Impact of abrupt deglacial climate change ontropical Atlantic subsurface temperaturesMatthew W. Schmidta,1, Ping Changa,b, Jennifer E. Hertzberga, Theodore R. Them IIa, Link Jia, and Bette L. Otto-Bliesnerc

aDepartment of Oceanography, Texas A&M University, College Station, TX 77843; bPhysical Oceanography Laboratory, Ocean University of China, Qingdao266003, People’s Republic of China; and cNational Center for Atmospheric Research, Boulder, CO 80307

Edited by James C. McWilliams, UCLA, Los Angeles, CA, and approved August 1, 2012 (received for review May 8, 2012)

Both instrumental data analyses and coupled ocean-atmospheremodels indicate that Atlantic meridional overturning circulation(AMOC) variability is tightly linked to abrupt tropical North Atlantic(TNA) climate change through both atmospheric and oceanicprocesses. Although a slowdown of AMOC results in an atmo-spheric-induced surface cooling in the entire TNA, the subsurfaceexperiences an even larger warming because of rapid reorganiza-tions of ocean circulation patterns at intermediate water depths.Here, we reconstruct high-resolution temperature records usingoxygen isotope values and Mg/Ca ratios in both surface- and sub-thermocline-dwelling planktonic foraminifera from a sedimentcore located in the TNA over the last 22 ky. Our results show sig-nificant changes in the vertical thermal gradient of the upperwatercolumn, with the warmest subsurface temperatures of the lastdeglacial transition corresponding to the onset of the YoungerDryas. Furthermore, we present new analyses of a climate modelsimulation forced with freshwater discharge into the North Atlan-tic under Last Glacial Maximum forcings and boundary conditionsthat reveal a maximum subsurface warming in the vicinity of thecore site and a vertical thermal gradient change at the onset ofAMOC weakening, consistent with the reconstructed record. To-gether, our proxy reconstructions and modeling results provideconvincing evidence for a subsurface oceanic teleconnection link-ing high-latitude North Atlantic climate to the tropical Atlantic dur-ing periods of reduced AMOC across the last deglacial transition.

Mg/Ca paleothermometry ∣ paleoclimate modeling ∣ Bonaire Basin ∣Heinrich Event ∣ sea surface temperature

Observational records of 20th century ocean-temperaturevariability in the tropical North Atlantic (TNA) show a

strong anticorrelation between surface cooling and subsurfacewarming over the past several decades that is thought to be as-sociated with recent variability in Atlantic meridional overturningcirculation (AMOC) (1). Furthermore, coupled atmosphere-ocean general circulation model (AOGCM) simulations indicatethat AMOC changes are tightly coupled to tropical Atlantic cli-mate (2–6), revealing a prominent subsurface warming in theTNA resulting from a major reorganization of intermediate-water circulation in response to AMOC weakening (3, 7, 8). Thesubsurface warming and associated change in the vertical thermalgradient in the western TNA have been identified as an importantfingerprint of AMOC variations (1). However, the validity of themodeling results during past abrupt climate events, when AMOCwas significantly weakened, has not been fully tested because ofa lack of high-resolution paleoproxy subsurface records. Mostproxy reconstructions in the TNA are for sea surface temperature(SST), and these records tell an inconsistent story: Some indicatea surface cooling during the Younger Dryas (YD) cold period (9,10) whereas others suggest SSTs increased (11–13). The only ex-isting deglacial proxy records of intermediate-water change in thewestern TNA are a benthic foraminiferal δ18O record from theTobago Basin at approximately 1.3 km depth (14) and a benthicforaminiferal Mg/Ca record from the Florida Straits at approxi-mately 750 m depth (15). Because observational and modelingstudies point to the subsurface between 300 and 600 m depth

in the western TNA as the region most sensitive to AMOC varia-bility (2), there is a dire need for high-resolution subsurface proxyrecords of temperature change within this depth range.

To reconstruct vertical changes in the thermal gradient of theTNA upper-water column over the past 22 ky, we measure δ18Ovalues and Mg/Ca ratios (as temperature proxies) in the plank-tonic foraminifera Globigerinoides ruber and Globorotalia crassa-formis from southern Caribbean sediment core VM12-107.Located between the Cariaco Basin and the Netherlands Antilles(Fig. S1), the core site is within a region influenced by coastalupwelling in the Bonaire Basin (Fig. 1).

ResultsThe age model for VM12-107 is based on 10 calibrated radiocar-bon dates from planktonic foraminifera spanning the upper270 cm of VM12-107, resulting in a deglacial sedimentation rateof approximately 18 cm∕ky (Table 1 and Fig. S2). Radiocarbonages were converted to calendar ages using CALIB 6.0 (16), usingthe standard 400–y reservoir age correction. Although recentstudies suggest reservoir ages in the nearby Cariaco Basin mayhave varied by several hundred years during periods of AMOCslowdown, such as the start of the YD and Heinrich Event 1 (H1)(17–19), it is possible that these changes were a local effect withinthe basin caused by restricted circulation during periods of lowersea level (18) and that reservoir age changes in the open Carib-bean/tropical Atlantic were much less. Furthermore, evidencesuggests that deglacial reservoir age changes in the tropical Atlan-tic were restricted to only brief periods at the beginning of theYD, between 13.0 and 12.53 ky (17), and during H1 (18). Omis-sion of the early YD radiocarbon date at 130 cm in the VM12-107age model (calibrated age of 12.67 ky) only changes the timing ofthe start of the YD in the VM12-107 record by approximately100 y (Figs. S2 and S3) and does not change the interpretationof the results.

G. ruber lives in the upper mixed layer in the southern Carib-bean (20, 21). Although the life cycle of deep-dwelling planktonicforaminifera can involve a vertical migration of several hundredmeters, G. crassaformis shell geochemistry indicates it mainly cal-cifies below the thermocline at a depth of 400–600 m in the south-ern Caribbean (20, 21). Based on shell oxygen isotope data,recent studies on the deep-dwelling planktonic foraminiferaG. crassaformis and Globorotalia truncatulinoides from the Flor-ida Straits and the Gulf of Mexico showed that, although G. trun-catulinoides may have migrated to much shallower water depths

Author contributions: M.W.S. and P.C. designed research; M.W.S., P.C., J.E.H., and T.R.T.performed research; B.L.O.-B. contributed new reagents/analytic tools; M.W.S., P.C.,J.E.H., T.R.T., and L.J. analyzed data; and M.W.S., and P.C., wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The data reported in this paper is archived at the National Oceanic andAtmospheric Administration Paleoclimatology database, www.ncdc.noaa.gov/paleo/paleo.html.1To whom correspondence should be addressed. E-mail: schmidt@ocean.tamu.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1207806109/-/DCSupplemental.

14348–14352 ∣ PNAS ∣ September 4, 2012 ∣ vol. 109 ∣ no. 36 www.pnas.org/cgi/doi/10.1073/pnas.1207806109

during the late deglacial and early Holocene, G. crassaformismaintained a more constant depth habitat near the base of thethermocline (22, 23). Therefore, we chose to use G. crassaformisas a proxy for recording subsurface conditions within a constantdepth range across the deglacial.

The G. ruber δ18O record displays a glacial-interglacial dif-ference of approximately 2.75‰, with the most positive valuesduring the Last Glacial Maximum (LGM) (Fig. 2A and Table S1).In comparison, the G. crassaformis δ18O record indicates a smal-ler glacial-interglacial amplitude of only approximately 1.2‰(Fig. 2B and Table S2). Comparison of the δ18O records showsa maximumΔδ18O gradient between the mixed layer and the sub-

thermocline over the last 7.2 ky. Because both temperature andsalinity affect foraminiferal δ18O values, this maximumHoloceneΔδ18O gradient suggests the vertical temperature/salinity gradi-ent was reduced during the LGM and the deglacial.

Mg/Ca ratios in G. ruber were converted to upper mixed layertemperatures using an Atlantic core-top calibration (21) (Fig. 2A,see SI Methods and Fig. S4). Results indicate a core-top tempera-ture of 25.7 °C, in agreement with the modern average annualtemperature of 25.6 °C at 30 m water depth at the core site (24)(Figs. 1 and 2C). Because the core is located within an upwellingregion, it is not surprising that cooler subsurface temperatureswould have a greater influence on the average mixed layer tem-perature. The G. ruber Mg/Ca record indicates an LGM coolingof 4.5 °C. Mixed layer temperatures initially increase at 18 ky andwarm to near modern values by 12.4 ky before abruptly cooling by2.2 °C midway through the YD (Fig. 3A).

