An improved North-South synchronization of ice core records around the 41 K beryllium 10 peak

Using new high resolution 10 Be measurements in the NGRIP, EDML and Vostok ice cores, together with previously published data from EDC, we present an improved synchronization between Greenland and Antarctic ice cores during the Laschamp geomagnetic excursion ~41 ky ago. We estimate the precision of this synchronization to be ± 20 years, 20 an order of magnitude better than our previous work. We discuss the implications of this new synchronization for making improved estimates of the difference between ice and enclosed gas of the same age (delta depth), difference between age of ice and enclosed gas at the same depth (delta age) in the EDC and EDML ice cores, spectral properties of the 10Be profiles and phasing between Dansgaard/Oeschger-10 (in NGRIP) and AIM-10 (in EDML and EDC).


Introduction
In a previous study, Raisbeck et al. (2007) synchronized the Greenland GRIP and Antarctic EPICA Dome C (EDC) ice cores using the structured peak of cosmogenic 10 Be resulting from a combination of the low geomagnetic intensity and variable solar activity during the Laschamp geomagnetic excursion ∼ 41 kyr ago (Lascu et al., 2016, and references therein). The estimated precision of this synchronization was 200 years, due mainly to uncertainties associated with the GRIP 10 Be record (sample time resolution of 25-50 years, corrections for loss of 10 Be on 0.4 µm filters for some samples, and several missing samples). We improve this situation significantly here by using a much higher resolution (5-10 years) 10 Be profile in the NGRIP ice core. In addition we report and synchronize high-resolution 10 Be profiles from the peak region of two other Antarctic cores, EPICA Dronning Maud Land (EDML) and Vostok. Maps showing the locations and a brief description of drilling operations for the ice cores studied here can be found in Jouzel (2013). With these improvements, we estimate the uncertainty on the synchronization to be ±20 years at our new tie points, and ±35 years over the whole Laschamp event. This precision is supported by the study of Svensson et al. (2013), who, on the basis of our earlier synchronization, identified three volcanic signals which they believe are common to the Greenland and Antarctic cores. 218 G. M. Raisbeck et al.: An improved north-south synchronization of ice core records

Methods
A continuous series of 55 cm long bag samples from the NGRIP core were available from 2102 to 2140 m. Each of these were cut into five 11 cm samples weighing ∼ 100 g and representing 5-10 years. The samples were treated and measured on the Gif-sur-Yvette Tandetron AMS facility as described in Raisbeck et al. (2007). AMS measurements were made relative to NIST 10 Be / 9 Be standard SRM 4320, assuming a nominal value of 2.68 × 10 −11 . While the value of 2.79 × 10 −11 has now been adopted by many groups for this standard, for consistency we continue to use the value that was used for all the previous measurements at the Gifsur-Yvette Tandetron. Since this factor is constant, it has no effect on our synchronization procedures. An average of 400 10 Be ions were counted for each sample, and the estimated precision (not including the uncertainty of the standard) was ∼ 7 %.
Vostok was one of the sites (the other being the original Dome C site) where the 10 Be peak discussed here was initially found (Raisbeck et al., 1987). Although the estimated date was younger than presently accepted, and the origin uncertain, it was already pointed out there that such a peak was a potentially interesting stratigraphic marker. Those initial measurements were made on discontinuous samples from the Vostok 3G core. We subsequently made measurements on a nearly continuous sequence of 1 m samples from the 4G core, which showed a double-humped peak (Raisbeck et al., 1992). It was in an effort to look for even finer structure that the measurements shown here were made. These samples were ∼ 10 cm in length and weighing > 500 g, taken from the 5G core between 580 and 620 m. As can be seen in Fig. 1d, the sampling depth is not centered on the peak. This is because the sampling was based on the initial identification in 3G and, as we discovered, there is apparently a ∼ 10 m offset between 3G and 5G. Since these measurements were carried out > 15 years ago, also at the Gif-sur-Yvette AMS facility, the sample preparation and measurement procedures are those described in Raisbeck et al. (1987). For almost all samples at least 1000 10 Be ions were counted, leading to an estimated measurement uncertainty of ∼ 6 %.
