Spatial distribution and post-depositional diffusion of stable water isotopes in East Antarctica

We have analysed the spatial variations in the mean stable water isotopic values, snow accumulation patterns and moisture sources along coast to inland transects in central Dronning Maud Land (cDML) and Princess Elizabeth Land (PEL) regions of East Antarctica. The δD and δO varied systematically from coastal to inland regions in cDML and PEL regions in response to the surface air temperature. While the elevation effect was not clearly visible, the isotope variations appeared to be associated with snow accumulation in cDML region and temperature in PEL region, which ultimately are associated 5 with elevation. Further, a clear influence of topography on the snow accumulation was observed in cDML region. Such an observation was not recorded in PEL transect, apparently due to the strong snow redistribution in this region due to katabatic winds. The moisture sources to the study areas were identified using HYSPLIT backtrajectory calculations. The major sources of precipitation during summer arrived from the south Atlantic ocean in the cDML and the Indian Ocean in PEL. During winter, the sources of precipitation in cDML extended upto Weddell Sea while in PEL, the sources extended upto 50◦S in the Indian 10 Ocean. In order to understand the post-depositional isotope diffusion processes in firn, a firn core which was drilled close to the cDML transect, five years after the snow core transect, was analysed in comparison with snow records. Our study showed a significant isotope amplitude diffusion with a diffusion length of 6 cm from the surface to 4 m depth in 5 years.

