Abrupt changes in the southern extent of North Atlantic deep water during Dansgaard Oeschger events

Transcription

1 SUPPLEMENTARY INFORMATION DOI: /NGEO2558 Abrupt changes in the southern extent of North Atlantic deep water during Dansgaard Oeschger events Julia Gottschalk 1*, Luke C. Skinner 1, Sambuddha Misra 1, Claire Waelbroeck 2, Laurie Menviel 3,4, Axel Timmermann 5 1 Godwin Laboratory for Palaeoclimate Research, Earth Sciences Department, University of Cambridge, Downing Street, CB2 3EQ Cambridge, UK; *jg619@cam.ac.uk 2 LSCE/IPSL Laboratoire, CNRS-CEA-UVSQ, Gif-sur-Yvette, France 3 Climate Change Research Centre, University of New South Wales, New South Wales 2052, Australia 4 ARC Centre of Excellence for Climate System Science, University of New South Wales, New South Wales 2052, Australia 5 International Pacific Research Center, Department of Oceanography, SOEST, University of Hawaii, Honolulu, USA NATURE GEOSCIENCE 1

2 Synthesis of high-resolution sediment cores in the Atlantic Ocean In Table S1, we have compiled sediment cores recording the Marine Isotope Stage (MIS) 3 interval with a temporal resolution better than 200 years, which has been suggested to permit the reconstruction of Dansgaard-Oeschger (D-O) climate variability 1. The spatial distribution of these records shows that hydrographic changes in the (sub-)surface- and deep ocean over D-O cycles are particularly well documented in the North Atlantic 1 (Table S1, Fig. 1). The distinct paucity of well-resolved and well-dated sediment records from the southern hemisphere has so far hampered our understanding of the nature and modes of interhemispheric climate variability and its underlying mechanisms. Chronology of sediment core MD Q The age model of sediment core MD Q described in detail in ref. 2 is based on 59 radiocarbon measurements 3 and has been extended by a stratigraphic alignment of minima in the abundance of the cold-water, planktonic foraminifer species Neogloboquadrina pachyderma (s.) with the first time derivative of the Antarctic temperature proxy D in the EPICA Dome C (EDC) ice core on the AICC2012 age scale 4 (Fig. S1). The planktonic radiocarbon ages have been transferred to calibrated calendar ages according the IntCal04 and Cariaco dataset references 5,6 using the Bayesian statistical software package BCHRON (ref. 7). Prior to calibration, planktonic radiocarbon dates have been corrected for variable reservoir ages corrections 3. The calibration with a different atmospheric reference dataset (and accordingly revised reservoir age corrections) has been shown to result in only minor differences in the chronology 8. We refer to the originally published age-scale 2 of the deglacial section of sediment core MD Q for clarity and simplicity. To establish chronostratigraphic constraints during the last glacial period based on an alignment of abundance minima of the cold-water planktonic foraminifer N. pachyderma (s.) 2

3 to the first derivative of the Antarctic temperature proxy 2, we determined the abundance of N. pachyderma (s.) by census counts of at least 300 whole planktonic foraminifera from a sample aliquot of the >150 µm size fraction. Abundances of N. pachyderma (s.) in surface waters and core-top sediments at present-day rapidly increase south of the sub-antarctic Front 9,10. Thus, recurrent minima of N. pachyderma (s.) abundances recorded in MD Q may reflect rapid shifts of the sub-antarctic Front during Antarctic warming intervals in response to North Atlantic climate perturbations, and associated rapid adjustments of the atmospheric circulation and the northern boundary of the Antarctic Circumpolar Current 9,11,12. This is consistent with the observation of a lead of N. pachyderma (s.) abundance variations with respect to planktonic foraminifer Mg/Ca-derived sea surface temperature (SST) changes in the Cape Basin 9, and arguably in MD Q (Fig. S1, S2). It has been proposed that planktonic Mg/Ca-derived SST variations more likely reflect changes in southern highlatitude SST in parallel with Antarctic temperature 9. To establish an age model which is in line with an in-phase relationship between Antarctic air temperatures and sub-antarctic SST which bears on the assumption of a thermal equilibrium of surface water- and air temperatures in the southern high-latitudes and accounts for the lead of N. pachyderma (s.) assemblages over Mg/Ca SST changes, we assume that the major latitudinal shift of the sub- Antarctic Front in the Southern Ocean (and thus the minima in polar species foraminifer assemblages at the core site) occured prior to the peak in Antarctic temperature while temperatures in Antarctica were undergoing maximum warming 9. Distinct minima in the abundance of N. pachyderma (s.) were therefore aligned with the maximum rate of warming over Antarctica 2, i.e. peaks in the first derivative of the (continuous) EDC D record 17 on the AICC2012 age scale 4. As both records exhibit different temporal resolutions, we smoothed and detrended the data to compensate for these differences 2 (Fig. S1, S2). Changes in the abundance of N. pachyderma 3

4 (s.) have been smoothed by a 300 year-running average, which approximates twice the average spacing of the planktonic foraminifer proxy data. The high-resolution EDC D record has been smoothed by a 1000 year-running average prior to derivation and was again smoothed by 300 year-running average thereafter. While derivation detrends the EDC D record, a piecewise linear trend has been subtracted from the N. pachyderma (s.) abundance record for the last deglacial (<18 ka BP) and glacial periods (>18 ka BP) to detrend the record. Our age model approach 2,3 is in close agreement with a direct stratigraphic alignment of abundance variations of G. bulloides with the EDC D record (Fig. S1b). Variations in G. bulloides abundances have been suggested to closely align with Antarctic temperature because the abundance of G. bulloides gradually increases towards sub-tropical waters and because this species may adapt their growth habitat and -season compensating for abrupt environmental changes 9,10,18. The low tiepoint density of the alternative age model approach based on a direct alignment of abundance variations of G. bulloides to EDC D leads to slight deviations from the applied age model 2,3 during the peak glacial (Fig. S3). However, the good correspondence of G. bulloides abundance variations with Antarctic temperature throughout the record (Fig. S1) is consistent with earlier propositions based on Cape Basin data 9, and therefore supports our age model. As shown in Fig. S3, the exact choice of alignment technique does not affect the conclusions of this study. Trace element analyses B/Ca ratios have been measured on 5 to 20 specimens of the benthic foraminifera Cibicides kullenbergi, also known as Cibicidoides mundulus 19, greater than 212 µm. It is widely 4

