Archival processes of the water stable isotope signal in East Antarctic ice cores

The oldest ice core records are obtained from the East Antarctic Plateau. Water isotopes are key proxies to reconstructing past climatic conditions over the ice sheet and at the evaporation source. The accuracy of climate reconstructions depends on knowledge of all processes affecting water vapour, precipitation and snow isotopic compositions. Fractionation processes are well understood and can be integrated in trajectory-based Rayleigh distillation and isotopeenabled climate models. However, a quantitative understanding of processes potentially altering snow isotopic composition after deposition is still missing. In low-accumulation sites, such as those found in East Antarctica, these poorly constrained processes are likely to play a significant role and limit the interpretability of an ice core’s isotopic composition. By combining observations of isotopic composition in vapour, precipitation, surface snow and buried snow from Dome C, a deep ice core site on the East Antarctic Plateau, we found indications of a seasonal impact of metamorphism on the surface snow isotopic signal when compared to the initial precipitation. Particularly in summer, exchanges of water molecules between vapour and snow are driven by the diurnal sublimation–condensation cycles. Overall, we observe in between precipitation events modification of the surface snow isotopic composition. Using high-resolution water isotopic composition profiles from snow pits at five Antarctic sites with different accumulation rates, we identified common patterns which cannot be attributed to the seasonal variability of precipitation. These differences in the precipitation, surface snow and buried snow isotopic composition provide evidence of post-deposition processes affecting ice core records in low-accumulation areas.


Introduction
Ice is a natural archive of past climate variations. The chemical and physical compositions of ice, and the air bubbles trapped inside it, are used as paleoclimate proxies (Jouzel and Masson-Delmotte, 2010). Over large ice sheets, the water isotopes in ice cores can provide reconstructions of historic temperatures as far back as the last glacial period in West Antarctica (up to 60 000 years ago), (WAIS Divide Project members, 2013), and the last interglacial period in Greenland (120,000 years ago) (North 5 Greenland Ice Core Project members, 2004;NEEM Community members, 2013). In East Antarctica, low accumulation rates enable the reconstruction of past climates over several interglacial periods, e.g. 420 000 years at Vostok (Petit et al., 1999), 720 000 years at Dome F (Kawamura et al., 2017) and 800 000 years at Dome C (EPICA, 2004(EPICA, , 2006. Even though reconstructions from ice cores from Greenland and West Antarctica do not extend as far back as from East Antarctica, high resolution analyses of these cores provide very fine temporal resolution from which the seasonal cycle can be resolved (Vinther et al.,10 2010; Markle et al., 2017). Seasonal variations are also imprinted in the snow isotopic composition of high accumulation sites in coastal areas of Antarctica (Morgan, 1985;Masson-Delmotte et al., 2003;Küttel et al., 2012). For low accumulation sites as found on the East Antarctic Plateau, there is no consensus whether ice core records can reveal the climatic signal at resolutions finer than multidecadal (Baroni et al., 2011) or not (Ekaykin et al., 2002;Pol et al., 2014;Münch et al., 2016). Ekaykin et al. (2002) analysed multiple pits from Vostok and identified large spatio-temporal variations caused by post-deposition processes 15 associated with surface topography and wind interactions. These phenomena create strong non-climate variability in a single core, which calls for stacking isotopic composition profiles from several snow pits to reveal the common climatic signal. Still, on the East Antarctic Plateau, a clear seasonal cycle is depicted in the isotopic composition of the precipitation (Fujita and Abe, 2006;Landais et al., 2012;Stenni et al., 2016) and of the surface snow (Touzeau et al., 2016). So far, whether this seasonal cycle is archived or not in buried snow, and thus, whether stacking an array of snow pits enables us to increase the signal to 20 noise ratio and depict a climatic record at the seasonal scale from water isotopic signal is an important open question (Ekaykin et al., 2014;Altnau et al., 2015;Münch et al., 2016Münch et al., , 2017. Several studies have focused on understanding how the climatic signal is archived in the isotopic composition of snow and ice on the East Antarctic Plateau. Since the early works of Dansgaard (1964) and Lorius et al. (1969), the relationship between 25 ice isotopic composition and local temperature has been attributed to the distillation associated with the successive condensation events on the path from the initial evaporation site to the deposition site (Criss, 1999). Nevertheless, the relationship between isotopic composition and surface temperature is not constant through time and space, due notably to processes within the boundary layer (Krinner et al., 1997), the seasonality of the precipitation between glacial and interglacial periods (Sime et al., 2009), variations in air mass transport trajectories (Delaygue et al., 2000;Schlosser et al., 2004) and in the evaporation 30 conditions (Vimeux et al., 1999). For East Antarctica, the glacial-interglacial isotope-temperature relationship appears quite close to the spatial gradient, but its validity for inter-annual variations (Schmidt et al., 2007) and warmer than present-day conditions (Sime et al., 2009) is unclear.