The EDML 10 Be measurements were performed at ETH Zurich with a sample preparation that is based on the common procedure already applied for the GRIP ice core (Yiou et al., 1997;Muscheler et al., 2004). Samples of 25 cm length corresponding to ∼ 15-year resolution and a typical weight of 110 g were melted and processed without any further filtering. The 10 Be / 9 Be ratios were measured using the compact AMS facility Tandy at ETH Zurich (Müller et al., 2010) resulting in uncertainties < 3 %. Results were normalized to the internal standard S2007N with a reference value of 10 Be / 9 Be = (28.1 ± 0.8) × 10 −12 (Christl et al., 2013). The EDC 10 Be measurements have been described in detail previously .
The climate records are provided by high-resolution isotopic profiles (either δD or δ 18 O) measured along these ice cores. δD is used as primary climate indicator for Vostok and Dome C, while δ 18 O is used for NGRIP and EDML. In addition to existing water isotopic records, new high-resolution measurements have been performed for this study on the Vostok 5G ice core (10 cm). These isotopic data are shown in Fig. 1a to d, with an additional smooth curve (see legend).
From the 10 Be concentrations, 10 Be fluxes of the four cores were calculated using accumulation rate estimates from water isotopes, ice thinning and dating constraints by Parrenin et al. (2007) for EDC, Ruth et al. (2007) for EDML, and Parrenin et al. (2004) for Vostok, and they were deduced from layer counting (Svennson et al., 2008) combined with the ss06bm thinning model of Johnsen et al. (2001) for NGRIP. Alternative accumulation rate determinations were obtained more recently through dating exercises such as for the AICC2012 timescale. These determinations do not show significant differences with previous determinations over our period of interest Veres et al., 2013;Andersen et al., 2006;Lemieux-Dudon et al., 2015). Only for the NGRIP ice core was it recently suggested that the mean accumulation rate of the MIS 3 could be overestimated by 20-30 % (Guillevic et al., 2013;Kindler et al., 2014;Lemieux-Dudon et al., 2015). Still, even when using this modified accumulation rate, the shape of the variations of 10 Be flux on the Laschamp event recorded at NGRIP is not significantly affected. This is because, for our purpose, it is the relative accumulation rates which are important.

Results
The results for the 10 Be concentrations (see Supplement) and fluxes of the four cores are shown in Fig. 1 as a function of their depth along with their stable isotope profiles. As previously noted (Yiou et al., 1997;Raisbeck et al., 2007) all cores show a highly structured peak of 10 Be centered on AIM-10 (EDC, EDML, Vostok) or DO-10 (NGRIP). It is this structure that allows us to make multiple synchronizations over the whole width of the peak, as shown in Fig. 2. This synchronization was done in two ways. Initially we selected five tie points (four peaks and one valley, Table 1 and Fig. 2) which appeared to us to be clearly common to all the profiles, and synchronized using the AnalySeries program of Paillard et al. (1996). Subsequently, in an effort to avoid the subjectivity of choosing tie points, we did the synchronization using the MATCH protocol of Lisiecki and Lisiecki (2002). The difference between these two procedures never differed by more than 42 years for NGRIP-EDC and 27 years for NGRIP-EDML. The maximum difference for the two procedures was for the period between tie points C and D, and was very small (< 10 years) near the tie points. This confirms our visual impression that the chosen tie points are indeed the most robust common features of the profiles. For Vostok, Figure 1. 10 Be concentrations, 10 Be flux and stable isotope profiles (δD or δ 18 O) and their running averages as a function of depth and age in four ice cores: (a) EPICA Dome C, EDC3 timescale ; (b) NGRIP, GICC05 timescale (Andersen et al., 2006); (c) EDML, EDML1 timescale ; and (d) Vostok, FGT1 timescale (Parrenin et al., 2004). The 10 Be flux has been calculated using accumulation rates used in the above timescales and smoothed with a 50-year binomial filter. AIM stands for Antarctic isotope maximum.
we were unable to get a satisfactory synchronization for the MATCH protocol without imposing severe restrictions. This is probably due to several missing sample sections for that record.
While we believe there is considerable evidence that flux is a better measure of 10 Be (and other trace species) deposition in polar cores, at the suggestion of a reviewer we have repeated the Match synchronizations of EDC and EDML with NGRIP using concentrations. We found that the four peaks and valley corresponding to the tie points shown in Fig. 2 did not vary by more than 20 years compared to those found using fluxes. In summary we believe our 10 Be synchronization is quite robust to the choice of raw or smoothed accumulations, or even the use of concentration.