5 assumed that C. kullenbergi prefers an epibenthic habitat and thus, records geochemical conditions of bottom waters 20. B/Ca ratios of epibenthic foraminifera have been shown to reflect the carbonate ion saturation state ( [CO 2-3 ]) of bottom waters 21, i.e. the deviation of in-situ carbonate ion concentrations ([CO 2-3 ] in-situ ) from saturation levels ([CO 2-3 ] saturated ); [CO 2-3 ] = [CO 2-3 ] in-situ - [CO 2-3 ] saturated. The B/Ca ratio of the epibenthic species C. kullenbergi records a 0.69 µmol mol -1 variation per μmol kg -1 change in [CO 2-3 ] (ref. 21). The uncertainty of reconstructed bottom water [CO 2-3 ] based on B/Ca ratios in C. kullenbergi is ±10 μmol kg -1 (ref. 21). Accurate determination of calcite matrix-bound boron can be biased by post-depositional alterations, e.g. preferential dissolution, and/or the presence of contaminant phases that are mostly attached or adsorbed onto foraminiferal shell tests, such as Fe-Mn oxyhydroxides, Mnrich carbonate overgrowths, clay minerals and organic detritus The samples were therefore chemically cleaned following the reductive-oxidative method to remove any contaminant phases from the test samples Foraminiferal test samples were split open with a scalpel to open all chambers without excessive fragmentation. The fragmented shells were ultrasonicated in milli-q water (3 times), methanol (2 times), and again in milli-q water (2 times) to remove clay materials. After clay removal, foraminiferal shells were reductively (Ammonia Ammonium Acetate buffer and Hydrazine) and then oxidatively (Sodium Hydroxide and Hydrogen Peroxide) cleaned 23,24, The chemically cleaned samples were transferred into fresh vials and were leached with weak (0.001 M) nitric acid in order to remove any chemisorbed ions from the calcite. Cleaned foraminifera were dissolved in 300 µl 0.1 M nitric acid, centrifuged and siphoned off from the top to leave the undissolved noncalcitic residues, e.g. silicates and pyrites, behind. The dissolved samples were first analyzed 5

6 on ICP-AES to determine their Ca 2+ concentration, then re-diluted to 10 ppm [Ca 2+ ] and analyzed by HR-ICP-MS according to the protocol outlined in ref. 30. B/Ca ratios from reductively cleaned samples ( Cd/Ca-cleaning ) are identical to oxidatively cleaned samples 23 ( Mg/Ca-cleaning ) within instrumental uncertainties. This rules out the potential problem of preferential dissolution of high-boron containing calcite during the reductive cleaning procedure 21,24,30. The uncertainty of B/Ca ratios measured from replicate samples (n=4) is ±1.9 µmol mol -1, or 2.5 % (1σ). The long-term reproducibility on B/Ca determination is 3.2% (2 ) on 180 measurements 30, which translates into an uncertainty of reconstructed bottom water [CO 2-3 ] of about 6.1 µmol kg -1. The efficiency of the removal of contaminant phases by chemical cleaning can be assessed through the determination of Fe/Ca, Al/Ca and Mn/Ca ratios of the samples 22,30. An influence of Fe-Mn oxyhydroxides on B/Ca ratios is reflected in a co-variation of Fe/Ca and Mn/Ca ratios with B/Ca values. Non-removed clay detritus generally leads to high Al/Ca ratios in the samples. High Mn/Ca ratios may indicate the presence of manganese-rich carbonate overgrowths 22. In sediment core MD Q, B/Ca values show no relationship with Al/Ca and Fe/Ca ratios, and a co-variance (R²=0.2) is observed between B/Ca and Mn/Ca ratios (Fig. S4). This implies that iron-manganese oxides and clay minerals have been successfully removed during reductive cleaning, which rules out their major impact on epibenthic B/Ca ratios. The correlation of Mn/Ca with B/Ca ratios may be caused by the presence of manganese-rich carbonate overgrowths 22. This phase is difficult to remove during cleaning as the solubility product of calcite (Ks(CaCO 3 )= 4.8*10-9, T=25 C; ref. 31) is higher than that of rhodochrosite (Ks(MnCO 3 ) = 2.6*10-11, T=25 C; ref. 32). Extensive cleaning will hence inevitably lead to 6

7 the leaching of foraminiferal calcite rather than the removal of the highly dissolution-resistant manganese-rich overgrowth carbonates 23. The quantification of a potential bias on epibenthic B/Ca ratios by these overgrowth phases is limited by the fact that the trace elemental composition of Mn-rich carbonate overgrowths is poorly constrained 33. If we assume that most of the Mn/Ca signal is caused by a boron-rich contaminant phase, that has a mean boron concentration of 350 ppm, which is equivalent to the boron concentration in Fe-Mn nodules 34 and regarded here as an upper limit, we can calculate the potential contribution of contaminated boron in our samples. The total amount of Mn in the samples has been derived from the weight of our samples after cleaning (~110 µg) and measured Mn/Ca ratios of the individual samples. Taking a concentration of 20 % of Mn (equivalent to Fe-Mn nodules; ref. 34) and the concentration of % of B in the contaminant phase into account, the calculated maximum concentration of contaminant boron in our foraminiferal test samples would amount to 0.6 µmol mol -1, which is less than 0.8 % of the total measured B/Ca levels (~132 µmol mol -1 ). The concentration of B in a potentially non-removed contaminant phase is lower than the analytical uncertainty of the present method. Hence, the potential contamination of non-removed carbonate overgrowth phases and/or Fe-Mn oxyhydroxides on calcite B/Ca ratios is minor. We thus argue that B/Ca is primarily driven by the foraminiferal response to changes in bottom water [CO 2-3 ] rather than by the presence of boron-bearing Fe-Mn oxyhydroxides or carbonate overgrowths. Calculation of carbonate ion concentrations Absolute changes [CO 3 2- ] has been calculated as the sum of the regional, pre-industrial [CO 3 2- ] level (7.8 µmol kg -1 ) and past [CO 3 2- ] changes derived from the difference of 7

8 down-core B/Ca values and the core-top reference based on a C. kullenbergi- specific sensitivity of 0.69 µmol mol -1 per µmol kg -1 change in [CO 2-3 ] (ref. 21): [CO 2-3 ] down-core B/Ca down-core - core-top ]/ [CO 2-3 ] pre-industrial (1) Past changes in bottom water [CO 2-3 ] at the study site are calculated as the deviation of [CO 2-3 ] down-core from the pre-industrial [CO 2-3 ] saturated at the core-site (87.3 µmol kg -1 ): [CO 2-3 ] down-core = [CO 2-3 ] pre-industrial, saturated + [CO 2-3 ] down-core (2) Carbonate parameters of the modern (i.e. pre-industrial) ocean have been computed with the CO 2 SYS program 35 using input data from the Global Ocean Data Analysis Project (GLODAP; ref. 36) and the World Ocean Atlas WOA09 (ref. 37) databases. The program applies the dissociation constants for carbonic acid and bicarbonate estimated by ref. 38 with a re-fit by ref. 39 and the dissociation constants for bisulfate of ref. 40. Phosphate and silicic acid concentrations have been neglected in the calculations. Calcium concentrations are derived from salinity via the relation [Ca 2+ ]= mol kg -1 *S/35 (ref. 41). The solubility product for calcite (K sp ) 42 is pressure-corrected 43. Seawater carbonate (i.e. calcite) saturation (Ω) as shown in Fig. 1a has been computed according to equation (3). Ω values smaller than 1 indicate carbonate undersaturated conditions, whereas Ω values greater than 1 represent carbonate oversaturation. Ω = [Ca 2+ ] in-situ * [CO 2-3 ] in-situ /K sp (3) Variations in mean oceanic [Ca 2+ ] are usually negligible due to the relatively small salinity changes. The carbonate saturation state of seawater is therefore primarily driven by [CO 2-3 ]. 8

