of δ 15 N-NO3in January, the simulated atmospheric δ 15 N-NO3values for January are greater than the measurements available from this study. The annual cycle of simulated atmospheric δ 15 N-NO3shows a 40 ‰ dip in spring to -32 ‰ from winter values which coincides with the simulated 425 atmospheric NO3mass concentration increase in spring. The highest atmospheric δ 15 N-NO3values (7 ‰) occur in winter. In the skin layer, the simulated NO3mass concentrations are an order of magnitude greater than our observations in January and we outline possible reasons for this discrepancy in the discussion (section 4.1). The simulated annual cycle of NO3mass concentrations in the skin layer steadily rise in spring and reach a peak in January when they begin to decline to the lowest concentration in winter. Simulated skin layer δ 15 N-NO3values in January are ~10 ‰ higher than our highest observations for 430 that month. They begin to decrease by 24 ‰ in spring at the same time as atmospheric δ 15 N-NO3values decrease. In October and November, the skin layer δ 15 N-NO3values begin to rise up to 14 ‰ in February.
The seasonality of NO3mass concentrations and δ 15 N-NO3values in the atmosphere and skin layer at DML is consistent with Dome C (Fig. 1). Similar to Dome C, NO3mass concentrations in the skin layer start to rise two months earlier than atmospheric NO3mass concentrations and the summer maximum is later. While the seasonality of δ 15 N-NO3in the skin layer 435 and atmosphere co-vary, simulated skin layer δ 15 N-NO3values are enriched relative to atmospheric values on average by 80 ‰.
The simulated NO3mass concentrations and δ 15 N-NO3values in the snow pit are illustrated in Fig. 7. Both the simulated depth profile of NO3mass concentration and δ 15 N-NO3for an accumulation rate of 6 cm yr -1 (w.e.) show seasonal variability in the first year with a range of of 380 ng g -1 and 20 ‰, which decreases with depth to a range of 95 ng g -1 and 10 ‰ in the fourth 440 year. Also plotted are the simulated NO3and δ 15 N-NO3depth profiles for accumulation rates of 2.5 cm yr -1 (w.e.) and 100 cm yr -1 (w.e.). As the accumulation rate increases, the annual layers of δ 15 N-NO3become thicker, the seasonal amplitude increases, the mean annual δ 15 N-NO3value decreases, and the δ 15 N-NO3values in the top 10 cm decrease. At very low snow accumulation rates, the seasonal cycle is smoothed, as in the case of Dome C. A similar pattern is observed for the simulated NO3mass concentrations with depth: seasonal cycles of NO3mass concentrations are more pronounced at higher snow 445 accumulation rates, while inter-annual variability is smoothed at Dome C. The simulated archived (i.e., annual average of the first year below 1 m) NO3mass concentration, δ 15 N-NO3 -, and NO3mass flux values are 120 ng g -1 , 130 ‰, and 210 pg m -2 yr -1 , respectively. The simulated annual average 15 εapp is -19 ‰ for the top 30 cm (i.e., active photic zone with an e-folding depth of 10 cm).

Validation of results
Our January 2017 NO3measurements agree well with values reported in the literature, and largely with the simulated results from the TRANSITS model with the exception of skin layer NO3mass concentration. While we made the first measurements of atmospheric, skin layer and snow pit δ 15 N-NO3 -, and skin layer NO3mass concentrations at DML, there are published measurements of NO3mass concentrations in snow pits and our concentrations agree well with those (Weller et al., 2004). 455 Our atmospheric mass NO3concentrations in January 2017 are lower than those observed in 2003 by Weller and Wagenbach (2007) which could be due to inter-decadal variability of atmospheric NO3 -(which varied by 30 ng g -1 over summer between 2003 and 2005) or reflect the different filter substrates used (Teflon/nylon versus glass fibre).
Overall, the simulated results are greater than our January observations, particularly the skin layer NO3mass concentrations (Fig. 6d). The discrepancy between the significantly higher simulated NO3mass concentrations than observations in the skin 460 layer was also found at Dome C. Erbland et al. (2015) suggested that this discrepancy could be related to either a sampling artefact, snow erosion or a modelled time response to changes in past primary inputs. We provide an alternative explanation for the extremely high simulated NO3mass concentrations in the skin layer using measurements of NO3mass concentration in diamond dust and hoar frost collected daily from Polyvinyl chloride (PVC) sheets at Dome C in summer 2007/08, i.e. new deposition. New deposition of diamond dust had NO3mass concentrations up to 2000 ng g -1 , which is four times greater than 465 that observed in natural snow from the skin layer at the same time (Fig. S4). Similarly, new deposition of hoar frost had NO3concentrations up to 900 ng g -1 , which is three times greater than the skin layer snow. The formation of surface hoar frost occurs by co-condensation, i.e. the simultaneous condensation of water vapour and NO3at the air-snow interface. Recent modelling suggests that co-condensation is the most important process explaining NO3incorporation in snow undergoing temperature gradient metamorphism at Dome C (Bock et al., 2016). Diamond dust can also scavenge high concentrations of 470 HNO3 at Dome C (Chan et al., 2018). Furthermore, the top layer of the snow pack is only 1 mm thick in the TRANSITS model, which is where we would expect the highest concentrations due to the exponential decay of NO3with depth (Fig. S4). If indeed the higher simulated values in the skin layer can be explained by hoar frost and diamond dust, then we can have greater confidence in the depth profile of NO3concentration. It is interesting to note that these higher simulated values in the skin layer do not impact the simulated depth profiles (Fig. 5). In summary, it is likely that we do not measure such high hoar frost 475 and diamond dust values in the skin layer because of sampling artefacts or blowing snow, which can dilute or remove the diamond dust and hoar frost.
While not yet observed elsewhere on the Antarctic continent, over the short intensive sampling period at DML we observe significant variability in NO3mass concentrations and δ 15 N-NO3values that resembles a diurnal cycle. Over 4 hours, the skin layer NO3mass concentrations varied by 46 ng g -1 , the skin layer δ 15 N-NO3by 21 ‰, and the atmospheric δ 15 N-NO3by 18 480 ‰. Other coastal studies have attributed daily variability to individual storm events (Mulvaney et al., 1998;Weller et al., 1999). We note that the sampling duration is too short to confirm any diurnal patterns but it would be interesting to investigate this further in future work.