In order to estimate the precision of our synchronization, we have used two procedures. Initially we adopted a procedure which involved looking at the correlation of EDC with EDML based on their 10 Be synchronization with NGRIP, and compared it with the direct correlation of EDC and EDML by Severi et al. (2007) using common volcanic peaks. The results are shown in Fig. 3a. As can be seen, the agreement is excellent. For the 6 volcanic peaks of Severi et al. (2007) that fall within our 10 Be tie points, the average difference between the observed and predicted depths corresponds to 4.7 ± 6.2 years (2σ standard deviation). One of the reviewers (C. Buizert) suggested that this "would provide a true test of the uncertainty in the 10 Be synchronization". However, we feel that, while it does strongly support our synchroniza-  (Lisiecki and Lisiecki, 2002) for EDC and EDML, or using five tie points (A-E) and AnalySeries (Paillard et al., 1996) for Vostok. For each site the detailed fluxes (histograms calculated using 11 cm accumulation rates for NGRIP, 55 cm for EDC, 50 cm for EDML and 10 cm for Vostok) and results smoothed with a 50-year binomial filter (curve) are shown. Table 1. The first two columns report the depth and GICC05 ages (Svensson et al., 2008) of our 10 Be tie points (BeA-BeE) and the volcanic markers (L1-L3) of Svensson et al. (2013) at the NGRIP site (see text). The following columns provide the depths of corresponding depths for EDML, EDC and Vostok ice cores either observed or predicted. tion procedure, it may underestimate the uncertainty between NGRIP and the Antarctic cores, as discussed below. As a second estimate of the precision of our synchronization, we can look at the volcanic correlations of Svensson et al. (2013). Starting from an earlier 10 Be synchronization, and layer counting results in both NGRIP and EDML, those authors identified three volcanic spikes they believed were common to NGRIP-EDML, two of which are linked to EDC by Severi et al. (2007). In Table 1 and Fig. 3b and c, we show the position of those events as found by our improved 10 Be synchronization. For EDML the differences are 66 cm (∼ 44 years) for L1, 51 cm (∼ 28 years) for L2 and 44 cm (∼ 24 years) for L3. In EDC the differences compared to those given by Svensson et al. (2013) are 27 cm (∼ 23 years) for L2 and 36 cm (∼ 31 years) for L3. As mentioned above, we believe our tie points have the most robust synchronized ages. It is therefore reasonable to expect the difference in predicted ages to increase with distance from those tie points. However, L1 is an exception, having the largest difference (66 cm) despite being the nearest (∼ 1 m) to a tie point (BeC). If that tie point is correct, it is thus reasonable to question whether L1 is indeed a bipolar event. In fact L1 was not listed as one of those used by Severi et al. (2007) to correlate EDC with EDML.
For the other two volcanic peaks, it can be noted that the age differences are systematic, having an average value of 27 ± 7 (2σ ) years for L2 and L3. Thus, the relative predicted ages between EDML and EDC for L2 and L3 dif-  Severi et al., 2007) and proposed bipolar volcanic points (full black squares from Svensson et al., 2013). We have indicated the uncertainty for the tie points used in Loulergue et al. (2007). For the new 10 Be and volcanic tie points, the uncertainties are smaller than their symbols. fer by less than 10 years, consistent with their identification as the same event by Severi et al. (2007). However, 27 ± 7 years is significantly larger than the 4.7 ± 6.2-year precision found above. We must therefore conclude that either L2 and L3 of Svensson et al. (2013) are also not bipolar, or that 4.7 ± 6.2 years underestimates our real uncertainty between NGRIP and EDC/EDML. We tend to favor the second explanation. The reason is that the 10 Be peaks involved are irregular and have durations of the order of 100 years. Given this, and the 10 Be sampling resolution, it seems unlikely to us that they can be synchronized to a precision of less than 10 years. Our synchronization procedure, either by eye (AnalySeries) or Match protocol, will maintain a tight correlation between EDC and EDML when synchronizing with any feature in NGRIP. However, if for some reason (higher and more variable ice accumulation rate, higher sampling resolution) the form of the 10 Be flux peaks at NGRIP differs slightly from those in EDC/EDML, then the correlation between NGRIP and the Antarctic cores may be offset by several decades. For this reason, we prefer to be conservative and adopt as our uncertainty the value implied assuming that the volcanic peaks L2 and L3 of Svensson et al. (2013) are indeed bipolar.
