lable at ScienceDirect

Quaternary Science Reviews 29 (2010) 399–409

Contents lists avai

Quaternary Science Reviews

journal homepage: www.elsevier .com/locate/quascirev

Millennial scale evolution of the Southern Ocean chemical divide

Christopher D. Charles a,*, Katharina Pahnke b, Rainer Zahn c,d, P.G. Mortyn d,e, Ulysses Ninnemann f,D.A. Hodell g

a Scripps Institute of Oceanography, UCSD, La Jolla, CA 92093-0244, USAb Department of Geology and Geophysics, University of Hawaii, 1680 East-West Road, Honolulu, HI 96822, USAc Institucio Catalana de Recerca i Estudis Avancats, ICREA, E-08193 Bellaterra, Spaind Universitat Autonoma de Barcelona, Institut de Ciencia i Tecnologia Ambientals, ICTA, Edifici Cn, Campus UAB, 08193 Bellaterra, Spaine Department of Geography, Universitat Autonoma de Barcelona, Bellaterra 08193, Spainf Bjerknes Center for Climate Research, UIB, Allegaten 55, 5007 Bergen, Norwayg Godwin Laboratory for Paleoclimate Research, University of Cambridge, Downing Site, Cambridge CB2 3EQ, UK

a r t i c l e i n f o

Article history:Received 20 January 2009Received in revised form18 September 2009Accepted 24 September 2009

* Corresponding author.E-mail address: ccharles@ucsd.edu (C.D. Charles).

0277-3791/$ – see front matter � 2009 Elsevier Ltd.doi:10.1016/j.quascirev.2009.09.021

a b s t r a c t

The chemical properties of the mid-depth and deep Southern Ocean are diagnostic of the mechanisms ofabrupt changes in the global ocean throughout the late Pleistocene, because the regional water massconversion and mixing help determine global ocean gradients. Here we define continuous time series ofSouthern Ocean vertical gradients by differencing the records from two high deposition rate deep seasedimentary sequences that span the last several ice age cycles. The inferred changes in vertical carbonand oxygen isotopic gradients were dominated by variability on the millennial scale, and they followedclosely the abrupt climate events of the high latitude Northern Hemisphere. In particular, the stadialevents of at least the last 200 kyr were characterized by enhanced mid-deep gradients in both d13C(dissolved inorganic carbon) and d18O (temperature). Interstadial events, conversely, featured reducedvertical gradients in both properties. The glacial terminations represented exceptions to this pattern ofvariability, as the vertical carbon isotopic gradient flattened dramatically at times of peak warmth in theSouthern Ocean surface waters and with little or no corresponding change d18O gradient. The availableevidence suggests that properties of the upper layer of the Southern Ocean (Antarctic IntermediateWater) were influenced by an atmospherically mediated teleconnection to high latitude NorthernHemisphere.

� 2009 Elsevier Ltd. All rights reserved.

1. Introduction

Deep sea sediment cores demonstrate that a defining charac-teristic of the late Pleistocene glacial episodes was the repeateddevelopment of a ‘‘chemical divide’’ that separated the interme-diate and deep ocean. The origin of this feature of cold climateintervals has yet to be resolved, despite its prominence andpotential importance to the climate system (Toggweiler, 1999). If itwere an expression of an altered geometry of deep ocean circu-lationdperhaps accompanying or even caused by reduced merid-ional overturningdthen this ‘‘chemical divide’’ would necessarilyimply a nearly global oceanic involvement in the abrupt climateshifts. Though the divide appears to varying extent in severalsedimentary tracers (e.g. Herguera et al., 1992; Marchitto andBroecker, 2006), the mid-to-deep separation effect is best

All rights reserved.

expressed in benthic foraminiferal stable carbon isotopic recordsfrom the Atlantic Ocean (Curry and Oppo, 2005). However, not allsedimentary tracers that should be sensitive to variable deep oceancirculation exhibit the same structure over the course of the latePleistocene climate cycles (e.g. Yu et al., 1996; Piotrowski et al.,2004, 2008); the difference among tracers raises the possibility thatfactors other than deep ocean circulation (air–sea and land–seaexchange of carbon, for example) might dominate the stable carbonisotopic distribution in the ocean. Furthermore, one could conceiveof a number of nutrient and carbon trapping schemes to explain the‘‘chemical divide’’ (e.g. Boyle, 1988) that would not necessarilyinvolve the variable rates of (physical) overturning of the water(Legrand and Wunsch, 1995). Consequently, the physical signifi-cance of one of the most prominent aspects of the ice age oceanremains in question.

As a region of both intermediate and deep water formation, theSouthern Ocean is obviously an especially pivotal region forunderstanding the phenomenon. And previous results from thehigh latitude Southern Hemisphere suggest a clear separation

Fig. 1. Benthic foraminiferal isotopic records from the mid-depth Southeast Pacific (top right panel) and the deep South Atlantic (bottom right panel) allow an assessment of verticalgradients in the Southern Ocean. The mid-depth record is a composite of measurements from several benthic foraminiferal genera, while the deep record is comprised ofmeasurements exclusively from Cibicidoides spp. For consistency, the d18O is expressed on a common Uvigerina scale (0.65& has been added to the values for ODP Site 1089). Thedata are from Pahnke and Zahn (2005), Hodell et al. (2002) and Mortyn et al. (2003). The panels on the left illustrate the oceanographic context of each site, shown as red circles onthe WOCE P15 and A12 salinity sections [adapted from the eWOCE atlas (Schlitzer 2000)].

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409400

between the characteristics of mid-depth and deep ocean duringcold periods (Ninnemann and Charles, 2002; Hodell et al., 2003).Thus, it is appropriate to consider the evolution of the verticalstructure of Southern Ocean isotopic gradients in as much detail aspossible. Here we refine and extend the observations of the verticalgradients in the carbon isotope composition of the Southern Oceanover the last 200 ka, as one step toward deeper mechanisticunderstanding of the separation between shallow and deep layersthroughout the global ocean. We construct time series of verticalcarbon and oxygen isotopic gradients by means of simplesubtraction of records from a deep South Atlantic and an inter-mediate depth Pacific sedimentary sequence. The distinction ofthese new gradient time series is that the constituent recordsresolve the millennial scale evolution of Southern Ocean isotopicdistribution, while also extending beyond the limits of typicalmillennial scale resolution piston core recordsdmost notably,through the penultimate deglaciation. The features of these quasi-

continuous indices are comparable to various other measures ofglobal climate variability and illustrate Southern Ocean processesthat may contribute to the manifestation of the ocean’s chemicaldivide on suborbital timescales.