In addition, under the exceptionally cold and dry conditions of East Antarctic drilling sites, the contribution of postdeposition processes to the isotopic composition of the snow cannot be neglected at the atmosphere/snow interface (Town et al., 2008;Sokratov and Golubev, 2009). Indeed, the relationship between temperature and isotopic composition of the surface snow is different from the one found in the precipitation (Touzeau et al., 2016). It has been recently evidenced that summer exchanges between snow and water vapour at the surface significantly affect the isotopic composition of the snow 5 both in Greenland (Steen-Larsen et al., 2014) and on the East Antarctic Plateau (Ritter et al., 2016). In this study, we intend to evaluate the different contributions to the snow isotopic composition signal in order to explain how the climatic signal is archived in ice isotopic composition. Therefore, we used Dome C as an open air laboratory to study the different contributions to the surface snow isotopic composition, combining : (1) direct precipitation input, (2) blowing snow, (3) exchanges with the atmospheric vapour and (4) exchange with the firn below the surface; as presented in Fig. 1.  Figure 1. Schematic of the different contributions to the snow isotopic composition (R i X stands for the composition of isotope i in the phase X: P stands for precipitation, V stands for vapour, S stands for surface snow, N stands for the deeper snow, 0 stands for the vapour at equilibrium with the snow): above the surface, both the precipitation and the sublimation/condensation cycles can contribute to the surface composition; in the open-porous firn below the surface, ice crystals can exchange with the air in the pores, influenced or not by wind pumping. Deeper in the firn, molecular diffusion in the interstitial air affects the snow isotopic composition.
At a site as far along the distillation path as Dome C (and in general on the East Antarctic Plateau), only a small amount of precipitable water is available (Ricaud et al., 2014) and precipitation is sparse and irregular (Genthon et al., 2015). As a result, accumulation of snow at Dome C is not a homogeneous process affecting a flat surface, and the snow deposition is patchy and strongly dependant of the surface roughness (Groot Zwaaftink et al., 2013;Libois et al., 2014;Picard et al., 2016a). As a result, a small but significant contribution to the annual mass balance comes from sublimation/condensation directly at the surface . Indeed, during the Austral summer, negative fluxes up to −0.3 mm w.e. month −1 , associated with sublimation, are observed whereas during the Austral winter, positive fluxes up to 0.1 mm w.e. month −1 , associated with condensation, are found. Integrated, these contributions for a site such as Dome C add up to 10% of the total annual accumulation. Even though the amount of water introduced by this contribution is limited, we expect a significant impact on 5 the isotopic budget of the snow.
At the diurnal scale, this asymmetry of temperature conditions during condensation/sublimation cycles is also observed in summer: even though the sun never goes down, the radiative budget creates a significant temperature cycle (up to 16 • C per day) during which the vapour isotopic composition evolves in parallel with the snow surface temperature (Casado et al., 2016;Ritter 10 et al., 2016). As sublimation occurs at higher temperature than condensation, the diurnal cycle results in a non-neutral isotopic budget. In addition, the dynamic of the atmospheric boundary layer is different during the day and during the night: during the day, active convection enables turbulent mixing throughout the boundary layer while during the night, the conditions are much more stratified as illustrated by the important temperature gradients that are observed (Casado et al., 2016;Vignon et al., 2017). 15 In the top first metres of the snowpack, isotope exchanges involved during snow metamorphism and diffusion within the porous matrix additionally affect the isotopic composition of the snow (Langway, 1970;Johnsen, 1977;Whillans and Grootes, 1985;Calonne et al., 2015). The diffusion length associated with these processes depends on the firn ventilation, the snow density, porosity and tortuosity and the exchange rate between the atmospheric water vapour and the surface snow (Johnsen et al., 2000;Gkinis et al., 2014). This wide range of processes hampers the interpretation of the isotopic signal. In particular it 20 is not clear how much of the original signal acquired during the formation of the precipitation is conserved during the burial of the snow .
Recent studies have focused on the monitoring of the isotopic composition of the snow pack on the East Antarctic Plateau (Touzeau et al., 2016), of the precipitation (Fujita and Abe, 2006;Landais et al., 2012;Stenni et al., 2016), and of the atmo-25 spheric water vapour (Casado et al., 2016;Ritter et al., 2016); exploring their links to climatic parameters and the implications for post-deposition processes during the archiving of the climatic signal by the snow isotopic composition. Here, we study the isotopic composition of the continuum between atmospheric vapour, precipitation, surface and buried snow. To do so, we combine different datasets from Dome C, on the East Antarctic Plateau, based on published studies and new results, in order to qualitatively characterise the different processes affecting surface snow isotope composition from diurnal to annual time scales. 30 We first report and compare the different methodologies applied for sampling surface snow, snow pits, precipitation and water vapour in the atmosphere (Section 2). Then, we deliver the results from different studies including surface snow measurements over several years, precipitation measurements, vapour and snow measurements and snow pits sampling (Section 3) before summarising our key conclusions.
2 Sites, material and methods

