Beryllium 10 in Antarctica over the last seven millennia

Abstract

Since the first measurements of beryllium 10 (¹⁰Be) in ice, in the seventies, numerous profiles of this cosmogenic isotope have been obtained both in Antarctica and Greenland. In this article, we focus on Antarctic data, available at nine different sites, covering a significant part of the Holocene, from 237 to 7101 yr BP. We show that correlating their ¹⁰Be profiles allows to synchronize these ice cores with an excellent accuracy and to document the spatial variability of ¹⁰Be concentration and flux. We then examine how this variability is taken into account by a simulation of ¹⁰Be fallout recently performed with the ECHAM6.3-HAM2.3 model. Except for a systematic underestimation of ¹⁰Be fluxes at high accumulation sites, these simulations are overall very satisfying. Finally the excellent accuracy of synchronisation based on ¹⁰Be profiles allows us to derive an Antarctic stack record over the last seven millennia. The reliability of the new ¹⁰Be stack is demonstrated by its superior correspondence with the accurately dated IntCal20 record based on ¹⁴C in tree rings.

Background & summary

Beryllium 10 (¹⁰Be) is a radionuclide (half-life of 1.39 • 10⁶ yr^1,2) formed by the interaction of cosmic rays with the Earth atmosphere and then precipitated out into various geophysical reservoirs such as loess, marine sediment and polar ice³. The ¹⁰Be flux is modulated by the heliomagnetic and geomagnetic fields^4,5; the higher these fields are, the more primary cosmic ray particles are deflected which leads to a decrease in cosmogenic isotope production with two consequences. First geomagnetic reversals⁶ and, for a shorter period, geomagnetic excursions⁷ - both characterized by lower geomagnetic intensity - are associated with ¹⁰Be peaks. Second, as there is a relationship between heliomagnetic activity and solar irradiance^8,9, reduced solar activity is associated with increased ¹⁰Be production, a relationship which can be used to estimate solar activity. More recently, sharp increases in ¹⁰Be^10,11 have been documented at the time of extreme solar particle events discovered from measurements of ¹⁴C - a radionuclide also formed by the interaction of cosmic rays with the Earth atmosphere - in trees¹².

Raisbeck et al.^13,14 reported the first measurements of cosmogenic ¹⁰Be in ice for samples from a 906 m core drilled during the 1977-78 Antarctic field season at the old Dome site, 50 km from the Dome Concordia station. They found a ¹⁰Be concentration that was larger in glacial-age ice by a factor of 2 to 3 than in Holocene ice. Among the possible explanations offered was that the precipitation rate in Antarctica during the glacial period was approximately 50% of that during the Holocene. In turn, measuring ¹⁰Be in the ice provides useful information about the change of accumulation with time along ice cores as illustrated, in the eighties and nineties, from the Dome C and Vostok deep ice core then extending respectively into the last and previous glacial periods^7,15,16,17; ¹⁰Be peaks⁷ - characteristic of periods of lower geomagnetic intensity such the one associated with the Laschamps event - were identified at these two sites.

Getting access to solar variability was one of the initial objective of ¹⁰Be measurements. Raisbeck et al.¹⁴ pointed out to an enhanced ¹⁰Be concentration about the time of the Maunder minimum. This enhancement, also observed in Greenland ice cores¹⁸, was confirmed by measurements covering the last millenium at South Pole¹⁹. That record closely correlated with the ¹⁴C measurements obtained on tree rings with periods corresponding to solar activity minima centred at about 1060 (Oort), 1320 (Wolf), 1500 (Sporer), 1690 (Maunder) and 1820 (Dalton) yr A.D.

Since both ¹⁰Be and ¹⁴C have the same origin (nuclear reactions of cosmic rays in the earth’s atmosphere), the idea of synchronizing their profiles, and thus their ages, was rapidly shown to have promise^20,21. This is particularly important for ice cores from the Antarctic plateau where low ice accumulation rates preclude layer counting. Based on a 180 m core drilled at the old Dome C site, Raisbeck and Yiou²¹ discussed the potential of such a comparison for dating Antarctic ice cores. Bard et al.^22,23 developed a more quantitative approach by using ¹⁰Be in a 127 m dated South Pole core as input to a 12-box carbon cycle model to simulate an atmospheric ¹⁴C.

Using a similar approach, Raisbeck et al.²⁴ derived the age for the bottom of the 178 m Vostok BH1 core and concluded that the EGT timescale^24,25 was too young by a gradually increasing amount which, at the bottom of this core, eventually reached ~400 years at 7000 yr BP (before 1950 AD). This study provided an absolute time marker which has later been systematically taken into account for Vostok^26,27 and Antarctic ice core²⁸ chronologies. While this time marker has thus been used extensively, the ¹⁰Be data on which it was based was never published. Doing that was in fact the initial motivation for the present paper.

Beyond this pioneering work, ¹⁰Be profiles covering part of the Holocene have now been measured in 9 different Antarctic sites (Fig. 1) with two adjacent ice cores, BH1 and BH2, near the Vostok station, two, PS1 and SPICE, near the South Pole station, three in the Dome C area - one, EPICA Dome C (EDC) near the Dome Concordia station and two, Old Dome C (ODC) and Little Dome C (LDC), respectively 50 and 40 km away - and single ice cores otherwise (Siple Dome, Taylor Dome, Wais Divide, Dome A, Dome F and EDML). Thanks to this ensemble of data, we hereafter document the geographical variability of the ¹⁰Be fallout and its relationship with the ¹⁴C record over a longer time period, the last ~7 ka. Our study is based on ¹⁰Be profiles measured by the Orsay team (who are authors of the present study) for two of the cores drilled in the Dome C area - old Dome C²¹ and EPICA Dome C - the two cores drilled at Vostok - BH1 and BH2 - and one, PS1, drilled at South Pole¹⁸. We also use a profile obtained in collaboration between this team and the Xi’an AMS Center along a core drilled at Dome A²⁹. In addition, our study includes profiles obtained by other teams at EDML, WAIS Divide (WD), Siple Dome (SD), Taylor Dome (TD), Dome F (DF), SPICE (SPI) at South Pole and Little Dome C (see below).