9 Additional influences on sedimentary partial dissolution proxies Along with changes in SST and in surface ocean export production, carbonate dissolution is the most important factor controlling the distribution of planktonic foraminifera in marine sediments 44,45. However, changes in export production may influence the assemblage and abundance of benthic foraminifera due to changes in the food quality and/or -quantity supplied to the deep sea 46 or may exert a substantial influence on the dissolution rate within the pore water of marine sediments 47. Therefore, these processes may have had an additional influence or our sedimentary partial dissolution indicators. Changes in export production have recently been shown to co-vary with dust deposition over the Antarctic continent and in the sub-antarctic Atlantic, and thus gradually decreased during Heinrich stadials 48,49. Decreases in the carbon export to the sea floor during Heinrich stadials may potentially lead to increased carbonate preservation 47 and a decrease of benthic faunal populations 46. However, this is inconsistent with observed changes in the Be/Pl ratio and foraminifer shell fragmentation in MD Q (Fig. 2). Furthermore, millennial-scale variations in planktonic Mg/Ca-based SST broadly parallel Antarctic temperature (Fig. S1, S2). Thus, the observed gradual changes in SST and export production can therefore not account for the observed dissolution patterns in sub-antarctic Atlantic sediments. Strong support for a predominant influence of the corrosiveness of bottom waters on the presented foraminifer census count parameters comes from additional (geochemical) measurements of the deep water carbonate chemistry. As outlined above, C. kullenbergi B/Ca predominately records changes in bottom water [CO 2-3 ] and its broad correspondence with foraminifer census counts during MIS 3 (Fig. 2) implies that changes in bottom water [CO 2-3 ] had a dominant influence on the planktonic foraminifer abundance, the Be/Pl ratio and the planktonic foraminifer shell fragmentation in sediment core MD Q. However, long- 9

10 term (glacial-interglacial) changes of the surface ocean hydrography may have had an influence on the sensitivity of the planktonic foraminifer dissolution proxies to changes in bottom water carbonate saturation. Large changes in the abundance and assemblage of planktonic foraminifera during the last deglaciation 2 (Fig. 2) may partially explain the decreased sensitivity of the foraminifer census count parameter to changes in bottom water [CO 3 2- ], which is instead documented in abrupt variations of C. kullenbergi B/Ca ratios. Correlation of sub-antarctic sedimentary dissolution changes with marine Dansgaard- Oeschger records from the northern hemisphere The marine expression of D-O cycles has often been observed to slightly differ from their counterparts in Greenland ice cores (Fig. S5). We therefore test the correlation between the obtained sedimentary partial carbonate dissolution proxies in the deep sub-antarctic Atlantic with high-resolution D-O records from the Iberian Margin 50, the Cariaco Basin 51 and the Arabian Sea 52 between 25 and 65 ka BP (Fig. S5). All data have been smoothed by a 300 year-running average and interpolated onto a common age scale. The obtained correlation coefficients R² range between 0.5 (Iberian Margin) and 0.3 (Cariaco Basin, Arabian Sea) and are statistically significant within the 95% confidence level, indicating a good temporal correspondence of the deep sub-antarctic Atlantic record with these marine northern hemisphere D-O records. Furthermore, we calculated the lead/lag correlation between the deep sub-antarctic Atlantic and the northern hemisphere records over time along a moving 8 ka-window. Maximum R² are achieved by shifting our deep sub-antarctic Atlantic carbonate saturation record with respect to the individual northern-hemisphere D-O records within the range of ±500 years over time (Fig. S5e,f). These offsets are within the age uncertainties of the analyzed marine sediment cores, which corroborates the tight link of D-O variability in the northern hemisphere and the deep sub-antarctic Atlantic. 10

11 Slight differences between the deep sub-antarctic saturation index and northern hemisphere records exhibiting D-O climate variability, e.g. the Iberian Margin 50, in the Cariaco Basin 51, in the Arabian Sea 52 or in Greenland 53,54, may be caused by chronostratigraphic uncertainties and the insensitivity of our sedimentary dissolution proxies to very short-lasting and lowamplitude events, e.g. D-O event 13 or 16 (Fig. 2, S5). Marine records from the tropical region 51 show that D-O event 6 has a smaller expression than the surrounding D-O events 5 and 7 (Fig. S5), which may also be related to the absence or the low-amplitude appearance of D-O 6 in the deep sub-antarctic Atlantic. Furthermore, D-O 9 and 10 appear to be combined to one long event in many marine records from the northern hemisphere, which also bears similarity to our record (Fig. S5). Transient palaeo-climate modeling Transient simulations of MIS 3 have been performed with the Earth system model of intermediate complexity LOVECLIM 55. The performance of the LOVECLIM model has been successfully validated for pre-industrial conditions and key intervals in the past 56,57. An equilibrium spin-up simulation was performed with an atmospheric CO 2 content of ppmv, orbital forcing and an estimated ice sheet extent and topography corresponding to 50 ka BP 58. The subsequent transient simulation uses time-varying external boundary conditions, i.e. orbital forcing parameters, atmospheric CO 2 as well as ice sheet extent and topography. The model imposes a time-dependent, anomalous North Atlantic freshwater flux 59, which mimics phases of ice-sheet disintegration/re-growth during MIS 3. The freshwater forcing function has been iteratively determined to maximize the fit of simulated Iberian Margin SSTs with respective sedimentary proxy data 59 (The Iberian Margin SST record and the model data have been placed on the GICC05 chronology 60. The GICC05 age scale is equivalent to the AICC2012 age scale 4, which has been applied to the study core chronology). In the absence of 11