Nitrate deposition
Here we discuss the various processes in which NO3can be deposited to the skin layer at DML. As we have just one month 485 of atmospheric and skin layer data, our ability to look at the deposition on seasonal scales is limited, however we provide new insights into the austral summer deposition processes.
While it is common to measure nitrogen species in snow and aerosol samples as the NO3ion using ion chromatography, nitrogen species can be deposited in various forms either by wet or dry deposition to the skin layer. We note that organic NO3plays are little role in determining snow concentrations (Jones et al., 2007;, and as such we focus our 490 discussion on inorganic NO3 -. The various nitrogen species include, i) a neutral salt (NO3co-deposition with sea salt or mineral dust; ), ii) NO3in air (HNO3 in gas-phase plus particulate NO3 -). Following the terminology of , this is referred to as "atmospheric NO3 -" and is represented by the NO3mass concentrations measured on our aerosol filters. Atmospheric NO3can either be deposited as dry deposition by adsorption to the snow surface as HNO3 has a strong affinity for ice surfaces (Abbatt, 2003;Huthwelker et al., 2006) or scavenged by precipitation as wet deposition, and iii) 495 co-condensation of HNO3 and water vapour onto snow crystals (Thibert and Domine, 1998).
Depending on the deposition pathway, NO3can either be predominantly incorporated into the bulk snow crystal or be adsorbed onto the surface of the snow crystal. Deposition pathways include co-condensation (formation of surface hoar frost), riming (deposition of supercooled fog droplets), and adsorption of HNO3 onto the snow crystal surface (dry deposition) (Röthlisberger et al., 2002). Both co-condensation (Bock et al., 2016) and dry deposition of HNO3, at very cold temperatures, can elevate 500 NO3mass concentrations in the skin layer. Furthermore, trace nitrogen impurities present in the interstitial air in the porous snow pack may be incorporated in snow crystals. While scavenging of NO3by snow (wet deposition) occurs sporadically throughout the year, dry deposition of particulate NO3or surface adsorption may take place continuously throughout the year.
We see both of these deposition processes taking place during January 2017.

Wet deposition 505
Precipitation at DML can occur either through sporadic cyclonic intrusions of marine air masses from the adjacent ocean associated with large amounts of precipitation, or clear sky diamond dust that contributes smaller amounts to the total precipitation (Schlosser et al., 2010). Overall, extreme precipitation events dominate the total precipitation (Turner et al., 2019). In austral summer, the transport of marine aerosol to DML is mediated by two synoptic situations, i) low-pressure systems from the eastern South Atlantic associated with high marine aerosol concentrations, and ii) persistent long-range 510 transport that provides background aerosol deposition during clear sky conditions (Weller et al., 2018). Weller et al. (2018) suggest that dry deposition of marine aerosol is dominant over wet deposition at DML. In contrast, Dome C receives predominantly diamond dust, and thus aerosol deposition is different there. More specifically, precipitation during our sampling campaign in January 2017 was relatively low compared to previous years.
Modelled daily precipitation at the nearest Regional Atmospheric Climate Model (RACMO2; Van Meijgaard et al. (2008)) 515 grid point (75.0014°S, 0.3278°W) is illustrated in Fig. 3a. The largest precipitation event of the month was on 1 January (0.27 mm) resulting from a low-pressure system in the South Atlantic (Fig. S5). For the rest of the month, half of the days had zero precipitation and the other half had very little precipitation (~0.05 mm per day).
We use the RACMO2 daily precipitation data to identify whether the cyclonic intrusions of marine air masses provide wet deposition of NO3to the site in January. In the skin layer, we observe that NO3mass concentrations and other sea salt ions 520 co-vary (Fig. S6) suggesting similar deposition pathways of these ions. Some peaks in the skin layer NO3concentration are accompanied by fresh snow laden with relatively high sea salt aerosol concentrations and atmospheric NO3mass concentrations, for example on 1, 13, and 18 January 2017. Such deposition events have also been observed on the Antarctic coast . During the formation of precipitation, essentially all HNO3 is removed from the gas-phase due to its high solubility in liquid clouds (Seinfeld and Pandis, 1998). Therefore, HNO3 can be scavenged from the atmosphere and 525 deposited as NO3in the skin layer. The uptake of HNO3 onto the snow and ice crystal surface during and after precipitation can also contribute further to the NO3mass concentrations found in the skin layer. On some precipitation days, we observe lower atmospheric NO3mass concentrations and higher skin layer NO3mass concentrations that could be a result of HNO3 scavenging. Mulvaney et al. (1998) observed higher skin layer concentrations in days when there was little snow accumulation and concluded that NO3is directly up taken onto the surface by dry deposition of particulate NO3and surface adsorption of 530 HNO3 (gas-phase) (Mulvaney et al., 1998). With only one month of data it is difficult to see the impact of wet deposition on the NO3concentration in the skin layer; i.e. whether fresh snowfall dilutes the NO3concentration in the skin layer or whether it scavenges HNO3 (gas-phase) resulting in higher concentrations of NO3in the skin layer. Most likely both processes are occurring. We note that due to post-depositional processes (section 3) any short-term signals observed in the skin layer are unlikely to be preserved. Even at the South Pole where the snow accumulation rate is slightly higher (8.5 cm yr -1 (w.e.); 535 (Mosley-Thompson et al., 1999) than DML deposition, NO3peaks are substantially modified after burial (Dibb and Whitlow, 1996).

Dry deposition
In order to investigate dry deposition of NO3 -, we first look at atmospheric NO3in relation to the wind direction and air mass back trajectories. The mean annual wind direction at the site is 65°, and January 2017 is no exception (Figs. 3 and S7). There 540 is an excursion from the predominant wind direction between 19-22 January, where the wind direction switches to the southwest. Although there are no studies indicating fractionation of δ 15 N-NO3in the atmosphere during atmospheric transport from the plateau to the coast, we do not see elevated NO3mass concentrations during this period nor do we see a marked difference in isotopic signature that is similar to Dome C at this time (Fig. 4). This, in line with air mass back trajectories (not shown) suggests that long-range transport of NO3re-emitted from inland sites of the Antarctic did not reach DML during our 545 campaign. We can also rule out any downwind contamination from the station. High concentrations of sea salt and mineral dust can promote the conversion of HNO3 (gas-phase) to aerosol, as well as trapping NO3 -(gas-phase) on salty snow surfaces. We see a relationship between sea salt aerosol and atmospheric NO3 -(R 2 = 0.59; p=<0.001) suggesting that even 550 km inland from the coast sea salt could promote the conversion of HNO3 to atmospheric NO3 -, although we acknowledge that our filters capture both aerosol NO3and HNO3, and sea salt concentrations are much 550 higher at Halley and coastal Antarctica where this mechanism sporadically occurs .
Scavenging of atmospheric NO3is largely responsible for the high mass concentrations observed in the skin layer. Variation in the concentration and isotopic signature of aerosol and surface snow at DML over January 2017 suggests atmospheric NO3is the source of NO3to the skin layer. Throughout the month, the increase in the skin layer concentration of summer NO3appears to be closely related to the decrease in the atmospheric NO3mass concentrations (Fig. 3). There is a lag between 555 atmospheric and skin layer NO3i.e. atmospheric NO3mass concentrations precede skin layer NO3mass concentrations by day or two, however a longer time series is required to confirm this. The lag suggests that atmospheric NO3is a source of NO3to the skin layer, in line with Dome C where the snow pack is the dominant source of NO3to the skin layer via the overlying air in summer. Furthermore, as atmospheric NO3is deposited to the snow surface, 15 N is preferentially removed first leaving the air isotopically depleted relative to the isotopically enriched snow ). Fig. 3 illustrates that the 560 δ 15 N-NO3in the atmosphere is depleted with respect to the δ 15 N-NO3in the skin layer snow. In the short time series, there are some periods where the δ 15 N-NO3in the snow and atmosphere are in phase, for example, 3-13 January 2017. During other periods, the δ 15 N-NO3in the snow and atmosphere switch to being out of phase emphasising NO3isotopic fractionation during those periods. Both HNO3 and peroxynitric acid (HNO4) can be adsorbed to the snow surface in tandem (Jones et al., 2014), and although we have no direct measurements of these during the campaign, based on previous studies we suggest that HNO3 565 is the most likely form of nitrogen to skin layer (Jones et al., 2007;Chan et al., 2018).
Furthermore, the adsorption of HNO3 on ice surfaces is temperature dependent with higher uptake at lower temperatures (Abbatt, 1997;Jones et al., 2014). However, there is only a relatively small temperature difference between Dome C and DML (summer mean temperature -30 °C and -25 °C respectively) which is not enough to drive a large difference in HNO3 uptake (Jones et al., 2014). In addition, the uptake is not dependent on the HNO3 concentration in the air (Abbatt, 1997). However, 570 the seasonal temperature difference at an individual site (i.e., DML or Dome C) is far greater, which could allow a seasonal dependence on the uptake and loss of NO3in the skin layer, which results in the retention of a greater proportion of NO3in summer (Chan et al., 2018).