Fig. 1

Map of Antarctica with the 9 sites where ¹⁰Be data are available over a part of the Holocene. Colored background indicates the elevation.

We note here a study of Nilsson et al.³⁰ aiming to reconstruct variations in average hemispherical ¹⁰Be production rates based on data obtained from both Greenland and Antarctic ice cores over the Holocene. For Antarctica these authors use data from EDML completed by short records from Dome F, Siple Dome and South Pole including a second profile recently measured³¹ along the SPICE core retrieved at this site starting ten years ago³². Our study is based on a larger ensemble of data with additional profiles at Vostok, Dome A (DA), WAIS Divide and three in the Dome C area which allows us to discuss the geographical variability of the ¹⁰Be fallout over Antarctica. For documenting this geographical variability, we account, in addition, for ¹⁰Be near surface data available along traverses conducted in the Dronning Maud Land area in the Atlantic sector of East Antarctica (Fig. 1). Near surface data are also available at Law Dome³³.

In this article, we first focus on the use of ¹⁰Be profiles as a dating tool comparing two approaches providing independent absolute timescales. First we use the ¹⁰Be - ¹⁴C correlation through a synchronization on the Δ¹⁴C calibration curve IntCal20³⁴ to derive an absolute timescale as initially suggested by Raisbeck and Yiou²¹ and developed by Bard et al.²², and apply it to the continuous and detailed Vostok profile back to ~ 7 ka; this timescale is referred to as the ¹⁴C timescale. Second we synchronize the ice cores on a common relative timescale by correlating their ¹⁰Be profiles, except for Taylor Dome and for the lower part of Siple Dome for which data are too sparse. We then derive a second Vostok absolute time scale - referred to as the ¹⁰Be timescale - using the fact that one of the cores, WAIS Divide, is dated by layer counting with a good accuracy³⁵. Based on this ¹⁰Be synchronization, we then examine the geographical variability of the ¹⁰Be fallout over Antarctica and compare this variability with results from a recent simulation of this ¹⁰Be fallout using an Atmospheric General Circulation Model (AGCM)³⁶. Finally, we propose a ¹⁰Be stack for the last 7 ka and a simulated ¹⁴C record based on our ¹⁰Be stack profile in order to test if this stack better accounts for the solar modulation than individual profiles. The method is resumed in the flowchart in Fig. 2.

Fig. 2

Flowchart resuming the method used in this study. ¹⁰Be ice core data from Orsay team and other teams are in blue and orange, respectively. The acronyms are detailed in Table 1.

Methods

¹⁰Be data from old Dome C, South Pole (PS1), Vostok, EPICA Dome C and Dome A

Key characterics of ¹⁰Be data used in this study are summarized in Table 1. In Fig. 3 are shown with respect to depth ¹⁰Be data measured in the Orsay laboratory (old Dome C¹⁹, PS1¹⁹, Vostok²⁴ and EPICA Dome C) and at the Xi’an AMS Center (Dome A²⁹). EPICA Dome C data have never been published by the authors of the present study, and the other records were used in previous publications, but not available in an open database (see Table 1). Samples from Old Dome C, South Pole (PS1) and Vostok were prepared by the Orsay team using the following analytical procedure. After melting the ice with 500 or 250 µg of ⁹Be carrier, beryllium was extracted by loading on a Dowex 50W-8X, 200-300 mesh ion exchange column and leached off with HCl. It was then precipitated as Be(OH)₂ and transformed into BeO by heating. For EPICA Dome C and Dome A, smaller quantities of ice (~60 g and ~100 g respectively) were used, which permitted direct precipitation of Be(OH)₂ without concentrating with the ion exchange resin. The BeO, together with carbon, silver or niobium powder, was pressed into Cu or Mo cathodes in order to measure the ¹⁰Be/⁹Be ratio by accelerator mass spectrometry. Knowing the weight of ice, weight of ⁹Be and the measured ¹⁰Be/⁹Be, the concentration of ¹⁰Be in the ice is calculated relative to the NIST standard for which the nominal ratio (2.68 • 10^-11) is adopted.

Table 1 Summary of key characteristics of ¹⁰Be ice core data used in this study.

Fig. 3

¹⁰Be concentration with respect to depth measured along cores drilled at (a) Vostok (BH1 and BH2), (b) Dome A, (c) PS1, and (d) Dome C (Old Dome C and EPICA sites).

For the old Dome C site, located 50 km from the EPICA Dome C site, a nearly continuous profile has been obtained on the lower part of a 180 m ice core drilled during the 1978-79 field season³⁷ about 10 m away from the 906 m ice core. About half of the samples from the upper part of the core were initially measured at the Grenoble cyclotron using six times higher ⁹Be carrier because it was necessary to fabricate self supporting Be pellets¹⁴. Although these samples were remeasured with the Tandetron, the larger ⁹Be carrier led to higher uncertainties. For this reason, we limit here the use of this old Dome C profile for the record below 56 m (Fig. 3) which from the dating derived by Raisbeck and Yiou²¹ extends between ~1 and 4 ka.

For the 127 m long PS1 core retrieved in 1984 near the South Pole station, measurements were performed using 1 kg samples every 0.8 m on average. This core, which covers slightly more than 1 ka (Figs. 3 and 4), was dated by recognizing 20 layers of impurities attributed to known volcanic eruptions and by correlation with a nearby core for which seasonal variations are observed for major ions and water isotopes³⁸.

Fig. 4

¹⁰Be concentration at Vostok (combined BH1 and BH2 see below), EDML⁴⁵ on the AICC2012 timescale and WAIS Divide³⁵ on the WD2014 timescale. Data from old Dome C¹⁹, Vostok²⁴, Dome F⁴⁹, South Pole (PS1¹⁹ and SPICE³¹), Taylor Dome⁴⁷, Dome A²⁹, Siple Dome⁴⁶, and EPICA Dome C (this study) are plotted on their original timescales. For Little Dome C we have used the timescale inferred from the comparison of ¹⁰Be profiles obtained at this site and at the Vostok site⁵⁰. The Taylor Dome data, which are low resolution and discontinuous, are represented by data points (squares). All these data are expressed with respect to the NIST standard (except for Siple Dome, Taylor Dome and SPICE, see text).