12 prescribed iceberg calving and anomalously fresh conditions, the Arctic Ocean and Nordic Seas are assumed to be anomalously saline instead, thus enhancing the formation of NADW and the strength of the AMOC. The implied freshwater forcing is consistent with the supply of ice-rafted detritus observed in North Atlantic sediment cores 61. Simulated Greenland air temperatures, SST anomalies at the Iberian Margin and in the southern high-latitudes are in close agreement with proxy-based temperature reconstructions 55,62 (Fig. 3). Simulated [CO 3 2- ] are calculated a posteriori from modeled temperature, salinity, pressure, dissolved inorganic carbon and alkalinity using the pressure-corrected 63 first and second acidity constant 38, the Henry solubility constant 64, the equilibrium constant of water 63, the equilibrium constant for bisulfate 65 and the equilibrium constant of boric acid 40. Simulated [CO 2-3 ] anomalies (Fig. 3) have been obtained by subtracting the long-term mean (50-30 ka) from absolute [CO 2-3 ] values at each model grid point. Modeled [CO 2-3 ] anomalies at the core site (Fig. 3a) have been calculated as an average between 48 S and 40 S, 18 W and 10 W, and between 3.3 and 4 km water depth. For interstadial periods as shown in Fig. 3b and Fig. S6, simulated [CO 2-3 ] and ocean currents have been calculated as average over , , , , and ka BP and for stadial intervals over , , , , and ka BP. While the simulated AMOC is strong during interstadial conditions, the deep Atlantic Ocean is filled with high-[co 2-3 ] NADW (Fig. 3b, S6). Similarly, the simulated AMOC suppression during stadial intervals is associated with the dominance of low-[co 2-3 ] AABW in the deep Atlantic (Fig. 3b, S6). This supports the proposition of a significant impact of diffusive ocean mixing and non-conservative effects due to the interaction of the organic carbon pump with ocean circulation changes on simulated [CO 2-3 ] variations in the deep sub-antarctic Atlantic. Simulated [CO 2-3 ] at the core location decreases, when NADW weakens, independent of 12

13 changes in AABW formation/transport rates, or the exact magnitude of changes in southern hemisphere westerly wind stress (Fig. S7). To illustrate the evolution of the [CO 2-3 ] anomalies in the Atlantic basin during a Heinrich type event in LOVECLIM, we have plotted [CO 2-3 ] anomalies and meridional ocean current anomalies (v') at 50 years after the start of a freshwater input in the North Atlantic from an idealized experiment performed with LOVECLIM under constant pre-industrial conditions (Fig. S8). It is shown along with the rate of meridional change of [CO 2-3 ] (v' d[co 2-3 ]/dy; Fig. S8b), to illustrate where changes occur first after a meltwater perturbation in the North Atlantic. Unfortunately, sub-centennially resolved [CO 2-3 ] data are not available for the glacial LOVECLIM simulation. In the pre-industrial LOVECLIM simulation, already after 50 years after the freshwater perturbation meridional currents in the entire Deep Western Boundary Current have weakened (Fig. S8a). Positive current anomalies at the core site, i.e. within the large [CO 2-3 ] gradient in the Southern Ocean, translate into the transport of more low-[co 2-3 ] AABW to the core site (positive rate of meridional [CO 2-3 ] change) during intervals of AMOC slow-down (Fig. S8b). This process accounts for about ~20% of the total [CO 2-3 ] adjustment at the core site at 100 years after the onset of freshwater hosing in the North Atlantic. As the thermohaline adjustment process is consistent in LOVECLIM for preindustrial and glacial boundary conditions, Fig. S8 illustrates that ocean wave processes have a notable impact on sub-antarctic Atlantic [CO 2-3 ] within a short time period after freshwater perturbations in the North Atlantic, when tracer advection anomalies still play a minor role in variations of deep sub-antarctic Atlantic [CO 2-3 ] 66,67. Calculating the lead/lag correlation between the modeled AMOC and simulated anomalies of [CO 2-3 ] at the core site during MIS3 using a sliding 8 ka-window (Fig. S9) we infer that modeled deep sub-antarctic Atlantic [CO 2-3 ] anomalies lag variations in the AMOC by 13

14 170±20 years on average. Corresponding correlation coefficients are high (R²> 0.8, statistically significant within the 95% confidence level), and imply a close causal coupling (Fig. S9). Age scale conversion for selected marine climate records in the Atlantic Ocean The chronologies of selected high-resolution marine proxy records documenting abrupt climate variability in key locations of the Atlantic Ocean (Fig. 2) have been converted to the GICC05 and AICC2012 age scales 4,60 in order to allow a direct comparison to proxy data of sediment core MD Q to be made. Within the uncertainties of the order of decades, the GICC05 and AICC2012 chronologies are identical during the last 60 ka BP 4. For TNO57-21, we use the most recently established chronology 68, which is based on the GICC05 age scale. The new age model of TNO57-21 has been applied to adjust the age scale of sediment core RC11-83 based on the proposed alignment of their benthic carbon isotope records 69. GICC05 ages of the benthic stable carbon isotope record from the Iberian margin have been obtained by the stratigraphic alignment of planktonic stable oxygen isotopes in the same sediment cores with Greenland temperature 70,71 documented by 18 O variations in the NGRIP ice core on the GICC05 age scale 4,53,60. The data from the Arabian Sea (SO KL) are shown on their original GICC05 age scale 52. The proxy data of sediment core ODP1002 (originally on the GISP2 age scale) 51 and the Iberian Margin sediment core MD (originally on the GRIP and GISP age scales) 50 have been transferred to the GICC05 age scale 60 for ages <32 ka BP and by a synchronization of calcium ion concentrations between GRIP/GISP2 and NGRIP for ages >32 ka BP 73. The GICC05modelext age scale 75 was applied for ages >60 ka BP. 14