Because the abundance of G. crassaformis decreased throughthe Holocene, the youngest measured G. crassaformis Mg/Ca ra-tio is at 5.43 ky (Fig. 2D). Although conversion of Mg/Ca ratios indeep-dwelling planktonic foraminifera is not as well-calibrated asin G. ruber, the calculated Mg/Ca temperature for this interval is10.0 °C. This temperature corresponds to modern conditions at400 m in the Bonaire Basin (24) and is in agreement with eco-logical studies (20, 25). Unlike the G. ruber temperature record,the deglacial G. crassaformis Mg/Ca–temperature record indi-cates LGM subsurface temperatures that were slightly warmerthan those in the early Holocene. Most surprising, the warmest

80°W 60°W 40°W 20°W 0° 20°E20°S

10°S

0°

10°N

20°N

30°N

35

35.235.4

35.435.6 35.6 35.6

35.8 35.8

36

8 9 10 11 12 13 14 15

Temperature (oC)

Longitude (oW)

10 °

˚La

titud

e (

N)

15°

26.527

2727.5

Colombian Basin

VM12-107

Laguna deLos Anteojos

Cariaco Basin

2824 25 26 27

85° 80° 75° 70° 65° 60°

VM12-107

Temperature (oC)

Fig. 1. Site location and modern surface and subsurface hydrography.Temperature (color) and salinity (contour) along density surface σ ¼ 1;026.8that varies from approximately 200 m near the equator to approximately600 m in the subtropics. The sharp subsurface temperature gradient alongthe boundary between the subtropical and tropical gyres is evident, separ-ating the warm SMW in the subtropical North Atlantic from the fresher tro-pical gyre water. This maximum subsurface temperature gradient forms atapproximately 300 m, extending to the western boundary and intersectingwith it near 10 °N. The two solid arrows indicate the southwestward sub-ducted flow in the TNA and the northward AMOC return flow, respectively.The dashed arrow indicates the equatorward western boundary flow result-ing from bifurcation of the subducted flow at the western boundary. Com-petition between this equatorward flow and the northward AMOC returnflow is a key element of the subsurface oceanic gateway mechanism. (Inset)Shows the annual mean temperature at 30 m depth in the southern Carib-bean and the location of VM12-107 (11.33 °N, 66.63 °W; 1,079 m), just outsidethe Cariaco Basin. Also shown are the locations of the Cariaco Basin and La-guna de Los Anteojos in northern Venezuela. The cooler temperatures nearsite VM12-107 are caused by coastal upwelling and are reflected in the cal-culated core-top Mg/Ca temperatures from this site. The temperature andsalinity data are based on World Ocean Atlas 2009 (24, 38).

Table 1. Radiocarbon dates from VM12-107 based on mixedsamples of G. sacculifer and G. ruber specimens

Depth(cm) 14C Age

Ageerror(y)

Calendarage(ky)

Calendarage error (−)

(ky)

Calendarage error (+)

(ky)

2.5 1140 35 0.69 0.05 0.0770.5 7900 55 8.37 0.13 0.1382.5 9380 40 10.22 0.07 0.1290.5 10050 65 11.05 0.26 0.14104.5 10400 50 11.46 0.22 0.23130.5 11200 50 12.67 0.10 0.16160.5 12750 50 14.37 0.35 0.52180.5 13450 60 15.74 0.56 0.64220.5 15600 60 18.47 0.42 0.18270.5 18500 90 21.59 0.26 0.47

Radiocarbon ages were converted to calendar ages using CALIB 6.0(16), using the standard 400-y reservoir age correction.

0.8

1

1.2

1.4

1.6

error

G. crassaformis

17.2

15.8

14.0

12.0

9.50.05

0.07

0.09

-44

-40

-36

0 5 10 15 20

18O (‰

VS

MO

W)

Calendar Age (kyr BP)

F

E

231 P

a / 23

0 Th

-2

-1

0

1

2

3

2.5

3

3.5

4

18O

c (‰

VP

DB

)

Mg

/Ca (m

mo

l/mo

l)

C

A

error

G. ruber

G. ruber

G. crassaformisB

26.2

24.7

23.0

20.9

Tem

p. ( oC

)

YD H1

warmer

cooler

reducedAMOC

D

Fig. 2. Oxygen isotope and Mg/Ca records from site VM12-107 over the past22 ky. Deglacial δ18O and Mg/Ca ratio records in G. ruber (upper mixed layer)(A, C) andG. crassaformis (lower thermocline) (B,D) from southern Caribbeancore VM12-107. The following equations were used to convert Mg/Ca ratiosto temperature: for G. ruber, Mg∕Ca ¼ 0.38exp½0.09 � Temp:� (21), and forG. crassaformis Mg∕Ca ¼ 0.339exp½0.09 � Temp:� (21). Note the abrupt sub-surface warming events during the YD and H1 in the G. crassaformis (orangeline) temperature record. Analytical error on replicate Mg/Ca measurementson G. ruber and G. crassaformis are also shown (C, D). (E) Bermuda Rise231Pa∕230Th record (27) indicating changes in AMOC strength across thedeglacial. (F) Greenland NGRIP ice core δ18O record (26). Gray bars indicatethe YD and H1. Black triangles on x axis show calibrated 14C-based ages inVM12-107.

Schmidt et al. PNAS ∣ September 4, 2012 ∣ vol. 109 ∣ no. 36 ∣ 14349

EART

H,A

TMOSP

HER

IC,

AND

PLANETARY

SCIENCE

S

subsurface temperatures correspond to the early YD, when sub-surface temperatures averaged 6.8 °C warmer (SD 0.5, n ¼ 6)than Holocene values (Fig. 2D).

DiscussionAcknowledging there is debate about the Mg/Ca–temperaturerelationship in deep-dwelling planktonic foraminifera (21, 25),the G. crassaformis Mg/Ca ratios during the early YD show a57% increase relative to Holocene ratios and an 18% increaseat the start of the YD (Fig. 3C). This provides convincing evi-dence for an abrupt subsurface warming associated with theinitiation of the YD at 13.01 ky (þ0.24; −0.15 ky) that is withinage model error of the start of the YD in the North Greenland IceCore Project (NGRIP) ice core record at 12.896 ky (�0.138 ky)(26). Although not as well resolved, G. crassaformis Mg/Ca ratiosalso increase by 15% at approximately 16 ky, likely reflectinga subsurface warming associated with H1 as well (Fig. 2Dand Fig. S5). It is not surprising that the magnitude of subsurfacetemperature change across H1 is not the same as during the YD,given that the boundary conditions and AMOC states bracketingthese events were significantly different. Going into H1, AMOCwas already in a weakened LGM state, and is thought to havedecreased even further during the event (27) (Fig. 2E). In con-trast, the period preceding the YD was characterized by a stron-ger AMOC state (27) (Fig. 2E). Thus, the greater subsurfacetemperature response at the start of the YD may reflect a largermagnitude of AMOC change. Additionally, AMOC remained ina significantly weakened state after H1 and into the “Mystery In-terval” for approximately another 1,000 y before rapidly increas-ing at the Bølling-Allerød transition at 14.5 ky (27) (Fig. 2E).Therefore, we would not expect to find a subsurface cooling trendon the transition out of H1, as observed in our record at the ter-mination of the YD.

A Tobago Basin benthic foraminiferal δ18O record was inter-preted to indicate intermediate-depth warming associated withAMOC slowdowns during the YD and H1 (14) (Figs. 3D and 4A).However, because benthic δ18O values are affected by variationsin water mass, salinity, and temperature, the cause of these δ18Ochanges remains uncertain. Unlike the abrupt warming between400–600 m at the onset of the YD suggested by the VM12-107G.crassaformis Mg/Ca record (Fig. 3C), the Tobago Basin benthicδ18O record suggests a much more gradual warming across theevent at a depth of almost 1.3 km (Fig. 3D). In fact, most of thewarming in the Tobago Basin record occurs during the late YD, aperiod characterized by subsurface cooling in the Bonaire Basin.Therefore, it is likely that different mechanisms were responsiblefor the temperature evolution at these two sites.

Changes in the 231Pa∕230Th ratio recorded in sediments fromthe Bermuda Rise are thought to reflect AMOC variability acrossthe last deglacial (27). As AMOC is reduced, the export of 231Paout of the Atlantic decreases, resulting in an increase in sediment231Pa∕230Th ratios (Fig. 3G). Because the estimated responsetime for this proxy is approximately 500 y (27), the 231Pa∕230Threcord will not reflect an abrupt change in AMOC. Therefore, theincrease in Bermuda Rise 231Pa∕230Th ratios at 12.7 ky is consis-tent with a major reduction in AMOC at the start of the YD at12.896 ky (�0.138 ky) in the NGRIP ice core record (Fig. 3I),followed by a gradual strengthening of AMOC during the lateYD. Comparison of the Bermuda Rise 231Pa∕230Th record withour G. crassaformis Mg/Ca record suggests that subsurface tem-peratures in the southern Caribbean warmed as AMOC abruptlyweakened at the start of the YD and that subsurface tempera-tures in the western tropical Atlantic gradually cooled as AMOCstrengthend during the late YD.