For the Vostok site, measurements were performed following the same methodology every 50 cm on two adjacent cores, BH1 and BH2 respectively 178 m and 51 m deep (Fig. 3). Both cores were drilled in the clean sector 300 m west of the main station during the 1989-90 season using a thermal drill. Using BH1 ¹⁰Be data, Raisbeck et al.²⁴ derived an age of ~7 ka at the bottom of this core.

For EPICA Dome C, continuous measurements have been performed from 7.7 m down to 51 m with either 1, 2 or 3 measurements along each 55 cm bag sample. For this study we have used 55 cm average values (Fig. 3). Based on the AICC time scale developed for this EPICA core^28,39 these ¹⁰Be data cover ~1 ka. The ¹⁰Be concentrations are higher at EPICA Dome C than at Old Dome C which (Table 2) is partly due to the lower accumulation at the EPICA site (by ~14%) despite the fact that the distance between the two drilling sites is only 50 km.

Table 2 Observed (obs.) and ECHAM6.3-HAM2.3 modeled (mod.) values at the nearest grid cells for altitude (Alt, m), ¹⁰Be concentration (Conc., 10⁵ at/g), accumulation (Accu., g/cm²/yr) and ¹⁰Be flux (Flux, at/m²/s).

For Dome A measurements were performed on ice chips collected along a ~ 110 m shallow ice core (Fig. 3) drilled ~ 300 m from the summit during the 21st Chinese Antarctic Research Expedition (DA2005 ice core, CHINARE 21) in the 2004/05 austral summer²⁹. Ice chips (~100 g) retained inside the drilling barrel were collected for each drilling run, weighed and transferred to plastic bottles together with the ⁹Be carrier. The 135 melted samples were sent to Orsay for sample preparation (see above) and then measured at the Xi’an AMS Center. Zhang et al.²⁹ compared this ¹⁰Be profile with the Δ¹⁴C profile in dendrochronologically dated tree rings assuming different values of the mean accumulation rate. They first showed that the value of 1.6 g H₂O/cm²/yr, suggested by Hou et al.⁴⁰ on the basis of the observed close off depth and a densification model⁴¹, leads to a very poor match. They then tried larger accumulation values, in the range of those observed at Vostok and Dome C with similar ¹⁰Be concentrations. They found the best fit with A = 2.31 g H₂O/cm²/yr and derived an age of ~3.2 ka at the bottom of this core. This choice was confirmed by similarities with the South Pole ¹⁰Be over their common part. A similar value of 2.31 g H₂O/cm²/yr was previously derived for the period 1260 - 2005⁴² (see also^43,44).

¹⁰Be ice core data from other sources

Our study - that we limit to the last seven millenia covered by our Vostok record - is also based on profiles obtained by other teams for five other Antarctic sites (Fig. 1). These include high resolution ¹⁰Be profiles covering a large part of this period at EDML⁴⁵ where there is no data for the last millenia, and WAIS Divide³⁵ where data are missing between 2.3 and 5.3 ka. They are complemented by data covering also a large part of this period at Siple Dome⁴⁶ but with a discontinuous record over its older half, and at Taylor Dome⁴⁷ in this case with low resolution, and on a profile covering a shorter period at Dome F^48,49. We also used ¹⁰Be profiles available at two sites previously documented, Little Dome C (located in the Dome C area 40 km from the EPICA Dome C core and 82 km from old Dome C), for which ¹⁰Be data have been measured on ice chips⁵⁰, and SPICE³² at South Pole. The ¹⁰Be records of WAIS Divide^51,52, SPICE³¹, Dome F⁵³, Taylor Dome⁴⁷, and Siple Dome⁴⁶ are stored in public databases. As the EDML and Little Dome C records were used in previous publications but not published in a public database, requests to obtain these records were submitted to the respective authors of these ¹⁰Be data (see Table 1). For EDML, data were provided only with respect to age after resampling with a 22-year time step. We thus have access to ¹⁰Be profiles covering part of the Holocene at 9 different sites Dome C, Vostok, South Pole, Siple Dome, Taylor Dome, EDML, WAIS Divide, Dome F and Dome A. with multiple cores at three of these sites, Vostok (BH1 and BH2), Dome C area (Old Dome C, EPICA Dome C and Little Dome C) and South Pole (PS1 and SPICE).

To account for the fact that different laboratories use different different ¹⁰Be/⁹Be standards, we have performed a recalibration (by 0.89 for Dome F, 0.96 for WAIS Divide and Little Dome C and 0.91 for EDML) in order to have all these data expressed with respect to the NIST standard adopted at the Gif sur Yvette Tandetron (¹⁰Be/⁹Be = 2.68 • 10^-11); such recalibration has not been applied for Siple Dome, Taylor Dome and SPICE due to the lack of relevant information. All these data (from Orsay team plus the other sources) are reported on Fig. 4 with respect to time.

Δ¹⁴C calibration curve IntCal20

For the ¹⁰Be - ¹⁴C correlation, the Δ¹⁴C calibration curve IntCal20³⁴, which is available in open access at https://intcal.org/curves.html⁵⁴, was used.

Third-party softwares

The version 2.3.1 of the Match software⁵⁵ (https://lorraine-lisiecki.com/match.html) was used for the synchronizations with a graphical user interface on MATLAB R2024b release. To simulate the ¹⁴C record based on ¹⁰Be stack, a 12-box carbon cycle model²², which is an hybrid of PANDORA⁵⁶ and the carbon cycle model used by Siegenthaler et al.⁵⁷, was used. These references^22,56,57 provide exhaustive information about the model geometry, specific equations and various parameters (carbon reservoir sizes, exchange fluxes and ¹⁴C steady state). The model built with the STELLA software has performances and characteristics (impulse response function and Bode plot for attenuation and phase) in agreement with those of other models^58,59.

