Stratigraphic noise and its potential drivers across the plateau of Dronning Maud Land, East Antarctica

Abstract

Stable water isotopologues of snow, firn and ice cores provide valuable information on past climate variations. Yet single profiles are generally not suitable for robust climate reconstructions. Stratigraphic noise, introduced by the irregular deposition, wind-driven erosion and redistribution of snow, impacts the utility of high-resolution isotope records, especially in low-accumulation areas. However, it is currently unknown how stratigraphic noise differs across the East Antarctic Plateau and how it is affected by local environmental conditions. Here, we assess the amount and structure of stratigraphic noise at seven sites along a 120 km transect on the plateau of Dronning Maud Land, East Antarctica. Replicated oxygen isotope records of 1 m length were used to estimate signal-to-noise ratios as a measure of stratigraphic noise at sites characterised by different accumulation rates (43–64 mm w.e. a${}^{-\mathrm{1}}$), snow surface roughnesses and slope inclinations. While we found a high level of stratigraphic noise at all sites, there was also considerable variation between sites. At sastrugi-dominated sites, greater stratigraphic noise coincided with stronger surface roughnesses, steeper slopes and lower accumulation rates, probably related to increased wind speeds. These results provide a first step to modelling stratigraphic noise and might guide site selection and sampling strategies for future expeditions to improve high-resolution climate reconstructions from low-accumulation regions.

1 Introduction

The East Antarctic ice sheet is a valuable climate archive. Stable water isotopologues in snow, firn and ice core records store information on past temperature variations e.g., especially on past glacial and interglacial periods . However, the interpretation of isotope records at subannual to decadal resolutions is hampered by significant uncertainties e.g. which limit the usability of this archive to quantify recent global warming impacts on the Antarctic Plateau . Several processes, such as precipitation intermittency , water vapour exchange and sublimation e.g., induce noise in the isotopic temperature imprint in snow, firn and ice cores. Additionally, accumulation does not occur as perfectly stratified layers but is affected by irregular depositions as well as wind-driven erosion and redistribution . These processes introduce a non-climatic variability into the isotope profiles, known as stratigraphic noise , which reduces the usability of single isotope profiles for climate reconstructions e.g.. Isotopic diffusion in snow and firn furthermore smoothes the overall isotopic variations, leading to a loss of high-frequency variations e.g..

While atmospheric circulation, temperature and precipitation intermittency introduce variations in the isotopic compositions on scales of hundreds of kilometres , stratigraphic noise has been defined as the uncorrelated part between two or more isotope profiles at local scales . found the decorrelation length of stratigraphic noise to be around 5 to 10 m on the plateau of Dronning Maud Land (DML), which means that the stratigraphic noise in one snow core is independent of the noise of an adjacent profile at a distance of more than 5–10 m.

Stratigraphic noise hampers the extraction and interpretation of climate signals, especially on subannual to decadal scales where accumulation rates are low e.g.. For instance, with accumulation rates of 240–600 mm w.e. a${}^{-\mathrm{1}}$ in Greenland, it is possible to extract seasonal and annual signals from the isotopic imprints in firn cores . However, in DML, with low accumulation rates of 40 to 90 mm w.e. a${}^{-\mathrm{1}}$ , seasonal snow layers can be completely eroded . Accordingly, discrepancies between isotope profiles and observed temperatures are smaller in coastal areas of East Antarctica compared to the drier inland plateau of DML . This is further evidenced by the signal-to-noise ratios (SNRs), which measure the ratio of the common (spatially coherent) isotopic signal over the independent stratigraphic noise: while found SNRs between 1.1 and 2.7 (at 140–520 mm w.e. a${}^{-\mathrm{1}}$) for annually resolved ice core records from Greenland, obtained a SNR of only 0.14 on the plateau of DML.

Although several studies have identified stratigraphic noise as a crucial limiting factor for high-resolution ice core signal interpretation, it remains unknown how stratigraphic noise differs spatially, e.g. across the East Antarctic Plateau (EAP), and how it is related to local environmental properties like the accumulation rate, slope inclination and surface roughness. Knowledge of such relationships would allow us to optimise the selection of sampling sites for extracting snow and firn cores for the purpose of high-resolution climate reconstructions. Less stratigraphic noise from optimal sites would enhance the effective resolution at which a climate signal can be extracted . Furthermore, this knowledge would enable stratigraphic noise to be simulated in proxy system models . An improved quantitative understanding would also result in more accurate estimates of past climate variability as it would allow us to correct for stratigraphic noise within the spectral domain e.g.. It might further enable the use of replicate cores as a proxy for past surface roughnesses and related wind speeds .

In this study, we use SNRs to quantify stratigraphic noise in high-resolution isotope records collected from seven sites in DML (EAP). We relate differences in stratigraphic noise to varying local environmental properties such as slope inclination, surface roughness and the accumulation rate in order to identify potential underlying environmental drivers.

Figure 1

(a) Study area on the plateau of Dronning Maud Land (DML), East Antarctica, with sampling sites D2, C4, C5, D7, D24 and D38 (red triangles) and Kohnen trenches (red circle) . Shading represents the surface elevation above sea level RAMP2;. (b) Example schematic illustrating the collection of replicate snow cores at each sampling site. (c) Wind speeds and direction at Kohnen Station from the local automatic weather station AWS 9, 2012–2018;.