Annual cycle of nitrate deposition
We use the simulated annual cycle of NO3from TRANSITS model to describe the seasonal evolution of NO3deposition to 575 DML. While NO3deposited to DML can be sourced from the sedimentation of polar stratospheric clouds in winter and we assume the atmospheric NO3loading is uniform under the polar vortex, in spring and summer NO3net deposition is related to local photochemistry and subsequent post-depositional processing rather than primary NO3sources. of NO3can be through the transport of re-emitted NO3from the surface snow at low accumulation regions of the polar plateau, or NO3produced in situ at DML in spring and summer. 580 The annual cycle of atmospheric NO3deposition (Weller and Wagenbach, 2007) indicates how much NO3is deposited to the skin layer from the atmosphere (Figs. 5 and 6). Year-round NO3mass concentrations have been measured in surface snow at the coastal sites of Halley (Mulvaney et al., 1998;Jones et al., 2011) andNeumayer Stations (Wagenbach et al., 1998), and the low snow accumulation site at Dome C (Fig. 1). An agreement with our simulated results, at all Antarctic sites the highest atmospheric NO3mass concentrations are found during summer when the solar radiation is close to its annual maximum and 585 NO3photolysis is strongest. The summer maximum at Dome C results from co-condensation of NO3 - (Bock et al., 2016). This intense uptake in the skin layer in summer is driven by the strong temperature gradient in the upper few centimetres of the snow pack, highlighting that both physical (deposition; Bock et al. (2016); Chan et al., 2018) and chemical (NO3re-emission; Erbland et al. (2015)) processes explain the cycling of NO3between the air and snow. The lowest NO3mass concentrations in the skin layer are found in winter. 590 Year-round atmospheric NO3data at DML and Dome C shows atmospheric NO3is at a minimum in April to June and reaches a maximum in late November, slightly out of phase with skin layer NO3 - Erbland et al., 2013) (Figs.   1 and 6). The fact that the seasonality of simulated skin layer and atmospheric NO3at DML matches observations at other sites in Antarctica gives confidence in our TRANSITS model results (Fig. 6).

Nitrate mass fluxes 595
Our two NO3mass flux scenarios in Fig. 5 highlight the importance of the skin layer in the air-snow transfer of NO3 -. Like Dome C, the greatest deposition flux of NO3is to the skin layer. The January dry deposition flux is greater than the annual mean flux estimated by Pasteris et al. (2014) and Weller and Wagenbach (2007) which is to be expected given the higher atmospheric NO3mass concentrations in summer (Fig. 6). The wet deposition flux, calculated for the greater DML region by Pasteris et al. (2014), falls within our two scenarios. Furthermore, the simulated archived NO3flux at DML of 210 pg m -2 s -1 600 over predicts the observed NO3archived flux of 110 pg m -2 s -1 due to the higher simulated archived NO3mass concentrations.
Interestingly, the simulated archived flux at Dome C (88 pg m -2 s -1 ) is lower than DML, yet the NO3deposition flux to the skin layer in January at Dome C is similar to DML. We continue our discussion focusing on the recycling and redistribution of NO3that occurs in the active skin layer emphasising its importance.

Recycling and redistribution 605
Post-depositional loss and redistribution of NO3is the dominant control on snow pack mass concentrations and δ 15 N-NO3isotopic signature on the Antarctic Plateau. Recycling of NO3at the air-snow interface comprises the following processes.
Nitrate on the surface of a snow crystal can be lost from the snow pack (Dubowski et al., 2001), either by UV photolysis or evaporation. UV-photolysis produces NO, NO2 and HONO while only HNO3 can evaporate. Both of these processes produce reactive nitrogen that can be released from snow crystal into the interstitial air and rapidly transported out of the snow pack to the overlaying air via wind pumping (Zatko et al., 2013;Jones et al., 2000;Honrath et al., 1999;Jones et al., 2001). Here, NO2 is either oxidised to HNO3, which undergoes wet or dry deposition back to skin layer within a day, or transported away from the site (Davis et al., 2004a). If HNO3 is re-deposited on the snow skin layer, it is available for NO3photolysis and/or evaporation again. Nitrate can be recycled multiple times between the boundary layer and the skin layer before it is buried in deeper layers of the snow pack. Photolysis and/or evaporation of NO3and subsequent recycling between the air and snow 615 alters the concentration and δ 15 N-NO3that is ultimately preserved in ice cores. Nitrate recycling therefore redistributes NO3from the active snow pack column to the skin layer via the atmosphere. Any locally produced NO2 that is transported away from the site of emission represents a loss of NO3from the snow pack.

Evaporation
The desorption of HNO3 from the snow crystal reduces the NO3concentration in the snow in coastal Antarctica (Mulvaney et 620 al., 1998). The evaporation of HNO3 is a two-step process, which involves the recombination of NO3 -+ H + -> HNO3 followed by a phase change to HNO3 (gas-phase). First, theoretical estimates indicated that evaporation of HNO3 should preferentially remove 15 N from the snow and release to the atmosphere leading to depletion in δ 15 N-NO3in the residual snow pack . Furthermore, recent laboratory experiments showed that evaporation imposes a negligible fractionation of δ 15 N-NO3 - Shi et al., 2019). However, we find that the snow pack is enriched in δ 15 N-NO3relative to the 625 atmosphere at DML (Figs. 3 and 6) and at Dome C (section 4.3.2). This fractionation observed in field studies cannot therefore be explained by evaporation, and must be attributed to different processes. It therefore follows that evaporation must be only a minor process in the redistribution of NO3between atmosphere and the snow pack above the Antarctic plateau.