15 References 1. Völker et al. Global distribution of centennial-scale records for Marine Isotope Stage (MIS) 3: a database. Quat. Sci. Rev. 21, (2002). 2. Gottschalk, J., Skinner, L. C. & Waelbroeck, C. Contribution of seasonal sub-antarctic surface water variability to millennial-scale changes in atmospheric CO 2 over the last deglaciation and Marine Isotope Stage 3. Earth Planet. Sci. Lett. 411, (2015). 3. Skinner, L. C., Fallon, S., Waelbroeck, C., Michel, E. & Barker, S. Ventilation of the deep Southern Ocean and deglacial CO 2 rise. Science 328, (2010). 4. Veres, D. et al. The Antarctic ice core chronology (AICC2012): an optimized multiparameter and multi-site dating approach for the last 120 thousand years. Clim. Past 9, (2013). 5. Reimer, P. J. et al. INTCAL04 terrestrial radiocarbon age calibration, 0 26 cal kyr BP. Radiocarbon 46, (2004). 6. Hughen, K., Southon, J., Lehman, S., Bertrand, C. & Turnbull, J. Marine-derived 14 C calibration and activity record for the past 50,000 years updated from the Cariaco Basin. Quat. Sci. Rev. 25, (2006). 7. Haslett, J. & Parnell, A. A simple monotone process with application to radiocarbondated depth chronologies. J. R. Stat. Soc. 57, (2008). 8. Skinner, L. C., Waelbroeck, C., Scrivner, a. E. & Fallon, S. J. Radiocarbon evidence for alternating northern and southern sources of ventilation of the deep Atlantic carbon pool during the last deglaciation. Proc. Natl. Acad. Sci. 111, (2014). 9. Barker, S. et al. Interhemispheric Atlantic seesaw response during the last deglaciation. Nature 457, 26 (2009). 10. Kucera, M. et al. Reconstruction of sea-surface temperatures from assemblages of planktonic foraminifera: multi-technique approach based on geographically constrained calibration data sets and its application to glacial Atlantic and Pacific Oceans. Quat. Sci. Rev. 24, (2005). 11. Rind, D. et al. Effects of glacial meltwater in the GISS coupled atmosphere-ocean model - 2. A bipolar seesaw in Atlantic Deep Water production. J. Geophys. Res. 106, (2001). 12. Mazaud, A. et al. Variations of the ACC-CDW during MIS3 traced by magnetic grain deposition in midlatitude South Indian Ocean cores: Connections with the northern hemisphere and with central Antarctica. Geochemistry Geophys. Geosystems 8, (2007). 13. Siani, G. et al. Carbon isotope records reveal precise timing of enhanced Southern Ocean upwelling during the last deglaciation. Nat. Commun. 4, 1 9 (2013). 14. Hayes, C. T. et al. A stagnation event in the deep South Atlantic during the last interglacial period. Science 346, (2014). 15. Caniupán, M. et al. Millennial-scale sea surface temperature and Patagonian Ice Sheet changes off southernmost Chile (53 S) over the past~ 60 kyr. Paleoceanography 26, PA3221 (2011). 16. Kaiser, J., Lamy, F. & Hebbeln, D. A 70-kyr sea surface temperature record off southern Chile (Ocean Drilling Program Site 1233). Paleoceanography 20, 1 15 (2005). 15

16 17. Jouzel, J. et al. Orbital and millennial Antarctic climate variability over the past 800,000 years. Science 317, 793 (2007). 18. King, A. L. & Howard, W. R. Planktonic foraminiferal flux seasonality in Subantarctic sediment traps: A test for paleoclimate reconstructions. Paleoceanography 18, 1019 (2003). 19. Loeblich, A. & Tappan, H. Foraminiferal genera and their classification. 970 (Van Nostrand Reinhold Company, New York., 1988). 20. Duplessy, J. C. et al. Deepwater source variations during the last climatic cycle and their impact on the global deepwater circulation. Paleoceanography 3, (1988). 21. Yu, J. & Elderfield, H. Benthic foraminiferal B/Ca ratios reflect deep water carbonate saturation state. Earth Planet. Sci. Lett. 258, (2007). 22. Boyle, E. A. Manganese carbonate overgrowths on foraminifera tests. Geochim. Cosmochim. Acta 47, (1983). 23. Barker, S., Greaves, M. & Elderfield, H. A study of cleaning procedures used for foraminiferal Mg/Ca paleothermometry. Geochemistry Geophys. Geosystems 4, 8407 (2003). 24. Yu, J., Elderfield, H., Greaves, M. & Day, J. Preferential dissolution of benthic foraminiferal calcite during laboratory reductive cleaning. Geochemistry Geophys. Geosystems 8, Q06016 (2007). 25. Lemarchand, E., Schott, J. & Gaillardet, J. How surface complexes impact boron isotope fractionation: evidence from Fe and Mn oxides sorption experiments. Earth Planet. Sci. Lett. 260, (2007). 26. Boyle, E. A. & Keigwin, L. D. Comparison of Atlantic and Pacific paleochemical records for the last 215,000 years: Changes in deep ocean circulation and chemical inventories. Earth Planet. Sci. Lett. 76, (1985). 27. Boyle, E. A. & Rosenthal, Y. in The South Atlantic: Present and Past Circulation (eds. Wefer, G., Berger, W. H., Siedler, D. & Webb, D. J.) (Springer Verlag, New York, 1996). 28. Rosenthal, Y., Boyle, E. A., Labeyrie, L. & Oppo, D. Glacial enrichments of authigenic Cd and U in subantarctic sediments: A climatic control on the elements oceanic budget? Paleoceanography 10, (1995). 29. Boyle, E. A. Cadmium, zinc, copper, and barium in foraminifera tests. Earth Planet. Sci. Lett. 53, (1981). 30. Misra, S. et al. Determination of B/Ca of natural carbonates by HR-ICP-MS. Geochemistry Geophys. Geosystems 15, (2014). 31. Patnaik, P. Handbook of Inorganic Chemicals. 28, 1125 (McGraw-Hill New York, 2003). 32. Johnson, K. S. Solubility of rhodochrosite (MnCO 3 ) in water and seawater. Geochim. Cosmochim. Acta 46, (1982). 33. Pena, L. D., Calvo, E., Cacho, I., Eggins, S. & Pelejero, C. Identification and removal of Mn-Mg-rich contaminant phases on foraminiferal tests: Implications for Mg/Ca past temperature reconstructions. Geochemistry Geophys. Geosystems 6, Q09P02 (2005). 34. Baturin, G. N. The Geochemistry of Manganese and Manganese Nodules in the Ocean. 342 (D. Reidel Publishing Company, Dordrecht, Holland, 1988). 16