[Figure omitted. See PDF]

2 Materials and methods

2.1 Study area

The sampling sites are situated on the EAP along a 120 km transect that rises gently from 2685 to 2892 m a.s.l. near Kohnen Station 75.002${}^{\circ }$ S, 0.007${}^{\circ }$ W; 2892 m a.s.l.; (Fig. a). Between 2012 and 2018, the annual mean temperature at Kohnen Station was $-\mathrm{40.9}$ ${}^{\circ }$C and the annual mean wind speed was 4.31 m s${}^{-\mathrm{1}}$ from an automatic weather station, AWS 9;. The region is characterised by katabatic winds consistently blowing downslope e.g. from north-easterly directions (Fig. c), with wind speeds in excess of 10 m s${}^{-\mathrm{1}}$ occurring about 10–20 times per year . Only a few precipitation events, introduced by low-pressure systems, bring most of the annual precipitation amount . Average accumulation rates over the past 200 years have been $\sim \mathrm{60}$ mm w.e. , possibly with a considerable increase during the last century . The snow surface around Kohnen Station is mainly comprised of small dunes and sastrugi with horizontal scales of the order of a few metres .

2.2 Snow core sampling

A set of snow profiles was sampled in December 2018 at six sites along a $\sim \mathrm{120}$ km transect south-west of Kohnen Station (Fig. a). At each of these sampling sites (D2, C4, C5, D7, D24 and D38) six 1 m snow cores were extracted along a line running perpendicularly to the dominant large-scale wind direction with a 10 m interprofile spacing – five of them were further processed and used in this study (Fig. b). The direction was chosen to allow a comparison with Kohnen trench studies . As the snow dunes in the study region are predominantly parallel to the wind direction, measuring perpendicularly to the wind ensures better sampling of the dunes along the 60 m overall distance. Each snow profile was extracted by vertically inserting a 1 m carbon fibre pipe into the sidewall of a snow pit. The collected snow profiles were cut horizontally into slices of 1.1 cm (for the upper 16.5 cm) and 3.3 cm (for the lower part), accounting for the diffusion length of $\sim \mathrm{3}$ cm at 1 m depth . Compression or expansion during handling, transport and cutting of the snow cores resulted in a maximum depth uncertainty of 2 cm and slight variations in the number of samples per profile (41–43). Combined with the maximum uncertainty of 1 cm resulting from the snow height measurements (Sect. ), the absolute depth values have a combined maximum uncertainty of 3 cm. All snow samples ($N=\mathrm{1249}$) were packed in plastic bags and transported in frozen state to Germany for further analysis.

2.3 Stable water isotope measurements

The stable water isotopic composition (${\mathit{\delta }}^{\mathrm{18}}$O, $\mathit{\delta }$D) of the snow samples was measured using a cavity ring-down spectroscopy instrument of Picarro, Inc. (model L2140-i) in the Laboratory for Stable Isotopes at the Alfred-Wegener-Institut in Potsdam, Germany. Post-run corrections were applied as described in . Scaling to the VSMOW–SLAP (Vienna Standard Mean Ocean Water–Standard Light Antarctic Precipitation) scale results in the $\mathit{\delta }$ notation which describes the ratio of heavy to light isotopes in per mille (‰). In-house standards were used for quality control. The mean combined measurement uncertainty is 0.07 ‰ for ${\mathit{\delta }}^{\mathrm{18}}$O and 0.5 ‰ for $\mathit{\delta }$D (root-mean-square deviation, RMSD). In the following we focus on the ${\mathit{\delta }}^{\mathrm{18}}$O values.

2.4 Trench isotope subset

To complement the datasets from the six sites along the transect, we use already published ${\mathit{\delta }}^{\mathrm{18}}$O profiles from Kohnen Station (Fig. a) derived in the years 2012/13 and 2014/15 by . Four snow trenches (Kohnen trenches) were excavated by a snow blower perpendicularly to the local snow dune direction. Snow profile samples were collected off the trench walls, resulting in snow profiles of high vertical as well as horizontal resolution (Table ). For the comparison of the trench data with the new dataset, we divided the trench data into 10 subgroups, each composed of four to five 1 m deep ${\mathit{\delta }}^{\mathrm{18}}$O profiles at distances of $\mathrm{10}\pm \mathrm{1}$ m from one another (Table ).

Table 1

Summary of the datasets which are used within this study. Subgroups of the Kohnen trenches, described in Sect. , are used to compare the resulting signal-to-noise ratios (SNRs) to the ones of locations D2, C4, C5, D7, D24 and D38.

Site	Sampling time	Spacing [m]	Depth [m]	Resolution [cm]	Number of profiles	SNR subgroups
D2, C4, C5, D7, D24, D38	Dec 2018	10	1	1.1–3.3	5 each	–
Trench T13-1${}^{\mathrm{a}}$	2012/13	0.1–2.5	1.2	3	38	5
Trench T13-2${}^{\mathrm{a}}$	2012/13	$\sim \mathrm{10}$–20	1.2	3	4	1
Trench T15-1${}^{\mathrm{b}}$	2014/15	5	3.4	3	11	2
Trench T15-2${}^{\mathrm{b}}$	2014/15	5	3.4	3	11	2

${}^{\mathrm{a}}$ . ${}^{\mathrm{b}}$ .