Data Records

The full dataset⁶⁰, including the compilation of ¹⁰Be concentration data, their synchronisation on Vostok and PS1 ¹⁰Be records, the stack calculations, the simulated Δ¹⁴C time series, and the configuration files used for the synchronizations with Match protocol, is publicly available in the Figshare repository at https://doi.org/10.6084/m9.figshare.29623730. This dataset consists of seven excel files and one zip file, which are described below.

The ¹⁰Be records prepared by the Orsay team (Vostok, Dome A, EPICA Dome C, old Dome C, and PS1; see the first subsection of Methods section and Table 1) are stored in the excel file named Be_holocene_data_orsay.xslx. The excel file contains a README tab with the references for each record and details on the structure of the file. The tab “Vostok (BH1 and BH2)” shows the ¹⁰Be time series in Vostok BH1 and BH2, their merging, and their new age after synchronization on the Δ¹⁴C calibration curve IntCal20. This age is used for the final stack. The tab “D14C derived from ¹⁰Be Vostok” is the Δ¹⁴C curve from ¹⁰Be Vostok on its old (EGT) and its updated age scale after synchronization on IntCal20 Δ¹⁴C time series. The tabs DA, EDC, ODC, and PS1 are the ¹⁰Be data info from the ice cores measured by the Orsay team. The depth, original age, ¹⁰Be concentrations, and the updated Δ¹⁴C age scale after synchronization on Vostok’s ¹⁰Be data are shown.

The ¹⁰Be records from other sources (Dome F, EDML, Little Dome C, Siple Dome, SPICE, Taylor Dome, and WAIS Divide; see the second subsection of Methods section and Table 1) are stored in the excel file named Be_holocene_data_others.xslx. As in the file Be_holocene_data_orsay.xslx, the references information, depth, original age, ¹⁰Be concentrations, and the updated Δ¹⁴C age scale after synchronization on Vostok’s ¹⁰Be data are shown. Both the original data values and the corrected ones after recalibration to NIST standard (noted with a “corr” notation) are shown.

The results of the synchronization of ¹⁰Be records (except for Taylor Dome) on Vostok one are displayed in the excel file synchronisations_on_Vostok.xlsx. The Vostok’s ¹⁰Be data is displayed in the first five columns of each tab (one tab per site), including the mean depth, original age scale, updated Δ¹⁴C age scale, ¹⁰Be concentrations, and normalized ¹⁰Be concentrations (by removing the mean and dividing by the standard deviation). In the subsequent columns, the original mean depth and age scales of the respective ¹⁰Be records, their depth scale after synchronization on Vostok’s ¹⁰Be data, the updated Δ¹⁴C age scale, the ¹⁰Be concentrations relative to the NIST standard, their normalized values, and, if applicable, the measured ¹⁰Be concentrations (for Dome F, WAIS Divide, Little Dome C, and EDML), are detailed. For Dome Fuji and Siple Dome, the synchronizations were made on the age scale due to the absence of depth information and the difficulty of performing this task using the depth scale, respectively. The resulting ¹⁰Be stack derived from the synchronized ¹⁰Be time series is shown in the file stack_calculation_on_Vostok.xlsx. For that, the ¹⁰Be time series were resampled every 22 years, normalized through division by their mean (“resampling and flux calc” tab), and averaged at each interpolated age (“stack conc” tab). The same work has been performed with fluxes in the “stack flux” tab.

A similar synchronization exercise has been performed on the PS1 South Pole’s ¹⁰Be data for the SPICE, EPICA Dome C, Dome A, WAIS Divide, Dome F, Siple Dome, Little Dome C, and Vostok ¹⁰Be records. The results are shown in the excel file synchronisations_on_PS1.xlsx. Similarly to the file synchronisations_on_Vostok.xlsx, the PS1’s ¹⁰Be data is displayed in the first four columns of each tab, including the mean depth, original age scale, ¹⁰Be concentrations, and normalized ¹⁰Be concentrations. In the subsequent columns, the original mean depth and age scales of the respective ¹⁰Be records, their depth scale after synchronization on PS1’s ¹⁰Be data, the updated age scale, the ¹⁰Be concentrations relative to the NIST standard, their normalized values, and, if applicable, the measured ¹⁰Be concentrations, are detailed. The resulting ¹⁰Be concentration stack derived from the synchronized ¹⁰Be time series on PS1’s ¹⁰Be data is shown in the file stack_calculation_on_PS1.xlsx.

The Δ¹⁴C time series based simulated with the 12-box ocean carbon cycle model by using Vostok and stack’s ¹⁰Be concentration time series are reported in the file C14_simulations.xlsx. A README describing each column of the sim_D14C result tab is included.

The configuration files to perform the synchronisations with the Match software are available in the zip file conf_file_match.zip.

Technical validation

¹⁰Be profiles as a dating tool

Applying the ¹⁰Be - ¹⁴C approach

In this work, we have repeated the approach followed by Raisbeck et al.²⁴ with the following differences. First, we have combined the BH1 and BH2 records over their common part but excluding the top 15 meters (covering the last 250 years) as these cores were obtained using a thermal drill with an associated risk of modifying the ¹⁰Be profiles over this upper part. Second, we have applied the Match protocol⁵⁵, a wiggle matching approach different from the one used by Raisbeck et al.²⁴. This protocol was first applied to correlate the BH1 and BH2 ¹⁰Be profiles over their common part - between 15 and 51 m depth - and derive an average profile over this part (Fig. 4). The match protocol appears quite accurate for this type of correlation as evaluated by the correlation to correspondences established from 6 volcanoes identified in both BH1 and BH2. The age difference between the two approaches has a mean value of 12 ± 8 years (1 σ).