2. Methods and core chronologies

As with other large scale climatic indices that exploit thedifference between time series (such as the Southern OscillationIndex), the purpose of our reconstructions is to highlight mosteffectively the depth-dependent changes by eliminating anypossible shared variability in Southern Ocean isotopic records. It isgenerally recognized that benthic foraminiferal carbon isotopictime series may result from the combined influence of oceancirculation, surface ocean productivity and the air–sea exchange ofcarbon (Broecker and Peng, 1982). And any given carbon isotopicrecord from Southern Ocean sediments no doubt may be subject to

Fig. 2. Result of iterative matching (by eye) of the millennial scale features of(G. bulloides) d18O and d13C. The reference chronology for both records is depicted inFig. 3. The d13C data from MD97-2120 was not published previously. All other datacome from Pahnke and Zahn (2005) and Mortyn et al. (2003).

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409 401

this potential complexity of processes: in fact, reasonable caseshave been made to support a major role for all of these factors indetermining benthic foraminiferal d13C variability over ice agecycles (Martin, 1990; Charles and Fairbanks, 1992; Mackensen et al.,1993; Stevens and Keeling, 2000). Our premise here is that theinfluence on d13C from some of these processesdthe reminerali-zation of organic matter produced in the surface ocean, for exam-pledshould not have a variable expression with depth (below thethermocline).Thus, while the cancellation of the common isotopicvariability should more precisely describe the evolution of any‘‘chemical divide’’ than is possible with any individual time seriesalone (illustrated in Fig. 1), the accurate construction of verticalgradients would also constrain the possible interpretations of thebasic regional carbon isotopic variability. The same general logicextends to the intermediate-deep difference in oxygen isotopiccomposition of benthic foraminifera, and, in fact, the relationshipbetween the vertical gradient time series (d13C and d18O) offersadditional clues to the processes that might separate the differentlayers of the ocean in warm or cold climates.

Though this principle is straightforward enough, the difficultylies in: (i) finding appropriately resolved records from the differentwater depth ranges; and (ii) correlating the individual sequences inthe absence of a continuous radiometric clock. For these reasons,depth transects of sediment cores are usually constructed fromcores within a confined region, where surface signals should be incommon and therefore useful for correlation. For our purposeshere, however, MD97-2120 from the mid-depth South Pacific(Pahnke and Zahn, 2005) and ODP Site 1089 from the deep SouthAtlantic (Hodell et al., 2002) provide the best available opportunityfor creating continuous time series of vertical isotopic gradients(Fig. 1). Despite the geographic separation of the cores, theyunderlie roughly equivalent surface circumpolar ocean regimes (inthe Northern Sub-Antarctic) and are characterized by approxi-mately the same sedimentation rate (the sediments correspondingto the Last interglacial period lie at about 16 and 20 m subbottomdepth, respectively). As a result, if the surface water variability inthe Pacific and Atlantic sectors of the Sub-Antarctic zone evolved inconcert, then the sedimentary sequences should be correlated with

one another on the basis of the independent surface water tracerssuch as isotopic records from the planktonic foraminifer Globigerinabulloides.

Our best attempt at such a stratigraphic exercise (Fig. 2) consistsof iterative optimization of the correlation between time series thatare, to varying degrees, independent of one another. First, weassumed a chronology (discussed in more detail below) for MD97-2120 and used this chronology as the benchmark for correlation.The choice of the reference sequence was important, because, ofthe two sites, the MD97-2120 sequence is the more continuous. (Asa single piston core, MD97-2120 is not subject to coring gaps as thespliced ODP sequence might be, and, furthermore, because it wasrecovered in shallower water, it was less influenced by carbonatedissolution.) Second, we anchored the timing of the most distinc-tive ‘‘abrupt’’ events in ODP 1089 planktonic d18O records to that oftheir apparent counterparts in the MD97-2120 record. Most ofthese events are undoubtedly the same as those recognized in theAntarctic ice cores, an observation that suggests the circumpolarextent of the anomalies. We then fine-tuned the chronology of Site1089 to maximize the correlation between the respective plank-tonic d13C records. This step was necessary, because there wereintervals in the planktonic d18O records that could not be tiedtogether reliably. There are also some intervals for which plank-tonic d13C variability is clearly correlated with planktonic d18O, and,for these intervals, the match of d13C simply insures stratigraphicconsistency. Finally, we used the benthic foraminiferal d18O asa cross-check on the correlation achieved from the planktonicrecords. A significant fraction of that benthic d18O signal must beshared at both sites, and any egregiously spurious correlationbetween the planktonic records should result in a fundamentaldiscrepancy between these benthic d18O records. However, noadjustments were necessary to accommodate the shared benthicd18O variability, because the optimization of the correlationbetween planktonic foraminiferal records did not violate theconstraints of benthic foraminiferal d18O variability at any point inthe sequence.

This full process resulted in slight modifications to the publishedchronologies for both sequences (Fig. 3), and, in the case of Site1089, the comparison with MD97-120 indicated the presence ofpreviously unrecognized coring gaps (these gaps do not materiallyaffect the conclusions drawn in previous papers; however they aresignificant for our purposes here). Though both records extend forseveral ice age cycles, the gaps in the 1089 sequences make itincreasingly difficult to achieve a consistent correlation prior toabout 200 ka. As a result, we truncated our comparison at thishorizon.

The reference chronology we adopted for MD97-2120 takesadvantage of the strong 23 kyr periodicity evident in the planktonicforaminiferal Mg/Ca SST record. We tuned this 23 kyr variability tothe 23 kyr component of Northern Hemisphere summer insolation,following the logic of Huybers and Denton (2008): the high latitudeSouthern Hemisphere surface temperatures should be sensitive tothe duration of Southern Hemisphere summer, which is in turnhighly correlated with the intensity of Northern Hemispheresummer insolation. This approach is relatively coarse, because weanchor the ordinal points of the filtered sea surface temperaturerecord (every 5–6 kyr) to the astronomical timescale. Theassumption of zero phase offset between Sub-Antarctic SST andboreal summer insolation is entirely compatible with the inde-pendent radiocarbon dating of the upper 50 ka of the MD97-2120record (Pahnke et al., 2003). This assumption of zero phase is alsogenerally compatible with the other possible strategy for marinecore chronological developmentdnamely, the transferral (via‘‘wiggle matching of temperature proxies’’) of the chronology fromDome Fuji, an ice core record for which the O2/N2 ratio in air

-40

-20

0

20

40

6

8

10

12

14

16

0 50 100 150 200

-62

-60

-58

-56

-54

-52

-50

6

8

10

12

14

16

18

0 50 100 150 200

0

50

100

150

200

0 500 1000 1500 2000 2500Depth (cms)

Age

(ka)

Age (ka)

Age (ka)

18O

(ice

) ‰

SST

(°C

)SS

T (°

C)

W/m

2•

Fig. 3. The reference chronology for MD97-2120 was constructed by tuning the 23 kyr component of the Mg/Ca SST reconstruction (Pahnke and Zahn, 2005) to that of NorthernHemisphere summer insolation (top panel). The results can be compared with the d18O ice record from Dome Fuji, placed on the Kawamura et al. timescale (middle panel). Theresulting age depth relationship for this chronology (black line) is also plotted with that of a previously published chronology for MD97-2120 (Pahnke and Zahn, 2005).