2.5 Definition and quantification of stratigraphic noise

The variations that are independent between adjacent isotope records are considered to be noise, while the signal is the isotope variations that those records have in common. The ratio of the common (signal variance) to the independent (noise variance) portion in the given data is the signal-to-noise ratio (SNR). This ratio can be derived based on the pairwise correlation coefficient, $r_{\mathrm{XY}}$, between two isotope profiles $X$ and $Y$, with

1$${\mathrm{SNR}}_{XY}={\displaystyle \frac{r_{XY}}{\mathrm{1}-r_{XY}}}$$ . In the absence of any noise, we would expect $r_{XY}=\mathrm{1}$, i.e. a perfectly stratified isotopic imprint. As the amount of stratigraphic noise increases, the pairwise correlations and hence the SNRs will decrease. The SNR thus provides a quantitative measure to objectively determine the proportion of stratigraphic noise in adjacent (intrasite) isotope records and to make intersite comparisons. At each sampling site, SNRs were estimated based on the mean pairwise correlation coefficient ($n=\mathrm{10}$) between the five ${\mathit{\delta }}^{\mathrm{18}}$O profiles (linearly interpolated to a 0.1 mm resolution) with respect to their absolute height reference (Appendix ). The SNR of the Kohnen trenches was determined from the mean of all pairwise correlations of all records of all subgroups ($n=\mathrm{95}$). Due to statistical uncertainty, SNR estimates can be negative; as this has no physical meaning, SNR estimates $<\mathrm{0}$ are set to 0. We assess the uncertainty in the SNR estimates by employing a bootstrap resampling procedure: we resample the pairwise correlation coefficients with replacement, calculate the SNR from the resamples and derive the confidence intervals from the distribution of the SNRs.

2.6 Environmental properties

We compare the amount of stratigraphic noise in the isotope records to the local slope inclination, accumulation rate and surface roughness.

Slope inclinations are hereby defined as the local inclination of 10 km long segments along the transect (azimuth of $\sim \mathrm{57.1}$${}^{\circ }$). They were derived using 200 m resolution data from the REMA digital elevation model DEM; and vary between 0.3 and 3.1 m km${}^{-\mathrm{1}}$. To assess the uncertainty in these estimates, we calculate the slope inclinations over 10 km segments with the same azimuth across 36 different points located at 200, 400 and 600 m around each study site (12 different directions in steps of 30${}^{\circ }$) and extract the SD of these slope inclinations.

We use the average accumulation rates over the last 200 years as determined by for the same transect using ice-penetrating radar. For sites D2, D7, D24 and D38, they vary between $\sim \mathrm{43}$ and $\sim \mathrm{59}$ mm w.e. a${}^{-\mathrm{1}}$ and are thus lower than the $\sim$ 64 mm w.e. a${}^{-\mathrm{1}}$ at Kohnen Station . To get an estimate for the uncertainty in these values, we use the accumulation records over the last 200 years from the B32 DML05 ice core at Kohnen Station . We calculate the SD of the 5-year-block-averaged record, since 5 years roughly represents the accumulation period of the new snow cores. For each site, we scale the SD to the local mean accumulation rate.

Snow heights were measured with a 2 m horizontal resolution and a height accuracy of $\pm \mathrm{1}$ cm along the line of the five snow cores at each site (60 m length; see Fig. b) using a geodetic levelling device. Surface roughness is defined by the standard deviation of these surface heights, SD${}_{\mathrm{SH}}$. Further, we resample the height values from each site 1000 times with replacement, estimate the surface roughness from the resamples and use the SD of these surface roughness values as a measure of uncertainty. To assess past snow surface heights and roughnesses, common isotopic extremes, i.e. isochrones, were manually traced wherever possible.

Figure 2

${\mathit{\delta }}^{\mathrm{18}}$O profiles at each of the new sampling sites plotted against snow depth. The colour scale indicates the ${\mathit{\delta }}^{\mathrm{18}}$O values. Black lines trace possible common isotope minima (solid lines) and maxima (dashed lines), wherever it was possible to identify common peaks (sites D2 and D38). Grey lines indicate the surface topography at 2 m (solid line) and 10 m (dashed line) horizontal resolutions. The dashed black boxes at site D24 indicate the depths of similar isotopic peaks, even if consecutive isochrones could not be assigned.

[Figure omitted. See PDF]

3 Results

3.1

${\mathit{\delta }}^{\mathrm{18}}$O profiles

Mean ${\mathit{\delta }}^{\mathrm{18}}$O values at the sampling sites range from $-\mathrm{43.9}$ ‰ to $-\mathrm{44.3}$ ‰ (Table , Fig. ) with no significant differences between sites (two-sided Student $t$ test; $p>\mathrm{0.05}$). The mean vertical variance within the arrays of the isotope profiles is ${\mathit{\sigma }}^{\mathrm{2}}=\mathrm{8.9}$ ‰${}^{\mathrm{2}}$ (SD 2.9 ‰) and thus slightly higher than the mean horizontal variance of ${\mathit{\sigma }}^{\mathrm{2}}=\mathrm{7.5}$ ‰${}^{\mathrm{2}}$ (SD 5.5 ‰). Between four and nine local ${\mathit{\delta }}^{\mathrm{18}}$O maxima were observed in the isotope profiles with a mean peak-to-peak amplitude of 5.0 ‰ (SD 3.0 ‰).