The Match protocol was then applied to correlate the atmospheric ¹⁴C record derived from the combined BH1 and BH2 ¹⁰Be profiles using the 12-box carbon cycle model of Bard et al.²², with the detrended IntCal20^{34 14}C tree ring record (Fig. 5). With an age of 7119 yr BP at 178 m depth (compared to 7160 for AICC), the resulting timescale, hereafter the Vostok ¹⁴C timescale, never differs by more than 60 years from the AICC2012 glaciological timescale currently used to interpret Vostok data (Fig. 6). It also is consistent with that estimated by Raisbeck et al.²⁴.

Fig. 5

(a) Oxygen-18 record (grey), a proxy of temperature change, measured on BH7 and BH8 cores located 100 m north of BH1⁹². (b) Combined BH1 and BH2 ¹⁰Be record on the AICC timescale (in blue) and the Δ¹⁴C derived from the ¹⁰Be data (in red). (c) The same ∆¹⁴C redated by matching (in red) with the detrended IntCal20 ∆¹⁴C derived from tree rings (black) using the Match protocol. Time series in panels (b) and (c) are normalized by substracting the mean and dividing by the standard deviation.

Fig. 6

Age difference between the ¹⁴C timescale and 1) in blue: the AICC2012 timescale 2) in orange: the ¹⁰Be timescale obtained applying the match protocol to correlate the Vostok ¹⁰Be profile with profiles at WAIS Divide and EDML and 3) in brown: the same comparison performed for the PS1 core over the last 1100 years.

Correlating the ¹⁰Be profiles between different sites

In a second step, we have derived a timescale for the other sites by synchronising the corresponding ¹⁰Be profile with the Vostok one. To validate this approach, we have evaluated the accuracy of applying the match protocol to correlate ¹⁰Be profiles by synchronizing the WAIS Divide and EDML ¹⁰Be records and then comparing the result with an independent volcanic synchronisation established between these two cores⁶¹. For the periods over which ¹⁰Be data are available at both sites, the two approaches (correlation using either ¹⁰Be or volcanoes) compare quite well (Figure S1) with a small difference of −2 ± 11 yr (1σ).

Interestingly, the WAIS Divide core is absolutely dated by counting of annual layers observed in the chemical, dust and electrical conductivity records³⁵. Over the past 2400 years, the uncertainty envelope of the WD2014 timescale is estimated to be smaller than 5 years. Prior to this period, the uncertainty on annual layer counting is estimated by inferring a ¹⁴C record from ¹⁰Be data and comparing it with the independent IntCal13 radiocarbon calibration curve⁶². Allowing for a possible offset of a couple of decades over the period covered by the brittle ice section between about 577 and 1300 m depth, the overall accuracy of layer counting is estimated to be better than 0.5%, e.g. a maximum of ~35 years at 7200 yr BP. For the period without ¹⁰Be data at WAIS Divide (between 2339 and 5369 yr BP) we use the correlation between the Vostok and EDML ¹⁰Be profiles and transferred it on to the WAIS Divide core based on the detailed and very precise correspondence established between EDML and WAIS Divide from volcanoes (Figure S1). We thus obtain a second absolute Vostok timescale, the ¹⁰Be timescale, tied to WD2014 from present-day back to 7200 yr BP (Fig. 6). The two absolute timescales are in excellent agreement with an average difference of −9 ± 14 yr (1σ).

For the more recent part of this period - back to ~1100 yr BP - the PS1¹⁰Be profile¹⁹ is also absolutely dated with a precision ranging between 2 and 10 years thanks to the identification of 20 eruptions³⁸ followed by a slight revison of six of these eruptions^22,58. Here also the agreement between the two approaches is very good (Fig. 6) with a difference of −3 ± 11 yr (1σ).

To sum up we can conclude that the accuracy of the absolute ¹⁴C Vostok timescale is probably better than ± 15 yr (1σ) as suggested from the comparison of its ¹⁰Be profile with those available at WAIS Divide, EDML and South Pole.

¹⁰Be fallout over Antarctica

Geographical variability

As shown in Fig. 4, there is between the different Antarctic sites a large spatial variability, up to a factor of 7, in the mean ¹⁰Be concentration with higher values for low accumulation sites such as Vostok, Dome A and Dome F. This spatial variability over the Antarctic continent has already been documented generally from surface samples with a first approach dealing with the Dome C and Vostok sites⁶³. For the first time, we have in our study an excellent coverage of the Antarctic continent with 5 sites well distributed over the East Antarctic Plateau, one inland in west Antarctica and 2 in coastal areas which offers the possibility to document the geographical variability of the ¹⁰Be fallout over a large part of the Holocene.

As expected and already well documented⁶³, the snow accumulation effect can be partly compensated by calculating ¹⁰Be fluxes. This requires to have access to snow accumulation and its variation through time. This is available for Vostok, EPICA Dome C and EDML sites for which age models have been developed deriving accumulation from temperature itself estimated for the isotopic composition of the ice³⁹. For the WAIS Divide site, the accumulation is estimated combining layer counting with an ice flow model⁶⁴ with, due to the limited changes in climate during the Holocene, only small variations in accumulation during this period. Over the last 7000 years, its standard deviation is limited to 5% which justifies using a constant value for sites where there is no other estimate. We use estimates provided in literature for Siple Dome and Dome A while for South Pole, SPICE, Old Dome C and Little Dome C an average value was calculated directly based on the ¹⁰Be timescale developed for these ice cores. These accumulation values are reported for each core in the file stack_calculation_on_Vostok.xlsx.

Still, as illustrated in Figure S2 and Table 2, the range of variation of ¹⁰Be fluxes is large - not far from a factor of 2 between Vostok and WAIS Divide - with lower ¹⁰Be fluxes for the low accumulation sites. We also note that the EPICA Dome C flux is higher than in the two other cores drilled in the Dome C area, by respectively 23% (Little Dome C) and 14% (Old Dome C).

Numerous other sites have been investigated - e.g.⁶⁵ and references therein - with a good coverage in the Dronning Maud Land area thanks to a traverse upstream Kohnen station (the EDML drilling site) performed in 2006 by the Alfred Wegener Institute⁶⁵ and to data obtained during the austral summer 2007- 2008 by a swedish japanese team in its western part⁶⁶. This expedition also allowed to extend the upstream Kohnen station german traverse roughly 400 km towards Dome F station⁶⁵. More recently, in 2017 and 2019, the Japanese Antarctic Research Expeditions sampled sites along an inland traverse in eastern Dronning Maud Land⁶⁷. The routes followed during these Dronning Maud Land traverses are indicated on Fig. 1 along with all the sites where ¹⁰Be data are currently available.