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409402

bubbles has been tuned directly to overlying insolation (Kawamuraet al., 2007). The orbitally tuned chronology adopted here forMD97-2120 does suggest a slight lead of SST with respect to theDome Fuji d18O ice record on the Kawamura time scale (Fig. 3b; seealso Huybers and Denton, 2008 for a discussion of this point). Sucha lead is not likely to be real, so slight adjustments would have to bemade in either the marine or ice core chronologies for stratigraphicconsistency. However, the discrepancies between the ice core-tuned chronology and the sediment core-tuned chronology areclose to the limits of stratigraphic resolution. And because there arelegitimate reasons to favor either approachdthe direct orbitaltuning of the sediment record vs. the transferral of the orbitallytuned ice core recorddfurther constraints are required to deter-mine whether either is more appropriate.

3. Results

The respective benthic d13C records ‘‘float’’ on the sediment corestratigraphic correlation (Fig. 4a) and are then differenced toproduce a vertical gradient time series (Fig. 4b). In making thisdifference time series, we use a simple subtraction of records toretain the per mil units, but the results would not be different if we

were to normalize the records first. The principle feature of thisindex is the millennial scale variability, which is highlighted at theexpense of the longer period fluctuations that are prominent in theconstituent records. Spectral analysis of this index (not shown)confirms the lack of power in the orbital frequencies, which wereevidently common to the two time series and were thereforecancelled in the differencing. The millennial scale variability of theindex is not only characteristic of the well-known ‘‘DansgaardOeschger’’ interval of 20–60 ka, but it also dominates intervals ofisotopic Stage 6, through to the penultimate deglaciation. In fact,there are only three well-defined intervals of persistently lowvertical gradient in d13Cdthe Holocene, marine isotope Stage 5a(70–80 kyr), and the Last interglacial period. Aside from theseintervals, the index jumps abruptly from high to low every fewthousand years. And even for the intervals of persistently weakvertical gradient, the development and termination of thoseintervals were seemingly set off by abrupt events (cf. the end of theLast interglacial episode).

Given the characteristics of this d13C difference time series, oneobvious question is whether any particular excursion might beartificialda product of erroneous stratigraphic correlation or theresult of other random errors. This question can be addressed

Fig. 4. Time series of the benthic foraminiferal carbon isotopic composition of themid-depth and deep Southern Ocean are differenced to create a continuous record ofvertical d13C gradient.

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409 403

directly in at least two separate ways. First, the entire procedure(including correlation of the planktonic foraminiferal records) maybe repeated, at least over the last 90 ka, using a different deepAtlantic sedimentary sequencedfor example, either TN057-21 orRC11-83 (Ninnemann et al., 1999). The results (not shown) arevirtually identical, regardless of the deep South Atlantic recordused, an observation suggesting that analytical errors or the idio-syncrasies of any one particular core do not influence the correla-tion of the records to any significant extent. One exception is themarine isotope Stage 5/4 transition, where various deep SouthAtlantic sedimentary sequences cannot be reconciled strati-graphically, probably because of intense dissolution problems andpossibly because of coring gaps. Thus, this is one clear interval thatshould be treated with suspicion. Apart from assessing specificanomalies, we can also lead or lag the constituent d13C series byseveral thousand years away from our ‘‘best guess’’ stratigraphiccorrelation (Fig. 5). These arbitrary shifts certainly do change thetiming and amplitude of many of the excursions in the d13Cdifference time seriesdfor example, over the penultimate degla-ciationdbut they do not alter its general structure. Thus, the basiccharacter of the gradient time series (though not necessarily itsdetails) is fairly robust to errors in stratigraphic correlation.

Closer examination of the individual response in both theintermediate and deep sequences reveals why this is the case. Theabrupt shifts towards increased d13C gradient (heightened ‘‘chem-ical divide’’) nearly always involved an increase in the mid-depthd13C during times of when the deep d13C was at its nadir or justbeginning to rise. Conversely, the abrupt decreases in the verticald13C gradient resulted from an increase in the deep d13C duringtimes when the mid-depth values had been stable for severalthousand years. Under these circumstances, the d13C gradient serieswould have the same general appearance, even if we were to shift

the relative timing of the individual records by 1500 years or so.The robustness of the difference series does not necessarily apply tothe major ice age transitions, where ‘‘edge effects’’ of miscorrela-tion can manifest themselves in a significant way (Fig. 4c).However, in general, the observations imply that the context andinterpretation (hence the ultimate value) of the reconstructedvertical gradients depend to a large extent on the accuracy of thereference chronology, as opposed to the stratigraphic correlationalone. This accuracy can probably only be judged in comparisonwith the chronology of other known events of the Earth’s climaterecord.

One additional systematic issue to consider is the extent towhich the two coring sites remained representative of the upperand lower layers of any vertical discontinuity. If water massboundaries migrated vertically past the fixed depths of the sedi-mentary sequences, then the reconstructed vertical gradientsmight be confounded. This ‘‘site stationarity’’ problem is probablymost relevant for the MD97-2120 sequences, because the core sitelies near the top surface of Upper Circumpolar Deep Water in themodern ocean (Fig. 1). In fact, the orbital scale variability incommon between the mid-depth and deep isotopic records isprobably a reflection of the common heavy influence of Circum-polar Deep Water at both sites. However, for the millennial scale,the site stationarity issue is one that requires additional recordsfrom various depths to assess properlyda prospect that may beimminent (Ninnemann et al., in prep).