3.2 Recent and past snow surfaces and surface roughnesses

The surface roughness SD${}_{\mathrm{SH}}$ varies between 3.5 cm (Kohnen trenches) and 8.6 cm (D24) (Table ). The variations at the Kohnen trench site are significantly smaller compared to the other sites (two-sided $F$ test, $p<\mathrm{0.05}$). Sites D24, D7 and C4 and sites D2, C5 and D38 form two distinct clusters, with no significant intracluster differences in surface roughness but significant intercluster variations.

At two sites (D2 and D38), it was possible to tentatively trace past snow surface variations by manually tracking local isotope extremes, i.e. isochrones (Fig. ). This was also done, with a higher degree of confidence, at one of the Kohnen trenches (T13-1; Table ) by . In contrast, here the assignment of common peaks was more ambiguous and uncertain at site D38 and particularly at D2, where one to two cycles might have been missed in the top 25 cm.

At all three sites (D38, D2 and Kohnen), the isochrones (horizontal black lines in Fig. ) exhibit a similar degree of roughness (SD${}_{\mathrm{past}}$) to the snow surface (dashed grey lines in Fig. ) with SD${}_{\mathrm{surface}}=\mathrm{3}$ cm and SD${}_{\mathrm{past}}=\mathrm{3.7}$ cm at the Kohnen trenches , SD${}_{\mathrm{surface}}=\mathrm{3.7}$ cm and SD${}_{\mathrm{past}}=\mathrm{3.5}$ cm at D2, and SD${}_{\mathrm{surface}}=\mathrm{4}$ cm and SD${}_{\mathrm{past}}=\mathrm{3.4}$ cm at D38. At site D2, the correlation between the isochrone profiles and the local surface heights (dashed grey line, Fig. ) ranges from 0.47 to 0.87, which suggests that the topography might have been preserved over the years. At site D38, the same test results in correlation coefficients between $-\mathrm{0.79}$ and 0.90, indicating an annual reorganisation of the stratigraphy, which is consistent with earlier findings at the Kohnen trenches .

At site D24, strong isotopic anomalies were found at depths of about $-\mathrm{20}$ and $-\mathrm{60}$ cm. Depending on the exact choice when tracing these isochrones, the resulting SD${}_{\mathrm{past}}$ varies between 1 and 2.8 cm, which differs significantly from the SD${}_{\mathrm{surface}}$ of 10 cm. However, as for sites C4, C5 and D7, consecutive ${\mathit{\delta }}^{\mathrm{18}}$O isochrones at D24 could not be traced with sufficient confidence due to strong irregularities in the isotopic cycles.

Figure 3

Signal-to-noise ratio (SNR, blue circles and crosses) with 95 % confidence intervals (blue lines) at the different sampling sites (a) together with the positions of the sites along the transect and (b) with elevations from the RAMP2 DEM yellow line;. The elevation data at C4 and C5 (crosses) were collected along a transect that was slightly ($\sim \mathrm{7}$ km) offset to the north (dashed line).

[Figure omitted. See PDF]

3.3 Signal-to-noise ratios

SNRs range from 0 (C5) to 0.77 (Kohnen trenches and D24) (Fig.  and Table ). The statistical uncertainty in SNR estimates is lower at sites with low SNRs (C4, C5, D7) as well as at the Kohnen trenches. For the latter, this is due to the larger number of available ${\mathit{\delta }}^{\mathrm{18}}$O profiles. The highest uncertainty was estimated for sites D38 and D24. Considering these uncertainties, the SNR at the Kohnen trenches is significantly higher ($p<\mathrm{0.05}$) compared to C4, C5 and D7. Furthermore, the SNR at D24 is significantly higher than at C5 and D7, while the SNR at D2 is significantly higher than at C5.

Table 2

Statistical properties of the ${\mathit{\delta }}^{\mathrm{18}}$O data and environmental parameters for all sampling sites. Listed are the mean and standard deviation of the isotope values, signal-to-noise ratios (SNRs), mean accumulation rates ($A$) derived from ground-penetrating radar in water-equivalent [mm w.e. a${}^{-\mathrm{1}}$] and snow-equivalent [cm snow a${}^{-\mathrm{1}}$] units, surface roughness (SD${}_{\mathrm{SH}}$ [cm]), the maximum height difference between two adjacent snow cores [cm], and slope inclinations [m km${}^{-\mathrm{1}}$]. For converting accumulation rates from water equivalent to snow equivalent, we assumed a snow density of 344 kg m${}^{-\mathrm{3}}$, which is the overall arithmetic mean of all sites ($n=\mathrm{19}$).

