In January 2013 the Vostok megadune area was visited for the first time. The accumulation-stake profile was established perpendicular to the megadune crest. The total number of stakes was 21 (named MD00 to MD20), the distance between adjacent stakes was about 100 m and the total length of the profile was 1983 m (Fig. 1). The samples of the upper 1.5 m of snow were also taken near each stake to be analysed for the concentration of the stable water isotopes (δD, δ18O and δ17O).

In January 2014 and January 2015 the stakes were revisited, and the repeated measurements of their heights and surface-snow density allowed us to obtain the amount of snow accumulated over two years (January 2013–January 2015). The snow samples were taken again for chemical and isotopic analyses.

In January 2015 we also drilled a 20 m borehole in point MD00 (Fig. 1). In the obtained firn core we measured snow density and took samples for stable water isotope analysis with a resolution of 10 cm.

The concentration of heavy water isotopes (δD and δ18O) in 42 surface-snow samples taken in 58th and 59th RAE, as well as in 183 samples from the MD00 core, were measured at the Climate and Environmental Research Laboratory (CERL) using a Picarro L2120-i analyser. Our working standard (VOS), measured after every five samples, was made of the light Vostok snow and calibrated against the International Atomic Energy Agency (IAEA) standards VSMOW-2, GISP and SLAP-2. The reproducibility of results defined by remeasurements of randomly chosen samples was 0.04 ‰ for δ18O and 0.2 ‰ for δD, which is 2 orders of magnitude less than the natural variability of the snow isotopic composition (see below) and thus satisfactory for the purposes of the study.

In October 2015 we also measured 17O-excess values in the samples collected during the 59th RAE using a Picarro L2140-i analyzer. For this, each sample was measured 15 times in the high-precision mode and we took an average of the last 10 measurements. For every three samples we measured the VOS standard previously calibrated against VSMOW-2, GISP and SLAP-2 (taking into account that they have 17O-excess values of, correspondingly, 0, 22 and 0 per meg; Schoenemann et al., 2013). The 17O-excess value of VOS was found to be 0 per meg, similar to SLAP. The reproducibility of the 17O-excess values of individual samples was 8 per meg.

All the data discussed in this paper are presented in Table 1.

The absolute altitude and location were determined with high accuracy along the profile using the geodetic GPS technique.

Therefore the Bernese GPS Software 5.1 (Dach et al., 2007) was used to process the two kinematic GNSS profiles (K58B and K58C) as a combined differential solution from the GPS and GLONASS observations. As reference stations for this network solution, we used two local receivers at Vostok station at a distance of 30 km and two additional reference stations at the coast with baseline lengths of about 1350 km (near the Russian research stations Progress and a station of the global IGS-network near the Australian Station Casey). After reducing the antenna positions to the snow surface we estimated the accuracy by calculating the height differences at track crossovers. Crossovers inside one track imply an internal RMS with differences of 2.9 cm (at 160 crossover points) for track K58B and 2.4 cm (at 9 crossover points) for track K58C. With a RMS of 7.7 cm at 5159 crossovers between the two tracks, we can estimate the absolute precision of a single kinematic surface height according to the variance propagation to about 5.4 cm.

During the austral summer field season of the 58th Russian Antarctic Expedition (January 2013), the ground penetrating radar (GPR) profiling was performed to study the snow-firn layer structure of the megadune area. The 200 MHz GSSI SIR10B GPR with 5106 200 MHz antenna was applied.

The GPR equipment was installed on two sledges towed by a Ski-Doo. The route of the GPR profiling was the same with the geodetic observations. In total, about 80 km of the GPR profiles were obtained (Fig. 1), though in this paper we only use the 2 km section obtained along the glaciological profile (points MD00-MD20 in Fig. 1).

The main problem with the processing and interpretation of the GPR data is the dielectric properties of the media where the electromagnetic waves propagate. We used the model published in Popov and Eberlein (2014) to transform the radio-echo time section (Fig. 2) into the depth section. To calculate the vertical electromagnetic wave speed, we used the firn density data measured in the MD00 core.

In Fig. 2 we show the GPR registration recorded along the glaciological profile (MD00-MD20).

Due to the subsequent measurements of the heights of the stakes established across the megadunes, we were able to define the snow accumulation between January 2013 and January 2015 (Fig. 3e). One may clearly see the regular spatial variability of the snow build-up. In accordance with the previous studies (see the review in Introduction), the snow is removed from locations with the increased surface slope (leeward side of the dunes) and deposited where the slope is decreased or inversed (hollow between dunes and windward side of the dunes). In a distance of few hundred metres the accumulation changes by an order of magnitude, from -0.5 to 16 cm of snow (or from -0.2 to 58 mm w.e. according to the surface-snow density, Fig. 3d). This range is larger than that reported by Frezzotti et al. (2002b) (from 7 to 35 mm). The mean annual accumulation over one dune wavelength is 22 mm, which is very similar to the accumulation at the Vostok stake network (23 mm). If the precipitation rates at Vostok station and in the megadune area were the same, then our result does not support the observation that over the megadune areas the accumulation is reduced due to the wind-driven sublimation (Frezzotti et al., 2004).