The G. ruber temperature record from VM12-107 also indi-cates an increase in mixed layer temperatures during the first600 y of the YD (Fig. 3A), consistent with previous temperaturereconstructions from the western TNA (11–13, 28) (Figs. 3Eand 4A). Remarkably, both the G. ruber and G. crassaformisMg/Ca records indicate a cooling midway through the YD at12.4 ky, well before the termination of the event in the Greenlandice core record (Fig. 3I), suggesting that subsurface temperatures

0.05

0.06

0.07

0.08

0.09

GreducedAMOC

231 P

a / 23

0 Th

Bermuda Rise

-42

-40

-38

-36

-34

40

50

60

70

80

90

9 10 11 12 13 14 15

H

% clastics

I

18O

(‰

VS

MO

W)

Calendar Age (kyr BP)

dry

wet Los Anteojos

Greenland

lateYD

earlyYD

25

27

29

24

26

28

Mg

/Ca-

SS

T (

oC

) E

Mg

/Ca-S

ST

(oC

)

westernequatorial

Atlantic

eastern equatorialAtlantic F

22

24

26

22

24

26

28

Mg

/Ca-

SS

T (

oC

)

A

Mg

/Ca-S

ST

(oC

)

VM12-107mixed layertemperature

Cariaco Basinmixed layertemperature

B

1

1.2

1.4

1.6

1.7

2.1

2.5

Mg

/Ca

(mm

ol/m

ol)

C

Cib

icido

ides ssp

. 18O

(‰ V

PD

B)

VM12-107sub-thermocline

DTobago Basin benthic1.3 km depth

Fig. 3. Tropical Atlantic climate evolution across the YD. (A) The G. ruberMg/Ca–temperature record from sediment core VM12-107. Note the earlyYD warming, followed by an abrupt cooling at 12.4 ky. (B) The G. ruberMg/Ca–temperature record from the Cariaco Basin (10) showing an abruptcooling at the onset of the YD. Note the coolest temperatures in the CariacoBasin correspond to the warmest temperatures at our study site. (C) Mg/Caratios in G. crassaformis from VM12-107 indicating an abrupt subsurfacewarming in the Bonaire Basin at the start of the YD. (D) The ice volume-corrected Tobago Basin benthic foraminiferal δ18O record (14), using an up-dated age model based on CALIB 6.0 (16), indicating a gradual warming at1.3 km water depth across the YD. (E) The G. ruber Mg/Ca–SST record fromthe western equatorial Atlantic (36) indicating significant warming duringthe early YD. (F) An eastern equatorial Atlantic SST record (28) based onMg/Ca ratios in G. ruber. (G) Bermuda Rise 231Pa∕230Th record (27) indicatingchanges in AMOC strength across the deglacial. (H) The percentage of clasticsediments in a Venezuelan Andes lake core record indicating the develop-ment of extremely arid conditions during the early YD and a gradual returnto wetter conditions during the late YD (29). (I) The Greenland NGRIP ice coreδ18O record, indicating the abrupt return to cold temperatures in Greenlandduring the YD (26). The gray and green bars indicate the early and late YD,respectively.

14350 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1207806109 Schmidt et al.

significantly influenced mixed layer conditions during this inter-val (Fig. 3A and C). As shown in a recent modeling study, subsur-face warming associated with a reduction in AMOC can influencemixed layer temperatures in zones of strong coastal upwelling (8).Therefore, intensification of coastal upwelling in the TNA duringthe YD is predicted to increase mixed layer temperatures at sitesalong the western boundary current, as reflected in the VM12-107 G. ruber Mg/Ca–temperature record. In contrast, a well-re-solved G. ruber Mg/Ca–temperature record from inside the Car-iaco Basin indicated a large cooling of 4 °C at the start of the YD(10) (Fig. 3B). Because the Cariaco Basin’s shallow sill (<100 mduring the deglacial) would inhibit the inflow of warm subsurfacewaters characteristic of the open TNA, the significant surfacecooling in the Cariaco Basin G. ruber Mg/Ca record is additionalevidence that regional mixed layer temperatures were influencedby a strong subsurface warming. Unlike in the TNA, the mixedlayer cooling inside the Cariaco Basin most likely reflects a com-bination of atmospheric-induced cooling and increased upwellingof cold intermediate waters restricted to inside the basin.

The Cariaco Basin G. ruber Mg/Ca–temperature record alsoindicates a two-phased YD, with the most intense cooling andstrongest upwelling occurring during the first 600 y (Fig. 3B). In-tensification of upwelling in the Cariaco Basin is consistent with asouthward displacement of the intertropical convergence zone(ITCZ) and decreased rainfall over northern Venezuela. A recentsedimentological study of a Venezuelan lake core (29) also sug-gested the most arid conditions in the southern Caribbean oc-curred during the first 600 y of the YD (Fig. 3H). Together, theseproxy reconstructions suggest the first half of the YD was mostextreme, when regional aridity and upwelling were at a maximum.

To determine if the subsurface temperature increase identifiedin our proxy record during the YD is consistent with the pre-viously identified mechanism found in an AOGCM water-hosingsimulation conducted under present-day conditions (3), we ana-lyze a new set of water-hosing simulations using LGM forcingsand boundary conditions (30). Observational data show thatwarm salinity maximum waters (SMW) from the North Atlanticsubtropical gyre are subducted and carried equatorward via theNorth Atlantic subtropical cell (STC). However, the modernequatorward pathway of the STC is blocked by strong northwardAMOC return flow along the western boundary, resulting in asharp subsurface temperature gradient separating the warm, saltysubducted SMW from the fresher tropical gyre water (31–33)(Fig. 1). A recent modeling study showed that when AMOC andits northward return flow weaken, the maximum subsurface tem-perature gradient zone at the western boundary gives rise to amaximum subsurface warming (2). The warming responds quicklyto the freshwater forcing because of a planetary wave-adjustmentprocess (1, 3, 34). If the AMOC slowdown is sufficiently strong,causing its return flow to be weaker than the equatorward branchof the North Atlantic STC (dashed arrow in Fig. 1), the warmSMW can flow into the equatorial zone and subsequently into thetropical South Atlantic (2). Through upwelling, the subsurfacewarming can influence SSTs, affecting the position and strengthof the ITCZ (3). However, this mechanism has only been identi-fied in AOGCM experiments under modern climate conditions(3) and has not yet been tested for past abrupt climate events.

Analyses of LGM water-hosing simulations confirm the opera-tion of the subsurface oceanic mechanism under LGM climateconditions (SI Methods, Figs. S6 and S7). Even with a relativelyweak freshwater forcing (0.1 Sv), the simulation produces asubstantial subsurface warming in the western TNA (Fig. 4Band Fig. S8). The averaged temperature between 300–600 mwarmed by approximately 5 °C near the maximum subsurfacetemperature gradient zone (Fig. 4B). At the surface, the modelsimulation shows patches of surface warming off the northernSouth American coast (Fig. 4A) produced by coastal upwellingthat brings the strong subsurface warming to the surface, counter-

acting the surface cooling induced by atmospheric processes (8).This finding supports our proxy reconstruction indicating thatsubsurface warming during the early YD significantly influencedmixed layer conditions in the Bonaire Basin, located within a up-welling zone (Fig. 3 A and C).

Our new high-resolution paleotemperature reconstructionsand modeling analyses provide evidence that abrupt changesin TNA climate were coupled to AMOC variability across the de-glacial. The most likely explanation for the abrupt subsurfacewarming in the southern Caribbean at the start of the YD is alarge change in horizontal heat advection near the maximum sub-surface temperature gradient zone caused by weakening of thewestern boundary current in response to a sudden AMOC reduc-tion. As the western boundary current continued to weaken andreach a threshold, warm SMW intruded into the equatorial zonealong the western boundary and subsequently spread into the tro-pical South Atlantic, producing surface warming south of theequator (3). Previous SST reconstructions from the eastern tro-pical South Atlantic suggest a warming at the start of the YD (35,36) (Figs. 3Fand 4A), consistent with the subsurface oceanic gate-way mechanism. Together with atmospheric-induced surfacecooling in the TNA, this caused the ITCZ to shift to its most ex-treme southern position during the early YD. Then, the SMWgradually cooled because of atmospheric processes at their sourceregion, and the subsurface warming diminished in the TNAduring the late YD. This would cause SSTs in tropical upwellingzones to also cool, affecting the position and strength of the ITCZduring the second half of the YD, well before the termination of

80°W 60°W 40°W 20°W 0° 20°E20°S

10°S

0°

10°N

20°N

0

0

0

0

0

0

0

0

0

0.50

1

1

0

0

0

0

0

0

0

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0

80°W 60°W 40°W 20°W 0° 20°E20°S

10°S

0°

10°N

20°N

5

0

0.5

0.5

0.51

1

1

1

1.5

1.5

1.5

5

0

1

2

3

4

Temp

erature A

no

maly ( oC

)Tem

peratu

re An

om

aly ( oC)

1

3

2

4

A

B

Fig. 4. Water-hosing simulation results under LGM forcings and boundaryconditions. Changes in (A) surface (averaged between 5 and 35 m), and(B) subsurface (averaged between 300 and 600 m) temperatures between aLGMwater-hosing and control simulation. The temperature changewas com-puted as the difference between: (i) the average temperature of the last 20 yof the 100-y hosing runwhere a freshwater forcing of 0.1 Svwasheld constant,and (ii) the average temperature of the same period of the control simulation.Time evolution of the subsurface temperature change averaged over themaximum subsurface temperature gradient zone indicated by the rectangle(Lower) is shown in Fig. S7. (A) Also shows the site locations (correspondingto encircled numbers) for: (i) VM12-107 in the Bonaire Basin, (ii) the TobagoBasin coreM35003-4 (14), (iii) the western equatorial Atlantic core GeoB3129-3911 (28), and (iv) the eastern equatorial Atlantic core MD03-2707 (36).