Photolysis
We focus our discussion on photolysis, which is the dominant process responsible for NO3loss and redistribution and 630 associated δ 15 N-NO3isotopic fractionation at low accumulation sites in Antarctica France et al., 2011).
Nitrate photolysis occurs in the photochemically active zone of the snow pack, known as the snow photic zone. Below this, NO3is buried. Nitrate photolysis in the active snow pack results in the production of NO2 leading to a reduction in the NO3concentration with depth in the snow pack (Fig. 4). In the photolysis-induced fractionation of NO3 -, 14 N is preferentially removed first resulting in an enrichment of δ 15 N-NO3in the snow pack. An individual snow layer is enriched when it is near 635 the surface during sunlit conditions, i.e. spring and summer. Therefore, spring snow layers undergo strong δ 15 N-NO3enrichment as they are exposed to UV near the surface for the longest; late summer and autumn layers experience less δ 15 N-NO3enrichment as they are exposed for less time before sunlight disappears at the start of polar winter, during which new precipitation buries existing snowfall.
We provide five lines of evidence that photolysis is the dominant process for NO3recycling and redistribution at DML. Firstly, 640 the highly enriched δ 15 N-NO3values of snow at DML and other Antarctic sites are among the most extreme observed on earth ( Fig. S8)  , and cannot be explained by any known anthropogenic, marine or other natural sources. The δ 15 N-NO3source signature of the main natural NOx sources (biomass burning, lightning, soil emissions is lower; δ 15 N-NO3 -<0 ‰) than anthropogenic NOx sources, which generally have positive δ 15 N-NO3values (-13< δ 15 N-NO3 -< 13 ‰) (Hastings et al., 2013;Kendall et al., 2007 and references therein). The δ 15 N-NO3observations of aerosol, skin layer and snow pit at 645 DML (-49< δ 15 N-NO3 -<99 ‰) lie outside of the range of natural and anthropogenic source end members, and thus cannot be explained by mixing of sources. Thus, our measurements at DML are unrelated to seasonal variations in NOx sources e.g.
increased springtime agricultural emissions, which has been observed in the mid-latitudes. The contribution of natural sources to the Greenland snow pack δ 15 N-NO3signature has also been discarded (Geng et al., 2014;Geng et al., 2015). Furthermore, the negative atmospheric δ 15 N-NO3values at DML (-20 to -49 ‰) are extremely low. Such low atmospheric δ 15 N-NO3values 650 have only been observed in Antarctica, and show marked difference to other mid-latitude tropospheric aerosol (-10< δ 15 N-NO3 -<10 ‰; Freyer (1991). We acknowledge that stratospheric NO3contributes to NO3mass concentrations in snow in Antarctica. Although its isotopic signature is uncertain, estimates of stratospheric δ 15 N-NO3are 19 ± 3 ‰ , and fall well outside of atmospheric observations at DML. The unique δ 15 N-NO3signature of low accumulation Antarctic snow and aerosol is thus related to post-depositional processes specific to low accumulation sites in Antarctica. 655 Secondly, denitrification of the snow pack is seen through the δ 15 N-NO3signature which evolves from the enriched snow pack (-3 to 99 ‰), to the skin layer (-22 to 3 ‰), to the depleted atmosphere (-49 to -20 ‰) corresponding to mass loss from the snow pack (Fig. 4). Denitrification causes the δ 15 N-NO3of the residual snow pack NO3to increase exponentially as NO3mass concentrations decrease. Additionally, although not the focus of the study, denitrification causes the δ 18 O-NO3values to increase in the residual NO3snow pack. 660 Thirdly, the application of TRANSITS to DML observations show that our observed atmospheric, skin layer and snow depth profiles of δ 15 N-NO3are similar to the simulated values where photolysis is the driving process (Figs. 6-7). Sensitivity analysis with TRANSITS is able to explain the observed snow pit δ 15 N-NO3variability (section 4.5). Nitrate isotope enrichment takes place in the top 25 cm, which is consistent with an e-folding depth of 10 cm. Here, the δ 15 N-NO3observations closely match the simulated δ 15 N-NO3values and show enrichment to this depth indicating NO3photolytic redistribution at DML in the 665 active photic zone of the snow pack (Fig. 7). Below the photic zone, δ 15 N-NO3values oscillate around a mean of ~125 ‰.
The mean values of the δ 15 N-NO3observations are lower than the simulated values, which could be related to uncertainties in a number of factors, for example: i) a shallower e-folding depth than modelled. During our field measurements, we derived a lower e-folding depth of 2-5 cm (Fig. S1) at DML which could explain the lower enrichment in δ 15 N-NO3 -(section 4.5.2), ii) lower JNO3values (NO3photolysis rate), which are related to a lower e-folding depth, and would lead to less enrichment of 670 δ 15 N-NO3in the snow pack, iii) higher atmospheric NO3input, however δ 15 N-NO3values are not sensitive to variable atmospheric NO3mass concentrations , and/or iv) variable accumulation which would shift the oscillations to the correct depth and lower the mean δ 15 N-NO3values below the photic zone (section 4.5.1). The difference between the simulated and snow pit values shows that DML site is less sensitive to photolysis than we expected from TRANSITS modelling of δ 15 N-NO3along an accumulation gradient .
Fourthly, we use Rayleigh isotopic fractionation to calculate apparent fractionation constants ( 15 εapp) associated with NO3fractionation between phases during evaporation-condensation processes. Nitrate evaporation from the snow pack has a 15 εapp of ~0 as determined by two independent studies Shi et al., 2019). This indicates that during NO3evaporation, the air above the snow is not replenished and thus there is only a small NO3mass loss. The isotopic fractionation of NO3evaporation is negligible across most of Antarctica at cold temperatures of <-24 °C (Shi et al., 2019) which is the case 680 for DML. However, evaporation of NO3at warmer temperatures (-4 °C) depletes the heavy isotopes of NO3remaining in the snow, and decreases the δ 15 N-NO3and the remaining snow by a few ‰ contrary to isotope effects of photolysis. In comparison, fractionation constants associated with laboratory studies and field observations of NO3photolysis are large: 15 εapp = -34 ‰ Meusinger et al., 2014) and -54 < 15 εapp < -60 ‰ Erbland et al., 2013), respectively.
The negative fractionation constant obtained from photolysis implies that the remaining NO3in the skin layer snow is enriched 685 in δ 15 N-NO3 -. In turn, the atmosphere is left with the source of NOx that is highly depleted in δ 15 N-NO3 -. This enrichment (depletion) is exactly what we observe in the snow pack (atmosphere) at DML (Figs. 4 and 6). The marked difference in values from the evaporation experiments and those observed in snow at Dome C allows us to separate out the isotopic signature of evaporation and photolysis processes.
Assuming a Rayleigh-type single loss process, we calculate a 15 εapp at DML of 12 ‰ using the active photic zone section of 690 the snow pack (top 30 cm). We also calculate a 15 εapp of -19 ‰ using our simulated results from the TRANSITS model. This simulated 15 εapp nicely matches the expected 15 εapp values (-59< 15 εapp< -16 ‰) within the "transition zone" of 5-20 cm yr -1 (w.e.) modelled by Erbland et al. (2015). As the two loss processes of evaporation and photolysis have different isotopic fractionation signatures, an 15 εapp of -19 ‰ cannot be explained by evaporation ( 15 εapp of 0) but rather photolysis albeit implying a weaker photolytic loss of NO3than Dome C ( 15 εapp < -59 ‰) . The discrepancy between our observed 695 and simulated 15 εapp is due to the larger snow accumulation rate, which preserves seasonality, and with a noisy signal, there is no pure separation of the loss processes assuming Rayleigh isotopic fractionation. The single-process Raleigh model does not work well at sites with annual signal in δ 15 N-NO3 -.
Lastly, we estimate the potential snow emission flux of NO2 (FNO2) from NO3photolysis in snow using Eq. (8).
where Jz(NO3 -) is the photolysis rate coefficient of reaction NO3 -+hν → NO2 + O − at depth, z, in the snowpack, and is derived by scaling surface measurements (section 2.6) with e-folding depth (= 2-10 cm), and [NO3 -]z is the amount of NO3per unit volume of snow at depth, z, in the snowpack. The calculated FNO2 is a potential emission flux assuming that all NO3within the snow grain is photo-available, no cage effects are present and NO2 is vented immediately after release from the snow grain to the air above the snowpack without undergoing any secondary reactions. based on the same model during 1 to 14 January 2012, were larger with 1.2-7.3 x 10 11 molecule m -2 s -1 , mostly due to larger J(NO3 -) values observed above the surface (section 2.6) as well as a larger e-folding depth (= 10 cm near 710 the surface). It should be noted that the observed FNOx was found to be up to 50 times larger than model estimates, which is attributed to the poorly constrained quantum yield of NO3photolysis in natural snow Frey et al., 2013). In summary, the weakened air-snow recycling at DML is due to i) the shallower e-folding depth compared to Dome C which implies reduced emission flux of NOx, and ii) the reduced UV exposure time of surface snow due to higher annual accumulation compared to Dome C. We estimate that NO3has a mean lifetime in the skin layer of 12 days to 3 years before it is photolysed 715 back to atmosphere.