In summary, if L2 and L3 are indeed bipolar events, and L1 is not, we estimate from the above that the uncertainty in our synchronization is ≤ 35 years over the whole Laschamp event, and ≤ 20 years at our tie points. This is an order of magnitude better than in Raisbeck et al. (2007). It can also be compared with the 2σ uncertainty in the methane interpolar synchronization (±73 years) estimated by Buizert et al. (2015) at this age. This shows that, at least in regions where there is significant structure in 10 Be profiles, it is possible to link Greenland and Antarctic climate records with decadal precision. As pointed out by Svensson et al. (2013), using such correlation as a framework, it may be possible to find other common volcanic signals in ice cores from the two hemispheres.

Spectral properties of the 10 Be profiles
As indicated in the Introduction, the centennial variations in the 10 Be profiles are believed to be caused by variations in solar activity. For example, during the Holocene, Steinhilber et al. (2012) found periodicities of 88 years (Gleissberg cycle) and 210 years (de Vries cycle) in a composite of tree ring 14 C and ice core 10 Be records. For the Laschamp period, Wagner et al. (2001) reported a ∼ 205-year periodicity of 10 Be in the GRIP core. It is thus interesting to do a spectral analysis on the higher resolution and better quality NGRIP 10 Be record reported here. This is shown in Fig. 4, where a very significant (> 99 %) 200-year signal is found in the Fourier spectrum (REDFIT program of Schulz and Mudelsee, 2002). As can be seen in the wavelet spectra, however (MATLAB package of Grinsted et al., 2004), this periodicity is really only significant for a short interval near the end of the time period studied. While there is also a short interval where an 88-year periodicity appears, this is not significant in the Fourier spectrum.
Since we now have the EDC and EDML 10 Be profiles synchronized on the NGRIP timescale, we might expect these to show similar periodicities. Somewhat to our surprise, however, this is only partially the case. While both show a ∼ 200year signal in the same time region of the wavelet analyses, these are barely significant (∼ 90 %) in the Fourier spectra. This is because that peak appears to be distorted by longer signals of ∼ 290 and 350 years in EDC and EDML, respectively. This is consistent with the observation of Cauquoin et al. (2014), who found that the 210-year periodicity in the EDML Holocene data used by Steinhilber et al. (2012), the only Antarctic core in their composite, was only significant over short and sporadic time periods. It is also consistent with the fact that Cauquoin et al. (2014) did not find any evidence for this periodicity in 10 Be from EDC during the interval 336-325 ka BP. It thus appears that there may be meteorological effects in the Antarctic (smaller ice accumulation rate?) which are relatively more important than in Greenland, and which tend to diminish or interfere with the 10 Be production signal recorded in the ice, compared to Greenland.

Implications for ∆depth and ∆age estimates
Using preliminary values of our previous 10 Be synchronization, together with methane correlations,  estimated the difference in depth ( depth) and age ( age) between the ice and gas records in EDC and EDML. The discrepancy they found at EDC compared with the classical firn densification model (Goujon et al., 2003) was in fact the first hard evidence that this model was inadequate for glacial ice from low accumulation sites in Antarctica. This had important implications for the EDC4 gas age construction. Loulergue et al. (2007) hence proposed that the age obtained by the Goujon et al. (2003) model using temperature and accumulation rate deduced from water isotopes (scenario 1 in Fig. 5 and Loulergue et al., 2007) was not appropriate. Instead, age for the EDC gas age was calculated with the Goujon et al. (2003) model but with an artificially higher accumulation rate than the one used in the construction of the EDC3 ice timescale (scenario 4 in Fig. 5 and Loulergue et al., 2007), so that the calculated age at 41 kyr was in agreement with the age determination using 10 Be and CH 4 synchronization.
