An ocean tale of two climates: Modern and LastGlacial MaximumRaffaele Ferrari ∗, Malte Jansen †, Jess F. Adkins ‡, Andrea Burke ‡ , Andrew L. Stewart ‡ , and Andrew F. Thompson ‡

∗Massachusetts Institute of Technology, Cambridge, MA,†Geophysical Fluid Dynamics Laboratory, Princeton, NJ, and ‡California Institute of Technology, Pasadena,CA

Submitted to Proceedings of the National Academy of Sciences of the United States of America

In the modern climate, the ocean below 2 km is mainly filled bywaters sinking into the abyss around Antarctica and in the NorthAtlantic. Paleo proxies indicate that waters of North Atlantic ori-gin were instead absent below 2 km at the Last Glacial Maxi-mum (LGM), resulting in an expansion of the volume occupiedby Antarctic origin waters. In this study we show that this re-arrangement of deep water masses is dynamically linked to theexpansion of summer sea ice around Antarctica. A simple theoryfurther suggests that these deep waters only came to the surfaceunder sea ice, which insulated them from atmospheric forcing,and were weakly mixed with overlying waters, thus being able tostore carbon for long times. This unappreciated link between theexpansion of sea ice and the appearance of a voluminous and in-sulated water mass may help quantify the ocean’s role in regulat-ing atmospheric carbon dioxide on glacial-interglacial timescales.Previous studies pointed to many independent changes in oceanphysics to account for the observed swings in atmospheric car-bon dioxide. Here it is shown that many of these changes aredynamically linked and therefore must co-occur.

Last Glacial Maximum | Southern Ocean | Sea ice | Carbon | Oceancirculation

Abbreviations: LGM, Last Glacial Maximum; SO, Southern Ocean; AABW,Antarctic Bottom Water; NADW, North Atlantic Deep Water; IDW, indian DeepWater; PDW, Pacific Deep Water

Significance StatementThe ocean’s role in regulating atmospheric carbon dioxide onglacial-interglacial timescales remains an unresolved issue inpaleoclimatology. Many apparently independent changes inocean physics, chemistry and biology need to be invoked toexplain the full signal. New understanding of the deep oceancirculation and stratification is used to demonstrate that the ma-jor changes invoked in ocean physics are dynamically linked.In particular the expansion of permanent sea ice in the South-ern Hemisphere results in a volume increase of Antarctic-originabyssal waters and a reduction in mixing between abyssal wa-ters of Arctic and Antarctic origin.

The Last Glacial Maximum (LGM) spanning the period betweenabout 25,000 and 20,000 years ago was characterized by global

atmospheric temperatures 3-6 degrees colder than today [1], atmo-spheric carbon dioxide (CO2) concentrations 80-90 ppm lower thanpreindustrial values [2], and extended ice sheets and sea ice [3, 4].Geochemical tracers suggest that the ocean volume filled by watersof Antarctic origin was nearly quadrupled at the LGM [5]. In thisstudy we apply new understanding of deep ocean dynamics to explainthe connection between the observed changes in the atmosphere, thecryosphere and the deep ocean water masses. This connection pro-vides a robust framework to quantify the drop in atmospheric CO2

concentrations, a critical gap in our understanding of glacial climates.Much attention has been paid to the inferred changes in the deep

ocean at the LGM, because they are believed to have played a keyrole in reducing atmospheric CO2 concentrations [6–8]. The deepocean contains 90% of the combined oceanic, atmospheric and ter-

-6000

-5000

-4000

-3000

-2000

-1000

0

)m( htpe

D

4.0

.0 4

.0 8

8.08.0

8.0

2.1

-6000

-5000

-4000

-3000

-2000

-1000

0

)m( htpe

D

-60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70

Latitude

8.0-

0- 4.

0

0

4.0

8.02.1

6.1

Fig. 1. Contour maps of δ13C for a section in the Western Atlantic [5] basedon water samples for the modern climate (top) and benthic foraminiferal stableisotope data using a variety of Cibicidoides spp for the LGM (bottom). The glacialreconstruction documents the shoaling of North Atlantic Deep Water to about1500 m, the expansion and northward penetration of Southern Ocean DeepWater/Antarctic Bottom Water.

restrial carbon and a rearrangement of deep water masses could havea large impact on the atmospheric carbon budget. Fig.1 shows the dis-tributions of δ13C of benthic foraminifera in the modern and glacialclimates along a north-south section from the western Atlantic basedon the data of [5]. In the modern section one can see the high δ13Ctongue of Arctic origin flowing over a low δ13C tongue of Antarcticorigin below 4 km. At the LGM, the Antarctic source water appearsto fill the whole ocean volume below 2 km squeezing a much reducedcomponent of Arctic source waters above 2 km.