Evidence for weaker recycling at DML
Only two studies have attempted to quantify the degree of NO3recycling between the air and snow (Davis et al., 2008;Erbland et al., 2015). Erbland et al. (2015) use the TRANSITS model to estimate that NO3is recycled 4 times on average before burial beneath the photic zone at Dome C, similar to the findings of Davis et al. (2008) for the same site. Using the approach of 720 Erbland et al. (2015), we find that NO3is recycled 3 times before it is archived at DML. A lower recycling factor than Dome C is consistent with spatial patterns of NO3recycling factors across Antarctica reported by Zatko et al. (2016). As Dome C and DML lie on the same latitude (75° S), incoming UV-radiation (except for cloud cover) should not impact the efficiency of photolysis and thus recycling at the two sites. While photolysis-driven NO3recycling can occur at all polar sites, the most intense enrichment of δ 15 N-NO3in the depth profile is seen at Dome C and Vostok (Erbland, 2011). Below we provide some 725 explanations for the weakened recycling at DML.

i.
Higher snow accumulation rate The TRANSITS modelling shows the influence of the snow accumulation rate on the depth profile of NO3concentration and δ 15 N-NO3 -, including the preservation of a seasonal cycle at higher snow accumulation rates (Fig. 7). At low accumulation sites, i.e. Dome C, NO3in the skin layer and thinner snow layers is exposed to sunlight (and the actinic flux) for longer 730 allowing more photochemistry and thus a very active snow pack with strong NO3recycling and δ 15 N-NO3enrichment. At DML, which has a higher snow accumulation rate, the skin layer is buried more rapidly, leaving less time to adsorb additional HNO3 from the atmosphere and less time for photolysis to redistribute snow pack NO3to the overlying air for re-adsorption to the skin layer. Following photolysis at DML, the recycling of NO3at the air snow interface alters the depth profile of δ 15 N-NO3in the top skin layer but below the skin layer δ 15 N-NO3in snow remains intact as there is less redistribution and a lower 735 loss of NO3than at Dome C. ii.
Shallower e-folding depth Based on measurements we derived an e-folding depth for DML ranging between 2 and 5 cm (Fig. S1). This estimate is similar to a modelled value at South Pole (3.7 cm; Wolff et al. (2002) which has a similar accumulation rate, and Alert, Canada (5-6 cm; King and Simpson, 2001). The e-folding depth at Dome C is considerably deeper, ranging between 10 cm to 20 cm 740 depending on the snow properties (France et al., 2011). The e-folding depth depends on the density and grain size of snow https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License. crystals, and the concentration of impurities. The larger e-folding depth at Dome C is due to the larger grain sizes and low impurity content. The impact of impurities in the range of observed polar snow concentrations on e-folding depth is small compared to the contribution from scattering by snow grains (France et al., 2011;Zatko et al., 2013). At Dome C, a larger efolding depth corresponds to a greater depth over which photochemistry occurs and thus stronger recycling and redistribution 745 of NO3 -. At DML, the lower e-folding depth of 2 to 5 cm lowers the mean δ 15 N-NO3value. iii.
Lower nitrate uptake at warmer temperatures Temperature can control skin layer NO3uptake and loss. At colder snow temperatures, there is greater adsorption of HNO3 to the skin layer (Abbatt, 1997;Jones et al., 2014). Although the difference in the mean annual temperature at Dome C compared to DML (~5 °C) is not large enough to explain the significantly higher skin layer NO3mass concentrations there. Compounding 750 this, NO3loss by evaporation is also dependent on temperature with maximum NO3loss at higher temperatures, i.e., diffusion of NO3in ice is slower at colder temperatures (Thibert and Domine, 1998). A compilation of NO3concentration data from Greenland and Antarctic ice cores showed that at very low accumulation rates lower temperatures lead to higher NO3mass concentrations preserved in the snow . Although the snow accumulation rate is closely linked to temperature, photolysis is the dominant NO3loss process at low snow accumulation sites in Antarctica. Therefore, any 755 differences in temperature between DML and Dome C could partly explain the greater uptake of HNO3 to the skin layer, higher mass concentrations of NO3in the skin layer, and stronger recycling at Dome C compared to DML.