In the construction of the AICC2012 chronology Veres et al., 2013), a different approach for the age calculation has been chosen. The construction of AICC2012 relies on the use of a Bayesian tool, DATICE , to optimize the chronology of five ice cores using absolute dating constraints and stratigraphic links in the gas and ice phases. The AICC2012 age calculation for EDC is hence constrained by the CH 4 and 10 Be data as in Loulergue et al. (2007). It also benefits from the recent finding of Parrenin et al. (2012) showing that the firn lock-in depth at Dome C during the last deglaciation was best determined from δ 15 N measurements in air trapped in ice cores rather than by outputs of firn modeling. As a conse- quence, the background scenario used for the lock-in depth and hence the age in the construction of the AICC2012 chronology for EDC was based on δ 15 N measurements performed at Dome C (Dreyfus et al., 2010;Landais et al., 2013), leading to a significantly smaller age than the firn model output for EDC during the glacial period. It should be emphasized that no firn model has been used for the background scenario of the EDC lock-in depth in the construction of AICC2012, which makes the comparison of age and depth between the AICC2012 and EDC3 gas timescales not straightforward.
Using the five tie points from the present synchronization, we have used a slightly modified version of the technique described by Loulergue et al. (2007) to calculate depth of EDC and EDML. Instead of correlating the methane profiles at a single point, we have correlated the whole profiles using the Match protocol (Fig. 6). Then, a age has been determined for each 10 Be peak. The resulting ages and depths (Fig. 5) are compared to the depth and age provided by Loulergue et al. (2007) for the construction of the EDC3 gas timescale. Our results are in reasonably good agreement with the age and depths calculated using scenario 4 in Loulergue et al. (2007). Our results are also in good agreement with the AICC2012 age at EDC, hence validating the chosen approach for delta age calculation. It is clear from Fig. 6 that the resolution of the methane data in EDC and EDML, particularly in the period 41.3-42.5 kyr, is the limiting factor in calculating the age uncertainty.

Implications for bipolar seesaw and stable isotope interpretation
The last glacial period is characterized by a succession of millennial-scale climatic events identified both in Greenland and in Antarctica. In Greenland, the climate shifts are very abrupt with a temperature shift between a cold Greenland stadial (GS) to a warm Greenland interstadial (GIS) designated as Dansgaard-Oeschger (DO) events. The DO succession and amplitude seems quite unpredictable even if it has already been suggested that the sequence of the abrupt variability of the last glacial period (length and frequency of GS and GIS) can be affected by the long-term (orbital) climatic variability . In Antarctica the temperature changes of the corresponding climatic events are much smoother, and are designated as Antarctic isotopic maxima (AIM). A clear relationship has been demonstrated between DO and AIM events: each DO event is associated with an AIM (EPICA community members, 2006;Jouzel et al., 2007) and the temperature/isotopic maximum of the AIM occurs approximately synchronously with the Greenland abrupt temperature increase (Fig. 7). The observed relationship between Greenland and Antarc- tica temperature evolutions led to the proposed theory of the bipolar seesaw (Broecker, 1998;Stocker, 1998). The global picture of the bipolar seesaw can be explained by a simple modeling of a slow thermal response of Antarctic temperature to Greenland abrupt warmings and coolings through a heat reservoir in the Southern Ocean (Stocker and Johnsen, 2003). The knowledge of the exact relative timing between DO events in Greenland and AIM events in Antarctica is limited by the ice core chronologies. Usually, Greenland and Antarctic ice cores are synchronized using the global atmospheric signals in the air trapped in ice core (CH 4 , δ 18 O of O 2 ). However, the climatic signals are recorded in the ice phase through water isotopic composition. Because the entrapped air is systematically younger than the entrapping ice, by centuries to millennia, accurately determining this age difference is crucial for the Greenland vs. Antarctica phasing.