Site
The East Antarctic Plateau is a high elevation area, over 2500 metres above sea level (m a.s.l.) covered with snow and ice spreading across most of the eastern continental part of Antarctica (Fig. 2). The East Antarctic Plateau is characterised by mean annual temperatures below −30 • C and accumulation below 80 kg.m −2 .yr −1 , as illustrated in Fig. 2 (Alley, 1980;Petit et al., 1982;Wendler and Kodama, 1984;Oerter et al., 2000;Ekaykin et al., 2002;van As et al., 2007;Lazzara et al., 2012;Casey et al., 2014;Genthon et al., 2015;Touzeau et al., 2016;Laepple et al., 2016) Site Location Altitude AWS mean 10 m firn Accumulation Mean wind (m a.s.l.) temperature (  Table 2). Because different teams were in charge of the different sampling activities, the protocols differ between the years.

5
The sampling protocol of the 2011 campaign (SUNITEDC) has been precisely described by Touzeau et al. (2016): the upper first millimetres of snow (1 to 5 mm) were gathered every 1-2 weeks using a metallic blade over a surface of 20 by 20 cm. This leads to samples of approximately 20 mL. The sampling areas were randomly picked, provided the surface was flat.

10
During the NIVO project (from 2013 to 2016), the surface snow was gathered by sampling roughly 15 mm of snow with a corning flask over a surface of 20 by 10 cm. This led to samples of approximately 50 mL. The sampling areas of the NIVO project were chosen randomly in a 100 by 100m "clean area" near the Atmospheric Shelter in parallel with density and specific surface area (SSA) measurements (see section 2.4). To address the problem of spatial variability, two samples separated by 10 to 50 m were gathered during each collection and we present here the average value of the two samples. In addition, during 15 summer 2013/14, regular samplings of surface and sub-surface snow were performed daily for almost 2 months. The surface samples were gathered using a corning flask from 0 to 3 cm depth. The sub-surface samples were gathered by the same tool  and measurement in Italy. For the precipitation samples, the protocol is detailed by Stenni et al. (2016). It is important to note that the protocol of surface snow sampling from the PRE-REC campaign differs greatly from the protocols from the NIVO and SUNITEDC programs due to the presence of the wood plate. We present profiles of isotopic composition sampled in snow pits at Dome C : two unpublished profiles from the first prelimi- presented here as well, which were previously described in Ekaykin et al. (2002Ekaykin et al. ( , 2004 and Ekaykin and Lipenkov (2009). We combine the results from six snow pits with depths varying from 2.5 m to 12 m and a minimum resolution of 5 cm. In addition, snow pits from the Explore-Vanish campaign are included comprising one 3.5 m deep snow pit from Vostok, one 2.6 m  Touzeau et al. (2016). Finally, we include two snow pits from the South Pole (Jouzel et al., 1983;Whitlow et al., 1992).
We do not expect any impact of the sampling technique when comparing the different snow pits considering the similarity of the protocols of all the snow pit samplings.

Atmospheric and snow surface monitoring
Water vapour isotopic composition has been measured at Dome C in 2014/15 (Casado et al., 2016). To reduce the noise, the dataset was averaged to hourly resolution and covers approximately one month. In parallel to water vapour isotopic composition monitoring, surface snow was sampled once to twice a day. For a period of 27 hours, the surface snow was sampled every hour to evaluate the diurnal cycle of both the vapour and the snow isotopic composition (see section 2.2).

10
We use temperature, wind speed and humidity measurements from the 45m meteorological profiling system described by Genthon et al. (2013). The temperature and humidity observations are performed using ventilated thermohygrometers HMP155 and are therefore free of radiation biases (Genthon et al., 2011). The temperature reanalysis product (ERA interim) has been compared to ventilated automatic weather station data (AWS) from Genthon et al. (2013) and we found a good agreement at 15 the seasonal scale and fairly good agreement at the event scale (not shown here). Wind speed and direction are measured using Young 05103 and 05106 aerovanes. Snow surface temperature is measured with a Campbell scientific IR120 infrared probe located 2 m above ground level.
In the field, snow metamorphism is difficult to quantify due to the large noise created by the spatial variability, requiring a 20 large number of samples every day. Therefore, we include grain index observations (Picard et al., 2012) obtained by satellite measurements using passive microwave satellite data. Picard et al. (2012) argue that the grain index is an indicator of the coarsening of snow grains and show its increase in summer to be anti-correlated with the integrated summer precipitation amount. When available, we include Surface Specific Area (SSA) measurements also as an indicator of metamorphism (Libois et al., 2015). These optical methods are completed with snow observations. Frost deposition was monitored with a time lapse of the growth of hoar at the surface (see the video at https://vimeo.com/170463778). An image processing script was used to characterise the growth of a few crystals at the surface of a sastruga.

Modelling approaches
To investigate the impact of post-deposition processes, it is necessary to present how the surface snow isotopic composition differs from the initial precipitation signal formed during the Rayleigh distillation. Here, we make use of the Rayleigh-type Mixed Cloud Isotope Model (MCIM) developed by Ciais and Jouzel (1994) which computes the Rayleigh distillation along the air mass trajectories. The model includes microphysical properties of clouds and in particular takes into account mixed 10 phase conditions. It is tuned with triple snow isotopic composition measured along a transect from Terra Nova Bay to Dome C (Landais et al., 2008). One of the main uncertainties in this model is the supersaturation. Indeed, Jouzel and Merlivat (1984) evaluate the impact on kinetic fractionation during the snow formation parametrised from the supersaturation. The tuning of the supersaturation has been proven suitable to evaluate the variations of isotopic composition at Dome C (Winkler, 2012).
This will provide a comparison between the spatial (at the scales from 10 to 1000 km) and the temporal slope of the isotopic 15 composition of precipitation at the seasonal scale. It will also be used to quantify the impact of post-deposition processes on the surface snow isotopic composition by providing a reference for the precipitation isotopic composition, for days with no precipitation.