Snow accumulation variability observed at the stakes during only two years of observations may not adequately represent the long-term average due to a very large random component. We used the GPR data (Fig. 3g) and data on firn density from the 20 m core in order to evaluate spatial variability of the multi-year average of the snow accumulation rate. The first internal reflection horizon (estimated age is about 130 years, see below) is located at the depth that varies from 3 to 11 m. Thus, the mean 130-year snow accumulation rate varies between 1 and 35 mm w.e. over one full dune wavelength, with an average of 21 mm w.e. Thus, the multi-year spatial variability of snow accumulation rate is considerably smaller than that obtained from 2-year stake measurements, but still larger than that reported by Frezzotti et al. (2002b).

The surface-snow density does not show any distinct spatial variability (Fig. 3d). The mean snow density (0.355 g cm-3) is slightly higher than that measured at Vostok stake network (0.33 g cm-3).

During the fieldwork seasons we also did not observe big differences between snow surface character in leeward and windward slopes of the dunes (Fig. 4). The erosion zone does not demonstrate the dominance of the glaze surface, and no big sastrugi are observed in the accumulation zone, as reported by Frezzotti et al. (2002b) for the Victoria Land megadune field. The small difference in snow morphology between low- and high-accumulation zones of the dune may be related to the relatively small dune size, although it does not explain rather big spatial variability of snow accumulation.

The spatial variability of the accumulation rate covaries well with the surface slope: the smaller the slope, the higher the accumulation (Fig. 3f and e).

In Fig. 3a–c we show the isotopic composition (δD, dxs and 17O-excess) of the upper 1.5 m of snow sampled twice near each point of our profile. One can see a wave in snow isotopic content (δD) with a magnitude of about 25 ‰ and a wavelength similar to that of the megadune. The spatial variability of the snow isotopic composition in the megadune area is significantly larger than in the area outside the megadunes. For example, the standard deviation of the δD values in the samples representing the upper 1.5 m of snow and taken along a 40 km profile in the southern part of Lake Vostok, where the glacier is characterized by a flat surface, is ± 4 ‰ (Ekaykin et al., 2012) against ± 6 ‰ in the megadune area. Another difference is that outside the megadunes the variability of snow isotopic composition is random, with no clear periodicity. At the same time, the average values of δD are similar in both cases, around -440 ‰.

There is a negative covariation between snow isotopic composition and accumulation rate (correlation coefficient, -0.38, is not statistically significant due to the small number of points).

We explain this behaviour of the snow isotopic composition by different post-depositional alteration of the initial isotopic composition of snow precipitation deposited in the low- and high-accumulation zones of the megadunes. Indeed, the erosion zone is characterized by “a long, multiannual, steep temperature–gradient metamorphism” (Frezzotti et al., 2002b), p. 8. Thus, the snow here should be enriched in heavy isotopes due to strong post-depositional modification (Town et al., 2008) that may be further facilitated by the increased permeability of the snow in such locations (Albert et al., 2004). We may speculate that in the megadunes described by Frezzotti et al. (2002b), characterized by a very strong modification of the snow physical properties, the isotopic transformation should be even stronger than in the Vostok dunes.

An alternative explanation of the negative spatial relationship between isotopic composition and snow accumulation was suggested by Ekaykin et al. (2002). Since in winter the snow crystals are smaller and wind speed is higher, this snow could be more easily redistributed by wind comparing to snow precipitated in summer. If so, in the erosion zone of the megadunes the proportion of summer snow is larger than in the accumulation zone.

We may use the isotopic data to determine which mechanism, post-depositional, redistributional or both, is mainly responsible for the anomaly of the snow isotopic composition in the megadune area.

The observed dxs/δD and 17O-excess / δD slopes (ratios between the standard deviations of the smoothed profiles shown in Fig. 3a, b and c) are correspondingly -0.2 ‰ ‰-1 and 0.9 per meg ‰-1. During the post-depositional changes of the snow isotopic composition, these slopes are -0.2 ‰ ‰-1 and 0.4 per meg ‰-1 (Ekaykin et al., 2016). During the seasonal cycle of the isotopic composition of snow precipitation at Vostok, these parameters are related by slopes -0.1 ‰ ‰-1 and 0.4 per meg ‰-1 (Landais et al., 2012).