Though ice core δD and δ 18 O records are often assumed to represent climatic signals of an entire region, spatial variation of these isotope records indicate complex signals that are often not straightforward to interpret. Factors such as accumulation rates, erosion and redistribution by wind contributes to such variations in spatial scale (Fisher et al., 1985;Town et al., 2008). Another 25 important factor that could influence the isotope based climate records in Antarctic ice is the role of extreme precipitation events (Turner et al., 2019). The deuterium excess (d), which is a second-order parameter defined by combination of δD and δ 18 O (d = δD -8.δ 18 O), reflects the behaviour of these two isotopic species occurring during the kinetic and equilibrium fractionation (Jouzel and Merlivat, 1984). The association between 'd' and the climatic parameters such as the surface temperature, relative humidity and the wind speed in the moisture source region have been established in previous studies (Jouzel et al., 1982;Jouzel 30 and Merlivat, 1984;Petit et al., 1991).
Paleoclimate studies based on ice cores primarily rely on stable water isotope records to provide paleoclimate information. However, these stable isotope records are modified post deposition by diffusion processes (Johnsen, 1977;Whillans and Grootes, 1985). Even though the stable isotope composition in top layers of snow and firn is influenced by diffusion of water vapour and snow densification (Cuffey and Steig, 1998;Hörhold et al., 2011), no net change in isotopic composition were 35 observed due to these processes.The diffusion rate in ice beneath the firn is negligible and therefore gets preserved for much longer time after surviving the diffusion in the top layers (Johnsen, 1977). Studies by Steen-Larsen et al. (2014) and Ritter et al. (2016) have shown isotopic exchanges in the boundary layer on daily scale and on diurnal scales in the polar regions.
The isotope signals on the near-surface snow and firn undergo fractionation due to sublimation and condensation processes as observed both in-situ and laboratory studies (Stichler et al., 2001;Sokratov and Golubev, 2009). Apart from the temperature-40 based fractionation, other factors such as relative humidity and wind speed also influence the post-depositional variations in the ice core isotope signals.
The climate records in ice cores reflect the conditions at which snowfall occurred. These records are influenced by various parameters such as the season at which the precipitation occur and variations in sources of moisture for precipitation. In order to better interpret the ice core stable isotope records, it is important to understand their present variability and interrelationships based on topography as mentioned in detail in our earlier publications (Mahalinganathan et al., 2012;Mahalinganathan and Thamban, 2016). The snow cores were collected using KOVACS Mark IV system and each core was typically 1 meter long and 14 cm diameter. After recovery, the snow cores were transferred directly into pre-cleaned high-density polyethylene bags and sealed immediately to avoid contamination during storage and shipped under frozen conditions to the National Centre for Polar and Ocean Research, India. The cores were stored at -20 • C till analysis. The snow cores were sub-sampled at a 5 cm resolution 60 under a clean laminar hood. A total of 667 sub-samples were processed where the outer layers were removed manually by a clean ceramic knife and the innermost cube was utilized for density measurements from which the snow accumulation rate values were calculated (Mahalinganathan et al., 2012). A portion of each sub-sample was transferred to a 10 mL teflon vials and sealed for further stable isotope analyses.
All sub-samples were melted and immediately analysed for δD and δ 18 O values in the Ice Core Laboratory of National Centre for Polar and Ocean Research using a dual inlet, Isoprime Isotope Ratio Mass Spectrometer, following standard procedures (Naik et al., 2010) which had an external precision of 0.05 ‰ . Fresh samples were also analysed using a unique laser-based Off-Axis -Integrated Cavity Output Spectroscopy (OA-ICOS) Triple Isotope Water Analyser (TIWA) by Los Gatos Research.
The melted samples were introduced directly into the TIWA using a Hamilton 1.2 µL zero dead volume syringe via an auto injector equipped with a heated ( 85 • C) injector block. In order to eliminate inter sample memory effect, four preparatory in-70 jections (injections which are not measured), followed by five measurement injections were run for each sample. The last five injections were averaged to produce a single, high-throughput sample measurement. The analytical precision of measurements for δD was ±0.5‰ and δ 18 O was ±0.1‰ using the TIWA system. Details of sampling locations and analytical results are provided in tables 1 and 2 for cDML and PEL, respectively.
The seasonality in snow cores was determined by establishing the summer and winter peaks in isotope records where a 75 minimum amplitude of 4‰ between summer and winter was used to differentiate these peaks as detailed in figure 2 our previous publications (Mahalinganathan et al., 2012;Mahalinganathan and Thamban, 2016). Near-surface temperature measurements were estimated for the year 2008 from Regional Atmospheric Climate Model version 2.3 (RACMO 2.3) output with a horizontal resolution of 27 km (Wessem et al., 2014). Isotope results from an ice core (IND33/B8, ∼101 m) drilled close to the cDML transect was used in order to estimate the diffusion of stable water isotopes in firn. The firn diffusion model originally described  et al. (2015) and Münch and Laepple (2018). These calculations were performed using the measurements from the ice core site (site specific parameters) with an average firn density of 340 kg m −3 , a mean surface temperature of -22 • C, local mean surface pressure of 800 mbar and a local accumulation rate of 300 kg m −2 yr −1 . The density and local accumulation rates were directly measured from the snow cores and ice core data while the temperature and local mean surface pressure values were 85 derived from the RACMO output.
Stable isotope values and related parameters for snow core sites from both cDML and PEL transects are provided in Tables 1 and 2. All snow cores from both transects showed clear seasonal variations in stable isotope values except for the cores at 95 240 km in cDML transect and 70 and 100 km in PEL transect, where seasonality could not be established, presumably due to core damage during transportation.
The relationship between δD and δ 18 O from all snow cores were assessed using linear regression analysis (Fig. 2). The analysis showed a strong correlation between δD and δ 18 O in both cDML (orange circles) and PEL (green circles) transects from coastal to inland region. The slope of the PEL transect (8.12) was close to that of the global meteoric water line (Craig,  Deuterium excess (d) was calculated for each 5 cm sub-sample of all the snow cores. The d values varied between -1.97‰ 105 and 17.18‰ in cDML and -3.43‰ and 11.10‰ in PEL region. The relationship between near-surface temperature and 'd' was analysed using linear regression method which showed a significant relationship for samples from cDML region with the regression equation: d = -0.11 x T + 2.78. (R = -0.41, p < 0.01). For the PEL region however, no significant relationship was found ( Fig. 4).
Results from the isotope firn diffusion calculations using the R package by Münch and Laepple (2018) revealed a diffusion 110 length of 6 cm over a period of 5 years (Fig. 5).
Five-day back-trajectory frequency maps of coastal, mountainous and inland locations showed vast differences in the sources between summer and winter (Fig. 6). During winter, the air parcels to cDML coast arrived from Weddell Sea and south Atlantic ocean while during summer the trajectories were mostly arriving from the Indian Ocean. Similarly, the air parcels to cDML inland arrived from the Weddell Sea, south Atlantic and the Indian Ocean. The air mass to PEL arrived predominantly from 115 south Indian Ocean during summer and winter.