Results and discussions
In this section, we review the results from the different datasets, illustrating the different steps of the archiving of the climatic 20 signal by the snow isotopic composition, from the precipitation to the buried snow (see Fig. 1). As we expect the precipitation input to be the primary contribution, we will first compare the variations of precipitation isotopic composition based on data from Stenni et al. (2016) and additional new data, to surface snow isotopic composition. Then, we will evaluate the exchanges with the atmosphere: first at the diurnal scale by reporting parallel measurements of water vapour and surface snow isotopic composition; and second, at the day-to-day and at the seasonal scales the difference between surface snow and precipitation 25 isotopic composition to the meteorological conditions and the grain index. Finally, we will compare the isotopic composition of the surface, sub-surface and buried snow to evaluate the exchanges within the firn. Compared to other year-long precipitation time series on the East Antarctic Plateau, this slope is lower than at Dome F (0.78 (Fujita and Abe, 2006)) and higher than at Vostok (0.26 ‰ • C −1 with R 2 = 0.58 (Touzeau et al., 2016)). First, we focus on the impact of local spatial variability (below 1 km) of the measurements. Indeed, significant differences can be found in the snow isotopic composition for distances ranging from 1 to 1000 m. To disentangle the local spatial variability from the temporal variations of the surface snow isotopic composition (Fig. 3), we compare the duplicate measurements 25 realised during the year 2014 in the project NIVO (red shade in Fig. 4 and see Section 2.2). We found that the spatial uncertainty of the surface snow isotopic composition measurements is 3.4‰ for δ 18 O s from 2 standard deviations calculated with the duplicates on the NIVO samples (randomly picked within 50 m). For that year, several sets of measurements are available from 3 different programs (PRE-REC in blue, NIVO in red and GLACIOLOGIE in black). Even though the PRE-REC sampling is more affected by heavy snowfall events due to the wood plate, apart from a limited number of outliers (5 over 59),  (Touzeau et al., 2016); Feb 2012 to Oct 2012, data from the project PRE-REC (this study); Nov 2013 to Jan 2016, data from the project NIVO (This study), the light green shaded area is the uncertainty due to spatial variability obtained from replicates (2σ, see Section 3.1.2) and the dark green line is the modelled surface snow isotopic composition from the toy model detailed in section 3.2; four years (2008 to 2011) of monitoring at Dome C of the variations of the isotopic composition of precipitation from the PRE-REC campaign (Stenni et al., 2016) (δ 18 Op, blue, dots: raw data, line: monthly average), for comparison we present the grain indices from satellite observations (Picard et al., 2012) (black fine line), and 2 m temperature measurements (red line) and precipitation (black bars) from ERA-interim reanalysis product. The blue shaded areas highlight the high grain index values (arbitrary threshold on the variations indicating when metamorphism is active).
with the isotopic composition of precipitation (Dreossi, personal communication). Apart from these outliers, when comparing the PRE-REC results to the NIVO results, the average difference is 1.5‰, which we attribute to both spatial variability and mixture of precipitation and surface snow. The comparison of the NIVO and GLACIOLOGIE datasets shows much smaller differences (on average 0.4 ‰) but the comparison was done on a limited period without any large synoptic event. These in- Second, we focus on the temporal variability at different time scales. The three years (2011, 2014 and 2015) present the typical temperature variations for the East Antarctic Plateau (see Fig. 3): a short, "warm" summer before a long, rather constant, winter as described by Van Den Broeke (1998). Over this cycle are imprinted short warm events often associated with advection of warmer air masses and precipitation; these warms events are particularly visible in winter. We observe a similar pattern for δ 18 O s of the surface snow: annual cycles with a steep maximum centred on January (roughly a month after the 10 temperature maximum) and a gradual decrease along most of the winter delayed by several months when compared to the local air temperature. Over these annual cycles, peaks of δ 18 O s of surface snow occur and some of them might be related to warm precipitation events as previously suggested by Touzeau et al. (2016). during summer, which will be addressed in the next section, and post-deposition processes which will be evaluated in Section.
3.3.  As the snow surface samples are 1.5 cm thick, they contain the accumulation of several precipitation events (roughly the thickness in accumulation expressed in snow equivalent). To evaluate the impact of the mixing over a 1.5 cm thickness, absolute values of the accumulate rate are important, and so we use precipitation estimates from ERA-interim products which have been renormalised in order to obtain a total amount of accumulation over 7 years matching the observations (Genthon et al., . The results of this modelled surface snow isotopic composition are presented in Fig. 3. We observe that the seasonal cycle of the modelled surface snow isotopic composition is in most cases delayed compared to the temperature seasonal cycle. In particular, the summer maximum of isotopic composition is observed roughly 2 months after the temperature high. We also observe that the modelled amplitudes of these summer excursions are not regular, some presenting high amplitudes (  The 7th of January was characterised by a large diurnal cycle in water vapour isotopic composition, humidity and temperature associated with a turbulent and convective atmospheric boundary layer enabling important exchanges of moisture between snow and vapour (Casado et al., 2016). This is a common situation in summer at Dome C due to weak katabatic winds. The without being impacted by meteorological events, we therefore focus on the night from 18:00 on the 6th of January to 5:00 on the 7th of January. In addition to the isotopic composition of the vapour and of the surface snow, in Fig. 5 we present 3m temperature measured by AWS, surface temperature measured by infrared sensing (Casado et al., 2016) and relative humidity calculated from the specific humidity measured by the Picarro laser instruments and the saturated vapour pressure at the surface temperature (Goff and Gratch, 1945). Note that due to intake of snow crystals in the inlet of the Picarro, relative humidity is 5 overestimated in very supersaturated conditions, but other hygrometers installed at Dome C confirmed supersaturated conditions with relative humidity ranging between 105% and 125% between 19:00 on the 6th of January and 06:00 on the 7th of January.
The evolution of water vapour and snow isotopic compositions is synchronous with observations of mist and solid conden-10 sation due to local large supersaturation. During this five hour period, water vapour δ 18 O v increased from -73‰ to -64‰ while snow δ 18 O s decreased by roughly 2‰. From 21:30 on the 6th of January (UTC time) to 01:40 on the 7th of January (UTC time), the volumes of three snow crystals were monitored by a script transferring the size in pixels of each crystals from the time-lapse sequence to surfaces using a length etalon and estimating the volume variations ∆V using a power law from the surface variations ∆S: ∆V ∝ ∆S 3/2 . This shows an increase by a factor from 1.5 to 3.9. The growth of the crystals observed 15 in the time-lapse suggests a large-enough ice deposition to significantly affect the isotopic composition. The solid condensation occurs simultaneously with the modification of the isotopic composition of the snow and of the vapour. We observe a small delay (2 to 3 hours) between the beginning of the vapour isotopic composition increase and the decrease of the surface snow isotopic composition. These observations can be explained by an exchange of molecules between the snow and the vapour, significantly affecting the snow isotopic composition and leading to an enrichment of the isotopic composition of the vapour 20 and a depletion in the snow (note that the origin of the vapour can be either from the free atmosphere or from the pores in the snow; we are not able to discriminate between the two) .