Site	${\mathit{\delta }}^{\mathrm{18}}\mathrm{O}$	SD${}_{{\mathit{\delta }}^{\mathrm{18}}\mathrm{O}}$	SNR	$A$	$A$	SD${}_{\mathrm{SH}}$	Max height	Slope
	[‰]	[‰]		[mm w.e. a${}^{-\mathrm{1}}$]	[cm snow a$-\mathrm{1}$]	[cm]	difference [cm]	[m km${}^{-\mathrm{1}}$]
D2	$-\mathrm{44.1}$	3.0	0.43	58.7	20.2	4.75	12.6	0.55
C4	$-\mathrm{44.6}$	3.3	0.10	–	–	7.23	11.3	1.84
C5	$-\mathrm{43.5}$	2.7	0	–	–	5.50	7.0	2.22
D7	$-\mathrm{44.6}$	2.9	0.07	43.6	15.0	7.35	17.0	1.97
D24	$-\mathrm{44.7}$	2.7	0.77	43.3	14.9	8.55	26.5	3.12
D38	$-\mathrm{45.2}$	2.9	0.57	52.8	18.1	5.36	7.5	0.35
Kohnen trenches	$-\mathrm{44.7}$	3.1	0.77	64.0	22.0	3.47	15.5	0.59

4 Discussion

We found strong variations in the amount of stratigraphic noise between the seven sites on the 120 km transect. The sites are also characterised by varying snow surface features, slope inclinations, accumulation rates and surface roughnesses. In this section, we discuss the observed differences and possible relationships and conclude with recommendations for future studies.

4.1 Isotope profiles and snow height evolution related to sastrugi and glazed surfaces

In shallow isotope profiles, which are not yet superimposed by diffusion , ${\mathit{\delta }}^{\mathrm{18}}$O minima represent snow that fell at colder atmospheric temperatures, i.e. in austral winter, while the maxima represent warmer temperatures from summer e.g.. Common isotopic peaks are therefore assumed to represent snow which accumulated during the same season or even the same accumulation event. Within these estimated isochrones, the isotope values between the snow cores show some variations. Isotope values within isochrones at locations D2 and D38 have a mean SD of 1.6 ‰. Such variations can be expected within the same season (e.g. summer or winter). They also result from isotopic diffusion, as the thickness of a snow layer with a certain isotopic value will affect how much the amplitude is reduced.

Based on the accumulation rate estimates , we expect the 1 m profiles to have accumulated within $\sim \mathrm{4.5}$ to $\sim \mathrm{6.7}$ years and to exhibit an equal number of isotopic maxima and minima if from the same site. However, the absolute count of isotopic peaks varied considerably, also between adjacent profiles. Furthermore, the isotope profiles exhibited strongly varying cycle lengths and amplitudes, which made it difficult to assign common isotopic peaks and which suggests a considerable redistribution and irregular accumulation. This is further confirmed by the analysis of the evolution of snow heights at D2 and D38. Particularly at D38, a refilling of troughs combined with a lower accumulation in elevated parts shows typical processes of snow deposition . At D2, D38 and the Kohnen trenches, the surface snow heights (solid grey lines in Fig. ) show similar variations to past snow heights, indicating similar surface roughnesses over time. Furthermore, these variations indicate the presence of pronounced snow features such as dunes (Fig. a) .

The highest surface roughness was observed at site D24, where isotopic peaks were too variable to assign consecutive isochrones. Still, similar isotopic values indicate a very low surface roughness at depths of around $-\mathrm{20}$ and $-\mathrm{60}$ cm. These nearly flat past surfaces are consistent with present snow surface features observed around the site, namely glazed surfaces between patches of large sastrugi and dunes (Fig. b). Glazed surfaces are characterised by flat and very dense snow, permeated by “thermal” cracks (Fig. c) . Glazed surfaces were found to occur when high wind speeds coincide with a hiatus in accumulation, e.g. on the steep slopes in the katabatic wind region , and could already be detected via satellite images within $\sim \mathrm{200}$ km to the south, south-east and north-west of the transect .

Figure 4

The snow surface at site C4 in panel (a), showing the typical sastrugi, and at site D24 in panels (b) and (c), showing a mix of sastrugi and glazed surfaces and a thermal crack with a glove placed next to it for scale.

[Figure omitted. See PDF]

4.2 Amount and structure of stratigraphic noise

For the newly collected isotope profiles, the low mean pairwise correlation coefficient of $r=\mathrm{0.19}$ (SD 0.31) indicates strong stratigraphic redistribution and irregular deposition. The pairwise correlations were independent of the spatial distance between profiles (Appendix ), which confirms that the decorrelation length of stratigraphic noise in the study area is $<\mathrm{10}$ m, as proposed by for Kohnen Station. The SNRs of $<\mathrm{1}$ indicate that stratigraphic noise explains more than 50 % of the ${\mathit{\delta }}^{\mathrm{18}}$O variability in the uppermost 1 m at all sites, which is consistent with previous findings for this region and confirms the low representativity of single isotope profiles . Along the 120 km transect, SNRs varied strongly between 0 and 0.77, which emphasises the importance of understanding their drivers.

4.3 Relationships between stratigraphic noise and environmental properties

Despite the low number of data points, the strong variations between sites allow us to formulate some hypothesis regarding the origin of stratigraphic noise. For this aim, we compare the SNRs with the local environmental characteristics, namely accumulation rates, slope inclination and surface roughnesses (Fig. ). We exclude spatial variations in precipitation or isotopic amplitudes as factors potentially affecting SNRs as we do not expect these features to differ strongly across the spatial scale considered here .