iv.
Lower photolysis rate At DML, NO3photolysis produces a lower NOx flux to the atmosphere and lower 15 εapp highlighting that the photolysis rate is lower thus the recycling strength is reduced (section 4.3.2). Furthermore, the large 15 εapp associated with NO3photolysis has 760 been determined for snow at Dome C Frey et al., 2009;Erbland et al., 2013) and DML. At both sites, δ 15 N-NO3is enriched in the remaining skin layer snow. However, at DML, the 15 εapp is lower due to less active photochemistry associated with a higher snow accumulation rate. Our results are consistent with Zatko et al. (2016) who suggest that the large fractionation constant associated with photolysis is greatest on the polar plateau where strong winds are most efficient at exporting NO3away from the site. 765 v.
Lower export of locally produced nitrate Export of locally produced NOx on the Antarctic Plateau leads to greater enrichment in the depth profiles of δ 15 N-NO3relative to the coast Zatko et al., 2016). Zatko et al. (2016) modelled the export of snow sourced NOx away from the original site of NO3photolysis, and found that the largest loss of NO3occurs in central Antarctica where most NO3is transported away by katabatic winds. At the coast, photolysis driven loss of NO3from the snow is minimal due to high snow 770 accumulation rates. Here, observations of enriched atmospheric δ 15 N-NO3show that NOx has been transported away from the location of its production on the Antarctic Plateau to the coast Morin et al., 2009). The greater export of NO3from Dome C allows efficient removal of recycled NO3from that site, resulting in a lower archived NO3mass flux and enriched δ 15 N-NO3signature in the surface snow. The enrichment of δ 15 N-NO3is due to the isotopic mass balance rather than an increase for photolysis intensity. With less export of NO3away from the DML site, locally sourced NOx is redeposited back 775 to the skin layer at the site of the emission and the depth profile of the δ 15 N-NO3is not as dramatically impacted as Dome C where there is substantial loss of NO3 -. Therefore, the degree of NO3recycling is also determined by the transport patterns across Antarctica.
Based on field, laboratory and theory, we conclude that NO3photolysis is the dominant post-depositional process on the Antarctic plateau controlling NO3mass concentrations and δ 15 N-NO3values in the snow and atmosphere. Nitrate photolysis 780 in snow causes δ 15 N-NO3fractionation of the magnitude needed to explain field and lab observations. The development of TRANSITS allows us to model the archived δ 15 N-NO3values taking into account all parameters in the air-snow system.

Preservation and archival
The photolysis-driven recycling of NO3is largely dependent on the time that NO3remains in the snow photic zone. Postdepositional loss of NO3at DML was quantified in a number of firn cores and snow pits by Weller et al. (2004) who found 785 that ~26 % of the NO3originally deposited to the snow pack was lost. The e-folding time for NO3at the site was reported as ~20 years, and NO3was archived after 5 to 6 years of deposition (or 1.1 -1.4 m depth) which is the time it takes for the NO3mean concentration to become representative of the last 100 years. At this point, the authors considered post-depositional loss of NO3to be negligible, and therefore archived. However, no skin layer measurements were made in the study and given how active the skin layer is NO3redistribution and recycling, we use our skin layer measurements to provide new constraints on 790 the archival values and time of NO3at DML.
Taking the high skin layer NO3mass concentrations into account (average of 230 ng g -1 in January for DML), we calculate a NO3loss of 60 ng g -1 (or 75 %) and enrichment of 170 ‰ from the snow pack. Assuming all NO3is archived below the photic zone, i.e., an e-folding depth of 5 cm, archival occurs below a depth of 15 cm, where NO3has a residence time of 0.75 years in the photic zone corresponding to one summer. At this point, the amplitude of the annual cycle of δ 15 N-NO3at DML does 795 not vary. Our archived values of 50 ‰ and 60 ng g -1 agree well with the mean values of the snow pit below the photic zone (30 cm), and are lower than the simulated archived values from TRANSITS (120 ng g -1 and 130 ‰) due to the stronger photochemistry in the model. Due to the larger e-folding depth and hence larger photic zone at Dome C, NO3has a longer residence time of 3 years (3 summers) in the photic zone. Here, archival of NO3occurs below a depth of 30 cm. Compared to Dome C, the archived values at DML have a similar concentration (Dome C: 35 ng g -1 ) but lower δ 15 N-NO3value (Dome C: 800 300 ‰), due to the thicker photic zone, stronger redistribution and recycling there.

Sensitivity of δ 15 N-NO3to deposition parameters and implications for interpreting ice core records of δ 15 N-NO3at DML
As first proposed by Frey et al. (2009) and later confirmed by field and lab studies Shi et al., 2019) it is UV-photolysis of NO3that dominates post-depositional fractionation of δ 15 N-NO3in snow and firn. Yet the 805 extent of photolytic fractionation and the δ 15 N-NO3ultimately preserved in firn and ice depends on the UV-spectrum of downwelling irradiance, on the time snow layers are exposed to incoming UV-radiation as well as on the snow optical properties. Previous studies showed that δ 15 N-NO3is sensitive to TCO but also to deposition parameters such as the annual accumulation rate (Shi et al., 2018;Noro et al., 2018;Erbland et al., 2013). Thus, if all deposition parameters remained constant or are wellconstrained it should be theoretically possible to use δ 15 N-NO3as an ice core proxy for past surface UV-radiation and 810 stratospheric ozone. Understanding the depositional parameters and their impact on δ 15 N-NO3is paramount for the interpretation of δ 15 N-NO3signals preserved in ice cores. As the interpretation of δ 15 N-NO3is site-specific, we investigate the sensitivity of the δ 15 N-NO3signature at DML to snow accumulation rate, e-folding depth and TCO. As the mean annual snow accumulation rate at DML is 6 cm (w.e.) yr -1 , we take this simulation as our base case.