Schmidt et al. PNAS ∣ September 4, 2012 ∣ vol. 109 ∣ no. 36 ∣ 14351

EART

H,A

TMOSP

HER

IC,

AND

PLANETARY

SCIENCE

S

the event in the Greenland ice core record (Fig. 3I). Therefore,the transition out of the YD may have started with a reorganiza-tion of the tropical hydrologic cycle. Although this mechanismmost likely explains the subsurface warming associated withH1, the evolution of the subsurface temperature response acrossH1 is not the same as during the YD because of the differingAMOC states bracketing each event. This study provides evi-dence that subsurface temperature changes along the TNAwestern boundary current were a distinctive feature of AMOC-induced ocean circulation changes across the abrupt climateevents of the last deglacial, forming a critical teleconnection be-tween high- and low-latitude climate change.

MethodsTo minimize intraspecific geochemical variations, specimens of G. ruber(white variety) and G. crassaformis were collected from the 250–350-μmand the 425–500-μm size fractions, respectively. Each G. ruber δ18O analysisis based on 20 individuals, and each Mg/Ca measurement was made on atleast 50 shells (run in duplicate). Because there is less seasonal variabilityin the thermocline and fewer deep-dwelling specimens, we used 4–6 indivi-duals of G. crassaformis for each δ18O analysis and 6–10 for trace metal ana-lysis, whenever possible. Samples for stable isotope analysis were sonicatedbefore analysis at Texas A&M University’s Stable Isotope Geosciences Facility.Analytical uncertainty for our reported δ18O measurements is better than0.07‰. Samples for trace metal analysis were first sonicated in rinses of ultra-pure water and methanol, cleaned in hot reducing and oxidizing solutions,and leached in dilute nitric acid. All clean work was conducted under tracemetal clean conditions. Samples were then analyzed using isotope dilution

on a High-Resolution Inductively Coupled PlasmaMass Spectrometer at TexasA&MUniversity. A suite of trace andminor elementmeasurements was madeon each sample, including Ca, Mg, Sr, Na, Ba, U, Al, Mn, and Fe. The pooled SDon replicate Mg/Ca measurements for G. ruber was 2.3% (df ¼ 86 based on103 analyzed intervals), and forG. crassaformis was 3.92% (df ¼ 52 based on 61analyzed intervals). Analyses with either anomalously high (>100 μmol∕mol)Al/Ca, Fe/Ca, or Mn/Ca ratios or low-percent recovery (<20%) were rejected.Analyses with high Al/Ca indicate the presence of detrital clays that werenot removed during the cleaning process. Elevated levels of Fe/Ca or Mn/Caindicate the presence of diagenetic coatings that were not removed duringthe cleaning process. Low-percent recovery values indicate the loss of shell ma-terial during the cleaning process, most likely caused by human error.

The Community Climate System Model, version 3.0, was used for the nu-merical simulations analyzed in this study. The model configuration is thestandard intermediate-resolution (T42x1), consisting of: (i) the CommunityAtmosphere Model, version 3.0, coupled to the Community Land Model, ver-sion 3.0, with a triangular spectral truncation at 42 wavenumbers (roughly2.8° in longitude and latitude); and (ii) the Parallel Ocean Program, version1.4.3, coupled to the Community Sea Ice Model, version 5.0, at 1° horizontalresolution (37).

ACKNOWLEDGMENTS. We thank A. Gondran for assistance with geochemicalanalyses and the Lamont Doherty Earth Observatory core repository forsediment core material. This project was funded by National Science Founda-tion Grant OCE-1102743 (to M.S. and P.C.) and by National Oceanic andAtmospheric Adminstration Grant NA11OAR4310154 (to P.C.). P.C. alsoacknowledges support from the National Science Foundation of China(Grants 41028005, 40921004, and 40930844) and the Chinese Ministry ofEducation’s 111 project (Grant B07036).

1. Zhang R (2007) Anticorrelated multidecadal variations between surface and subsur-face tropical North Atlantic. Geophys Res Lett 34:1–6.

2. Zhang R, Delworth TL, Held IM (2007) Can the Atlantic Ocean drive the observedmulti-decadal variability in Northern Hemisphere mean temperature? Geophys Res Lett34:1–6.

3. Chang P, et al. (2008) Oceanic link between abrupt changes in the North AtlanticOcean and the African monsoon. Nat Geosci 1:444–448.

4. Otto-Bliesner BL, Brady EC (2010) The sensitivity of the climate response to the mag-nitude and location of freshwater forcing: Last glacial maximum experiments.Quat SciRev 29:56–73.

5. Wen CH, Chang P, Saravanan R (2010) Effect of Atlantic meridional overturning circu-lation changes on tropical Atlantic sea surface temperature variability: A 2 1/2-layerreduced-gravity ocean model study. J Clim 23:312–332.

6. Cheng W, Bitz CM, Chiang JCH (2007) Adjustment of the global climate to an abruptslowdown of the Atlantic meridional overturning circulation. Past and Future changesof the Ocean’s Meridional Overturning Circulation: Mechanisms and Impacts, edsA Schmittner, JCH Chiang, and SR Hemming (AGU Monograph, Washington, DC),Vol 173, pp 295–314.

7. Chiang JCH, Cheng W, Bitz CM (2008) Fast teleconnections to the tropical Atlanticsector from Atlantic thermohaline adjustment. Geophys Res Lett 35:1–5.

8. Wan XQ, Chang P, Saravanan R, Zhang R, Schmidt MW (2009) On the interpretation ofCaribbean paleo-temperature reconstructions during the Younger Dryas. Geophys ResLett 36:1–6.

9. Guilderson TP, Fairbanks RG, Rubenstone JL (2001) Tropical Atlantic coral oxygen iso-topes: Glacial-interglacial sea surface temperatures and climate change. Mar Geol172:75–89.

10. Lea DW, Pak DK, Peterson LC, Hughen KA (2003) Synchroneity of tropical high latitudeAtlantic temperatures over the last glacial termination. Science 301:1361–1364.

11. Rühlemann C, Mulitza S, Muller PJ, Wefer G, Zahn R (1999) Warming of the tropicalAtlantic ocean and slowdown of thermohaline circulation during the last deglaciation.Nature 402:511–514.

12. Hüls M, Zahn R (2000) Millennial-scale sea surface temperature variability in thewestern tropical North Atlantic from planktonic foraminiferal census counts. Paleo-ceanography 15:659–678.

13. Schmidt MW, Spero HJ, Lea DW (2004) Links between salinity variation in the Carib-bean and North Atlantic thermohaline circulation. Nature 428:160–163.

14. Rühlemann C, et al. (2004) Intermediate depthwarming in the tropical Atlantic relatedto weakened thermohaline circulation: Combining paleoclimate data and modelingresults for the last deglaciation. Paleoceanography 19:1–10.

15. Came R, et al. (2007) North Atlantic intermediate depth variability during theYounger Dryas: Evidence from benthic foraminiferal Mg/Ca and the GFDL R30coupled climate model. Past and Future Changes in the Ocean’s Overturning Circu-lation: Mechanisms and Impacts on Climate and Ecosystems, eds A Schmittner,JCH Chiang, and SR Hemming (American Geophysical Union, Washington, DC).

16. Stuiver M, Reimer PJ, Reimer R (2011) CALIB Radiocarbon Calibration (Queen’s Univer-sity, Belfast, Ireland), Version 60.

17. Hua Q, et al. (2009) Atmospheric 14C variations derived from tree rings during theearly Younger Dryas. Quat Sci Rev 28:2982–2990.

18. Southon J, Noronha AL, Cheng H, Edwards RL, Wang Y (2012) A high-resolution recordof atmospheric 14C based on Hulu Cave speleothem H82. Quat Sci Rev 33:32–41.

19. Muscheler R, et al. (2008) Tree rings and ice cores reveal C-14 calibration uncertaintiesduring the Younger Dryas. Nat Geosci 1:263–267.

20. Steph S, Regenberg M, Tiedemann R, Mulitza S, Nürnberg D (2009) Stable isotopes ofplanktonic foraminifera from tropical Atlantic/Caribbean core-tops: Implications forreconstructing upper ocean stratification. Mar Micropaleontol 71:1–19.

21. Anand P, Elderfield H, Conte MH (2003) Calibration of Mg/Ca thermometry in plank-tonic foraminifera from a sediment trap time series. Paleoceanography 18:1–15.

22. Cléroux C, Lynch-Stieglitz J, Schmidt MW, Cortijo E, Duplessy JC (2009) Evidence forcalcification depth change of Globorotalia truncatulinoides between deglaciationand Holocene in the Western Atlantic Ocean. Mar Micropaleontol 73:57–61.

23. Cléroux C, Lynch-Stieglitz J (2010) What caused G.truncatulinoides to calcify in shal-lower water during the early Holocene in the western Atlantic/Gulf of Mexico? PAGESFirst Young Scientist Meeting (IOP Conference Series: EES, Philadelphia).

24. Antonov JI, et al. (2006) World ocean atlas 2005. NOAA Atlas NESDIS 6 (USGPO,Washington, DC) p 182.

25. RegenbergM, Steph S, Nürnberg D, Tiedemann R, Garbe-Schönberg D (2009) Calibrat-ing Mg/Ca ratios of multiple planktonic foraminiferal species with delta O-18-calcifi-cation temperatures: Paleothermometry for the upper water column. Earth Planet SciLett 278:324–336.