Thermodynamics of the isotopic exchanges between the snow and atmosphere during a frost deposition in a closed box system
We evaluate the exchanges at the diurnal scale between the atmospheric vapour and surface snow using a simple thermodynamic during this period (∆P v = 50P a), this transfer requires the height of the atmospheric box in which the vapour is removed to be: where R is the molar gas constant and T the air temperature. This height of 12 m is in agreement with independent estimate of the boundary layer thickness considering the summer dynamic of the boundary layer at Dome C   (2) where ∆n 18 cond is the number of molecules condensed during the frost deposition, ∆n 18 ren is the number of molecules renewed in the atmospheric box by either advection, molecular or turbulent diffusion; and ∆n 18 v is the variation of heavy isotope molecules in the atmospheric box. Note that both the closed box contribution ∆n 18 v and the renewal ∆n 18 ren are here defined as positive 10 contributions to the total amount of heavy isotopes condensing ∆n 18 cond . In this framework, ∆n 18 v is estimated by: Where ∆R 18 v is the variation of isotopic composition in the boundary layer. The contribution of heavy isotopes towards the surface snow in a closed box-like system is thus: ∆n 18 v = 5.6 10 −4 mol.m −2 . It is interesting to note that the contribution associated with the fractionation n v ∆R 18 v accounts for less than 10% of the contribution of the closed box system. 15 The renewal of the atmospheric box with the free atmosphere can not be directly inferred here as we only measured the water vapour isotopic composition at 2 m height. Based on summer isotopic composition profiles in Polar Regions (Berkelhammer et al., 2016), we assume that the value of R 18 v measured at a height of 2m is representative for the whole lower atmospheric box. We consider that during the 'day time', active turbulence has homogenised the vapour isotopic composition of the atmospheric 20 column, and that during the 'night', due to more stable conditions, the isotopic composition above the atmospheric box has remained constant. Thus, in the particular case of a frost deposition, the term ∆n 18 ren is calculated as a fraction of ∆n 18 v by: where is a parameter (varying between 0 and 1) which depends on advection and turbulent processes, and n 18 v (init) is the initial number of heavy isotope set for the entire boundary layer before turbulence stopped.  Figure 6. Schematic of the thermodynamic model for the exchanges between the snow and the vapour during condensation: the snow exchange with the vapour through equilibrium fractionation at the phase transition and the atmospheric vapour in the atmospheric boundary layer is only weakly renewed by advection of air masses (0 % : no advection, closed box system in blue; 50 % meaning that the contribution of the advection is equal to half the amount of water vapour involved in the phase transition (purple) and 100 % in orange.) We apply this simple model to evaluate the variation of isotopic composition of the 1.5 cm of surface snow associated with 5 the condensation of 0.3 mol.m −2 of water vapour for the cases of a closed box system ( = 0%) or with the renewal of 100% of the heavy isotopes (see Fig. 6). We find that for a closed box system ( = 0), the modelled variation of surface snow isotopic composition is 1.91 ‰ close to the observed value of 1.99 ± 0.3 ‰ in the surface snow δ 18 O (see Fig. 5). In the case of a renewal of 100% of the heavy isotopes in the atmospheric box ( = 1), we obtain a value of 2.87 ‰, which overestimates the changes of surface snow isotopic composition. 10 This box model shows that at Dome C, the surface snow isotopic composition can be significantly affected by the deposition of frost in one example during one "night". At NEEM, Steen-Larsen et al. (2014) showed that the isotopic compositions of the snow and the vapour present a parallel increase in warm conditions in between precipitation events up to 7 ‰. Similar observations have been made in Antarctica at Kohnen (Ritter et al., 2016). However, at the diurnal scale, no parallel evolution isotopes in the vapour): the vapour is enriched in heavy isotopes while snow is depleted during frost deposition events. We attribute the difference of behaviour compared to the parallel evolution observed at Kohnen (Ritter et al., 2016) and NEEM (Steen-Larsen et al., 2014) to the position of the station on the top of a dome. Indeed, at Dome C, the weakness of the katabatic winds (3.3 m.s −1 ) decreases the renewal of air masses long enough for the exchanges with the snow to be detected.
At Kohnen and NEEM, stronger winds are observed (4.5 and 4.1m.s −1 , respectively) leading to a more efficient renewal of 5 air masses able to exchange with the surface (larger values). It is important to note that the humidity content at Dome C is lower than at Kohnen and NEEM, resulting in a smaller reservoir of water vapour with which the snow can exchange. Despite these low humidity levels, a significant impact of the sublimation/condensation cycles on the snow isotopic composition is observed. Similar studies measuring isotopic composition of vapour and snow at sites with similar temperatures but larger wind speeds (such as at Vostok), could provide more robust insights on the impact of wind on the renewal of air masses 10 compared to humidity levels. This study relies on one event of attested solid condensation and the monitoring of more events is necessary to be able to quantitatively evaluate the fractionation processes involved, however because of the asymmetry of the dynamic of the atmospheric boundary layer between night and day (stable at night during condensation, turbulent during the day during sublimation), we expect a more important renewal of the air masses during the day, leading to an anomaly of isotopic composition after several sublimation/condensation cycles.