Recent studies have allowed for the age uncertainty on Greenland vs. Antarctica climatic sequences to be significantly decreased. This has been done through an increase in the number of stratigraphic links between Greenland and Antarctic ice cores (e.g., Schüpbach et al., 2011;Svensson et al., 2013) and new high-resolution Antarctic records at relatively high accumulation sites, and hence with small ice-air age differences (WAIS Divide Project Members, 2015). These recent improvements have revealed a refined sequence of Greenland vs. lower latitudes and Antarctic climatic changes during the last glacial period. First, for GS associated with an occurrence of Heinrich events, it has been evidenced that a decoupling exists between the Greenland or high-latitude climate and lower latitudes, with Greenland remaining in a cold stable phase, while lower or Southern Hemisphere latitudes exhibit significant climatic variations (Barker et al., 2015;Guillevic et al., 2014;Rhodes et al., 2015;Landais et al., 2015). Second, a comparison of several climatic records in Antarctica with the Greenland sequence has highlighted regional differences in the shape of the AIM events and in the phasing between the temperature increase in Antarctica and the abrupt temperature change in Greenland over the long GSs that are associated with a strong discharge in the North Atlantic, i.e., Heinrich stadials (Buiron et al., 2012;Landais et al., 2015). Third, an accurate dating exercise on the west Antarctica WDC ice core over the last 60 kyr suggests that the decrease in Antarctic temperature occurs ∼ 200 years after the abrupt temperature increase in Greenland (WAIS Divide Project Members, 2015). This has been interpreted as a "northern push for bipolar seesaw" (van Ommen, 2015). Still these studies suffer from the same limitation associated with the ice-air age difference and rely on determination of the past depth of the firn, based on firn densification and/or measurements of δ 15 N of N 2 in the trapped air.
Here we use the direct synchronization of NGRIP with EDC, EDML and Vostok based on the 10 Be records. In Fig. 8 we show the 10 Be-synchronized climate records of the four cores, as represented by the δD and δ 18 O profiles. Comparing the three Antarctic records, one can observe that, while they have the same general features, there are also significant differences in detail. These are probably due to meteorological effects such as seasonal variation in isotopic ratios and depositional effects. This shows that one must be careful to not over-interpret details of individual stable isotope profiles in low-accumulation regions of the Antarctic as a climate proxy. This is particularly evident in the Vostok 5G profile, which differs significantly from the previously published profile from the 3G core .
We confirm the Greenland vs. Antarctica relationship observed by Raisbeck et al. (2007) with the increase in δD or δ 18 O in Antarctica leading the abrupt δ 18 O increase in NGRIP by several centuries, which is a classical feature of the bipolar seesaw (Blunier and Brook, 2001), operating here for a small DO (AIM) event. The high resolution of our water isotopes and 10 Be records also enables us to look at the finescale Greenland vs. Antarctica temporal relationship around the abrupt warming recorded at NorthGRIP. Although it is impossible to identify by eye any clear inflection point corresponding to AIM 10 in Vostok, a significant peak of δ 18 O is observed at WDC and EDML about 200 years before the abrupt δ 18 O increase at NorthGRIP, while the most significant peak in EDC is observed about 150 years after the abrupt Greenland δ 18 O increase. The signal is not unambigu- BREAKFIT software (Mudelsee, 2010) EDML (this study) −213 ± 40 years −207 ± 50 years EDC (this study) +130 ± 30 years +150 ± 50 years Vostok (this study) Non-significant Non-significant WDC, WAIS Divide Project Members (2015) −156 ± 50 years −129 ± 50 years Figure 8. Phasing between the GS-GIS transition in Greenland (vertical orange line) and the maximum of the AIM 10 in 4 different Antarctic ice cores on the GICC05 age scale. The synchronicity between Vostok, EDC, EDML and NGRIP is based on the 10 Be records while the WDC isotopic record has been drawn on the GICC05 timescale using the correspondence between the WSD timescale and GICC05 given in Buizert et al. (2015).  (2015), we have used the same statistical approaches, i.e., the BREAKFIT software by Mudelsee (2010) and a similar automated routine (referred to as MATLAB routine in Table 2) using a second-order polynomial (rather than a linear for BREAK-FIT) fit to the data. The water isotope profiles for EDML and EDC show several wiggles associated with a variability of ∼ 100-200 years. Despite this variability, it is still possible to identify a maximum for AIM 10 using the aforementioned statistical tools, keeping in mind limitations linked to the chosen range of detection and short-term variability in δ 18 O signal. The two statistical tools identify inflection points on the EDML and EDC δ 18 O records. Using both methods, the maximum is reached at EDML, ∼ 210 years before the abrupt Greenland warming, while the maximum detected at Dome C appears to be ∼ 140 years later. To complete this study, using the same statistical tools we have determined the shift between the abrupt Greenland warming and the maximum in the water isotopic record at WDC, and found that the WDC δ 18 O maximum is ∼ 142 years earlier. Such analysis leads to the conclusion that all ice cores do not display the same isotopic signal over an AIM, a result in line with previous studies highlighting a strong regional variability at the AIM scale in Antarctica (e.g., Buiron et al., 2012;Landais et al., 2015).