17 35. Lewis, E. & Wallace, D. W. R. Program developed for CO 2 system calculations. Oak Ridge, Oak Ridge National Laboratory ORNL/CDIAC-105 (Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, US Dept. of Energy, 1998). 36. Key, R. M. et al. A global ocean carbon climatology: Results from Global Data Analysis Project (GLODAP). Global Biogeochem. Cycles 18, GB4031 (2004). 37. Locarnini, R. A. et al. World Ocean Atlas 2009, Volume 1: Temperature. NOAA Atlas NESDIS 68 68, (U.S. Government Printing Office, Washington, D.C., 2010). 38. Mehrbach, C. Measurement of the apparent dissociation constants of carbonic acid in seawater at atmospheric pressure. Limnol. Oceanogr. 18, (1973). 39. Dickson, A. G. & Millero, F. J. A comparison of the equilibrium constants for the dissociation of carbonic acid in seawater media. Deep Sea Res. 34, (1987). 40. Dickson, A. G. Standard potential of the reaction: AgCl(s) + 1/2 H 2 (g) = Ag(s) + HCl(aq), and and the standard acidity constant of the ion HSO 4 - in synthetic sea water from to K. J. Chem. Thermodyn. 22, (1990). 41. Riley, J. P. & Tongudai, M. The major cation/chlorinity ratios in sea water. Chem. Geol. 2, (1967). 42. Mucci, A. The solubility of calcite and aragonite in seawater at various salinities, temperatures, and one atmosphere total pressure. Am. J. Sci. 283, (1983). 43. Ingle, S. E. Solubility of calcite in the ocean. Mar. Chem. 3, (1975). 44. Anderson, D. M. & Archer, D. Glacial-interglacial stability of ocean ph inferred from foraminifer dissolution rates. Nature 416, (2002). 45. Mix, A. C. Influence of productivity variations on long-term atmospheric CO 2. Nature 337, (1989). 46. Herguera, J. C. & Berger, W. H. Paleoproductivity from benthic foraminifera abundance: glacial to postglacial change in the west-equatorial Pacific. Geology 19, (1991). 47. Le, J. & Shackleton, N. J. Carbonate dissolution fluctuations in the western equatorial Pacific during the late Quaternary. Paleoceanography 7, (1992). 48. Martínez-García, A. et al. Iron Fertilization of the Subantarctic Ocean During the Last Ice Age. Science 343, (2014). 49. Anderson, R. F. et al. Biological response to millennial variability of dust and nutrient supply in the Subantarctic South Atlantic Ocean. Philos. Trans. R. Soc. 372, (2014). 50. Shackleton, N. J., Hall, M. A. & Vincent, E. Phase relationships between millennialscale events 64,000-24,000 years ago. Paleoceanography 15, (2000). 51. Peterson, L. C., Haug, G. H., Hughen, K. A. & Röhl, U. Rapid changes in the hydrologic cycle of the tropical Atlantic during the last glacial. Science 290, (2000). 52. Deplazes, G. et al. Links between tropical rainfall and North Atlantic climate during the last glacial period. Nat. Geosci. 6, (2013). 17

18 53. North Greenland Ice Core Project Members. High-resolution record of Northern Hemisphere climate extending into the last interglacial period. Nature 431, (2004). 54. Dansgaard, W. et al. Evidence for general instability of past climate from a 250-kyr ice-core record. Nature 364, (1993). 55. Menviel, L., Timmermann, A., Friedrich, T. & England, M. H. Hindcasting the continuum of Dansgaard Oeschger variability: mechanisms, patterns and timing. Clim. Past 10, (2014). 56. Goosse, H. et al. Description of the Earth system model of intermediate complexity LOVECLIM version 1.2. Geosci. Model Dev. 3, (2010). 57. Menviel, L. Climate-Carbon cycle interactions on millennial to glacial timescales as simulation by a model of intermediate complexity. (PhD thesis, University of Hawaii, 2008). 58. Abe-Ouchi, A., Segawa, T. & Saito, F. Climatic conditions for modelling the Northern Hemisphere ice sheets throughout the ice age cycle. Clim. Past 3, (2007). 59. Martrat, B. et al. Four climate cycles of recurring deep and surface water destabilizations on the Iberian margin. Science 317, (2007). 60. Svensson, A. et al. A year Greenland stratigraphic ice core chronology. Clim. Past 4, (2008). 61. Van Kreveld, S. et al. Potential links between surging ice sheets, circulation changes, and the Dansgaard-Oeschger cycles in the Irminger Sea, kyr. Paleoceanography 15, (2000). 62. Menviel, L., Spence, P. & England, M. H. Contribution of enhanced Antarctic Bottom Water formation to Antarctic warm events and millennial-scale atmospheric CO 2 increase. Earth Planet. Sci. Lett. 413, (2015). 63. Millero, F. J. Thermodynamics of the carbon dioxide system in the oceans. Geochim. Cosmochim. Acta 59, (1995). 64. Weiss, R. F. Carbon dioxide in water and seawater: the solubility of a non-ideal gas. Mar. Chem. 2, (1974). 65. Dickson, A. G. & Goyet, C. Handbook of methods for the analysis of the various parameters of the carbon dioxide system in sea water (version 2). Oak Ridge, Oak Ridge National Laboratory ORNL/CDIAC-74 (Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, US Dept. of Energy, 1994). 66. Goodman, P. J. Thermohaline Adjustment and Advection in an OGCM. J. Phys. Oceanogr. 31, (2001). 67. Friedrich, T. & Timmermann, A. Millennial-scale glacial meltwater pulses and their effect on the spatiotemporal benthic δ 18 O variability. Paleoceanography 27, PA3215 (2012). 68. Barker, S. & Diz, P. Timing of the descent into the last ice age determined by the bipolar seesaw. Paleoceanography 29, (2014). 69. Piotrowski, A. M., Goldstein, S. L., Hemming, S., Fairbanks, R. G. & Zylberberg, D. R. Oscillating glacial northern and southern deep water formation from combined neodymium and carbon isotopes. Earth Planet. Sci. Lett. 272, (2008). 18

19 70. Skinner, L. C., Elderfield, H. & Hall, M. in Ocean Circulation: Mechanisms and Impacts (eds. Schmittner, A., Chiang, J. C. H. & Hemming, S. R.) (American Geophysical Union, Geophysical Monograph Series, 2007). 71. Lynch-Stieglitz, J. et al. Muted change in Atlantic overturning circulation over some glacial-aged Heinrich events. Nat. Geosci. 7, (2014). 72. Rasmussen, S. O. et al. A new Greenland ice core chronology for the last glacial termination. J. Geophys. Res. 111, (2006). 73. Rasmussen, S. O. et al. Synchronization of the NGRIP, GRIP, and GISP2 ice cores across MIS 2 and palaeoclimatic implications. Quat. Sci. Rev. 27, (2008). 74. Andersen, K. K. et al. The Greenland ice core chronology 2005, ka. Part 1: Constructing the time scale. Quat. Sci. Rev. 25, (2006). 75. Wolff, E. W., Chappellaz, J., Blunier, T., Rasmussen, S. O. & Svensson, A. Millennialscale variability during the last glacial: the ice core record. Quat. Sci. Rev. 29, (2010). 76. Cacho, I. et al. Dansgaard-Oeschger and Heinrich event imprints in Alboran Sea paleotemperatures. Paleoceanography 14, (1999). 77. Zarriess, M. & Mackensen, A. Testing the impact of seasonal phytodetritus deposition on δ13c of epibenthic foraminifer Cibicidoides wuellerstorfi: A 31,000 year highresolution record from the northwest African continental slope. Paleoceanography 26, PA2202 (2011). 78. Mulitza, S. et al. Sahel megadroughts triggered by glacial slowdowns of Atlantic meridional overturning. Paleoceanography 23, (2008). 79. Hodell, D. A., Channell, J. E. T., Curtis, J. H., Romero, O. E. & Röhl, U. Onset of Hudson Strait Heinrich events in the eastern North Atlantic at the end of the middle Pleistocene transition (~ 640 ka)? Paleoceanography 23, PA4218 (2008). 80. Hodell, D. A., Evans, H. F., Channell, J. E. T. & Curtis, J. H. Phase relationships of North Atlantic ice-rafted debris and surface-deep climate proxies during the last glacial period. Quat. Sci. Rev. 29, (2010). 81. Margari, V. et al. The nature of millennial-scale climate variability during the past two glacial periods. Nat. Geosci. 3, (2010). 82. Elliot, M., Labeyrie, L. & Duplessy, J.-C. Changes in North Atlantic deep-water formation associated with the Dansgaard-Oeschger temperature oscillations (60-10 ka). Quat. Sci. Rev. 21, (2002). 83. Dokken, T. M., Nisancioglu, K. H., Li, C., Battisti, D. S. & Kissel, C. Dansgaard- Oeschger cycles: Interactions between ocean and sea ice intrinsic to the Nordic seas. Paleoceanography 28, (2013). 84. Sachs, J. P. & Anderson, R. F. Increased productivity in the subantarctic ocean during Heinrich events. Nature 434, (2005). 85. Diz, P. & Barker, S. Linkages between rapid climate variability and deep-sea benthic foraminifera in the deep Subantarctic South Atlantic during the last 95 kyr. Paleoceanography 30, (2015). 86. Hemming, S. R. Heinrich events: Massive late Pleistocene detritus layers of the North Atlantic and their global climate imprint. Rev. Geophys. 42, RG1005 (2004). 19