Figure 5

Scatter plots of signal-to-noise ratios (SNRs) against accumulation rate $A$ [mm w.e. a${}^{-\mathrm{1}}$], surface roughness SD${}_{\mathrm{SH}}$ [cm] and slope inclination [m km${}^{-\mathrm{1}}$]. Most sites were dominated by sastrugi, while at D24 (red dots), we observed a mix of sastrugi and glazed surfaces. The latter was therefore excluded in the linear regression analysis (dashed lines). Vertical lines indicate the 95 % confidence intervals of the SNR estimates, while horizontal lines represent the uncertainty in the environmental properties (2 $\cdot$ SD). Uncertainties in 10 km slope inclinations are very small such that they are not visible for most sites.

[Figure omitted. See PDF]

At the sites dominated by sastrugi (all except D24; see blue points, Fig. ), SNRs show a strong positive correlation with accumulation rates ($r=\mathrm{0.89}$, $p=\mathrm{0.11}$) and strong negative correlations with surface roughnesses ($r=-\mathrm{0.81}$, $p=\mathrm{0.05}$) and slope inclinations ($r=-\mathrm{0.92}$, $p<\mathrm{0.05}$). This is consistent with earlier findings which showed that isotope records at coastal sites in East Antarctica with higher accumulation rates contain a more consistent climate signal than those at lower-accumulation sites in DML or near Dome Fuji . The negative correlation of the SNRs and surface roughness, thus higher undulations related to a higher noise level, is intuitive. It indicates that, while the surface roughness represents only a snapshot of a single summer season, it might be at least partly representative of the general surface roughness during the past years. If this is confirmed, the surface roughness, which is easy to measure, could be used as an indicator of the stratigraphic noise. The correlation between SNRs and slope inclinations is robust to the spatial scales over which the inclination is calculated (Appendix ).

At D24, past and present surface roughnesses were inconsistent, which could be related to the surface consisting of a mix of sastrugi and glazed surfaces that can alternate both spatially and temporally. This feature was (visually) unique compared to the other sites and coincided with the lowest accumulation rate, the highest slope inclination and the highest surface roughness. The relationships between SNR and environmental properties, as proposed for sites dominated by sastrugi, therefore do not seem to hold at a site with such extreme environmental properties and the related occurrence of glazed surfaces e.g.. We speculate that D24 represents a discontinuous record in which the two flat anomalies or isochrones (Fig. ) represent a summer and a winter layer, respectively, from any time within the nearly 7 years of accumulation. The SNR, as defined in this study, is the ratio of the shared signal between the records and the local variations. Therefore, the estimated SNR at site D24 is high, driven by the two anomalies. Between the two anomalies, the past surface roughnesses could have been as high as the present surface roughness. We therefore assume that the spatially coherent part at this site does not represent a seasonal or annual climatic signal and that the isotope records therefore do not provide any useful information for climate signal interpretations.

4.4 Relationships between slope inclination, surface roughness and the accumulation rate

The different environmental properties slope inclination, surface roughness and the accumulation rate also exhibit strong covariations (Fig. 1). Surface roughness is higher at sites with higher slope inclinations ($r=\mathrm{0.82}$, $p<\mathrm{0.05}$) and smaller accumulation rates ($r=-\mathrm{0.97}$, $p<\mathrm{0.05}$), while the latter correlates negatively with slope inclination ($r=-\mathrm{0.82}$, $p=\mathrm{0.09}$). These findings partly contrast , who proposed that sastrugi heights would be proportional to accumulation rates in Greenland. furthermore question simple relationships between slope inclination, surface roughness and the accumulation rate across large spatial scales. However, close links were frequently confirmed at smaller scales: for example, accumulation rates tend to be lower in areas with steeper slopes in various EAP locations e.g., often associated with the underlying bedrock topography e.g. or wind-driven sublimation e.g. and redistribution . Steeper slopes are associated with stronger wind speeds e.g., which in turn affect snow surface features e.g.. High wind speeds increase both the propagation speed and the height of dunes and thereby increase surface roughness , at least within the physical constraints of maximum snow heights e.g. about 1.5 m for sastrugi;.

4.5 Implications for future studies on stratigraphic noise

Based on the presented dataset, we conclude that the assessed environmental properties affect the amount of stratigraphic noise. However, due to the low number of sampling sites and replicates, as well as shallow sampling depths, the SNR estimates are subject to considerable uncertainty, which renders our results somewhat speculative, even if they are supported by previous studies. Furthermore, the significant covariance among the different environmental properties makes it difficult to disentangle their individual contributions. Additional studies, ideally including more replicates and deeper profiles that cover a wider range of depositional conditions, are needed to test and refine the proposed relationships: in a first-order approximation, an increase in the number of snow cores or the length of the cores by a factor of $n$ will reduce the standard error of the correlations by $\mathrm{1}/\sqrt{n}$. That means that increasing the number of cores by a factor of 4 already decreases the uncertainty in the pairwise correlations by half. Given the significant cost and work associated with collecting samples in situ, it would be useful to test whether high-frequency ground-penetrating radar e.g. could serve to estimate stratigraphic noise. This would allow us to cover much larger spatial scales, which would provide information on possible large-scale drivers such as precipitation patterns . Moreover, continuous monitoring of the snow surface and accumulation using a laser scanner or photogrammetry could provide more detailed knowledge about variations in snow stratigraphy related to depositional conditions including wind speeds.