3.1. Southern Ocean vertical gradients

Regardless of possible complications and chronological adjust-ments to the d13C vertical gradient time series, it seems clear that itbears the typical stamp of Northern Hemisphere climate fluctua-tions. Over the last 100 ka, the resemblance of this index to theGreenland ice core record of climate is particularly striking, in bothtiming and scale (Fig. 6a). Using the well-established millennial scaleevents of the circum-North Atlantic as a reference, the abruptstrengthening of the vertical d13C gradient in the Southern Ocean (inessence, an increase in the mid-depth d13C values) seeminglyoccurred in conjunction with the all the major stadial events in theice core, especially those also featuring Heinrich events (McManuset al., 1999). This association was emphasized in the originaldescription of the MD97-2120 d13C record (Pahnke and Zahn, 2005).On the other hand, the decreases in the d13C gradient (in essence, anincrease in deep d13C values) seemingly occurred during the tran-sitions to all the interstadial events of the ice core record. Thus, whilethe associationwith the abrupt climate events of the last 100 kyr wasdistinct in the deep (vs. mid-depth) Southern Ocean, the verticalchemical gradient maintained a relatively consistent relationshipwith canonical records of Northern Hemisphere climate events, inboth the cooling and warming phases.

This relationship between the Southern Ocean and NorthernHemisphere climate was not confined to the limits of the GreenlandIce core record. The Chinese speleothem record of monsoon vari-ability is now acknowledged as a precisely dated archive mani-festing the influence of high latitude Northern Hemisphere climatechange in Southeast Asia; in the speleothem records, anomalousmonsoon activity shows a close temporal link to high latitudeNorthern Hemisphere climate on millennial timescales (Wanget al., 2008). Fig. 6b demonstrates that there is also a relatively closetemporal match between the ‘‘anomalous’’ fluctuations in monsoonstrengthddefined here by the deviations away from the dominantprecessional variabilitydand the Southern Ocean vertical isotopicgradients throughout at least the last 190 ka reflecting that both arelinked via atmospheric connections to North Atlantic variability. Itis interesting to note, for example, interstadial-like configurations

0

0.5

1

1.5

20 50 100 150 200

1

1.5

2

2.5

3

3.5

4120 125 130 135 1400 5 10 15 20

0.5

1

1.5

2

2.5

3

3.5

4

4.5

0

0.5

1

1.5

20 5 10 15 20 120 125 130 135 140

0

0.5

1

1.5

2

13C

‰

Age (ka)

13C

‰

13C

‰•

Age (ka) Age (ka)

18O

‰

18O

‰•

a

b c

ed

Fig. 5. The basic characteristics of the vertical gradient time series are fairly robust to details of stratigraphic correlation. Top panel (a) illustrates the effects of stratigraphiccorrelation, as the 1089 time series is arbitrarily lagged þ1.5 kyr (red line) and �1.5 kyr (gray line) away from the ‘‘best guess’’ correlation (black line). Middle panels (b, c) show anexpanded view of the time series in (a) for Terminations I and II. Bottom panels (d, e) illustrate the effect on the correlation with MD97-2120 planktonic foraminiferal d18O referenceseries (black line), assuming an arbitrary lag of the 1089 time series by þ1.5 kyr (blue line) for Terminations I and II. The orange line in both (d) and (e) represents our ‘‘best guess’’correlation, as in Fig. 2.

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409404

at about 135 ka, well in advance of Termination II, in both themonsoon record and the record of Southern Ocean carbon gradient.Both records also subsequently feature strong anomalies duringHeinrich event 11 at 130 ka. These comparisons of course dependon the orbitally tuned deep sea sediment chronology and its rela-tionship to the U/Th dated speleothem chronology. However, thematch of full records is compelling enough to suggest thatmillennial scale Northern Hemisphere climatic events may be tiedwith the Southern Ocean vertical structure over multiple ice agecycles.

Aside from the temporal signature, additional clues to the originof this Southern Ocean carbon isotopic divide come from

comparison to the vertical oxygen isotopic gradient and the proxiesfor Sub-Antarctic surface temperature. Increases in the strength ofthe vertical gradient in d13C were generally accompanied byincreases in both the overlying surface temperature and increasesin the vertical gradient of benthic foraminiferal oxygen isotopes ().The increased oxygen isotopic gradient during times of heightenedcarbon isotopic gradient resulted from the fact that the mid-depthforaminiferal d18O shifted to lower values while the deep d18Oremained relatively constant. (The marine isotope Stages 2–4interval offers perhaps the clearest example of this phenomenon).These foraminiferal d18O changes of course could represent eithertemperature fluctuations or shifts in the isotopic composition of

Fig. 6. The vertical gradient in Southern Ocean benthic foraminiferal carbon isotopescompared to the record of high latitude Northern Hemisphere climate from the NorthGRIP ice core (NGRIP members, 2004) and the Chinese speleothem record from SanBaocave (Wang et al., 2008). The ‘‘high pass filter’’ of the speleothem record removesfrequencies lower than 1/10 kyr, to isolate the millennial scale variability from thedominant precessionally-forced response. The intervals characterized by Heinrichevents (McManus et al., 1999) stand out as times of prominent excursions in all therecords. Note that the horizontal scale is split between the top and bottom panels.

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409 405

seawater, but their correlation with foraminiferal Mg/Ca based SSTestimates (Pahnke and Zahn, 2005) and the Antarctic ice corefluctuations (Fig. 6b) strongly suggests that the millennial scalevariations in the vertical oxygen isotopic gradient were the productof upper ocean warming/cooling cycles. We note that sea ice-related changes in the salinity of Circumpolar Deep Water-dcommonly hypothesized for the ice age periods (Stevens andKeeling, 2000; Adkins et al., 2002, 2005)dwould probably notmanifest themselves in the deep foraminiferal d18O record becauseof the relatively insignificant oxygen isotopic fractionation associ-ated with sea ice formation.

Prominentexceptionstothe prevailing pattern of variabilityoccurredduring the deglacial/interglacial transitions, intervals when the timeseries of SSTand the vertical oxygen isotopic gradient were uncorrelated.During these episodes, deep and intermediate oxygen isotopic profilesexperienced shifts of similar magnitude, and, as a result, the verticalgradient did not change significantly along with the surface warming.The implication here is that, during these intervals, the warming pene-trated to the deepest depths of the Southern Ocean. Throughout the iceage periods, however, the enhanced Southern ‘‘chemical divide’’ alsofeatured an enhanced Southern Ocean temperature divide, at least forthe northern periphery of the circumpolar region.