Sensitivity of the ice core δ 15 N-NO3signal to accumulation rate 815
The δ 15 N-NO3signal is indeed sensitive to the snow accumulation rate at DML. Here, the accumulation rate varied between 2.5 and 11 cm yr -1 (w.e.) over the last 1000 years . Figs. 7a-b shows the potential impact of this variability in the snow accumulation rate on the NO3concentration and δ 15 N-NO3signature at DML calculated with the TRANSITS model. Considering that the actual snow accumulation rate varied between 3.5 and 7.1 cm yr -1 (w.e.) in our snow pit, our δ 15 N-NO3measurements fall within the simulated δ 15 N-NO3depth profile for the accumulation rates over the past 1000 years. 820 Although the mean snow pit δ 15 N-NO3is ~50 ‰ lower, the snow pit depth profile parallels the base case profile for the top 30 cm. Here, there is a clear enrichment of δ 15 N-NO3in both the snow pit and base case profiles corresponding to the depth of the photic zone (30 cm), and demonstrating that NO3photolysis is taking place in this section of the snow pack. Below the photic zone, the seasonal variability of the base case δ 15 N-NO3depth profile is constant between 100-153 ‰ indicating that no further enrichment or NO3redistribution is taking place in the archived section of the snow pack. 825 Despite the relatively high NO3mass concentrations and enriched δ 15 N-NO3in the skin layer at DML, clear seasonal cycles remain in the depth profile in contrast to the lower snow accumulation site of Dome C where the depth profile is relatively constant below the photic zone. Figs. 7a-b indicate that at higher snow accumulation rates, the seasonality of atmospheric NO3and δ 15 N-NO3is preserved due to faster burial. Even at 6 cm yr -1 (w.e.), the snow layers remain in the active photic zone for 0.75 years and the weaker recycling factor is low enough to conserve the seasonality. Whereas at Dome C, snow layers remain 830 within the photic zone for longer (about 3 years), and NO3loss and redistribution continues until the seasonal cycle becomes smoothed (Figs. 7a-b). Thus, NO3recycling is strongest in the lowermost snow accumulation regions.
Below the active photic zone, there is an offset between the base case and snow pit δ 15 N-NO3depth profile in terms of i) the amplitude of the summer and winter δ 15 N-NO3values, and ii) the mean δ 15 N-NO3value (Fig. 7). To account for this offset we investigated how the timing of snow deposition altered the δ 15 N-NO3depth profile. Rather than assuming a constant 835 accumulation rate of 6 cm yr -1 (w.e.), as in the base case, we find that a variable snow accumulation rate, based on our observations from the snow pit, alters the depth of the summer and winter δ 15 N-NO3peaks (Fig. 7b.). Using the actual annual accumulation rates improves the model fit (~10 cm depth; Fig 7a). Furthermore, the timing of the snow accumulation throughout the year has a significant control on the amplitude of the seasonal δ 15 N-NO3cycle. Snowfall at DML has a bimodal distribution with higher accumulation in austral autumn and early austral summer (Fig. S9). In Fig. 7c,  of the snow accumulation during the year by depositing 90 % of the annual snowfall in i) the first week of winter, and ii) the first week of summer, which represents the upper bound for snow accumulation in winter and summer respectively. The remaining 10 % of the annual snowfall is distributed evenly across the rest of the weeks of the year. Summer snow accumulation results in a higher δ 15 N-NO3enrichment compared to winter snow accumulation, as the exposure of summer layers to UV is longer and thus NO3photolysis is stronger. Therefore, the timing and rate of snowfall can explain the 845 misalignment between snow pit observations and base case simulation, which shifts the depth and amplitude of the δ 15 N-NO3peaks in the depth profile.
On centennial to millennial timescales, the snow accumulation rate has varied in regions of Antarctica (Thomas et al., 2017), which could potentially modify the degree of post-depositional processing and thus impact the archival and temporal variability of δ 15 N-NO3in ice cores. Interestingly, Geng et al. (2015) found that post-depositional loss of NO3in Greenland could fully 850 account for the large difference between the glacial and Holocene δ 15 N-NO3signature. At DML, higher snow accumulation rates would result in lower NO3mass concentrations and more depleted δ 15 N-NO3values in the skin layer, thus reducing the recycling strength and lowering the sensitivity of the UV proxy recorded in the ice over time, and vice versa. TRANSITS modelling predicts that the upper and lower bounds of δ 15 N-NO3values in a 1000-year ice core from DML that has an accumulation rate between 2.5 and 11 cm yr -1 (w.e.) to be 70 -360 ‰. Furthermore, δ 15 N-NO3values could range between 855 90-110 ‰ depending the timing of snowfall and extreme precipitation events, which are known to play a dominant role in snowfall variability across Antarctica (Turner et al., 2019). At DML, snow pit observations suggest that the variation of δ 15 N-NO3between the polar day and polar night is 20 ‰. This seasonality is within the range of values expected for changes in snow accumulation rates over time (Fig. 7). Therefore, any seasonal variation in ice core δ 15 N-NO3will need to be accounted for in order to observe decadal, centennial and millennial scale trends in δ 15 N-NO3 -. 860

Sensitivity of the ice core δ 15 N-NO3signal to e-folding depth
We measured an e-folding depth at DML (2-5 cm) which is lower than that employed in the TRANSITS model (10 cm).
Furthermore, a range of e-folding depth values, between 3.7 and 20 cm, have been reported for Antarctica. The positive bias of the TRANSITS simulation in archived δ 15 N-NO3at DML may be due to e-folding depth being smaller than at Dome C as indicated by direct observations. In order to test this assumption, the sensitivity of archived δ 15 N-NO3to the parameter e-865 folding depth needs to be quantified, which has not been done before as far as we know. Zatko et al. (2016) modelled the efolding depth over Antarctica and investigated the impact of snow-sourced NOx fluxes but not on δ 15 N-NO3 -. The e-folding depth has a large influence on the δ 15 N-NO3depth profile in terms of i) depth of the photic zone and thus depth of the δ 15 N-NO3enrichment, and ii) the mean archived δ 15 N-NO3value below the photic zone (Fig. 7d). A larger e-folding depth strengthens the δ 15 N-NO3enrichment in the photic zone and archived mean δ 15 N-NO3value. For example, an e-folding depth 870 of 10 cm at DML gives δ 15 N-NO3enrichment down to 25 cm and an archived mean δ 15 N-NO3value of 125 ‰ in the snow pack compared to an e-folding depth of 20 cm, which enriches the snow pack to 45 cm and more than doubles the archived mean δ 15 N-NO3value to 320 ‰. Meanwhile, an e-folding depth of 2 cm gives minimal enrichment and a low archived mean https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License. δ 15 N-NO3value of 25 ‰. In comparison to the base case simulation, which has an e-folding depth of 10 cm, a lower e-folding depth of 5 cm decreases the archived mean δ 15 N-NO3in the snow pack to ~50 ‰, closely matching our snow pit observations. 875 Hence, a shallower e-folding depth of 5 cm can explain the more depleted δ 15 N-NO3snow pit profile, relative to the base case simulation, as NO3photolysis occurs in a shallower depth. Therefore, e-folding depth knowledge is required to understand the sensitivity of archived δ 15 N-NO3at specific sites. A lower e-folding depth and variable snowfall throughout the year can explain the misalignment between the snow pit observations and simulated δ 15 N-NO3depth profiles. 4.5.3 Sensitivity of ice core δ 15 N-NO3signal to TCO 880 Fig. 8 shows the strong sensitivity of δ 15 N-NO3to variations in decreasing TCO. A decrease in TCO will increase UV radiation reaching the surface at an ice core site. As a result, stronger photolysis enhances NO3loss, redistribution and recycling from the snow pack and ultimately decreases the archived NO3concentration. Furthermore, a decrease in TCO enriches the δ 15 N-NO3signature as the snow is exposed to a greater UV dose. We expect that a change of 100 Dobson Units (DU), i.e. the amount that ozone now decreases each spring as a result of stratospheric ozone destruction processes, will result in a 22 ‰ 885 change in δ 15 N-NO3at DML. The variability in δ 15 N-NO3induced by TCO is similar to the seasonal variability of δ 15 N-NO3recorded in the snow pit (20 ‰) and less than the predicted variability of δ 15 N-NO3due to variability in snow accumulation (340 ‰), thus the development of a large ozone hole is unlikely to be observed above the natural background δ 15 N-NO3variability in the ice core at this site. The sensitivity of δ 15 N-NO3to TCO is greater at Dome C than DML.