These studies largely focus on the relationship between the ¹⁰Be concentration and accumulation change or other parameters relative to the sampling site, latitude, elevation or distance from the coast⁶⁷. The link with accumulation or related parameters is not surprising because, as for other aerosols, there is for ¹⁰Be a combination of wet and dry depositions. The ¹⁰Be concentration measured in the surface snow is the sum of its concentrations deposited by precipitation-related (wet deposition) and turbulent (dry deposition, which is inversely proportional to the accumulation) processes in the atmosphere. This results in a well documented linear relationship between ¹⁰Be concentration and the inverse of the accumulation both in western⁶⁵ and eastern Donning Maud Land⁶⁷. There is however a difference with the linear trend derived for the entire Antarctica⁶⁵ and in eastern Dronning Maud Land with a noticeable change in this linear trend at around 75°S (elevation 3500 m) as illustrated in Fig. 8 for the JARE traverse. This change is attributed to the dominance of wet deposition north of 75°S and of dry deposition south of this boundary⁶⁷.

In the following, we examine the ¹⁰Be fallout variability (see Table 2 and Fig. 8, S2 and S3) as illustrated from the 11 dated cores (BH1 and BH2 being merged in one single record) covering part of the Holocene, thus excluding Taylor Dome (difficult to date). We use the average values of ¹⁰Be concentration and flux calculated over the period from 237 to 985 yr BP (Table 2). When there is no data over this period (EDML and old Dome C) we take the value over the previous millennium and scale it using sites where data are available for the last two millennia (scaling factor 1.121). We proceed in the same way for Little Dome C where data are only available over a part of the last millenium (scaling factor 0.952). In turn, we avoid uncertainties associated with measurements covering too short periods with very large local associated variability⁶⁶ and the need for correction owing to different conditions of cosmogenic production variability⁶⁵.

For these 11 ¹⁰Be records, we confirm the existence of a linear relationship between the ¹⁰Be concentration and the inverse of the accumulation⁶⁵ from data covering shorter time periods with however a slightly higher slope (Figure S3). This higher slope results from the inclusion of Dome A data and significantly higher concentration than previously measured at Dome F which is confirmed by surface data obtained on samples collected during the JARE traverse⁶⁷.

Data-model comparison

The geographical distribution of these 11 cores allows a relevant comparison with simulations of ¹⁰Be fallout using AGCMs. This is interesting because a simulation using the state-of-the-art ECHAM6.3-HAM2.3 AGCM has been recently published^36,68 combining this aerosol-climate model with the latest ¹⁰Be production model (CRAC: Be⁶⁹). This ensures that ¹⁰Be fallout modeling is improved with respect to results previously based on simulations performed with the GISS model⁷⁰ and the previous ECHAM5 version^71,72 with lower resolution and less elaborated ¹⁰Be production models.

The main purpose of this AGCM modeling - and this is also holds true for the study of Zheng et al.³⁶ - is to answer the question of whether the ¹⁰Be deposition is proportional to global or local production rates, and in particular to investigate the “polar bias” debate. Hereafter, we will not discuss this important aspect but rather limit the use of the recent ECHAM6.3-HAM2.3 simulation results⁶⁸ to test the ability of such a modeling approach to correctly simulate the ¹⁰Be fallout over Antarctica both for concentrations and fluxes. Observed and simulated concentrations and fluxes, together with their ratios are reported for each site in Table 2. Their geographical distribution is illustrated in Figs. 7a and 7b. Both the observed inland increase of ¹⁰Be concentration and inland decrease for fluxes are well represented by the model. However there are some significant differences. This is not surprising given - as documented by⁷³ - the variability of aerosol transport processes between different sites which may be difficult to capture in a GCM simulation.

Fig. 7

Comparison between observed (colored circles) and ECHAM6.3-HAM2.3 results (colored background) for a) ¹⁰Be concentration and b) ¹⁰Be flux, respectively. The observed and modeled values for each station is reported on Table 2.

For concentrations, the comparison is satisfying (within ± 25%) for 10 out of the 11 ¹⁰Be profiles for which data have been measured or estimated over the last millennium (period 237 - 985 yr BP when possible). Observed concentration value at WAIS Divide is not correctly captured by ECHAM6.3-HAM2.3 with a modeled value 63% too large. For fluxes, the average of simulated values for the 11 cores is in reasonable agreement with the observed average (73.2 and 81.9 at/m2/s, respectively). There is a good agreement (within ± 25%) between simulated and observed fluxes for Siple Dome, EDML, little Dome C and EPICA Dome C while simulated fluxes are ~30% too high for Old Dome C (due to too high modeled accumulation rate), WAIS Divide - where the too large simulated concentration is partly compensated by a higher simulated accumulation – PS1, and SPICE. Instead, simulated values are systematically lower (at 43 and 42%, respectively) than observed ones for the three highest elevation sites (>3400 m), Vostok, Dome A and Dome F. Finally, the comparison between Dome F and EPICA Dome C sheds light on a noticeable weakness of the simulation: while the measured ¹⁰Be flux is 74% higher at Dome F than at EPICA Dome C, simulated values are quite similar (Table 2). Several reasons can explain such deficiency. The relatively coarse resolution means that the orography of Antarctic continent is smoothed compared to reality (Table 1), which influences the deposition processes, moisture transport and the precipitation patterns. Moreover, post-deposition effects such the redistribution of snow by the wind are not modeled. Also, the ECHAM6.3-HAM3.2 simulation has been performed for the period 2005-2013, whereas the observed values for the present study were considered for the period 237 to 985 yr BP, when the anthropogenic influence on climate was still limited.