It is important to establish whether the enhanced vertical gradi-ents in the Southern Ocean also preconditioned the other oceanbasins to heightened divides. This issue is difficult to assess empiri-cally, because high deposition rate records that can be correlated withthe Southern Ocean sequences are rare; and we should be clear thatwe are referring here to the millennial scale changes in gradients, asopposed to the vertical structure of ‘‘time slices’’ such as the LGM.Thus, full assessment of the issue must await a more extensivenetwork of records. In any case, records from the Brazil margin (Cameet al., 2003), the Coral Sea (Bostock et al., 2004), and the Arabian Sea(Jung et al., 2009) feature comparable isotopic evolution as MD97-2120, though the radiocarbon chronologies are difficult to align withsufficient precision. And at least one additional Northern IndianOcean record (published in Naqvi et al., 1994) displays carbon andoxygen isotopic trends that are virtually identical to that of MD97-2120, despite the fact that the absolute values of the d18O and d13C arequite different (Fig. 8). The Southern Ocean is the only plausible sourcefor ventilation of mid-depth Indian Ocean water that fills the Anda-man Basin, which has an effective sill depth of 1500 m (Naqvi et al.,1994); thus, the similarity between sedimentary sequences fromthese two regions indicates that Southern Ocean processes made animprint on the characteristics of the bulk interior of the ocean.

4. Discussion

One common conception of vertical temperature and nutrient(carbon) gradients in the ocean’s interior over ice age cycles is thatthe main changes must have resulted from the variable operation ofthe Atlantic meridional overturning circulation. The logic is that asoverturning in the north weakened, the upper layer would come tobe dominated by warm, salty and nutrient poor northern sourcewater just as cold salty and nutrient rich southern source waterinvaded a greater area of the deep ocean (e.g. Alley and Clark, 1999).In some formulations of the ice age ocean (e.g. Stevens and Keeling,2000) the separation between the upper and lower layers would lieat about 2700 m, the effective sill depth of the Drake Passage. Thesegeneral ‘‘conveyor’’ concepts seem especially applicable to thedistribution of sedimentary tracers in the Atlantic (Curry and Oppo,2005; Marchitto et al., 2002). But they have also been applied to thechanges observed in other ocean basins as well (e.g. Waelbroecket al., 2006).

Furthermore, the variable strength of the meridional over-turning circulation relationship is most often invoked to explain thegeographic pattern of abrupt surface climate events globally (e.g.Broecker, 2006). For example, the anti-phased warming of the highlatitudes of the Northern and Southern Hemisphere has beendescribed as the result of the cross equatorial heat piracy of theNorth Atlantic overturning (the ‘‘bi-polar seesaw’’; Crowley, 1992)or as the result of ocean adjustment to freshwater input in theNorth (the ‘‘thermal-freshwater seesaw’’; Knutti et al., 2004). Foreither of these explanations of global abrupt climate change, thegradients of temperature and nutrients in the ocean interior(including the Southern Ocean) should vary in accordance with theperturbations to the sinking of NADW in the north.

Our records of the chemical and thermal divide of the SouthernOcean fit some, but not all aspects of this general conception of theAMOC. For example, the strong resemblance between the SouthernOcean vertical gradient time series and the Greenland ice corerecord of climate might be taken as evidence for the propagation ofanomalies via the ‘‘conveyor’’. On the other hand, it is important torecognize that the Southern Ocean vertical gradients that we havereconstructed here are actually a composite of two distinctresponses that may have separate origins and therefore separateimplications: Heinrich events and other strong North Atlanticcooling events were characterized by prominent excursions

0

0.2

0.4

0.6

0.8

1

1.2

0

0.5

1

1.5

2

2.50 10 20 30 40 50 60

6

8

10

12

14

16

18

0

0.4

0.8

1.2

1.6

0 50 100 150

6

8

10

12

14

16

18

0.5

1

1.5

2

2.5

0 50 100 150

SST

(°C

)SS

T (°

C)

Age (ka)

Age (ka)

13C

‰18

O ‰

18O

‰

13C

‰

a

b

c

Fig. 7. (a) The vertical gradient in Southern Ocean oxygen isotopes (black line) follows the reconstruction of Sub-Antarctic SST (gray line) at MD97-2120 on millennial timescales,except during deglaciation. (b) The same is generally true of the vertical carbon isotopic gradient (black line), but the divergence between Dd13C and SST (gray line) is especiallypronounced during the terminations. (c) The relationship between the vertical carbon isotopic gradient (black line) and oxygen isotopic gradient (gray line) over the last 60 kahighlights the difference in pattern between the ‘‘Dansgaard Oeschger’’ interval and Termination I.

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409406

towards higher d13C and lower d18O (higher temperatures) in themid-depth South Pacific record, while the warming episodes in theNorth Atlantic were marked by significant excursions to high d13Cand modest increases (if anything) in the d18O of deep SouthAtlantic record. These responses of the upper and lower layers ofthe Southern Ocean were anti-correlated to some extent, but therewere prominent exceptions to this rulednotably, during theterminations. During these episodes, the upper Southern Ocean(and the Antarctic continent) warmed when the vertical carbonisotopic gradient flattened dramatically to its modern state, anassociation of events that is exactly the opposite of expectationsfrom a simple ‘‘bi-polar seesaw.’’

And though the ocean seesaw concepts seem capable ofexplaining the millennial scale variability in the cold periods(isotope Stages 2–4, for example), the mechanisms centered solely

on the Northern Hemisphere perturbations to the thermohalinecirculation fail to explain why the terminations would be different(Wolff et al., 2009). For example, the return to modern verticalgradients in carbon at about 14 kadif this were a product of theflux and/or geometry of NADWdshould have been accompanied bystrong cooling in the high southern latitudes. Such a strongresponse is not observed in either marine or ice core records.Furthermore, it is important to emphasize that, even during the‘‘Dansgaard Oeschger’’ events of the Stages 2–4, the verticalgradients in both carbon and oxygen isotopes maintained a strongcorrelation to Sub-Antarctic surface ocean proxy measurements;and sediment cores that record the variations in surface Sub-Antarctic surface water properties during Heinrich events stronglysuggest a southward displacement of the influence of subtropicalwaterdi.e. not just heat. Such fluctuations in turn demand either

Fig. 8. Comparison between the benthic foraminiferal records over the last deglacia-tion from MD97-2120 in the Southwest Pacific (Pahnke and Zahn, 2005) and RC12-344in the Andaman Basin, North Indian Ocean, with an effective sill depth of 1.3 km (Naqviet al., 1994). The values in parentheses on the d13C scale refer to MD97-2120. Thesimilarities between these oxygen and carbon isotopic sequences suggest that vari-ability of the Southern Ocean mid-depth propagated northward into the interior of theocean (though the amplitude of response is attenuated in the northern core).The RC12-344 sequence was not radiocarbon dated; its age scale is estimated here on the basis ofthe oxygen isotopic variability.