26. Rasmussen SO, et al. (2006) A new Greenland ice core chronology for the last glacialtermination. J Geophys Res [Atmos] 111:1–16.

27. McManus JF, Francois R, Gherardi J-M, Keigwin LD, Brown-Leger S (2004) Collapse andrapid resumption of Atlantic meridional circulation linked to deglacial climatechanges. Nature 428:834–837.

28. Weldeab S, Schneider RR, Kölling M (2006) Deglacial sea surface temperature and sali-nity increase in the western tropical Atlantic in synchrony with high latitude climateinstabilities. Earth Planet Sci Lett 241:699–706.

29. Stansell ND, et al. (2010) Abrupt Younger Dryas cooling in the northern tropicsrecorded in lake sediments from the Venezuelan Andes. Earth Planet Sci Lett293:154–163.

30. Liu Z, et al. (2009) Transient simulation of last deglaciation with a new mechanism forBolling-Allerod warming. Science 325:310–314.

31. Fratantoni DM, Johns WE, Townsend TL, Hurlburt HE (2000) Low-latitude circulationand mass transport pathways in a model of the tropical Atlantic ocean. J Phys Ocea-nogr 30:1944–1966.

32. Jochum M, Malanotte-Rizzoli P (2001) Influence of the meridional overturning circu-lation on tropical-subtropical pathways. J Phys Oceanogr 31:1313–1323.

33. Hazeleger W, Drijfhout S (2006) Subtropical cells and meridional overturning circula-tion pathways in the tropical Atlantic. J Geophys Res [Oceans] 111:1–13.

34. Kawase M (1987) Establishment of deep ocean circulation driven by deep-water pro-duction. J Phys Oceanogr 17:2297–2317.

35. Mulitza S, Rühlemann C (2000) African monsoonal precipitation modulated by inter-hemispheric temperature gradients. Quat Res 53:270–274.

36. Weldeab S, Lea DW, Schneider RR, Andersen N (2007) 155,000 years of West Africanmonsoon and ocean thermal evolution. Science 316:1303–1307.

37. Collins WD, et al. (2006) The community climate cystem model, version 3 (CCSM3).J Clim 19:2122–2143.

38. Locarnini RA, et al. (2006) World Ocean Atlas 2005, (USGPO, Washington, DC),Temperature, Vol 1, p 182.

14352 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1207806109 Schmidt et al.

Supporting InformationSchmidt et al. 10.1073/pnas.1207806109SI TextSI Methods. Salinity Effect on Shell Mg/Ca in G. ruber.Although bothculturing studies (1–5) as well as core-top and sediment trap stu-dies (6–14) show that temperature is the primary control on Mg/Ca ratios in foraminiferal calcite, recent studies found a relation-ship between Mg/Ca ratios in core-top planktonic foraminiferaand salinity (15–18). By analyzing G. ruber Mg/Ca ratios andδ18O values in core tops across an Atlantic meridional transect,a recent study developed multivariate equations to calculatemean annual surface temperature that take into account the ef-fect of salinity on G. ruber shell Mg/Ca ratios. Their results sug-gest a 27% increase in G. ruber shell Mg/Ca-per-1 salinity unitincrease when seawater salinity was above 35 (18). The multivari-ate temperature equation uses both the measured shell Mg/Caratio and δ18O values to calculate surface temperature:

Temp:ð°CÞ ¼ 16.06þ 4.62 � lnðMg∕CaÞ− 3.42ðδ18OcalciteÞ − 0.1ðΔCO3

2−Þ:

Although it is still not clear why elevated salinity would havesuch a large impact on G. ruber Mg/Ca ratios, we wanted to de-termine how the use of the new multivariate temperature equa-tion would affect our G. ruber temperature record. Because thesalinity effect has not been found in subthermocline-dwellingplanktonic foraminifera, we did not attempt to calculate subsur-face temperatures using the multivariate equation. However, werecalculated the deglacial mixed layer temperature record basedon G. ruber and compared the results with those obtained usingthe Sargasso Sea sediment trap calibration for planktonic for-aminifera (the equation used in this manuscript) (7) (Fig. S5).Application of the ΔCO3

2− correction term calculated usingthe estimated in situ ΔCO3

2− (76 μmol∕kg) for our core site re-sults in a core-top temperature of 18.1 °C, much too cold to beconsidered realistic. Furthermore, given the relatively shallowdepth of VM12-107 (1,079 m) and the good degree of carbonatepreservation in the core, we did not include the dissolution cor-rection term.

The resulting multivariate equation results in a calculatedcore-top temperature of 29 °C for the upper mixed layer atour study site. This is 3 °C warmer than the modern average an-nual upper mixed layer (30 m depth) temperature for the BonaireBasin, and 1.4 °C warmer than the maximum seasonal sea surfacetemperature of 27.6 °C during the fall. For comparison, the cal-culated upper mixed layer temperature using the Sargasso SeaMg/Ca:temperature calibration (7) is 25.7 °C, in excellent agree-ment with the modern annual mean mixed layer temperature of25.6 °C at 30 m depth and with the annual mean sea surface tem-perature of 26.7 °C (19). The coolest temperatures calculatedusing the multivariate equation are 21.3 °C at approximately16 ky and 22.2 °C during the Last Glacial Maximum (LGM)(Fig. S5), suggesting a maximum glacial-interglacial temperaturegradient of almost 7 °C at our site. This amount of cooling in thewestern tropical Atlantic is much larger than the estimated cool-ing of only approximately 3 °C in the western tropical Atlantic atthe LGM based on the results of the MARGO project (20).Although the Bonaire Basin may have cooled more than the aver-age western tropical Atlantic because of the influence of upwel-ling, a cooling of 7 °C at the LGM is too large to be consideredrealistic. Given that neither the calculated core-top temperaturenor the estimated glacial-interglacial temperature gradient seemsreasonable using the multivariate temperature equation, we

chose to use the Sargasso Sea sediment trap calibration equationwhen calculating upper mixed layer temperatures for this study.

Regardless, even if salinity is the dominant influence on ourdeglacial planktonic foraminiferal Mg/Ca ratios, our results arestill consistent with the proposed oceanic gateway mechanismin this manuscript. Atlantic meridional overturning circulation(AMOC) weakening would allow the warm, salty salinity maxi-mum waters (SMW) from the subtropical gyre to flow southward,influencing conditions at our study site. Therefore, it is possiblethe increase inG. crassaformisMg/Ca ratios at the start of the YDreflects a combination of both increased subsurface temperaturesand salinity.

Climate Model and Numerical Simulations. The Community ClimateSystem Model, version 3.0 (CCSM3), is used for the numericalsimulations analyzed in this study. The model configuration isthe standard intermediate-resolution (T42x1), consisting of: (i)the Community Atmosphere Model, version 3.0 (CAM3),coupled to the Community Land Model, version 3.0 (CLM3),with a triangular spectral truncation at 42 wavenumbers (roughly2.8° in longitude and latitude); and (ii) the Parallel OceanProgram, version 1.4.3 (POP), coupled to the Community SeaIce Model, version 5.0 (CSIM5), at 1° horizontal resolution(21). The POP has a nonuniform horizontal grid with the finestresolution at the equator (0.27°), and has 40 vertical levels. In thestudy region, the POP’s meridional resolution is less than 0.5°.

Two control simulations were analyzed. The preindustrial (PI)control simulation was forced with the estimated climate condi-tions of 1870, following the protocols of the Paleoclimate Mod-eling Intercomparison Project, phase 2 (PMIP-2; more informa-tion online at http://pmip2.lsce.ipsl.fr/), and was initiated from along present-day control simulation forced with the estimated cli-mate conditions of 1990 (22). The simulation ran for multicen-turies and the data analyzed were from years 400 to 699 ofthe PI simulation. The LGM control simulation was forced withLGM forcings and boundary conditions, as described by thePMIP-2, and ran for more than 500 y (see ref. 23 for details).The data analyzed were from years 400 to 500. Fig. S4 showsthe mean AMOC streamfunctions averaged between years 400and 420 of the PI and LGM control simulations. Fig. S6 showsthe temperature and salinity along density surface of σ ¼1;026.8 in the PI simulation and of σ ¼ 1;028 in the LGM controlsimulation. Comparing the PI simulation to the observationshown in Fig. 1, it is clear that the model captures the circulationsin the subtropical and tropical North Atlantic reasonably well,reproducing the essential structure of observed maximum sub-surface temperature gradient (STG) zone along the boundarybetween the subtropical and tropical gyres. A similar maximumSTG zone is observed in the LGM control simulation (Fig. S6).