20 Table S1. Proxy records in the Atlantic Ocean region with a mean temporal resolution <200 years documenting abrupt climate variability in the equatorial (surface) ocean a, in Greenland ice cores b, the North Atlantic surface ocean c, both the North Atlantic surface and deep ocean d, the deep North Atlantic e and the South Atlantic surface ocean f during Marine Isotope Stage 3 (plotted in Fig. 1) Longitude Latitude Depth (m) Core Reference ' ' 767 GeoB3104 compiled in ref. 1 a ' ' 772 GeoB3912 compiled in ref. 1 a ' ' 893 ODP1002 compiled in ref. 1 a ' ' 847 MD Ref. 52 a ' ' 3240 Dome compiled in ref. 1 b ' 65 11' 2037 Dye compiled in ref. 1 b ' 72 36' 3208 GISP2 compiled in ref. 1 b ' 72 35' 3232 GRIP compiled in ref. 1 b ' 3770 MD Q this study ' 49 53' 3884 DSDP609 compiled in ref. 1 c ' ' 1308 GIK23071 compiled in ref. 1 c ' 58 13' 3380 HU compiled in ref. 1 c ' 50 12' 3348 HU compiled in ref. 1 c ' ' 3000 KNR140-2 JPC37 compiled in ref. 1 c ' 36 9' 1841 MD Ref. 76 c ' ' 2174 MD compiled in ref. 1 c ' ' 2094 MD compiled in ref. 1 c ' ' 3448 MD compiled in ref. 1 c ' ' 2925 MD compiled in ref. 1 c ' ' 2465 MD compiled in ref. 1 c ' ' 1123 MD compiled in ref. 1 c ' 3375 SU90-09 compiled in ref. 1 c -21 4' 54 5' 2900 SU90-38 compiled in ref. 1 c ' 52 34' 3955 SU90-39 compiled in ref. 1 c ' 54 15' 2393 V23-81 compiled in ref. 1 c ' 49 27' 3935 V28-82 compiled in ref. 1 c ' ' 1020 ENAM93-21, MD compiled in ref. 1 c ' ' 2452 ENAM97-09 compiled in ref. 1 d -18 3' 12 26' 3233 GeoB Ref. 77 d ' ' 2384 GeoG Ref. 78 d ' ' 2472 GIK23415 compiled in ref. 1 d ' 49 53' 3883 IODP U1308 Ref. 79 d ' 53 3' 3082 KN JPC-13 Ref. 80 d ' 23 45' 542 KNR GGC Ref. 71 d ' ' 546 KNR JPC Ref. 71 d 20

21 ' KNR31-GPC-5 compiled in ref. 1 d ' ' 4758 KNR31-GPC-9 compiled in ref. 1 d ' ' 2637 MD Ref. 59, 70, 81 d 4 34' 66 41' 1230 MD compiled in ref. 1 d ' ' 4461 MD compiled in ref. 1 d ' ' 4462 MD compiled in ref. 1 d ' ' 2159 MD compiled in ref. 1 d ' ' 3146 MD compiled in ref. 1 d ' ' 1841 MD compiled in ref. 1 d ' 37 48' 3146 MD K Ref. 70 d ' 57 29' 2161 Na87-22 Ref. 82 d ' ' 2996 ODP1059 compiled in ref. 1 d ' ' 2263 ODP658 compiled in ref. 1 d ' ' 777 PS2644 compiled in ref. 1 d ' ' 1099 SO75-26KL compiled in ref. 1 d ' ' 1416 SO82-5 compiled in ref. 1 d ' 58 13' 2100 SU90-16 compiled in ref. 1 d ' 62 40' 2085 SU90-24 compiled in ref. 1 d ' 50 6' 4255 SU90-44 compiled in ref. 1 d ' ' 1849 V compiled in ref. 1 d ' ' 1500 MD Ref. 83 d ' ' 1994 ODP983 compiled in ref. 1 e ' -41 1' 4981 TNO57-21 Ref. 84, 85 f 21

22 Fig. S1. Age model for sediment core MD Q. a, Sedimentation rate. b, Depth-age tie-points derived from radiocarbon dates (grey) 3, the stratigraphic alignment 2 of the abundance lows of N. pachyderma (s.) with maxima of the first derivative of the EPICA Dome C (EDC) D record 17 on the AICC2012 age scale 4 (blue) as well as the alternative stratigraphic alignment of abundance peaks of G. bulloides with maxima in the EDC D record (rose); envelopes show the associated age model uncertainty (95% high posterior density region) calculated with the Bayesian statistical software BCHRON (ref. 7). c, Mg/Ca- 22

23 derived sea surface temperatures (SST) obtained from G. bulloides (red) and N. pachyderma s. (blue) (envelopes show 500 year-window 1σ standard deviations). d, Antarctic temperature D record of the EDC D ice core 17. e, Variations in the abundance of N. pachyderma (s.) (envelopes show 500 year-window 1σ standard deviations). f, Detrended and smoothed (300 year-running average) record of abundance changes of N. pachyderma (s.) (black; 500 yearmoving average in darkred) and smoothed (300 year-running average) record of the first derivative of Antarctic EDC D temperature record (grey); arrows in b and f indicate age markers of the final core chronology and their associated uncertainties 2,3 ( 14 C dates in grey, stratigraphic alignment of N. pachyderma (s.) abundance variations with the rate of change in Antarctic temperature in blue). g, G. bulloides abundance variations in MD Q (note the different y-axis for the deglacial period) based on the final core chronology 2,3 and the EDC D temperature record 17 (grey), stippled line shows G. bulloides abundance variations based on the alternative stratigraphic alignment shown in b (tiepoints and their associated uncertainties are indicated below). 23