4.6 Suggestions for optimal site selection for high-resolution climate reconstructions from the Late Holocene

In most cases, the SNR and not the measurement resolution is the limiting factor for the temporal resolution of the climate signal that can be recovered from snow, firn and ice cores . The higher the noise level, the more averaging in time is needed to reduce the uncorrelated (white) noise while preserving the more persistent (red) climate signal. As an example, when assuming a climate signal with a power spectral density of the form $f^{-\mathit{\beta }}$, with $\mathit{\beta }=\mathrm{1}$ and where $f$ denotes frequency, and uncorrelated noise, a reduction in the noise by a factor of $n$ would increase the attainable climate resolution by the same factor. Assuming that 50 % of the noise is stratigraphic noise , we would expect up to 4 times the attainable climate resolution at Kohnen (SNR of 0.77) relative to C4 (SNR $=$ 0.1). As stratigraphic noise strongly varies across $\sim \mathrm{100}$ km, we suggest that an optimal site selection at such small spatial scales can already allow us to improve the SNRs in snow and firn cores and to considerably increase the effective resolution of climate reconstructions from the East Antarctic Plateau.

Site selection can be done by filtering for the most suitable environmental properties: while data on small-scale accumulation rates and surface roughnesses e.g. are scarce, slope inclination data are available at high spatial resolutions for the entire Antarctic continent e.g. and can easily be assessed. Furthermore, we assume that areas characterised by glazed surfaces are poorly suited sampling locations, as the archive of the climate signal might be intermittent. Such locations can be excluded by in situ observations, by employing remote sensing approaches to snow surface classification , or by selecting locations with low slope inclinations and a certain amount of accumulation e.g.. Considering that the ice sheet topography was fairly constant over large parts of the EAP during the past millennium, related to the bedrock topography , we hypothesise that the amount of stratigraphic noise imprinted in the snow over the past decade is representative of past centuries. This could be tested by collecting and analysing long (e.g. 100 m) high-resolution firn cores. Future studies should however also account for the fact that ice properties such as stratigraphic noise are advected horizontally due to ice flow by checking for similarly suitable environmental properties further upstream. The size of that area can be determined based on ice flow velocity e.g. and the desired snow or firn core length.

The sampling setup at the selected sites should then follow the suggestions of : the distance of replicate cores should be larger than the expected decorrelation length of stratigraphic noise, for example 10 m in the DML plateau area. The number of cores should be chosen based on the expected amount of stratigraphic noise and the intended signal resolution. Based on the findings by , we suggest taking five replicates at locations with similar environmental properties to Kohnen Station. The sample direction should be perpendicular to the overall wind direction if the surface roughness is measured across the sampled cores as in this study.

Signal interpretation should further consider influences on the isotopic composition from, for example, sublimation , snow metamorphism and precipitation intermittency. The latter can be responsible for up to 50 % of the noise variance across large spatial scales . The sampling strategy we propose here could therefore be expanded by replicate cores taken at optimal distances to account for precipitation intermittency, as suggested by .

5 Conclusions

In this work, we assessed stratigraphic noise and its spatial variations along a 120 km transect to the south-west of Kohnen Station, Dronning Maud Land, the East Antarctic Plateau. We analysed the local, non-climatic variability in ${\mathit{\delta }}^{\mathrm{18}}$O compositions at high vertical resolution across spatial scales ranging from local (50 m) to regional ($\sim \mathrm{120}$ km), assessing their dependency on the following local environmental properties: the accumulation rate, surface roughness and slope inclination. Within the study area, we found that stratigraphic noise dominates the seasonal to interannual isotopic signal. The amount of noise also varies significantly across the different sites. At sites that are dominated by sastrugi, stratigraphic noise is lower if the terrain is flatter, the surface less rough and accumulation rates higher. All these environmental characteristics are typically associated with lower wind speeds. Sites characterised by these properties are likely more suitable for collecting isotope profiles that provide meaningful climate signals. Assuming that the proposed relationships are stationary over time, these findings could be applied to snow, firn and ice cores that are several hundred metres in length and thus increase the effective resolution of Late Holocene climate reconstructions from the East Antarctic Plateau.

Appendix A Height reference

When the snow surface variations stay similar during the deposition of the sampled snow, e.g. when dunes and troughs are persistent in size and location, snow samples can be processed at depths relative to each other. This is done, for example, for surface snow samples and shallow cores e.g.. If the surface height variations are expected to have changed throughout the time of accumulation, snow cores are instead processed at absolute heights, which was decided for the Kohnen trenches . We tested this method for the new dataset by calculating pairwise correlation coefficients for all possible pairs of isotope profiles at each site (Fig. ), using both the absolute and the relative heights. We would expect the correlation to be stronger with more common signal located at a similar depth, at either absolute or relative heights.

Figure A1

Pairwise correlation coefficients, $r$, of isotope profiles from different sites using absolute (black) and relative (red) heights. Vertical lines show the mean correlation coefficients.

[Figure omitted. See PDF]

We obtained a mean correlation coefficient of $r=\mathrm{0.19}$ (SD 0.31) when using absolute heights and 0.17 (SD 0.32) for relative heights. The difference in the mean correlations was statistically not significant for any of the sites ($p<\mathrm{0.05}$, Fig. ). Additionally, we were unable to identify common isotopic peaks at most sites, which also indicates that local snow heights generally changed fast. At the same time, we were able to confirm highly irregular snow accumulations, e.g. at site D38. These findings are consistent with previous findings e.g.. The snow cores in this study are therefore processed at the absolute height reference.