Comparison between two long-term climate simulations forcedby preindustrial conditions (1870) and by LGM forcings andboundary conditions, as described by the PMIP-2, reveals thatthe global mean temperature was colder by approximately 4.5 °Cduring the LGM, with stronger cooling at high latitudes (23, 24).Associated with the colder LGM climate (23, 24), AMOC wasweaker and shallower, whereas the Antarctic BottomWater over-turning cell was stronger. The cooling also resulted in a strongerwind-stress curl over the North Atlantic, producing an intensifiedand southward-shifted, wind-driven North Atlantic subtropicalgyre in the LGM simulation compared to the PI simulation.As a consequence, the North Atlantic subtropical ventilation ratewas enhanced, resulting in a stronger North Atlantic subtropical

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 1 of 10

cell (STC) during the LGM (23) (Fig. S4). Although the changesin the North Atlantic Ocean circulation were substantial andsignificant, the equatorward pathway of the North Atlantic STCremained closed in the LGM control simulation, as evidenced bythe asymmetry in the South and North Atlantic STCs (Fig. S4).Similar to the PI simulation, the North Atlantic STC in the LGMcontrol simulation was nearly invisible in the upper AMOCstreamfunction (Fig. S4), because of the influence of the strongcross-equatorial interhemispheric flow associated with AMOCreturn flow. Accordingly, the maximum STG zone remained in-tact under the LGM condition, separating the SMW from the tro-pical gyre water (Fig. S6), similar to what is observed in thepresent climate (Fig. 1). The southward-shifted, wind-driven gyreand intensified North Atlantic STC, however, caused the maxi-mum STG zone to expand equatorward, resulting in a strongergradient during the LGM than during the PI period. These con-ditions would be in favor of a stronger subsurface warming in re-sponse to a weakening of AMOC, because the stronger gradientgives a stronger convergence of heat at the maximum STG zonefor the same amount of reduction in the western boundary flow.

A set of LGMwater-hosing experiments were conducted at theNational Center for Atmospheric Research, with freshwater-for-cing strength varying between 0.1–1.0 Sv at either the subpolarNorth Atlantic or the Gulf of Mexico (24). All of the hosing runsanalyzed in this study were started at year 400 of the LGM controlsimulation and forced with a constant freshwater input in the sub-polar North Atlantic between 50 °N and 70 °N for 100 y. Fig. S7shows the circulation and subsurface temperature responses tothe freshwater forcing. Even when the freshwater forcing is setat 0.1 Sv, AMOC reduces its strength significantly (by approxi-mately 30%), as indicated by the time series of the maximumstrength of AMOC computed as the maximum steamfunction va-lue for each monthly mean data between 20 °N and 60 °N andbetween the depth range of 500 m and 3000 m (Fig. S7, LowerLeft). More symmetric STCs in the upper tropical Atlantic Oceanbegin to emerge as AMOC settles to a weaker state (Fig. S7,Upper Left). Note that in the LGM control simulation, the equa-torward lower branch of the North Atlantic STC has a steadynorthward (positive) flow of approximately 2.5 Sv (Fig. S7, LowerLeft, lower green line), which is computed as the zonally averagedtransport across the Atlantic basin at 9 °N in the depth range be-tween 100 and 400 m. This flow weakens as AMOC reduces instrength as a part of planetary wave adjustment in response tothe freshwater forcing in the North Atlantic. At approximately60 y into the simulation, when the maximum strength of AMOCdrops below 10 Sv, the North Atlantic STC flow (Fig. S7, LowerLeft, lower red line) changes its sign from positive (northward) tonegative (southward). This indicates that the North Atlantic STCstarts to transport SMW equatorward, suggesting the opening ofthe equatorward gateway of the North Atlantic STC. Meanwhile,the subsurface temperature near the maximum STG zone in-creases steadily as AMOC weakens, and reaches a value of9.5 °C near the end of the 100-y constant freshwater forcingrun from its initial value of 5 °C. The rapid subsurface warmingis caused by the large change in horizontal heat advection nearthe maximum STG zone because of the weakening of the westernboundary current as a result of planetary wave adjustment in re-sponse to the freshwater forcing. As the freshwater forcingstrength increases to 0.25 Sv, AMOC weakens more rapidly(Fig. S7, Lower Left, upper red line), causing a more rapid in-crease in the subsurface temperature (Fig. S7, Lower Right, blueline). It also causes the flow in the equatorward lower branch ofthe North Atlantic STC to reverse its direction earlier (20 y into

the simulation) than in the 0.1 Sv case (60 y into the simulation),suggesting a much earlier opening of the equatorward gateway.Interestingly, the timing of the gateway opening in this case againcorresponds to the point at which AMOC decreases below 10 Sv.Therefore, the modeling results suggest that in the CCSM3, thesubsurface oceanic teleconnection mechanism would operate un-der LGM conditions, provided that AMOC strength was reducedbelow 10 Sv.

Of particular note are the patches of surface warming alongthe western boundary, north of the maximum STG zone (Fig. 4A).These surface warmings are produced by coastal upwelling thatbrings the strong subsurface warming to the surface, counteract-ing the surface cooling induced by atmospheric processes (25).This modeling result is significant because it has important im-plications for the interpretation of proxy records in the region.If a sediment core is located within an upwelling zone, as inthe case of sediment core VM12-107, then surface temperaturechanges from the proxy record during a period of weakenedAMOC are expected to follow changes in the subsurface tem-perature. Previous sea surface temperature reconstructions fromthe eastern tropical South Atlantic showed a warming during theYD (26, 27), consistent with this mechanism. Thus, the LGMwater-hosing simulations support our hypothesis that the oceanicteleconnection mechanism was operating during the abrupt cli-mate events of the last deglacial.

The modeling study presented here represents advancementsover previous modeling studies in several aspects. First, the sub-surface oceanic teleconnection mechanism was identified in a setof present-day climate water-hosing simulations using a differentatmosphere-ocean general circulation model (AOGCM) (28).The modeling results shown here are a unique demonstrationthat this mechanism operates under LGM forcing and boundaryconditions. The fact that two different AOGCMs demonstratethe operation of the teleconnection mechanism under differentclimate forcings and boundary conditions gives further confi-dence that the mechanism is robust and not model-dependent.Second, although subsurface warming in the western tropicalNorth Atlantic during periods of reduced AMOC has been shownin previous AOGCM water-hosing simulations under both mod-ern (29) and LGM climate (30, 31) forcings and boundary con-ditions, the detailed warming mechanism was not studied and nolink was established between the subsurface warming and theoceanic teleconnection mechanism. Finally, the modeling setupused in this study, and the forcings and boundary conditions usedfor the LGM climate, are more advanced than those used in pre-vious studies. The horizontal and vertical resolution of the oceanmodels used in the prior studies (29–31) range from 2°, 3.5°, and2.8° and from 18, 11, and 26 vertical levels, respectively. In con-trast, the model used in this study has a horizontal resolution of 1°in longitude and less than 0.5° in latitude in the study region and avertical resolution of 40 levels with finer resolution in the upperocean. The improved model resolution enabled us to have a de-tailed examination of the warming and teleconnection mechan-isms. The model used in this study also has much improvedphysics compared to the models used in the previous studies(29–31). CCSM3 has a comprehensive sea-ice model that in-cludes both realistic sea-ice dynamics and thermodynamics.The ocean component model includes a set of sophisticated phy-sics parameterizations. Furthermore, the LGM forcings andboundary conditions used in our simulations follow the protocolsof PMIP-2, which has improved estimates of LGM boundaryconditions.

1. Lea DW, Mashiotta TA, Spero HJ (1999) Controls on magnesium and strontium uptake

in planktonic foraminifera determined by live culturing. Geochim Cosmochim Act

63:2369–2379.

2. Nürnberg D, Bijma J, Hemleben C (1996) Assessing the reliability of magnesium in

foraminiferal calcite as a proxy for water mass temperatures. Geochim Cosmochim

Act 60:803–814.

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 2 of 10

3. Russell AD, Hönisch B, Spero HJ, Lea DW (2004) Effects of seawater carbonate ionconcentration and temperature on shell U, Mg, and Sr in cultured planktonic forami-nifera. Geochim Cosmochim Act 68:4347–4361.

4. Mashiotta TA, Lea DW, Spero HJ (1999) Glacial-interglacial changes in Subantarctic seasurface temperature and δ18O-water using foraminiferal Mg. Earth Planet Sci Lett170:417–432.

5. Kisakürek B, Eisenhauer A, Bohm F, Garbe-Schönberg D, Erez J (2008) Controls on shellMg/Ca and Sr/Ca in cultured planktonic foraminiferan, Globigerinoides ruber (white).Earth Planet Sci Lett 273:260–269.

6. Dekens PS, Lea DW, Pak DK, Spero HJ (2002) Core top calibration of Mg/Ca in tropicalforaminifera: Refining paleotemperature estimation. Geochem Geophys Geosys3:1–29.

7. Anand P, Elderfield H, Conte MH (2003) Calibration of Mg/Ca thermometry in plank-tonic foraminifera from a sediment trap time series. Paleoceanography 18:1–15.

8. Lea DW, Pak DK, Spero HJ (2000) Climate impact of Late Quaternary Equatorial Pacificsea surface temperature variations. Science 289:1719–1724.

9. Rosenthal Y, Boyle EA, Slowey N (1997) Temperature control on the incorporation ofmagnesium, strontium, fluorine, and cadmium into benthic foraminiferal shells fromLittle Bahama Bank: Prospects for thermocline paleoceanography. Geochim Cosmo-chim Act 61:3633–3643.

10. Elderfield H, Ganssen G (2000) Past temperature and δ18O of surface ocean watersinferred from foraminiferal Mg/Ca ratios. Nature 405:442–445.