The expansion of Antarctic-origin abyssal waters, richer in nu-trients and metabolic carbon than the deep Arctic-origin watersit replaced, is believed to have reduced atmospheric CO2 by 10-

Reserved for Publication Footnotes

www.pnas.org/cgi/doi/10.1073/pnas.????????? PNAS Issue Date Volume Issue Number 1–8

20 ppm [9]. Concomitant changes in the ocean circulation, strati-fication, biological nutrient uptake and carbonate compensation havebeen invoked to account for the full 80-90 ppm drawdown [10]. Thephysical feedbacks that have been identified are (I) a change in theocean circulation that isolates abyssal waters from the surface, possi-bly associated with an equatorward shift of the Southern Hemispherewesterlies [11–13], (II) an increase in abyssal stratification acting asa lid to deep carbon [14], (III) an expansion of sea ice acting to re-duce the CO2 outgassing over the Southern Ocean [15], and (IV)a reduction in the mixing between waters of Antarctic and Arcticorigin which is a major leak of abyssal carbon in the modern cli-mate [16]. Current understanding is that some combination of all ofthese feedbacks, together with a reorganization of the soft-tissue andcarbonate pumps, is required to explain the observed glacial drop inatmospheric CO2 [17]. Here we apply new theories of the deep oceancirculation to illustrate that there is a direct dynamical link betweenthe drop in temperature, the expansion of sea ice around Antarctica,the rearrangement of the ocean deep water masses and the change incirculation at the LGM. Hence these are not independent feedbacks,rather they are expected to co-occur in each glacial cycle and may ex-plain why the atmospheric CO2 concentrations dropped by the sameamount for at least the last four glacial cycles.

The modern deep ocean stratification and circulationMunk [18], in one of the most influential papers on the subject, ar-gued that the dense ocean waters that sink through convection into theabyss at high latitudes return back to the surface as they are mixed byturbulence with lighter overlying waters. In this conceptual modelhigh latitude convection sets the rate of the overturning circulation,while mixing determines the deep ocean stratification through a bal-ance between upwelling of dense waters and downward diffusion oflight waters. This view ignored the pivotal role of the Southern Ocean(SO) in controlling both the stratification and the overturning of deepwater masses [19–21]. In this section we review the essential ele-ments of the new paradigm that has emerged over the last twentyyears. In the next section we discuss its implications for interpretingthe changes observed in the ocean at the LGM.

Fig.2 shows the zonally averaged distribution of density1 in theglobal ocean based on the World Ocean Circulation Experiment(WOCE) hydrographic atlas [23]. Density surfaces, herein isopyc-nals, are approximately flat between 50◦S and 50◦N, except for thetwo bowls of light waters in the upper 500 m, on the two sides of theequator, which are the result of wind-driven circulations [24]. Southof 50◦S, in the latitude band of the SO, and north of 50◦N in theNorth Atlantic, the isopycnals develop a substantial slope. Keepingin mind that the large scale oceanographic flows are directed primar-ily along isopycnals, it follows that the SO and the North Atlanticare the primary conduits through which the deep ocean exchangesproperties with the atmosphere.

The Ocean Stratification. The slope of isopycnals in the SO is set bya balance between winds and geostrophic eddies [25]. The SouthernHemisphere westerlies drive a clockwise overturning circulation with

Latitude

Dep

th [m

]

−80 −60 −40 −20 0 20 40 60 80−4000

−3000

−2000

−1000

0

Neu

tral

den

sity

[kg/

m3 ]

25262727.727.827.92828.128.228.3

Fig. 2. Zonally averaged neutral density surfaces based on the WOCE hydro-graphic atlas [23]. The vertical line marks the 45◦S latitude, the northernmostlatitude reached by the Antractic Circumpolar Current. The dashed line is thetheoretical prediction based on equation (1) for the shape of the density surface27.9 kg m−3 outcropping at the summer sea ice edge in the Southern Hemi-sphere.