Evidence for isotopic modifications linked to snow metamorphism
We investigate whether, at the seasonal scale, the exchanges of heavy isotopes during sublimation/condensation cycles contribute to the seasonal signal, and therefore, whether the surface snow isotopic composition variations are linked to exchanges with the atmospheric water vapour, independent of the precipitation input. To do so, we use the grain index as an indicator of grain coarsening. Indeed, the grain coarsening in the surface snow is due to successive sublimation/condensation with the pore 20 vapour. Figure 3 presents the grain index estimated from satellite data (Picard et al., 2012) to evaluate the impact of metamorphism in the surface snow. Periods of strong metamorphism identified during the summer are highlighted (blue shaded areas). We (and partially in 2012), the large summer increase of δ 18 O s is associated with small summer increase of grain index, in this case delayed after mid-January. The same pattern is not observed for precipitation (Fig. 3) whose seasonal isotopic variations appear more regular and in phase with temperature.  (Picard et al., 2012;Libois et al., 2015); the rise usually starts during the first week of December. Rapid falls of the grain index result from important precipitation events and the input of small snow grains from precipitation. For instance, during summer 2015, we observe significant variations of the surface snow isotopic composition while no precipita-tion input was identified, as described in section 3.2. During this period, intense metamorphism was observed, as highlighted by the grain index rise in Fig. 3. The variations of roughly 8 ‰ observed in the surface snow isotopic composition are in phase with the temperature variations. This supports the hypotheses of an input from the exchanges with the vapour: indeed, the snow is porous at Dome C, and thus the exchanges with the vapour could affect the first centimetres globally. This is discussed more thoroughly in section 3.4. Finally, the slow decrease during winter is explained by the accumulation of new small snow 5 flakes by precipitation onto the coarse grains formed during the summer. Winter metamorphism is too slow to impact the snow structure.
Finally, we find two features of interest regarding the amplitude of the variations of δ 18 O s of the surface snow in Fig. 3.
First, the inter-annual variability of the summer surface snow isotopic composition seems directly related to the strength of 10 the metamorphism as suggested by the negative correlation between the amplitude of the grain-index increase in summer and the maximum δ 18 O s (R 2 = 0.54). However, this correlation appears to be independent of the link between metamorphism and the amount of precipitation (not shown), but a larger dataset is required to validate these preliminary results. The summer increase of δ 18 O s seems to be very sensitive to the date at which the intense summer metamorphism starts. The large val- is no apparent relationship between the isotopic composition of precipitation and the grain index from 2008 to 2011 (see Fig. 3).

Surface and sub-surface snow isotopic exchanges
During the summer 2013/14, regular sampling of surface (0 to 3 cm) and sub-surface (3 to 6 cm) snow were carried out at Dome C. Once a day, two samples of snow at each level were taken along with specific surface area (SSA) measurements to 25 evaluate the size of the grains and therefore, how much metamorphism has taken place (Picard et al., 2016b). These data are presented in Fig. 7 with temperature and precipitation from reanalysis and in-situ observations of precipitation. We observe that overall, during these two summer months, the sub-surface isotopic composition is almost systematically lower than the surface one. This is found as well in most of the snow pits presented in Section 3.5.

30
From the end of November to the 15th of December, we observe similar values of surface and sub-surface snow isotopic composition. This is a period during which metamorphism has not started yet as indicated by the large values of SSA (Fig. 7).
The surface snow isotopic composition is low (around -55‰) and the SSA is high which is typical of winter snow. From the 16th of December, we observe large differences between the surface and the sub-surface snow isotopic composition (up to 5‰ higher at the surface) and the first decrease of SSA indicating the effects of metamorphism. This feature is not associated with significant precipitation input, and the difference between the surface and the sub-surface isotopic composition could be linked to post-deposition processes. Until the 31st of December, numerous drift events mix the snow and therefore cause strong spatial variability. Finally, due to a large precipitation event around the 2nd of January, we observe a significant increase of snow δ 18 O s of 18 ‰ at the surface first and at the sub-surface no more than two days later. The increase of the sub-surface 5 δ 18 O cannot be explained only by the accumulation of snow as this event only accounts for roughly 1cm w.e. (which is already large compared to the annual accumulation of roughly 2.3 cm w.e.). Here, it is clear that the surface snow isotopic composition is directly affected by this precipitation event, whereas it seems that the subsurface snow isotopic composition is less sensitive to this impact and that it only changes as a delayed reaction to the surface changes. is important to note here that the variations of isotopic composition include both spatial and temporal variations as only one sample per day was taken, therefore some of the variability might be due to spatial variability.
This sampling campaign suggests that: (1) snow metamorphism alters surface snow isotopic composition even in the absence of precipitation, and (2) precipitation isotopic composition can rapidly be transferred to the sub-surface. The most likely 5 candidate for this signal transfer is molecular diffusion throughout the interstitial air, but more extensive time series will be necessary to quantify the processes involved.

Signal archived in the snow pack
We have observed that the surface snow isotopic composition signal is composed of several contributions, from the precipitation input to the exchanges of the atmospheric vapour. Both leave on a seasonal signal in the snow isotopic composition, 10 which could be visible in vertical profiles of snow isotopic composition. We now focus deeper in the firn to evaluate whether the signal imprinted in the snow isotopic composition at the surface is preserved after the burial when accounting for isotopic diffusion as suggested by Münch et al. (2017) for Kohnen Station.