11. Hastings DW, Russell AD, Emerson SR (1998) Foraminiferal magnesium in Globigeri-noides sacculifer as a paleotemperature proxy. Paleoceanography 13:161–169.

12. Rosenthal Y, Lohmann GP, Lohmann KC, Sherrell RM (2000) Incorporation and preser-vation of Mg in Globigerinoides sacculifer implications for reconstructing the tem-perature and 18O∕16O of seawater. Paleoceanography 15:135–145.

13. McKenna VS, Prell WL (2004) Calibration of the Mg/Ca of Globorotalia truncatuli-noides (R) for the reconstruction of marine temperature gradients. Paleoceanography19:1–12.

14. McConnell MC, Thunell RC (2005) Calibration of the planktonic foraminiferal Mg/Capaleothermometer: Sediment trap results from the Guaymas Basin, Gulf of California.Paleoceanography 20:1–18.

15. Ferguson JE, Henderson GM, Kucera M, Rickaby REM (2008) Systematic change offoraminiferal Mg/Ca ratios across a strong salinity gradient. Earth Planet Sci Lett265:153–166.

16. Mathien-Blard E, Bassinot F (2009) Salinity bias on the foraminifera Mg/Ca thermome-try: Correction procedure and implications for past ocean hydrographic reconstruc-tions. Geochem Geophys Geosys 10:1–17.

17. Arbuszewski JA, deMenocal PB, Kaplan A, Klinkhammer G, Ungerer A (2009) Towardsa global calibration of the G. ruber (white) Mg/Ca paleothermometer. Geochim Cos-mochim Act 73:A49.

18. Arbuszewski J, deMenocal P, Kaplan A, Farmer E (2010) On the fidelity of shell-derivedδ18Oseawater estimates. Earth Planet Sci Lett 300:185–196.

19. Antonov JI, et al., eds (2010)World Ocean Atlas 2009 (USGPO, Washington, DC), Vol 2(Salinity), p 184.

20. Waelbroeck C, et al. (2009) Constraints on the magnitude and patterns of ocean cool-ing at the Last Glacial Maximum. Nat Geosci 2:127–132.

21. Collins WD, et al. (2006) The Community Climate System Model version 3 (CCSM3).J Clim 19:2122–2143.

22. Otto-Bliesner BL, et al. (2006) Climate sensitivity of moderate- and low-resolutionversions of CCSM3 to preindustrial forcings. J Clim 19:2567–2583.

23. Otto-Bliesner BL, et al. (2006) Last Glacial Maximum and Holocene climate in CCSM3.J Clim 19:2526–2544.

24. Otto-Bliesner BL, Brady EC (2010) The sensitivity of the climate response to the mag-nitude and location of freshwater forcing: Last glacial maximum experiments.Quat SciRev 29:56–73.

25. Wan XQ, Chang P, Saravanan R, Zhang R, Schmidt MW (2009) On the interpretation ofCaribbean paleo-temperature reconstructions during the Younger Dryas. Geophys ResLett 36:1–6.

26. Mulitza S, Rühlemann C (2000) African monsoonal precipitation modulated by inter-hemispheric temperature gradients. Quat Res 53:270–274.

27. Weldeab S, Lea DW, Schneider RR, Andersen N (2007) 155,000 years of West Africanmonsoon and ocean thermal evolution. Science 316:1303–1307.

28. Chang P, et al. (2008) Oceanic link between abrupt changes in the North AtlanticOcean and the African monsoon. Nat Geosci 1:444–448.

29. Came R, et al. (2007) North Atlantic intermediate depth variability during the YoungerDryas: Evidence from benthic foraminiferal Mg/Ca and the GFDL R30 coupled climatemodel. Past and Future Changes in the Ocean’s Overturning Circulation: Mechanismsand Impacts on Climate and Ecosystems, eds Schmittner A, Chiang JCH, Hemming SR(American Geophysical Union, Washington DC).

30. Rühlemann C, et al. (2004) Intermediate depthwarming in the tropical Atlantic relatedto weakened thermohaline circulation: Combining paleoclimate data and modelingresults for the last deglaciation. Paleoceanography 19:1–10.

31. Cheng W, Bitz CM, Chiang JCH (2007) Adjustment of the global climate to an abruptslowdown of the Atlantic Meridional Overturning Circulation. Past and Futurechanges of the Ocean’s Meridional Overturning Circulation: Mechanisms and Impacts,eds Schmittner A, Chiang JCH, Hemming SR (AGU Monograph, Washington, DC),Vol 173, pp 295–314.

Fig. S1. Bathymetric map showing the location of VM12-107 in the southeastern Bonaire Basin. The Bonaire Basin is connected to the Caribbean throughseveral channels, with the deepest sill at approximately 1,500 m. Therefore, intermediate-water changes in the western boundary current will affect hydro-graphic conditions within the basin. Also note the location of the nearby Cariaco Basin to the southeast.

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 3 of 10

Fig. S2. The age model for core VM12-107 (blue line) is based on 10 calibrated radiocarbon dates (using the standard 400-y reservoir age correction) onplanktonic foraminifera (mixed Globigerinoides sacculifer and Globigerinoides ruber specimens) spanning the upper 270 cm of the core. Error bars representthe two-sigma calibration uncertainty. The resulting age model indicates nearly linear sedimentation rates and an average deglacial sedimentation rate ofapproximately 18 cm∕ky. Because uncertainty in reservoir age changes at the beginning of the Younger Dryas (YD) between 13.0 and 12.53 ky (1) makes itdifficult to calibrate the radiocarbon date at 130 cm (calibrated age of 12.67 ky, red diamond), an alternate age model was constructed by omitting thisdatapoint (black line). This resulted in shifting the abrupt subsurface warming at the beginning of the YD in the Globorotalia crassaformis record (shownin Fig. 2) from 13.01 (−0.15; þ0.24 ky) to 13.12 ky (−0.29; þ0.40 ky), still within age model error of the start of the YD in the North Greenland Ice Core Project(NGRIP) ice core record at 12.896 ky (�0.138 ky) (2).1 Hua Q, et al. (2009) Atmospheric 14C variations derived from tree rings during the early Younger Dryas. Quat Sci Rev 28:2982–2990.2 Rasmussen SO, et al. (2006) A new Greenland ice core chronology for the last glacial termination. J Geophys Res [Atmos] 111:1–16.

Fig. S3. The G. crassaformis Mg/Ca record (blue diamonds) plotted on the alternate age model (developed by omitting the radiocarbon age control point atthe beginning of the YD) and the NGRIP oxygen isotope record (black line) (1). The abrupt subsurface warming at the beginning of the YD in theG. crassaformisrecord (blue diamond outlined in red) occurs at 13.12 ky (−0.29; þ0.40 ky), within age model error of the start of the YD in the NGRIP ice core record at12.896 ky (�0.138 ky) (1). Purple diamonds on x axis are the calibrated age control points with their associated two-sigma error.1 Rasmussen SO, et al. (2006) A new Greenland ice core chronology for the last glacial termination. J Geophys Res [Atmos] 111:1–16.

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 4 of 10

Fig. S4. Comparison of calculated G. ruber temperatures across the deglacial using the new multivariate equation (18) (blue line with diamonds) and thecalculated temperatures using the Sargasso Sea Mg/Ca–temperature relationship (7) used in this study (red line with circles). Although the timing of tem-perature changes remains the same, note the unrealistically large glacial-interglacial temperature gradient of almost 7 °C using the multivariate equation.Red and blue stars on the y axis are the modern mean annual temperatures at 30 m depth and at the sea surface at the core site, respectively.

Fig. S5. Comparison of the timing of the VM12-107 subthermocline warming events during Heinrich Event 1 (H1) and YD (using the original age modelincluding all 10 calibrated radiocarbon dates) and absolute-dated Hulu Cave speleothem oxygen isotope records (1) across the deglacial. The age modelfor VM12-107 is based on calibrated radiocarbon dates (black triangles on x axis) and the age model for the Hulu Cave records is based on U-Th dates(1). Close agreement between the timing of the two deglacial subsurface warming events (gray bars) and the Hulu Cave oxygen isotope record acrossthe deglacial provides additional confidence in the radiocarbon-based age model developed for VM12-107.1 Wang YJ, et al. (2001) A high-resolution absolute-dated Late Pleistocene monsoon record from Hulu Cave, China. Science 294:2345–2348.

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 5 of 10

Fig. S6. Mean AMOC streamfunction simulated by the PI and LGM control runs. The averages were computed from years 400 to 420 of the simulations. (Right)Shows the AMOC streamfunctions over the full depth of the ocean in the Atlantic basin; (Left) shows the streamfuctions in the upper 500 m of the tropicalAtlantic Ocean, indicated by the rectangles (Right). The contour interval is 2.0 Sv. Positive values indicate a clockwise circulation.

34.6

34.8

34.8

35

35

35.2

35.435.635.8

36

36.2

35.6

35.6

35.8

35.8

36

36.236.436.6

36.8

37

37

37.2

4 6 8 10 12 14 16

Temperature (oC)

Fig. S7. Temperature (color) and salinity (contour) along density surface σ ¼ 1;026.8 (Left) of the PI simulation and along density surface σ ¼ 1;028.0 (Right) ofthe LGM control simulation. Both of these isopycnal surfaces are in the same depth range, varying from a depth of approximately 150 m near the equator to adepth of approximately 550 m in subtropics. Note the presence of the subsurface temperature gradient at nearly the same location under both climate bound-ary conditions.