Thus we may conclude that the mixing of the summer and winter precipitation in different proportions cannot explain the variability of snow isotopic composition observed in the megadune area, since the variability of 17O-excess and of dxs in this case would be significantly smaller. The post-depositional factor would better explain the observed snow isotopic composition in the megadunes, but the variability of 17O-excess still seems to be too strong. Note, however, that the data by Ekaykin et al. (2016) were obtained in laboratory experiments, not in natural conditions, so the 17O-excess / δD slopes reported there may be underestimated.

We should also note that the 17O-excess values positively covariate with the accumulation rate (Fig. 3c and e), although one would expect a negative covariation if the snow isotopic composition variability is due to the post-depositional processes. At present we do not know whether this positive covariation is caused by errors in the 17O-excess values, or if it suggests another mechanism that creates isotopic anomalies in the megadune area.

We used the GPR data to reconstruct the previous positions of the dunes and calculate the velocity of the dune drift.

In the GPR profile taken across the megadunes (Fig. 2), we see several distinct internal reflection horizons (IRH). For seven of them we defined the depths (Table 1) and, subtracting these depths from the ice sheet elevation, we could define the absolute altitude of each IRH (Fig. 3g). Thus we can see the buried surfaces of our dune and may trace its drift in time.

For this, we first need to date each IRH. We used the density profile obtained from the MD00 core and the average snow accumulation rate in the megadune area in order to calculate the depth–age function and determine the age of each IRH (Fig. 3g). The uppermost IRH marks the surface of the dune about 130 years ago and the lowermost 530 years ago.

We chose three fold hinges (the summit of crests and two lowest points of the fold dips) to trace the dune drift. On average, the dune drifts upwind at a rate of 4.6 ± 1.1 m yr-1. This corresponds very well with the value reported by Frezzotti et al. (2002b), 5 m yr-1. With this velocity, a dune drifts by one full wavelength in about 410 years.

As pointed by Arcone et al. (2005), the apparent dune drift velocity observed in the GPR images is a combination of the real dune velocity and the ice movement, and the real dune velocity is higher than the observed one. According to (Richter et al., 2013), ice flow velocities in this region do not exceed 2 m yr-1, so the real dune drift velocity could be up to 6.6 m yr-1. However, the ice is moving almost in parallel with the dune crests (from north-west to south-east). So, the projection of the ice speed vector on the MD profile is close to 0 m yr-1, and the correction to the dune drift velocity due to the ice movement should be close to zero, too. Finally, for the purposes of our study we need not real, but the resultant dune velocity observed by GPR, so in the further calculations we use the apparent dune velocity of 4.6 m yr-1.

When the dune drifts, the spatial anomalies in snow physical properties, accumulation rate and isotopic composition drift accordingly. For example, the point MD00, now located at the leeward side of the dune with reduced accumulation and enriched isotopic composition, was in the hollow between dunes with increased accumulation and lower heavy isotope content about 300 years ago. If one drills an ice core in the megadune area, he would see a quasi-periodic (with the period of 410 years) oscillations in snow accumulation and isotopic composition related to the dune drift.

Let us simulate an oscillation such as we would see in a core drilled at the MD00 point.

In Fig. 5a we showed a temporal variability of snow accumulation (blue) and isotopic composition (red) that should be seen in point MD00 when the dune crosses it. To construct these curves, we simply divided the distance of each point in Fig. 3a and e by the above-mentioned velocity of the dune drift. We also showed the climatic variability in the Vostok region over the same time interval (Ekaykin et al., 2014) in Fig. 5a in purple. Note that the amplitudes of the both components are similar.

Combining the dune-related and climatic components, we obtain the expected temporal variability of snow isotopic composition in MD00 (Fig. 5b). Then we transform it to a vertical isotopic profile using the depth–age function presented in Fig. 5c. This function takes into account the significant dune-related variability in snow accumulation rate (Fig. 5a), which is why it is essentially non-linear. The resulting simulated vertical isotopic profile is presented in Fig. 5d by a red line. In the upper part of the profile the isotopic oscillations are compressed due to the low accumulation rate, and when the accumulation is higher, they are stretched.

In our simulations we do not take into account that the megadune snow may experience enhanced diffusive smoothing due to the increased ventilation of the snow column. In this case the isotopic profile could be substantially smoothed.

In Fig. 5d we also showed the real isotopic profile from the MD00 core with the blue line. One can see a good resemblance between them, except for the very bottom part of the MD00 core. Indeed, at a depth of 20 m the core has a δD isotopic composition of -420 ‰. Neither the climatic record nor the isotopic profile from the megadunes (Fig. 5a) demonstrate such high isotopic value, which is why we could not reproduce it in our simulations.