In Fig. 8, ¹⁰Be concentration are reported with respect to the altitude, both for sites examined in this study and along the JARE59 traverse. As a common characteristic, these two data sets exhibit a steeper increase for higher elevation sites. Focusing on the six sites on the East Antarctic Plateau (the three sites in the Dome C area, Vostok, Dome F and Dome A), there is a well observed (r² = 0.80) linear increase of 0.480 • 10⁵ at.g^-1 per km for the measured ¹⁰Be concentrations (dark blue line in Fig. 8).

Fig. 8

¹⁰Be concentration with respect to the altitude for sites examined in this study with data (blue circles) and model results (red circles) and for sites along the JARE59 observational dataset (green circles). Regression lines are drawn for sites examined in this study (observed and modeled) and for JARE59 sites below and above 3500 m⁶⁷.

For simulated values, a weaker correlation (r² = 0.57) and a linear regression slope lower by 35% (0.320 • 10⁵ at.g^-1 per km) are found. This can be explained by the fact that the same grid cell encompassed the proximate station locations EDC and LDC, a consequence of the relatively coarse horizontal resolution of the ECHAM6.3-HAM2.3 simulation. Data obtained along the JARE59 route are organized along two different linear trends. A very low slope for the sites below 3500 m is found (0.060 • 10⁵ at.g^-1 per km). In contrast, for JARE59 sites above 3500 m, the slope is very steep (1.64 • 10⁵ at.g^-1 per km) and is much higher than for the 6 East Antarctic sites. Horiuchi et al.⁶⁷ concluded that the latitude of 75°S (or elevation of 3.5 km) is likely the most important boundary for ¹⁰Be deposition along this JARE inland traverse route in eastern Dronning Maud Land with a deposition regime dominated by wet deposition north of 75°S and by dry deposition south of this latitude. These authors also evoked the seasonal enhancement of stratosphere-troposphere exchange with an associated increase of ¹⁰Be concentration in austral summer.

Along this line, one can note that simulated fluxes are systematically underestimated for our high elevation sites located southward of 75°S (Fig. 7a, Vostok, Dome A and Dome F) while they are, for our other sites, correctly simulated or overestimated (Fig. 7a, WAIS Divide and Little Dome C). This could be explained by the difficulty of the ECHAM6.3-HAM2.3 simulation to account for such a boundary, the existence of which is thus somewhat supported by our data and their comparison with ECHAM6.3-HAM2.3 simulation. However, the ECHAM6.3-HAM2.3 simulation is overall very satisfying accounting correctly for the observed distribution of the ¹⁰Be fallout over Antarctica. It would be thus quite justified to use this ¹⁰Be version ECHAM6.3-HAM2.3 AGCM for other time spans such as glacial periods for which data are available at various sites (Dome C, Vostok, Dome F, EDML and WAIS Divide).

Towards a ¹⁰ Be stack for the last 7200 years

Following previous authors^45,58, we stacked the ¹⁰Be records from the various sites. One advantage of stacking at moderate resolution (22 yr) multiple records with various sample spacings is to smooth out abrupt and large ¹⁰Be spikes occurring during single years, such as extreme solar particle events^10,11,12 and volcanic eruptions^74,75. To derive a stack record aiming to faithfully represent the average ¹⁰Be fallout over Antarctica, it should be, in principle, better to combine ¹⁰Be fluxes rather than concentration. However, there are additional uncertainties linked with estimates of the accumulation along the core, estimated through its link with water isotopes for Vostok, EPICA Dome C and EDML (AICC 2012 timescale²⁸), derived by combining layer counting and a glaciological model at the WAIS Divide site⁶⁴ or limited to an average value due to the lack of data for the other sites. For all these reasons, we have calculated two stacks - one based on ¹⁰Be concentrations and the other one based on ¹⁰Be fluxes – which appears to be very similar.

We first calculated the mean value and standard deviation for each ¹⁰Be concentration time series interpolated on a common time scale as described in the previous section. The standard deviation, expressed in % of the mean value, varies between 10% for Vostok and 16.7% for WAIS Divide considered as a single series. It was previously broken down into two series due to a large gap (2327-5385 yr BP). The shallow part would lead to a standard deviation of 14.7% and the bottom part to 11.5%. The average of relative standard deviations is 13.5%. Each time series was then normalized to its own average value, a procedure which retains its relative variability.

The 11 normalized time series were then averaged at each interpolated age, leading to the stack record (Fig. 9), which exhibits a standard deviation of 9.9% of the mean. This stack variability is thus somewhat smaller than observed for individual time series as illustrated by comparing this mean value with the time series from Vostok, the only site which covers the entire period. This is expected because observed variations are partly due to uncorrelated noise in the different time series (analytical uncertainties and genuine site-to-site differences).

Fig. 9

22-yr time step ¹⁰Be concentration time series (in grey) after matching with the Vostok profile (in red) using the Match protocol. The resulting stack is in black. Time series are normalized through division by their mean value.

It is difficult to quantify uncertainties of the stack record because the raw time series have neither the same length nor the same measurement resolution, even if time series are interpolated at the same ages. In particular, the total number of records drops to two during a millennium-long interval (5300-4200 yr BP) covered only by the Vostok and EDML records. This makes it precarious to calculate uncertainties for each age based on a variable and sometimes low number of ¹⁰Be concentrations. As a surrogate we use the average variability of the standard deviations calculated from the individual time series. The standard deviation of these values is 2%, which can be applied to the value at each age of the stack. In summary, the stack curve exhibits a variability (i.e., standard deviation) of 10% with an uncertainty range between 8 and 12%.

A similar stack has been calculated with the flux time series (Figure S4). Both concentration and flux stacks are very similar, which is due to limited variations of the ice accumulation over the last millennia. Furthermore, only some sites have independent estimations of the time variation of ice accumulation, whereas a constant accumulation is assumed at other sites (ODC, PS1, SPI, SD, DA, LDC). In that case, the relative variations of the flux are exactly the same as for the concentrations.