Northern hemisphere sea/land ice

anomaly

Shoaling/weakening of Northern

Meridional Overturning(NADW)

Southwardexcursion of

ITCZ

Warming of surface southern hemisphere through

reduction of cross equatorial heat transport

Southward shiftof southern hemisphere

subpolar andpolar atmospheric

boundaries

Altered vertical gradients in

oceanic properties

Warmer and less nutrient-rich

Antarctic Mode and Intermediate Water

A B

Fig. 9. Flow chart of mechanisms that might account for not only the altered verticalgradients in the Southern Ocean over millennial timescales but also the temporalconnection with Northern Hemisphere events. The hypothetical oceanic pathway, ‘‘A’’is the most popular in published literature (e.g. Waelbroeck et al., 2006; Clark et al.,2007), but the atmospheric pathway, ‘‘B’’, deserves serious consideration. These twopathways need not be mutually exclusive.

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409 407

a change in the westerly wind field or a change in eddy activity thatdetermines the balance of subtropical (vs. polar) water near thefrontal zones (Ninnemann et al., 1999; Sachs and Anderson, 2005).

These considerations lead us to propose an alternative expla-nation for much of the behavior we observe: that the millennialscale variability in at least the upper Southern Ocean water masseswas a product of atmospheric teleconnections to the NorthernHemisphere (depicted in Fig. 9b) as opposed to effects of the AMOC‘‘conveyor’’. Chiang and Bitz (2005) demonstrate that the imposi-tion of sea ice anomalies in the Northern Hemisphere has thepotential to influence winds and pressure patterns in the SouthernHemisphere, through a southward shift in the convergence zones inthe tropical Pacific. While these and other analogous modelingexperiments (Velinga and Wood, 2002) emphasize the connectionbetween sea ice in the high northern latitudes and the ITCZ, it islogical to expect that the ultimate effects would extend to highersouthern latitudesdespecially if the tropical Pacific was affectedstrongly (Clark et al., 2007). And though the water mass trans-formations that produce intermediate water in the Southern Oceanare not well characterized even for the modern ocean, changes inwind-driven mixing and gradients near the polar frontal zone, orlarge scale meridional shifts in the precipitation-evaporationbalance could certainly give rise to variations in the ‘‘preformed’’characteristics of AAIWdincluding its temperature, salinity,nutrient content and carbon isotopic composition.

One main advantage of explaining the variability of the upper layerof the Southern Ocean ‘‘chemical divide’’ as an atmospheric tele-connection, as opposed to an oceanic propagation, is that the termi-nations would pose no contradiction: if the teleconnection dependedon a link between sea ice and the tropical Pacific convergence zone,

then that link could have been overwhelmed by other more directinfluences on the atmospheric circulation during deglaciation.

What would be unique about terminations in this regard? Forone, the exact times of strongest fall/winter [the most criticalseason for ENSO development (Clement et al., 1999)] insolationforcing on the equatorial Pacific were the very same intervals thatthe normal anti-correlation between the Northern Hemisphere andthe upper Southern Ocean broke down. Other forces such as largechanges in the sea level could also interfere with the sea ice-ITCZlink (Bush and Fairbanks, 2003). Direct greenhouse gas forcing inthe high latitudes of both hemispheres might also be strong enoughto dominate any atmospheric teleconnection; in fact, the typicalpattern of correlation in the vertical gradients broke down whenatmospheric carbon dioxide rose to nearly its full interglacial level.Finally, J.R. Toggweiler (pers. communication) points out anotherpossible unique aspect of the terminationsdthat these may havebeen times of maximum spin-up of the Antarctic CircumpolarCurrent as the westerlies shifted southward to their maximumextent, in which case, the mixing of heat and carbon at sites justnorth of the ACC may appear to be anomalous with respect to otherintervals of less extreme atmospheric forcing. Thus, unlike the

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409408

AMOC, a variety of atmospheric effects could explain the isotopicmanifestation of the terminations in the Southern Ocean.

For the most part, these arguments can be applied most easily tothe upper layer, but it is conceivable that the isotopic composition ofthe lower layer of the Southern Ocean would also be affected byatmospheric teleconnections to the Northern Hemisphere. Histori-cally, the coincidence between increases in deep d13C and NorthernHemisphere stadial/interstadial transitiondfor example, the Bolling/Allerod transition at about 14.5 kyrdhas been explained as a conse-quence of increased NADW input to the Circumpolar Deep Water. Thisexplanation does, however, conflict with other interpretations of d13Cin low-resolution deep Atlantic sediment cores that argue fora minimum flux of NADW during the main stadial/interstadial inter-vals (Lisiecki et al., 2008).This apparent paradox might be reconciled ifone argues that the density of deep water south of the ACC influencesthe admixture of deep water at sites north of the ACC (i.e. at Site 1089).Then the increases in deep d13C at Site 1089 would be sensitive to thenorth/south density contrast as opposed to the flux of NADW, and,consequently, could be subject (through changing Southern Oceansea ice field) to the same atmospheric teleconnection patterns that weenvision for the upper layer.

In any case, this atmospheric teleconnection hypothesis for theupper ocean is testable in a number of respects, because if it wereresponsible for altering the characteristics of AAIW, then, byextension, it may explain much of the appearance of the upper layerof the chemical divide during Heinrich events. For one thing, sucha mechanism allows for significant decoupling between d13C andother nutrient or water mass tracers in the upper layer of the‘‘chemical divide’’, because gas exchange imparts a significantinfluence on d13C in the Sub-Antarctic zone (Lynch-Stieglitz et al.1995; Charles et al., 1993). Secondly, the mechanism might actuallypredict a decoupling between the normal (modern) association ofphysical tracers such as sortable silt and nutrient tracers in the mid-depth ocean (Hoogakker et al., 2007). Thirdly, significant variabilityin the carbon isotopic composition of intermediate water formed inthe Southern Ocean should impart a signature in the atmosphere aswell (for example, during Heinrich events, this mechanism shouldlead to relative depletion of atmospheric radiocarbon or carbon-13)despecially, if, as implied by the records in Fig. 7, the SouthernOcean processes influenced a substantial volume of the upperocean. Finally, the characteristics of specific intervals of the ice agecycles should be diagnostic of mechanism. For example, the initialchanges in Southern Ocean vertical gradients across Terminations Iand II bear strong resemblance to those of the MIS 4/3 boundary,yet an anti-phased pattern of warming in the Southern Oceanupper layer (relative to Northern Hemisphere climate) was ulti-mately maintained during the Stage 4/3 transition, unlike duringTerminations I and II.