Implications for interpreting ice core δ 15 N-NO3 -890
Site-specific air-snow transfer studies provide an understanding of the mechanisms that archive δ 15 N-NO3in ice cores, thus allowing for the interpretation of longer records of δ 15 N-NO3from the site. Ice core records of archived NO3mass concentrations and δ 15 N-NO3at DML are a result of three uptake and loss cycles that occur in the top 30 cm during sunlit conditions. While we do not observe further redistribution of NO3in layers deeper than the photic zone, we cannot rule out any further NO3diffusion within the firn or ice sections of an ice core. This redistribution unlikely results in a loss of NO3 -895 but could migrate NO3to different layers, for example in acidic layers around volcanic horizons (Wolff, 1995).
There are a number of factors that will control the variability of the archived δ 15 N-NO3signature in ice cores recovered from DML. The δ 15 N-NO3signature in the snow pack is most sensitive to changes in the snow accumulation rate and e-folding depth, with snowfall timing and TCO also playing a smaller role. The snow accumulation rate and e-folding depth could influence the archived δ 15 N-NO3composition by up to 300 % over the last 1000-years. This magnitude is comparable to 900 modelled enrichment in ice-core δ 15 N-NO3 -(0 to 363 ‰) due photolysis-driven loss of NO3at low accumulation sites in Antarctica by Zatko et al. (2016). While the timing of snowfall and changes in TCO will have a smaller impact of 20 ‰ on archived δ 15 N-NO3 -. Ice core δ 15 N-NO3records at DML will be less sensitive to changes in UV than those at Dome C (Fig.   8), however the higher snow accumulation rate and more accurate dating at DML allows for higher resolution δ 15 N-NO3records. We acknowledge that in addition, other factors such as light absorbing impurities (Geng et al., 2015), local 905 https://doi.org/10.5194/acp-2019-669 Preprint. Discussion started: 10 September 2019 c Author(s) 2019. CC BY 4.0 License. meteorology, source of emissions and transport of NOx and NO3 -, atmospheric oxidant concentrations, and polar NO3formation can influence the rate of recycling and export of snow sourced NOx. We discussed above that atmospheric δ 15 N-NO3values are unlikely to be influenced or sourced from snow exported up wind from the polar plateau due to the local meteorology at DML at least for the duration of the campaign. Yet these factors may have changed over time.
Given a variable accumulation rate and smaller e-folding depth, which we provide evidence for at DML, the TRANSITS model 910 is able to reproduce our snow pit observations, justifying our previous assumption that photolysis is the main driver of NO3post-depositional processes at DML. In fact, TRANSITS does such a good job at simulating NO3recycling in Antarctica that we recommend that this tool is employed before the commencement of future ice core δ 15 N-NO3studies to understand the sensitivity of the signal to various factors. Taking changes snow accumulation into account, it may be possible to reconstruct past UV and TCO from the δ 15 N-NO3signal in DML ice cores provided other factors such as the e-folding depth have remained 915 the same.

Conclusions
Our key findings are: -Isotopes are a powerful tool for unpicking post-depositional processes affecting ice core signals of NO3at low accumulation sites; 920 -At DML, post-depositional loss of NO3is controlled predominantly by photolytic loss; -Photolysis redistributes NO3between the snow pack and atmosphere resulting an enrichment of δ 15 N-NO3in the skin layer; -TRANSITS, a photolysis driven model, modelling suggests that NO3is recycled three times before it is archived in the snow pack below 15 cm and within 0.75 years; 925 -Once archived, the seasonal variability of δ 15 N-NO3values and NO3mass concentrations oscillate between -1 to 80 ‰ and 30 to 80 ng g -1 , respectively; -TRANSITS can explain the observed snow depth profiles of δ 15 N-NO3constrained by an e-folding depth of 5 cm, the observed snow accumulation rate, and variable snowfall timing.
-TRANSITS sensitivity analysis showed that the δ 15 N-NO3signature in the snow pack is most sensitive to changes in 930 the snow accumulation rate (up to 300 %) and e-folding depth (up to 300 %), with snowfall timing (~20 %) and total column ozone (~20 %) also playing a smaller role; -Constraints on e-folding depth are critical for calculating photolytic loss of snow pack NO3and for interpreting δ 15 N- -The NO3recycling process at DML is weaker than Dome C, largely because of the higher snow accumulation rate and lower e-folding depth; -TRANSITS has now been tested at two sites in Antarctica, namely DML and Dome C, and we recommend applying this model to new ice core sites to understand the sensitivity of the δ 15 N-NO3signal before embarking on new ice 940 core projects; -By accounting for variability in the snow accumulation rate and assuming a constant e-folding depth, it may be possible to reconstruct past UV-radiation at ice core sites with very a low accumulation rate and low accumulation variability, as low accumulation variability will have little effect on δ 15 N-NO3in comparison to the UV dose reaching Experience Project (REP) that contributed to this manuscript. We would like to thank British Antarctic Survey (BAS) and Alfred Wegener Institute (AWI) staff for their field and logistics support at Halley Station and Kohnen Station, respectively. 950 Technical support for nitrate isotope analysis at the Institut des Géosciences de l'Environnement (IGE), Grenoble was provided by Joris Leglise, Ines Ollivier and Ilan Bourgeois. We thank Joseph Erbland for providing the TRANSITS model. Field samples collected at Dome C was possible through the program SUNITEDC/CAPOXI (grant 1011/1177) funded by the Institut Polaire Français IPEV. J.S and N.C thank the ANR (Investissements d'avenir ANR-15-IDEX-02 and EAIIST grant ANR-16-CE01-0011-01) and the INSU program LEFE-CHAT for supporting the stable isotope laboratory. This is publication 1 of PANDA 955 platform on which isotope analysis were performed. PANDA was partially funded by the LabEx OSUG@2020 (ANR10 LABX56). All winter over personal who collected the year-round Dome C samples in extreme conditions, years after years, are deeply acknowledged. In addition, we thank Emily Ludlow, Shaun Miller, Catriona Sinclair, Rebecca Tuckwell, and Neil Brough for technical support at BAS. Thanks to James France for discussions around the e-folding depth measurements and interpretation, and to John Turner for discussions of the local meteorology. We acknowledge Utrecht University who supplied 960 the AWS data for AWS9 at DML05/Kohnen (https://www.projects.science.uu.nl/iceclimate/aws/files_oper/oper_20632), and the precipitation data from the RACMO2 model (https://doi.org/10/c2pv). We would like to thank Bodeker Scientific, funded by the New Zealand Deep South National Science Challenge, for providing the combined NIWA-BS total column ozone database. Wind roses were plotted using the openair package in R. The data set for the DML nitrate isotopic ratios and nitrate mass concentrations in aerosol, skin layer and snow pits is available through the Polar Data Centre 965 https://doi.org/10.5285/1467b446-54eb-45c1-8a31-f4af21e60e60, and supporting data are also included as figures and tables in the supplement.