Observation of apparent cycles
15 At Dome C, variations of δ 18 O with depth are quite large within the top of the firn (typically of the order of 5 ‰) and irregular ( Fig. 8). This feature has been confirmed by isotopic records on two 1 m deep snow pits dug in 2014/15 at Dome C (Fig.   9). The two isotopic composition profiles, separated by roughly 100 m, are not well correlated (R = 0.15), as expected from stratigraphic noise . 20 In order to evaluate whether these variations could be explained by intermittent precipitation events, and thus reflect climatic variability, we generated a profile of isotopic composition by accumulating snow using the relationship between precipitation isotopic composition and temperature, and the reanalysis products, as previously described in section 3.2 and in Supplementary Materials A. The results of this model include the precipitation intermittency and the temperature anomaly associated with large precipitation events. The results are presented in Fig. 9. In the synthetic snow isotopic composition profiles, we observe is actually observed on the power spectra , so these 20 cm variations can only be referred as 'apparent cycles'.
Thus, neither seasonal variability nor multi-year accumulation variations explain the δ 18 O N variability in snow pits observed at Dome C.

Similarity of the apparent cycles across the East Antarctic Plateau
We extend the study of snow pit profiles found at Dome C to four other sites on the East Antarctic Plateau (Kohnen, S2, South Pole and Vostok) which are characterised by different meteorological and glaciological parameters such as mean annual temperature, elevation, wind speed and direction, accumulation or sastrugi height. A representative subsection of the profiles of isotopic composition from the different sites is presented in Fig. 12 in Supplementary Materials B.

10
We evaluate the similarity of variability between the different sites by manual counting of the successive maxima. We again find a 20 cm spacing on average between δ 18 O N maxima across sites of the Antarctic Plateau (Table 4,   For sites with high accumulation such as the South Pole and Kohnen (around 20 cm of snow equivalent accumulation), seasonal variability should be evident in snow isotopic composition variations (Jouzel et al., 1983). In this case, the observed cycle lengths could simply reflect the imprint of seasonal variations in annual layers. Yet, the profiles are highly variable, exhibiting strong differences in between sites and as well as in between pits from a single site, even if sampled the same year. This can be attributed to the mixture of the potential climate signal and non-climate noise (Fisher et al., 1985;Münch 24 The Cryosphere Discuss.  Laepple et al., 2016). Here, when we compare several snow pits dug the same year at the same sites (for instance at Kohnen as presented on Fig. 12 and observed on more profiles), we do not observe synchronous peaks in the profiles of isotopic compositions. Since multi-year cycles of the climatic conditions and thus isotopic composition of precipitation would globally affect one site, this finding suggests that the features we observe are the expression of non-climatic (post-deposition) processes, resulting in spatial variability smoothed by diffusion .

5
For sites with lower accumulation such as Vostok and S2, we observe similar results to those from Dome C: the spacing between δ 18 O N maxima are larger than expected from the annual accumulation rates. For these sites, in order to observe the seasonal variations (with an associated length of the order of the accumulation between 6 and 7 cm), a resolution finer than 3 cm is necessary to resolve the seasonality (Nyquist, 1924;Shannon, 1949). The limited resolution of the S2 profile may thus explain why no seasonal cycle of isotopic composition is visible. However, in the case of Vostok and Dome C, the vertical 10 resolution of the isotopic composition profile is fine enough to establish the lack of seasonal cycle.
This indicates that non-climatic (post-deposition) processes are preponderant for the formation of these δ 18 O N variations in the firn. A forward model to evaluate how this signal is formed within the firn is out of the frame of this study and has been thoroughly evaluated in the pair study from Laepple et al. (2017). 15 3.6 Relations between isotopic composition and temperature in the precipitation and in the snow One major limitation in the isotopic paleothermometer is the uncertain and potentially variable relationship between isotopic composition and temperature. This is reflected in the range of slopes which are used to reconstruct temperature from isotopic composition including slopes obtained from precipitation at the seasonal scale, spatial slopes from samples covering several years or temporal slopes from independent calibration at large temporal scale (higher than 100 years). Here, we investigate 20 how the isotope-temperature slope is affected by the post-deposition processes described above, and if this could explain the different slopes found in the literature from different type of samples (precipitation, surface snow, buried snow).
First, we compare the relationship between the isotopic composition of precipitation and surface snow and temperature in the model and the datasets. Figure 10a presents the isotopic composition -temperature relationship in the dataset of isotopic 25 composition of precipitation and computed by the Mixed Cloud Isotopic Model (MCIM, see section 2.5). Except in summer (December, January, and February), the MCIM is able to faithfully simulate the isotopic composition of precipitation. The simulated relationship between δ 18 O p and temperature in the model shows a slope of 0.95‰ • C −1 (see Table 5), similar to the one found from the data from the transect between Terra Nova Bay and Dome C which were used to tune the model . For the entire seasonal cycle, we observe for the isotopic composition of precipitation a slope below 0.46‰ • C −1 .