Spatial variability of snow accumulation and stable water isotopes
Spatial variations of snow accumulation in Antarctica are primarily due to the presence of physical barriers during snowfall and snow redistribution post deposition (Melvold et al., 1998;Vaughan et al., 1999). The accumulation rates in cDML re-120 gion showed a large spatial variation with the near-coastal section having a substantially high average accumulation of 354 kg m −2 yr −1 (Table 1). Compared to this, the interior section of the transect had less than half of the accumulation rate than the coast (average 155 kg m −2 yr −1 ), while the mountainous section showed moderately high accumulation rate (average 235 kg m −2 yr −1 ). Such large spatial variability in cDML region can be attributed to the presence of extensive mountain chains existing parallel to the coast. These mountain chains in cDML act as a physical barrier to the air masses arriving from the 125 Southern Ocean impacting the snow accumulation and redistribution. As a result, the study area could be separated into three distinct accumulation regimes. The physiography and topography of the cDML region evidently influenced the snow accumulation rates showing a strong correlation with distance and elevation (Table 3). On the contrary, the PEL transect showed moderately high accumulation with little variation between the coastal (276 kg m −2 yr −1 ) and the inland (260 kg m −2 yr −1 ) sections. Although there exist substantial slope (> 8 m km −1 ) in the coast of PEL (Mahalinganathan et al., 2012), it did not 130 affect the overall accumulation rates in the region. Such uniform accumulation pattern in PEL transect could be explained by snowdrift and redistribution induced by strong katabatic winds in this region (Allison, 1998;Ding et al., 2020). With no orographic barriers to influence the flow of katabatic winds, the accumulation pattern in PEL tend to get smoothened by the wind scouring and snow redistribution.
The mean values of δ 18 O and δD from each core location decreased from coastal to inland in both cDML and PEL regions 135 suggesting the typical continental effect, i.e. depletion of δ 18 O and δD with the increasing distance and elevation from the moisture source (Tables 1 and 2). However, the multiple regression models using the geographical parameters and δ 18 O showed negligible variance with distance and elevation (Table 4). Based on these regressions in cDML region, the snow accumulation changes were found to be the primary driver for spatial variations in δ 18 O and δD values. Surface temperature also had noticeable effect on isotopic values in this region (Table 4). However, in the PEL region, the snow accumulation did not play 140 any role in the spatial variations of δ 18 O and δD while temperature played a crucial role. The distance from coast and elevation are known to impact the isotopic composition variation in Antarctic snow (e.g. Huybrechts et al., 2000;Masson-Delmotte et al., 2008). However, there are large regional variations and within short transects, such relationships may not be linear. Since accumulation and temperature parameters are directly impacted by the distance from the coast and elevation, they have indirect influence on the isotopic change.

145
The relationship between δ 18 O and δD was determined using all the snow cores and the local meteoric water lines (LMWL) for this region were calculated (Fig. 2). The δ 18 O values in cDML region (orange circles) were more depleted than that of the PEL region (green circles). Though the coastal sections in cDML and PEL appears to be almost on the same latitude (∼ 69 • S), a difference of 6 ‰ was observed between the coasts. This could be due to the fact that the sampling in cDML begin from 110 km inland while the sampling in PEL begins 10 km from the open ocean. The slope of the LMWL in cDML (7.9) is lower 150 5 https://doi.org/10.5194/tc-2020-77 Preprint. Discussion started: 7 April 2020 c Author(s) 2020. CC BY 4.0 License. than that of the global meteoric water line (GMWL) while the slope of LMWL in PEL (8.12) had a slope close to GMWL. The cDML transect, therefore, have a slope closer to that of the Antarctic meteoric water line (AMWL), while the PEL transect have a slope closer to that of the GMWL. All the samples were utilized to observe the δ 18 O and δD relationship which did not appear to show any deviation from the AMWL.
The δ 18 O-temperature relationship vary depending on the location of the sampling sites. An Antarctic wide study by covers a small region from coast and inland. In spite of such complications, the results from the present study area (Fig. 3) showed that with an increase of every δ 18 O‰ value, there was an increase in the temperature by 0.7 • C in cDML and 1 • C in PEL. Therefore, the proposed spatial slope (0.80‰ / • C) by Masson-Delmotte et al. (2008) seems to be reasonable.
The deuterium excess (d) values did not show any correlation with any geographical parameters in both cDML and PEL transects (Table 3). An Antarctic-wide compilation of datasets by Masson-Delmotte et al. (2008) showed that even though 165 both elevation and distance appear to generally control d, results from locations below 2000 m elevations did not produce any satisfactory correlation. The PEL transect in this study had only two sampling sites above 2000 m elevation, while the d above 2000 m altitude in cDML transect did not show any meaningful correlation with the geographic parameters. This could be due to involvement of other factors such as regional differences in moisture sources (Simmonds et al 2003). A significant negative correlation was observed between surface temperature and d in cDML (Fig. 4). This negative correlation could be 170 associated with various factors such as the changes in moisture sources, kinetic fractionation on ice crystals at supersaturation point (Jouzel and Merlivat, 1984) and increase of d along the distillation path (Masson-Delmotte et al., 2008).