This last point serves as a reminder that the Southern Oceanvertical gradients we have reconstructed here have effectivelyfiltered out the longer period (orbital scale) fluctuations in prop-erties that are common to the depths of these specific sedimentarysequences. The gradient records are therefore blind to possiblechanges in the vigor or geometry of the thermohaline circulation onthese longer timescales that might have driven mid-depth anddeep Southern Ocean in parallel. Furthermore, with only two sites,we obviously cannot claim that the behavior we observe capturesall aspects of the ‘‘chemical divide’’dfor example, its expressionbetween 2000–3000 m water depth in time slice reconstructions ofthe LGM (Marchitto and Broecker, 2006). Nevertheless, it is worthconsidering the processes that gave rise to the changes we observeand the extent to which they should operate in the same way onvarious timescales. Given that the Southern Ocean millennial scalevariability analyzed here is demonstrably different from longerperiod behavior; and given that the terminations are apparently

unique (in their pattern of associations) from the rest of the ice agecycle, the evolution of Southern Ocean vertical gradients must havebeen a product of more than just the variable operation of the NorthAtlantic thermohaline circulation.

Regardless of origin, changes in the vertical gradients in theSouthern Ocean would certainly influence the appearance of watermass mixing in sensitive regions of the world ocean (Skinner et al.,2003), the heat storage and therefore the sensitivity of the climatesystem to perturbation (Adkins et al., 2005), and the CO2 balance inthe atmosphere (Toggweiler, 1999; Sigman and Boyle, 2000). Thusthe time series created here provide a necessary step towardintegration of these various aspects of ‘‘abrupt change’’. Further-more, the comparison between the Southern Ocean ‘‘chemicaldivide’’ and the U/Th dated speleothem record (Wang et al., 2008)suggests that a unified chronology of abrupt events throughoutmuch of the ocean/atmosphere system may be achievable.

Acknowledgements

This work was an outgrowth of projects supported previously byNSF and the Ocean Drilling Program (grants to CDC and DAH). Wethank numerous colleagues for discussion, including Ralph Keeling,Jess Adkins and Jeff Severinghaus. We also thank Robbie Toggweilerand an anonymous reviewer for their especially helpful commentson an earlier draft.

References

Adkins, J.F., McIntyre, K., Schrag, D.P., 2002. The salinity, temperature, and delta O-18 of the glacial deep ocean. Science 298, 1769–1773.

Adkins, J.F., Ingersoll, A.P., Pasquero, C., 2005. Rapid climate change and conditionalinstability of the glacial deep ocean from the thermobaric effect and geothermalheating. Quaternary Science Reviews 24, 581–594.

Alley, R.B., Clark, P.U., 1999. The deglaciation of the northern hemisphere: A globalperspective. Annual Review of Earth and Planetary Sciences 27, 149–182.

Bostock, H.C., Opdyke, B.N., Gagan, M.K., Fifield, L.K., 2004. Carbon isotope evidencefor changes in Antarctic Intermediate Water circulation and ocean ventilation inthe southwest Pacific during the last deglaciation. Paleoceanography 19.doi:10.1029/2004PA001047 PA4013.

Boyle, E.A., 1988. The role of vertical chemical fractionation in controlling latequaternary atmospheric carbon-dioxide. Journal of Geophysical Research-Oceans 93, 15701–15714.

Broecker, W.S., 2006. Abrupt climate change revisited. Global and Planetary Change54, 211–215.

Broecker, W.S., Peng, T.-H., 1982. Tracers in the Sea. Eldigio Press.Bush, A.B.G., Fairbanks, R.G., 2003. Exposing the Sunda shelf: tropical responses to

eustatic sea level change. Journal Of Geophysical Research-Atmospheres 108.doi:10.1029/2002JD003027.

Came, R.E., Oppo, D.W., Curry, W.B., 2003. Atlantic Ocean circulation during theYounger Dryas: insights from a new Cd/Ca record from the western subtropicalSouth Atlantic. Paleoceanography 18 (4), 1086. doi:10.1029/2003PA000888.

Charles, C.D., Fairbanks, R.G., 1992. Evidence from Southern Ocean sediments for theeffect of North Atlantic deep water flux on climate. Nature 355, 416–419.

Charles, C.D., Wright, J.D., Fairbanks, R.G., 1993. Thermodynamic influences on themarine carbon isotope record. Paleoceanography 8, 691–698.

Chiang, J.C.H., Bitz, C.M., 2005. Influence of high latitude ice cover on the MarineIntertropical Convergence Zone. Climate Dynamics 25, 477–496.

Clark, P.U., Hostetler, S.W., Pisias, N.G., Schmittner, A., Meissner, K.J., 2007. Mecha-nisms for a w7-kyr climate and sea-level oscillation during marine isotopestage 3. In: Schmittner, A., Chiang, J., Hemming, S. (Eds.), Ocean Circulation:Mechanisms and Impacts. Geophysical Monograph 173. American GeophysicalUnion, Washington, D.C, pp. 209–246.

Clement, A.C., Seager, R., Cane, M.A., 1999. Orbital controls on the El Nino/SouthernOscillation and the tropical climate. Paleoceanography 14, 441–456.

Crowley, T.J., 1992. North Atlantic Deep Water cools the Southern Hemisphere.Paleoceanography 7, 489–497.

Curry, W.B., Oppo, D.W., 2005. Glacial water mass geometry and the distribution ofd13C of SCO2 in the Western Atlantic Ocean. Paleoceanography 20 PA1017.

Herguera, J.C., et al., 1992. Evidence for a bathyl front at 2000-m depth in the glacialPacific, based on a depth transect on Ontong Java Plateau. Paleoceanography 7,273–288.

Hodell, D.A., Charles, C.D., Curtis, J.H., Mortyn, P.G., Ninnemann, U.S., Venz, K.A.,2002. Data Report: Oxygen Isotope Stratigraphy of ODP Leg 177 Sites 1088,1089, 1090, 1093, and 1094, Proc. ODP, Sci. Results 177, Ocean Drilling Program,College Station, TX.

C.D. Charles et al. / Quaternary Science Reviews 29 (2010) 399–409 409

Hodell, D.A., Venz, K., Charles, C.D., Ninnemann, U.S., 2003. Pleistocene verticalcarbon isotope and carbonate gradients in the South Atlantic sector of theSouthern Ocean. Geochemistry, Geophysics, Geosystems. doi:10.1029/2002GC000367.

Hoogakker, B.A.A., McCave, I.N., Vautravers, M.J., 2007. Antarctic link to deep flowspeed variation during Marine Isotope Stage 3 in the western North Atlantic.Earth and Planetary Science Letters 257, 463–473.

Huybers, P., Denton, G., 2008. Antarctic temperature at orbital timescales controlledby local summer duration. Nature Geoscience 1, 787–792.

Jung, S.J.A., Kroon, D., Ganssen, G., Peeters, F., Ganeshram, R., 2009. EnhancedArabian Sea intermediate water flow during glacial North Atlantic cold phases.Earth and Planetary Science Letters 280, 220–228.