water moving equatorward at the surface. This flow acts to steepenisopycnals, but instabilities develop that generate a counter-clockwiseoverturning circulation that slumps isopycnals back to the horizontal.An equilibrium is achieved between the wind and eddy driven over-turning circulations when the isopycnal slope s is equal to,

s =τ

ρ0fK, [1]

where τ is the local surface wind stress, ρ0 is a reference surface den-sity (seawater is weakly compressible and the density is everywherewithin a few percent of 1027 kg m−3), f is the Coriolis frequencyequal to twice the Earth’s rotation rate multiplied by the sine of thelatitude, and K is an eddy diffusivity which quantifies the efficiencyof instabilities at slumping isopycnals. This formula holds in a zon-ally averaged sense or, more precisely, for an average along the pathof the Antarctic Circumpolar Current (ACC) that flows around thewhole globe in the SO.

To illustrate the power of the scaling law (1), we plot the pre-diction for the slope of isopycnal 27.9 kg m−3, which outcrops inthe surface mixed layer at 65◦S, for a characteristic wind stress of0.1 N m−2 and eddy diffusivity K =1000 m2 s−1 [25]. The blackdashed line in Fig.2 shows that with these values the scaling (1) cor-rectly predicts the slope of isopycnals shallower than 3000 m, i.e.above major topographic ridges and seamounts which modify some-what the slope of deeper isopycnals. In the upper few hundred me-ters, the isopycnals become flat at latitudes where winter ice meltsand creates a shallow layer of fresh water. This is a transient summerphenomenon. Our scaling applies to isopycnals below this layer.

Given the surface distribution of density, the scaling for the slopedetermines the zonally averaged distribution of density below 1 km.The argument is purely geometrical and it is illustrated by the dashedblack line in Fig.2. The isopycnals slope downward from the latitudewhere they intersect the surface (more appropriately the base of thesurface mixed layer) to approximately 45◦S, the northernmost lati-tude reached by the ACC where equation (1) holds. North of 45◦S,the isopycnals are essentially flat, because the presence of lateralboundaries does not permit strong zonal flows which would resultin tilted isopycnals. Density surfaces lighter than 28 kg m−3 cometo the surface also at high latitudes in the North Atlantic with a verysteep slope in response to upright convection driven by strong cool-ing.

These simple scaling arguments provide sufficient informationto rationalize the most conspicuous changes observed in the oceanat the LGM. However they are not a self-contained theory of the

Longitude

Latitu

de

mm2 s!3

0 50 100 150 200 250 300 350

!75

!70

!65

!60

!55

!50

!45

!40

!0.02

!0.01

0

0.01

0.02

Fig. 3. Annual mean buoyancy flux from a state estimate that combines 3 yearsof available observations with an ocean model [26]. The black line denotes the70% quantiles of annual mean sea ice concentration, essentially the area of theocean covered by ice 70% of the time. The change in sign of the buoyancy fluxjust north of the Antarctic continent is roughly co-located with the 70% quantileof sea ice coverage.

1Herein by density we mean neutral density, that is the density corrected to eliminate dynam-ically irrelevant compressive effects that increase the water density at depth [22]. Values arereported in kg m−3 subtracting 1000 kg m−3, so a neutral density of 28 kg m−3 is actually1028 kg m−3 .2Buoyancy is defined as b = −g(ρ − ρ0)/ρ0 , i.e. the departure of density ρ from areference value ρ0=1027 kg m−3 and multiplied by the gravitational acceleration. The buoy-ancy flux is the flux of density across the air-sea interface generated by heating/cooling andevaporation/precipitation, except for a change in sign. A positive buoyancy flux means thatthe ocean is becoming lighter because of warming, precipitation or ice melting. A negativebuoyancy flux represents cooling, evaporation, or brine rejection by ice freezing.

2 www.pnas.org/cgi/doi/10.1073/pnas.????????? Footline Author

ocean stratification, because they require knowledge of the surfacedensity distribution and the maximum latitude reached by the ACC.Also they ignore important departures from zonality. The ACC itselfextends further north in the Pacific sector than in the Atlantic one andas a result the same isopycnal tends to be nearly 500 m deeper in thePacific Ocean than in the Atlantic Ocean.