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 6 of 10

Fig. S8. AMOC streamfunction of the 0.1-Sv LGM hosing run in the full depth of the ocean in the Atlantic basin (Upper Right) and in the upper 500 m of thetropical Atlantic Ocean (Upper Left) indicated by the rectangle (Upper Right). (Lower Right) Shows the subsurface temperature change near the maximum STGzone averaged over the rectangle indicated in Fig. 4 (Lower) and a depth range between 300 m and 600 m. (Lower Left) Shows the maximum strength of theAMOC (upper three curves) and the flow of the lower branch of the North Atlantic STC (lower three curves) in the depth range between 100 m and 400 mindicated by the vertical line (Upper Left). The green, red, and blue lines (Lower) show the results from the LGM control, 0.1-Sv, and 0.25-Sv hosing runs,respectively.

Table S1. VM12-107 G. ruber stable oxygen isotope and Mg/Ca data

Depth(cm)

δ18O(‰ VPDB)

Mg/Ca(mmol∕mol)

Count of tracemetal analyses SD

0.5 −1.97 3.830 3 0.0672.5 3.730 14.5 3.687 16.5 3.526 2 0.0398.5 3.793 2 0.1079.5 −2.10 3.650 2 0.01214.5 3.753 2 0.00316.5 3.678 2 0.00218.5 3.783 2 0.03321.5 −1.86 3.904 2 0.03422.5 3.810 2 0.12326.5 3.561 2 0.10328.5 3.757 2 0.08331.5 −1.86 3.514 2 0.07832.5 3.469 2 0.03334.5 3.456 138.5 3.828 2 0.05341.5 −1.82 3.544 2 0.06342.5 3.670 2 0.05444.5 3.247 2 0.032

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 7 of 10

Depth(cm)

δ18O(‰ VPDB)

Mg/Ca(mmol∕mol)

Count of tracemetal analyses SD

46.5 3.320 2 0.00848.5 3.400 2 0.00351.5 −1.7352.5 3.444 2 0.04156.5 3.485 158.5 3.344 160.5 −1.70 3.562 2 0.00762.5 3.704 2 0.04864.5 3.640 2 0.00968.5 3.655 2 0.03570.5 −1.24 3.617 2 0.01172.5 3.816 2 0.13078.5 3.391 2 0.00180.5 −0.9082.5 −1.01 3.383 2 0.07784.5 −0.86 3.329 2 0.03886.5 −0.78 3.466 2 0.08088.5 −1.30 3.376 2 0.06790.5 −1.16 3.100 2 0.02692.5 −0.97 3.294 2 0.03294.5 −0.62 3.336 2 0.06096.5 −1.2198.5 −0.54 3.260 2 0.088101.5 −0.63 2.999 2 0.100102.5 −0.15 3.221 3 0.044104.5 −0.30 2.934 3 0.050106.5 −0.75 3.129 3 0.045108.5 −0.32 2.990 3 0.035110.5 −0.53112.5 −0.70 3.207 2 0.126114.5 −0.60 3.002 2 0.118116.5 −0.69118.5 −0.37120.5 −0.43 3.069 2 0.062122.5 −0.75124.5 −0.71 3.722 2 0.060126.5 −0.61 3.717 2 0.060128.5 −0.54 3.435 2 0.055130.5 −0.38 3.568 2 0.081132.5 −0.29 3.242 2 0.052134.5 −0.55 3.417 2 0.055136.5 −0.45 3.541 2 0.088138.5 −0.62140.5 −0.49 3.295 2 0.057142.5 −0.43 3.333 2 0.084144.5 −0.40 3.415 2 0.055146.5 −0.16 3.150 1148.5 0.08 2.910 1150 3.101 1151.5 −0.12 3.070 1152.5 0.24 3.165 1154.5 0.23 2.845 2 0.067156.5 −0.25 2.913 1158.5 −0.17160.5 0.01 3.191 2 0.072162.5 0.01 3.124 2 0.080164.5 −0.04 3.088 2 0.121166.5 0.08 3.147 1168.5 0.19 2.983 1170.5 −0.07172.5 0.38174.5 0.54 2.748 2 0.036176.5 0.43 2.949 1178.5 0.26180.5 0.13 2.910 2 0.095182.5 0.57 2.571 2 0.013184.5 0.58 2.446 2 0.067186.5 0.60 2.609 2 0.078188.5 0.54 2.782 2 0.066190.5 0.44 2.804 2 0.124192.5 0.54 2.658 2 0.040

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 8 of 10

Depth(cm)

δ18O(‰ VPDB)

Mg/Ca(mmol∕mol)

Count of tracemetal analyses SD

194.5 0.59 2.699 2 0.029196.5 0.38 2.740 2 0.042198.5 0.06200.5 0.30 2.593 2 0.021202.5 0.84 2.690 1204.5 0.55 2.745 1206.5 0.45 2.532 2 0.086208.5 0.49 2.575 2 0.001210.5 0.38212.5 0.35 2.590 2 0.060214.5 0.19 2.429 2 0.023216.5 0.31220.5 0.39222.5 0.61224.5 0.29226.5 0.70228.5 0.72230.5 0.37 2.404 2 0.024232.5 0.56234.5 0.51236.5 0.49238.5 0.72240.5 0.48 2.527 2 0.053242.5 0.84248.5 0.75251.5 0.55 2.776 2 0.141260.5 0.48

VPDB, Vienna Pee Dee belemnite.

Table S2. VM12-107 G. crassaformis stable oxygen isotope andMg/Ca data

Depth(cm)

δ18O(‰ VPDB)

Mg/Ca(mmol∕mol)

Count of tracemetal analyses

SD(Mg/Ca)

0.5 1.809.5 1.7621.5 1.4931.5 1.7242.544.5 1.94 0.830 148.5 1.0952.5 0.891 160.5 1.61 0.942 162.5 1.126 164.5 1.7366.5 1.42 1.096 2 0.04268.5 1.3170.5 1.2872.5 1.5076.5 1.5580.5 1.86 1.238 182.5 1.024 2 0.05484.5 1.77 1.153 3 0.05186.5 1.246 4 0.05888.5 2.0490.5 1.053 4 0.05792.5 2.05 1.096 2 0.02894.5 2.21 1.270 196.5 2.09 1.272 3 0.03198.5 1.217 2 0.00998.5 2.05101.5 1.93102.5 1.85104.5 1.69106.5 2.26 1.294 1108.5 1.374 4 0.052108.5 1.99110.5 2.02112.5 2.22 1.297 2 0.059114.5 2.15

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 9 of 10

Depth(cm)

δ18O(‰ VPDB)

Mg/Ca(mmol∕mol)

Count of tracemetal analyses

SD(Mg/Ca)

116.5 2.35 1.417 2 0.005118.5 1.96 1.377 1120.5 2.14 1.396 2 0.064122.5 2.04124.5 1.28 1.584 1126.5 1.544 2 0.104128.5 1.71130.5 1.80 1.581 2 0.092132.5 2.06 1.466 2 0.019134.5 2.21 1.439 2 0.087136.5 1.59 1.632 2 0.058138.5 2.09 1.282 1140.5 1.88 1.307 2 0.013142.5 2.28 1.384 2 0.045144.5 2.51148.5 2.39 1.254 1151.5 2.33 1.358 2 0.032152.5 1.88154.5 2.13 1.383 1158.5 2.75160.5 2.04 1.500 2 0.035164.5 2.16 1.453 1166.5 1.460 2 0.028166.5 2.39168.5 1.428 1170.5 1.419 1172.5 2.43 1.429 1178.5 2.67 1.440 2 0.114180.5 2.39 1.387 2 0.033182.5 2.43 1.428 1184.5 1.304 1188.5 2.50 1.243 2 0.011192.5 2.58196.5 2.54 1.307 1198.5 2.28200.5 2.88 1.229 1204.5 2.78206.5 1.286 2 0.087208.5 3.01 1.256 2 0.000210.5 3.09 1.187 2 0.001212.5 3.02214.5 3.11 1.257 2 0.038216.5 2.67218.5 2.98 1.102 2 0.023220.5 1.152 2 0.013220.5 3.28222.5 2.66 1.069 1224.5 2.75226.5 2.67 1.160 2 0.041228.5 2.84 1.105 2 0.069230.5 2.60232.5 2.72234.5 1.093 4 0.043236.5 2.71238.5 2.98240.5 2.77 1.205 4 0.044243.5 1.217 2 0.038246.5 2.98248.5 2.66 1.226 2 0.035251.5 2.77252.5 3.12 1.277 2 0.001254.5 2.59 1.115 2 0.006256.5 2.69258.5 2.93260.5 3.16 1.229 2 0.056262.5 2.93264.5 2.92266.5 2.99268.5 2.81270.5 2.57

VPDB, Vienna Pee Dee belemnite.

Schmidt et al. www.pnas.org/cgi/doi/10.1073/pnas.1207806109 10 of 10