As suggested by a referee (C. Buizert), a stack can be drawn using the 3 δ 18 O records from EDML, EDC and WDC (Fig. 8). Because we could not get a 10 Be synchronization with the Match protocol, and unlike EDC and EDML have no independent evidence from volcanic spikes to support our estimated precision, we prefer not to include Vostok in the stack, although tests including it show no significant influence on our conclusions. We emphasize that constructing such a stack is appropriate only if one assumes "a priori" that the "true" climatic part of the δ 18 O records for the three core locations are synchronous at the studied timescale, which remains to be shown. This stack clearly shows the δ 18 O decrease observed 200 years after the abrupt δ 18 O increase in Greenland, but it also depicts an inflection point ∼ 200 years (185-219 years according to the MATLAB routine) before the Greenland abrupt δ 18 O increase, so that on the stack, δ 18 O displays a ∼ 400-year-wide plateau at the time of the abrupt Greenland δ 18 O increase. Such a plateau is not unexpected and follows from our observation that in some ice cores (WDC, EDML) the maximum of the AIM was before the abrupt δ 18 O increase in Greenland, while in another (EDC) the maximum of the AIM occurs after the abrupt Greenland δ 18 O increase.
This result seems at first sight to nuance the conclusions presented in WAIS Divide Project Members (2015) using the same approach for the determination of the breakpoint. In both the EDML and WDC ice cores, we find that the maximum in Antarctica for AIM 10 is reached more than 140 years before the Greenland warming. The stack of Antarctic δ 18 O records over AIM 10 also suggests that the maximum δ 18 O level is reached ∼ 200 years before the Greenland warming. This suggests one must be cautious in drawing a conclusion about a general mechanism for AIM vs. DO dynamics based on a stack of several cores for a given DO-AIM event. Similarly, one must be cautious about drawing conclusions based on a stack of different DO-AIM events in the same core, as proposed in WAIS Divide Project Members (2015). Indeed, the lead/lag of Greenland vs. Antarctica may be different from one event to another. This is evident from Fig. 7, where the maximum of AIM 7, 8 and 10 clearly leads the abrupt warming in Greenland both at WDC and EDML. Both stacking several AIM together on one core or stacking one AIM using several cores can be misleading when discussing short-term variability that may be different from one core to another, or one event to another.
Finally, even if the location of the maximum of the AIM, hence the statement of "northern push for bipolar seesaw", can be disputed, we confirm that the main δ 18 O decrease in Antarctica occurs ∼ 200 years after the Greenland temperature increase. This particular result is robust in different Antarctic sectors and for different DO/AIM events and should be retained to refine our understanding of bipolar seesaw.

Conclusion
We have shown here that, in periods where there are significant production variations of 10 Be, it is possible to synchronize Greenland and Antarctic ice cores with decadal precision. This in turn means that the climate and other environmental parameters registered in these ice cores can be synchronized at this same precision, thus allowing different models and mechanisms to be more finely tested. The Supplement related to this article is available online at doi:10.5194/cp-13-217-2017-supplement.
Acknowledgements. This project was initiated during discussions between the first author and Sigfus Johnsen at an EPICA meeting at San Feliu, Spain, in 2003. Sigfus was always a strong supporter of 10 Be measurements in ice cores, and enthusiastically responded positively to the idea of using high-resolution NGRIP samples to overcome limitations encountered with GRIP samples. We therefore dedicate this paper to Sigfus' memory. We thank C. Buizert and an anonymous reviewer for helpful comments. We would like to acknowledge the many people involved in logistics, drill development and implementation, ice core processing and analysis; without them this work would not have been possible. This includes NorthGRIP, directed and organized by the Department of Geophysics at the Niels Bohr Institute for Astronomy, Physics and Geophysics, University of Copenhagen, and supported by funding agencies and institutions from participating countries, and EPICA, the European Project for Ice Coring in Antarctica, a joint European Science Foundation-European Commission scientific program, funded by the EU and by national contributions from Belgium, Denmark, France, Germany, Italy, the Netherlands, Norway, Sweden, Switzerland, and the UK. We thank all Russian and French participants of the Vostok drilling, field work and ice sampling. The research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/20072013)/ERC grant agreement no. 30604.
Edited by: K. Goto-Azuma Reviewed by: C. Buizert and one anonymous referee