24 Fig. S2. Age model of the deglacial (left) and late MIS 3 (right) section of MD Q. a, Mg/Ca-derived temperatures obtained from G. bulloides (red) and N. pachyderma s. (blue) (envelopes show 500 year-window 1σ standard deviations). b, EDC D temperature record 17 on the AICC2012 age scale 4, stippled lines indicate similar broad features in both records. Variations in the abundance of N. pachyderma (s.) c, in percent (envelopes show 500 yearwindow 1σ standard deviations) and d, detrended and smoothed by a 300 year-running average (darkred solid line shows a 500 year-moving average). e, Smoothed (300 yearrunning average) record of the first derivative of Antarctic EDC D temperature record 17 (smoothed by a 1 ka-running average prior to derivation); arrows at the bottom indicate age marker of the final core chronology and their associated uncertainties 2,3 ( 14 C dates in grey, stratigraphic alignment of N. pachyderma (s.) abundance variations with the rate of change in Antarctic temperature in blue) 24

25 Fig. S3. Comparison of stratigraphic alignment approaches. a, Variations in the abundance of G. bulloides based on a stratigraphic alignment of minima in N. pachyderma (s.) with maxima in the rate of change, i.e. first derivative, of Antarctic temperature 2 (solid line) and based on an alternative stratigraphic alignment of G. bulloides abundance variations to Antarctic temperature (stippled line, arrows show applied tiepoints). b, Variations in the abundance of N. pachyderma (s.) (%Nps) mainly based on a stratigraphic alignment of %Nps with the rate of change, i.e. first derivative, of Antarctic temperature 2 (solid line, arrows show applied tiepoints 2 : grey - radiocarbon 3, blue- %Nps-d D/dt) and based on the alternative stratigraphic alignment (stippled line); Antarctic temperature is approximated by the EDC D record 17 (grey, AICC2012 age scale 4 ). c, Reconstructed carbonate saturation index based on an alignment of %Nps-d D/dt 2 (solid line) and the alternative stratigraphic alignment (stippled line), NGRIP 18 O record 53 on the GICC05 age scale 60,75 (grey). 25

26 Fig. S4. Contamination levels of reductively cleaned epibenthic foraminiferal test samples. Mn/Ca (diamonds), Al/Ca (triangles) and Fe/Ca (circles) ratios versus B/Ca ratios obtained from the reductively cleaned test samples of C. kullenbergi 26

27 Fig. S5. Comparison of Dansgaard-Oeschger (D-O) variability in the deep sub-antarctic Atlantic and other ocean regions. a, G. bulloides 18 O variations in Iberian Margin sediment core MD b, Color reflectance of sediment core ODP1002 from the Cariaco Basin 51 (smoothed by a 5 point-running average). c, Color reflectance of Arabian Sea sediment core SO KL 52 (smoothed by 500 point-running average). The deep sub-antarctic Atlantic carbonate saturation index of sediment core MD Q is shown in grey. d, NGRIP 18 O record 53. e, Offsets between the deep sub-antarctic Atlantic and the northern hemisphere records (red: Iberian Margin, orange: Cariaco Basin, blue: Arabian Sea) at f, maximum correlation (R²) along a moving 8 ka-window (calculated values are plotted at the mid-point of the applied window); all proxy records are reported on the GICC05 or the equivalent AICC2012 age scale 4 ; numbers in d, denote D-O events 54 (top) and Heinrich stadials (HS; bottom) 86 27

28 Fig. S6. Simulated [CO 2-3 ] and circulation changes in the Atlantic (70 W-20 E). Zonal [CO 2-3 ] averages (shaded) with contours showing the maximum Atlantic meridional overturning streamfunction (Sv) (solid lines indicate a clockwise circulation, stippled lines indicate an anti-clockwise circulation) during a, stadial and b, interstadial (bottom) conditions between ka BP simulated with a transient LOVECLIM model simulation 55 (please refer to the text for details on the averaged time intervals); star marks location of the sediment core MD Q. 28

29 Fig. S7. Link between AABW overturning, southern hemisphere westerly wind stress and deep sub-antarctic Atlantic [CO 2-3 ]. a, Maximum overturning circulation in the North Atlantic (Sv). b, Changes in AABW overturning (Sv; plotted as absolute values). c, Southern hemisphere westerly wind stress averaged between S. d, [CO 2-3 ] anomalies averaged at the location of sediment core MD Q in the deep sub-antarctic Atlantic. Data are simulated across HS4 in two transient experiments (black: standard transient glacial experiment, grey: transient glacial experiment with enhanced AABW due to additional Southern Ocean buoyancy forcing) performed with LOVECLIM

30 Fig. S8. Evolution of Atlantic [CO 2-3 ] after a freshwater perturbation in the North Atlantic. a, Meridional ocean current anomalies (v', in m s -1 ; shading) and total mean [CO 2-3 ] (µmol l -1 ; contours) and b, [CO 2-3 ] anomalies (µmol l -1 ; shading) and the rate of meridional change of [CO 2-3 ] (contours) at 3000 m ocean depth in the Atlantic Ocean after 50 years after freshwater input in the North Atlantic. Red star marks the location of the study core MD Q within the steep sub-antarctic [CO 2-3 ] gradient. Blue and red contours in b show a northward (positive) and southward (negative) meridional shift in [CO 2-3 ], respectively. Northward mass transport is schematically indicated by a black arrow. 30

31 Fig. S9. Lead/lag correlation of deep sub-antarctic Atlantic [CO 2-3 ] anomalies and the AMOC index in LOVECLIM. a, Freshwater forcing time series of the model simulation 55. b, Simulated (black) and reconstructed (grey) air temperature anomalies ( T) over Northeast Greenland 53,55. c, Modeled (black) and reconstructed (grey) SST anomalies off the Iberian Margin 55,59. d, Simulated maximum Atlantic Meridional Overturning Circulation (AMOC) streamfunction 55. e, Simulated meridional current velocities (negative: southward). f, Modeled [CO 2-3 ] anomalies at the study site. g, Offset between simulated [CO 2-3 ] anomalies from the modeled AMOC index at h, maximum correlation (R²) along a moving 8 ka-window (calculated values are plotted at the mid-point of the applied window). All records are reported on the GICC05 age scale. Numbers denote Heinrich stadials (HS) 86 and D-O events