Extent and impact of isotope diffusion
Stable water isotope records undergo rapid diffusion in snow and firn than in solid ice (Johnsen, 1977;Whillans and Grootes, 1985). As a result, the seasonal amplitude of these records is homogenized in firn -resulting in a considerable reduction in the 175 summer and winter δ values. Since these records are the primary tool in understanding the relationship between isotope ratios and temperature, it is important to estimate the amount of isotope diffusion.
In order to understand the extent of such diffusion in a high accumulation region like the cDML transect, we compared the isotope records of a snow core (cDML 9) with that of a shallow ice core close to the snow transect, IND-33/B8 (Fig. 1). This ice core was drilled 5 years after the snow cores were retrieved (2013-14 Summer). The detailed stable isotope records and 180 chronology of this ice core is discussed in an upcoming paper (Tariq et al., 2020, unpublished). A direct comparison of δ 18 O values of the ice core to a snow core from the transect showed a considerable amount of diffusion in the seasonal amplitude of δ 18 O (Fig. 5). The snow cores originally had an amplitude >5‰ while the ice core seasonal amplitude was reduced to 2‰ .
Further, to analyse the extent and impact of isotope diffusion in this core site, the diffusion length was estimated using a firn diffusion model (Johnsen et al., 2000), using the site specific temperature (-22 • C), surface firn density (340 kg m −3 ), local 185 mean surface pressure (800 mbar) and the local accumulation rate (300 kg m −2 yr −1 ). The degree of smoothing of the stable water isotope record depends on the isotopic diffusion length which is shown to increase in the topmost firn layer (Johnsen et al., 2000). Differential diffusion (σ) was calculated using the equation σ = σ 2 (z 2 ) − σ 2 (z 1 ), where z 1 and z 2 are initial and final depths (Münch et al., 2016;Münch and Laepple, 2018). Results from firn diffusion model on the 33/B8 ice core showed a diffusion length of around 6 cm in the surface to a depth of 4 m (Fig. 5, inset). The calculation showed a rapid increase in 190 diffusion length from the surface before attaining a maximum at 30 meters. This findings can be used during interpretation of stable isotope record of deep ice cores by removing the effects of diffusion on inter-annual variability.

Sources of moisture for precipitation in the region
The sources of moisture in Antarctica is widely distributed and strongly influenced by topography, sea-ice conditions and midlatitude land-ocean contrasts (Sodemann and Stohl, 2009). Moisture source regions could be identified using several techniques.

195
For instance, d was used as a tracer in isotope models by Petit et al. (1991) and Ciais and Jouzel (1994). An attempt was made to understand the moisture origin by tracing backwards in time, the air parcels from different sampling sites.
Five-day air parcel backward trajectory frequency map showed moisture source regions close to 60 • S latitude during summer and up to 50 • S latitude during winter (Fig. 6). The cDML region had moisture sources from the Weddell sea and south Atlantic Ocean during winters while during summer, the sources were much restricted to the southern Atlantic and Indian Ocean. This  Competing interests. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.