Kawamura, K., Parrenin, F., Lisiecki, L., Uemura, R., Vimeux, F., Severinghaus, J.P.,Hutterli, M.A., Nakazawa, T., Aoki, S., Jouzel, J., Raymo, M.E., Matsumoto, K.,Nakata, H., Motoyama, H., Fujita, S., Goto-Azuma, K., Fujii, Y., Watanabe, O., 2007.Northern Hemisphere forcing of climatic cycles in Antarctica over the past360,000 years. Nature 448, 912–916.

Knutti, R., Fluckiger, J., Stocker, T.F., Timmermann, A., 2004. Strong hemisphericcoupling of glacial climate through freshwater discharge and ocean circulation.Nature 430, 851–856.

Legrand, P., Wunsch, C.,1995. Constraints from paleotracer data on the North-Atlanticcirculation during the last glacial maximum. Paleoceanography 10, 1011–1045.

Lisiecki, L.E., Raymo, M.E., Curry, W.B., 2008. Atlantic overturning responses to LatePleistocene climate forcings. Nature 456, 85–88.

Lynch-Stieglitz, J., Stocker, T.F., Broecker, W.S., Fairbanks, R.G., 1995. The influence ofair–sea exchange on the isotopic composition of oceanic carbon: observationsand modeling. Global Biogeochemical Cycles 9 (4), 653–665.

Mackensen, A., Hubberten, H.-W., Bickert, T., Fischer, G., Futterer, D.K., 1993. The d13Cin benthic foraminiferal tests of Fontbotia wuellerstorfi (Schwager) relative to thed13C of dissolved inorganic carbon in Southern Ocean deep water: implicationsfor glacial ocean circulation models. Paleoceanography 8 (5), 587–610.

Marchitto, T.M., Broecker, W.S., 2006. Deep water mass geometry in the glacialAtlantic Ocean: a review of constraints from the paleonutrient proxy Cd/Ca.Geochemistry Geophysics Geosystems 7, Q12003.

Marchitto, T.M., et al., 2002. Paired benthic foraminiferal Cd/Ca and Zn/Ca evidencefor a greatly increased presence of Southern Ocean water in the glacial NorthAtlantic. Paleoceanography 17 (3), 16.

Martin, J.H., 1990. Glacial–interglacial CO2 change: the iron hypothesis. Paleo-ceanography 5, 1–13.

McManus, J.F., Oppo, D.W., Cullen, J.L., 1999. A 0.5 million year record of millennial-scale climate variability in the North Atlantic. Science 283, 971–975.

Mortyn, P.G., Charles, C.D., Ninnemann, U.S., Ludwig, K., Hodell, D.A., 2003. Deep seasedimentary analogs to the Vostok Ice Core. Geochemistry Geophysics Geo-systems 4, 8405.

Naqvi, W.A., Charles, C.D., Fairbanks, R.G., 1994. Carbon and oxygen isotopic recordsof benthic foraminifera from the Northeast Indian Ocean: implications onglacial–interglacial atmospheric CO2 changes. Earth and Planetary ScienceLetters 121, 99–110.

NGRIP Members, 2004. High resolution Climate Record of the Northern Hemisphereback into the last Interglacial Period. Nature 431, 147–151.

Ninnemann, U.S., Charles, C.D., Hodell, D., 1999. Origin of global millennial scaleclimate events: constraints from the Southern Ocean Deep Sea SedimentaryRecord. In: Clark, P., Keigwin, L., Webb, R. (Eds.), Mechanisms of Global ClimateChange at Millennial Timescales, AGU Geophysical Monograph Series, vol. 112,p. 99–112.

Ninnemann, U.S., Charles, C.D., 2002. Changes in the mode of Southern Oceancirculation over the last glacial cycle revealed by foraminiferal stable isotopicvariability. Earth and Planetary Science Letters 201, 383–396.

Pahnke, K., Zahn, R., Elderfield, H., Schulz, M., 2003. 340,000-year Centennial-scalemarine record of southern hemisphere climatic oscillation Science 301, 948–952.

Pahnke, K., Zahn, R., 2005. Southern Hemisphere water mass conversion linkedwith North Atlantic climate variability. Science 307, 1741–1746.

Piotrowski, A.M., et al., 2004. Intensification and variability of ocean thermohalinecirculation through the last deglaciation. Earth and Planetary Science Letters225 (1-2), 205–220.

Piotrowski, A.M., Goldstein, S.L., Hemming, S.R., Fairbanks, R.G., Zylberberg, D.R.,2008. Oscillating glacial northern and southern deep water formation fromcombined neodymium and carbon isotopes. Earth and Planetary Science Letters272, 394–405.

Sachs, J.P., Anderson, R.F., 2005. Increased productivity in the Subantarctic oceanduring Heinrich events. Nature 434, 1118–1121.

Schlitzer, R., 2000. Electronic atlas of WOCE hydrographic and tracer data nowAvailable. EOS, Transactions American Geophysical Union, 81 (5), 45.

Sigman, D.M., Boyle, E.A., 2000. Glacial/interglacial variations in atmosphericcarbon dioxide. Nature 407, 859–869.

Skinner, L.C., Shackleton, N.J., Elderfield, H., 2003. Millennial-scale variability ofdeep-water temperature and delta O-18(Dw) indicating deep-water sourcevariations in the Northeast Atlantic, 0–34 cal. ka BP. Geochemistry GeophysicsGeosystems 4, 1098. doi:10.1029/2003GC000585.

Stevens, B.B., Keeling, R.F., 2000. The influence of Antarctic sea ice on glacial–interglacial CO2 variations. Nature 404, 171–174.

Toggweiler, J.R., 1999. Variation of atmospheric CO2 by ventilation of the ocean’sdeepest water. Paleoceanography 14, 571–588.

Vellinga, M, Wood, RA, 2002. Global climatic impacts of a collapse of the Atlanticthermohaline circulation. Climatic Change 54, 251–267.

Waelbroeck, C., Levi, C., et al., 2006. Distant origin of circulation changes in theIndian Ocean during the last deglaciation. Earth and Planetary Science Letters243, 244–251.

Wang, Y.J., Cheng, H., Edwards, R.L., et al., 2008. Millennial- and orbital-scalechanges in the East Asian monsoon over the past 224,000 years. Nature 451,1090–1093.

Wolff, E.W., Fischer, H., Rothlisberger, R., 2009. Glacial terminations as southernwarmings without northern control. Nature Geoscience. doi:10.1038/NGEO442.

Yu, E.F., et al., 1996. Similar rates of modern and last-glacial ocean thermohalinecirculation inferred from radiochemical data. Nature 379, 689–694.