Appendix B Pairwise correlations with interprofile spacing

Pairwise correlation coefficients for isotope profiles collected at Kohnen Station were found to increase as the interprofile spacing drops to below 5–10 m , indicating that less distant isotope profiles contain more dependent noise, probably due to the spatial scales of snow surface features like sastrugi. In order to quantify independent noise, we used a minimum interprofile spacing of 10 m in this study. With this setup, we did not find any relationship between spacing and pairwise correlation coefficients (Fig. ), which indicates that the decorrelation length of stratigraphic noise which was found to be 5–10 m at Kohnen Station is valid for larger areas across the plateau of DML.

Figure B1

Pairwise correlation coefficients, $r$, of the ${\mathit{\delta }}^{\mathrm{18}}$O profiles as a function of their interprofile spacing. Colours indicate the different sites.

[Figure omitted. See PDF]

Appendix C Slope inclination scales

We calculated slope inclinations on spatial scales ranging from 1 to 15 km REMA DEM; and tested how this affects the relationship with SNR, surface roughness and the accumulation rate. Regarding SNR, the absolute values of the correlation coefficient varied slightly for the different spatial scales, increasing from $r=-\mathrm{0.82}$ ($p<\mathrm{0.05}$) at 1 km to $r=-\mathrm{0.95}$ ($p<\mathrm{0.05}$) at 15 km (Fig. ). A similar increase in correlation with increasing spatial scale was obtained for surface roughness and accumulation rates. Previous studies have already proposed the existence of relationships between the accumulation rate, slope inclination and snow surface features for slopes calculated at scales of 1–2 km e.g. and 16 km . While a link between wind speed and slope inclination has been established using coarse datasets e.g., this link was expected to break down with higher resolutions e.g. 50 km;. Others proposed that small slope changes at scales $<\mathrm{10}$ km and the commensurate changes in wind speed would be able to explain differences in accumulation rates , e.g. by sublimation and redistribution , as well as snow surface features . In this study, we found that correlation coefficients between SNR, the accumulation rate and surface roughness increased with increasing spatial scales. Yet considering the small size of our sample, the sensitivity of the correlation coefficients to different slope scales could be dependent on a single surface undulation in the sampling area. We therefore take into account earlier results from wind simulations e.g. and variations in accumulation rates and surface features across smaller scales e.g. and do not exceed 10 km for calculating slope inclinations.

Figure C1

Correlation coefficients, $r$, between SNRs, surface roughness (SD${}_{\mathrm{SH}}$) and accumulation rates ($A$) with slope inclinations calculated using different scales (1–15 km). Stars indicate values with statistical significance ($p<\mathrm{0.05}$).

[Figure omitted. See PDF]

Appendix D Relations between slope inclination, surface roughness and the accumulation rate

Possible relationships between slope inclination, surface roughness and the accumulation rate are discussed in Sect. .

Figure D1

Comparisons between the accumulation rate $A$ [mm w.e. a${}^{-\mathrm{1}}$], surface roughness SD${}_{\mathrm{SH}}$ [cm] and slope inclination [m km${}^{-\mathrm{1}}$]. Linear regression lines (dashed) suggest possible relationships. Vertical and horizontal lines represent 2 $\cdot$ SD of the according environmental property as an indication of uncertainty. Uncertainties in 10 km slope inclinations are very small such that they are not visible for most sites.

[Figure omitted. See PDF]

Data availability

All measurements are available in the PANGAEA database under 10.1594/PANGAEA.956273 and 10.1594/PANGAEA.956663 .

Author contributions

TL, MH and JF designed the expedition, and TL designed the sampling strategy. NH and TL designed the study. JF, RD, TL and MH carried out the sampling on the EAP. NH conducted the isotope measurements with the help of AZ and TM. All authors, especially TL, AZ and TM, contributed to the scientific analysis. NH performed the analysis and wrote the manuscript, which was reviewed by all authors.

Competing interests

The contact author has declared that none of the authors has any competing interests.

Disclaimer

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Acknowledgements

We thank the scientists, technicians and support staff at Kohnen Station for their assistance, especially Klaus Trimborn for his skilful support during sample collection. Furthermore, we would like to thank Hanno Meyer and Mikaela Weiner for their work in the isotope laboratory at AWI Potsdam and Christoph Schneider for scientific supervision of the initial draft. Data analysis was performed in R, a language and environment for statistical computing. The Antarctic map is based on Quantarctica datasets in QGIS, kindly provided by the Norwegian Polar Institute .

Financial support

This project received financial support from the Helmholtz Association through the Polar Regions and Coasts in the Changing Earth System (PACES II) programme (COMB-I project) and from the European Research Council (ERC) under the EU's Horizon 2020 Research and Innovation Programme (grant agreement no. 716092). It was furthermore supported by the Informationsinfrastrukturen Grant of the Helmholtz Association as part of the DataHub of the Research Field Earth and Environment.The article processing charges for this open-access publication were covered by the Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research (AWI).

Review statement

This paper was edited by Xavier Fettweis and reviewed by Massimo Frezzotti and two anonymous referees.