Due to the limited availability of ¹⁰Be antarctic data for the Holocene, the focus of related studies has, long been limited to the last millenium with time series available at South Pole^19,22,23 and then Dome F^48,49,58. Four additional time series covering this millenium are now available (Vostok, WAIS Divide, Dome A and EPICA Dome C). The mean value defined using these records is shown in Figure S5 after defining a common timescale based on correlation with the well dated South Pole ¹⁰Be record. The two curves are quite similar but correlating with South Pole allows calculation of the mean value back to 1977 AD (instead of 1713 AD as in Figure S2). This millenium stack is very similar to the PS1 South Pole record¹⁹ initially used to estimate solar irradiance over the last 1200 years^22,23.

Towards a simulated ¹⁴ C record based on ¹⁰ Be

Both stacks for ¹⁰Be concentration and flux and their uncertainty ranges were used as input forcing curves for the 12 box-model²². Calculations start at 7100 yr BP with a model at preindustrial steady state (atmospheric ∆¹⁴C = 0). The ¹⁰Be record is introduced as a perturbation of the cosmogenic production relative to its mean. Figure 10 shows the resulting simulated ∆¹⁴C time series based on the ¹⁰Be concentration. As expected, the simulations based on concentration and flux are similar, taking into account the estimated uncertainties (Figure S6).

Fig. 10

Simulated ∆¹⁴C time series (black) based on the ¹⁰Be concentration stack used as an input of the 12-box model²². The blue and orange time series correspond to simulations with an enhanced variability of 8 and 12%, respectively, aiming to represent minimum uncertainties on the resulting ∆¹⁴C. The yellow curve corresponds to a simulation with the ¹⁰Be stack with an enhanced long-term trend to take into account the polar bias for geomagnetic changes and a reduced high frequency variability to correct for the polar amplification of heliomagnetic changes³⁶. An additional reduction of 5% of the variability has been applied to account for production sensibility differences for ¹⁴C compared to ¹⁰Be⁶⁹.

All ∆¹⁴C records are characterized by a long-term trend decrease of 20‰, notably between 7000 and 3000 yr BP. A similar multi-millennial trend exists in the ¹⁰Be stacks, but it is less obvious because large centennial peaks are superimposed (Fig. 9). This contrast is a direct consequence of the different behaviors of ¹⁰Be and ¹⁴C, variations of ¹⁴C being low-pass filtered by the carbon cycle (e.g. Figure 3 in²²).

The observed ∆¹⁴C record measured on tree-rings³⁴ is also characterized by a long-term trend over the Holocene. However, it is much steeper than the simulated trend based on ¹⁰Be. Between 7000 and 2000 yr BP, atmospheric ∆¹⁴C changes by about 100‰ and 20‰ in observed and simulated records, respectively. This long-term trend is linked to several cumulative phenomena as proposed in previous studies^{30,76,77,78,79}, in particular the geomagnetic field variations over the Holocene and the memory effect of the carbon cycle which integrates higher cosmogenic production before 7000 yr BP. Another source of variation are carbon-cycle changes that occurred during the Holocene as testified by atmospheric CO₂ variations⁸⁰.

The box-modeling approach assumes that ¹⁰Be measured in polar ice is representative of the mean global cosmogenic production. This assumption is mainly based on the efficient mixing of ¹⁰Be in the stratosphere, which partly homogenizes large production gradients. In reality, the atmosphere is incompletely mixed and correction factors have been suggested for a polar bias (e.g.^22,81) which can also be evaluated with 3D models^36,65,70,82. The further complexity is that this polar bias is different for the heliomagnetic and geomagnetic modulation of cosmogenic production: solar variations are enhanced at the poles whereas geomagnetic changes are subdued at high latitudes^20,22,83. The exact magnitudes of polar biases linked to geomagnetic and solar modulations are still unsettled. Indeed, some model simulations indicate that ¹⁰Be is well-mixed in the atmosphere⁸², while other model experiments^65,70 show significant polar enhancement of solar variations and attenuation of geomagnetic changes in polar regions.

It is likely that centennial variations are due to the Sun whereas the multi-millennial trend is due to slow changes of the geomagnetic field^{30,45,76,77,84,85}. As a first step to tackle these problems, we provide an additional ∆¹⁴C simulation obtained by applying simple corrections to the ¹⁰Be concentration stack: its long-term trend (as calculated over the 3-7 ka BP period) has been enhanced by 35% to take into account the maximum polar bias for geomagnetic changes and its high frequency variability has been reduced by 5% to correct for the maximum polar amplification evaluated for heliomagnetic changes³⁶. An additional reduction of 5% of the variability has been applied to account for production sensibility differences for ¹⁴C compared to ¹⁰Be⁶⁹.

The corrected ∆¹⁴C simulation is shown as a light-green line in Fig. 10. As expected, its long-term trend is enhanced by about 10‰ and its high frequency structures slightly reduced with respect to the standard case illustrated by the three other curves in Fig. 10. Nevertheless, the corrected ∆¹⁴C record is close to the variability upper bound (±12%), suggesting that our first order uncertainties are reasonable.

Finally, we have in Fig. 11 compared the detrended ∆¹⁴C simulation obtained using either the ¹⁰Be stack or the Vostok record alone with the detrended IntCal20 record. The Pearson’s correlation coefficient r and the Kendall’s τ coefficient of the ¹⁰Be stack with IntCal20 ∆¹⁴C calibration curve are equal to 0.62 (p = 6.89 • 10^-146) and 0.35 (p = 1.29 • 10^-82), respectively. For the Vostok record alone with IntCal20 calibration curve, those coefficients are lower (r = 0.50 (p = 1.29 • 10^-82) and τ = 0.28 (p = 6.12 • 10^-54), respectively). The difference between the two correlations is statistically significant at more than 99% according to a Steiger’s Z test (Z = 9.38, p-value = 8.99 • 10^-17). This suggests that the ¹⁰Be stack provides a more representative global record than using Vostok alone.

Fig. 11

Detrended ∆¹⁴C modeled values obtained using either the ¹⁰Be concentration stack (red curve) or the Vostok record alone (blue curve) with the detrended IntCal20 record (black curve).