30
Strong differences are not unexpected between temporal and spatial slope of precipitation (Ekaykin, 2003;Landais et al., 2012;Touzeau et al., 2016). It is interesting to note here that the winter temporal slope of isotopic composition of precipitation (0.76 ‰ • C −1 ) matches the spatial slope of isotopic composition for the East Antarctic Plateau (0.77 ‰ • C −1 for low isotopic composition area) as illustrated in Fig. 10a. Here, the low slope of the entire seasonal cycle is due to the summer isotopic  Fig. 10a) resulting in a slope of 0.41‰ • C −1 for the entire year. One way to explain such a low slope would be to introduce additional fractionation linked with re-evaporation during the precipitation events which can affect the snow flakes' isotopic composition (Koster et al., 1992) and therefore decrease the slope with temperature. This may also result from changes in air mass trajectory, and therefore in the Rayleigh distillation. Backtrajectory calculations for the East Antarctic Plateau indicate strong asymmetry 5 of the moisture sources for austral summer and winter (Sodemann and Stohl, 2009;Winkler et al., 2012). Finally, in the MCIM, the condensation temperature is estimated through a linear relationship with the local surface temperature (Ciais and Jouzel, 1994). The reduced summer temperature inversion at Dome C (Ricaud et al., 2014) is thus not taken into account in the MCIM which could also lead to a reduced slope.  By contrast, links between temperature and surface snow isotopic composition are not clear (Fig. 10) by the surface snow isotopic composition cycle is sometime coherent with the precipitation cycle (2014, see Fig. 10 c)) and sometimes not (2011 see Fig. 10 b); 2015, not shown). The difference of methods to sample the surface snow through the year does not explain this behaviour as for instance, the same protocol was applied to both 2014 and 2015. The surface snow isotopic composition is compared with the output of the MCIM model (Fig 10). These results confirm that in 2014, the snow isotopic composition spans all the range predicted by the model for this range of temperature, as for the isotopic composition  Because the timeseries of surface snow δ 18 O s and of temperature are not in phase, it is not possible to directly estimate the corresponding temporal slope by linear regression. This is particularly important in 2014 when the amplitude of the isotopic composition cycle is greatest. We therefore estimate the relationship by comparing the peak to peak range in temperatures and isotopic composition. As the phase lag is smaller in 2011, we use that year to compare the peak-to-peak slope to the linear  Stenni et al. (2016) and Table 5).
In conclusion, the signal observed in the precipitation isotopic composition is already not entirely reflected in the first centimetres of the surface snow. Averaging due to the sampling process and precipitation intermittency are both expected to 5 impact the isotopic signal, however, our model suggests that they are not sufficient to explain all the difference in signal. Here, we identify that redistribution of the snow (for instance, by wind) can explain spatial and temporal short scale variations of surface snow isotopic composition but may not be sufficient to explain the signal obtained in the surface snow at the seasonal scale, and therefore prevents from using the slope of precipitation isotopic composition against temperature to reconstruct climatic signal from ice records from Dome C at a seasonal scale.

Conclusions
In this study, we explored the post-deposition processes affecting the archiving of water isotopic composition from precipitation to the snow pack focusing on Dome C, a low accumulation site on the East Antarctic Plateau.
First, we demonstrated that surface snow isotopic composition at Dome C is affected by post-deposition processes, in particular exchanges between the atmosphere and the snow pack, leading to a seasonal signal in the surface snow isotopic composition different from the one expected from the precipitation signal. The amplitude of this isotopic signal seems to be associated with the strength of the surface metamorphism. The post-deposition effect influences the relationship between δ 18 O and temperature, which has important consequences for the interpretation of deep ice core water isotopes signal. 20 Second, we showed that at Dome C, similarly to other low-accumulation sites in East Antarctica, variations of the δ 18 O signal with depth in shallow firn cores does not correspond to past climatic seasonal variations. The typical length associated with these variations or 'apparent cycles' is on the order of 20 cm, and differ for low-accumulation sites with the expected seasonal cycles. 25 Third, we illustrated how different can be the slope between surface snow isotopic composition and temperature compared to the slope between precipitation isotopic composition and temperature. We attributed this behaviour to the different fractionation and meteorological conditions involved in the different inputs that affect the surface snow isotopic composition between precipitation and various post-deposition processes.

30
Our study is a qualitative demonstration of the importance of post-deposition effects for isotopic signal at the surface and sub-surface of snow. A second step would be to address this process in a more quantitative way through controlled laboratory experiments, field studies and use of snow models equipped with water isotopes. The combined use of water isotopes (d − the IPEV staff that made the campaigns possible, LGGE and LIPHY for providing logistic advice and support, PNRA-PRE-REC project for the 2014 surface snow data at Concordia station.

Appendix A: Simulation of the precipitation isotopic composition
The precipitation isotopic composition is simulated from the ERA-interim temperature and snowfall products and the relationship between precipitation isotopic composition and temperature obtained from section 3.1.1: The simulated precipitation isotopic composition is compared to field measurements in Fig. 11: we observe that the simulated precipitation content matches the observations at the seasonal scale except for the very low values of isotopic composition observed in winter. This is striking in winter 2010 where the modelled δ 18 O p is capped down to -65 ‰ whereas values below -70 ‰ are frequency observed. The simulated seasonal cycle only captures 85 % of the observed seasonal cycle in terms of amplitude.

5
At the day-to-day scale, high values of δ 18 O p associated with warm synoptic events are captured by the simulated precipitation product in winter. In summer, the modelled δ 18 O p fails to capture these warm events impact on δ 18 O p .
Overall, the simulated isotopic composition signal captures most of the seasonal cycle observed in the datasets, in particular, it successfully models the variability level in winter, the temporality of the high δ 18 O p events and the lack of lag between precipitation isotopic composition and temperature.

Appendix B: Signal in the snow pits across the East Antarctic Plateau
We analyse the typical variations observed in the snow pit by manually counting the successive local extrema with a threshold of minimum 1.5‰ for δ 18 O N and 10‰ for δD N for the difference between a minimum and a maximum (in both cases, the 15 thresholds are chosen higher than the measurement precision and lower than the annual variations of surface snow isotopic composition; sensitivity tests have been carried out that show the impacts are not significant). For each snow pit, the mean cycle length is estimated by counting the number of maxima over the length of the pit. We present the average of the cycle length of the different pits for each site (Table 4).  Figure 12. Isotopic composition profiles from 2 pits from Kohnen (Yellow), 2 pits from Dome C (Purple), 2 pits from Vostok (Blue), one pit from S2 (Green) and two pits from South pole (Red) and counting of cycles (circles) for each profile with a threshold of 1.5‰ in δ 18 ON between the successive local minima/maxima to prevent noise from artificially being counted as cycles.