The Southern Ocean Overturning Circulation. The density distri-bution implied by the scaling (1) and shown in Fig.2 can be used todiagnose the zonally averaged overturning circulation in the SO. Webegin from the meridional flow at the ocean surface. Fig.3 shows theannual averaged air-sea buoyancy flux2 based on an ocean state esti-mate that combines available observations [26]. The buoyancy fluxis negative around coastal Antarctica where the relatively warm sub-surface waters that upwell in the SO are cooled to the freezing pointand become saltier through brine rejection as new ice is formed. Thebuoyancy flux is positive further north where sea ice is only seasonaland the ocean is warmed in summer by the atmosphere and fresh-ened by ice melting. The sign of the yearly average buoyancy fluxconstrains the direction of the mean surface meridional flow. Watersmust become denser to move poleward where the surface density ishigher. A surface density gain is achieved when waters are exposedto a negative buoyancy flux, while a surface density loss demands apositive buoyancy flux. Thus the surface meridional flow in a steady

80S

l1

0km

2km

4km

0km

2km

4km

50S 80N

Last Glacial Maximuml2

ICE

ICE

Modern

Fig. 4. (Top) Schematic of the overturning circulation for the modern climate.The ribbons represent a zonally averaged view of the circulation of the majorwater masses; blue is AABW, green is NADW, red are IDW and PDW, and or-ange are Antarctic Intermediate Waters. The dashed vertical lines representsmixing-driven upwelling of AABW into NADW and IDW/PDW respectively. Thereis also some mixing between NADW and IDW/PDW in the Southern Ocean. Thedashed black line represents the isopycnal that separates the upper and loweroverturning branches present in the Southern Ocean. `1 is the distance betweenthe northernmost latitude reached by the ACC, indicated by a solid grey line, andthe quasi permanent sea ice line. The grey lines are the crest of the main bathy-metric features of the Pacific and Indian ocean basins: mixing is enhanced belowthese line. (Bottom) Schematic of the overturning circulation for the LGM. Theextent of the quasi permanent sea ice line has shifted equatorward compared tomodern climate (`2 < `1). Mixing-driven upwelling of abyssal waters is confinedbelow 2 km and it cannot lift waters high enough to upwell north of the ice line.As a result the abyssal overturning circulation closes on itself, leaving above asmall overturning cell of North Atlantic waters.

state must be directed poleward where the buoyancy flux is negativeand equatorward where the buoyancy flux is positive. This is a keyinsight for the rest of our argument.

The meridional surface flow, together with the realization thatwater masses flow along isopycnals below the surface [25], can beused to diagnose the two branches of the SO overturning. Today theisopycnal 27.9 kg m−3, marked as a black dashed line in Fig.2, in-tersects the surface approximately where the air-sea buoyancy fluxchanges sign. An upward isopycnal flow must develop to feed thedivergent surface flow: surface waters south of the isopycnal experi-ence a negative buoyancy flux and thus flow poleward, while surfacewaters north of the isopycnal flow equatorward in response to the pos-itive buoyancy flux. The resulting circulation pattern is composed oftwo overturning branches. The lower branch comes to the surfaceparallel, but below, the dividing isopycnal, flows poleward along thesurface and then sinks into the abyss around Antarctica completingthe lower branch (along the Antarctic continental margin brine re-jection makes the waters so dense that they plunge into the abyssalong boundary overflows.) The upper branch, instead, comes to thesurface above the dividing isopycnal, flows equatorward at the sur-face and then sinks along isopycnals as intermediate waters, belowthe upper ocean thermoclines. The two branches of the overturningcirculations are sketched in Fig.4 and are identified in ocean observa-tions [27–29]. This diagnostic argument suggests that the patterns ofthe overturning circulation can be reconstructed from knowledge ofthe line where the air-sea buoyancy flux changes sign in the SO.

This simple diagnostic model can be used to interpret the changesin deep water masses at the LGM, if we find a proxy for the surfacebuoyancy flux. The negative buoyancy flux around Antarctica ariseunder permanent sea ice where the heat fluxes are weak and salinityfluxes are strong due to brine rejection. The solid white line in Fig.3confirms that the transition between negative and positive buoyancyflux closely coincides with the extent of the quasi-permanent sea iceline, here defined as the ocean area covered by sea ice 70% of theyear. This suggests that the area of negative buoyancy flux scales withthe extent of the quasi-permanent sea ice. Hence we surmise that theseparation between the upper and lower overturning branches shiftswith the expansion and contraction of sea ice in different climates.This is the key hypothesis pursued below. But first we must describehow the two SO overturning branches complete their journey throughthe three ocean basins north of the SO.

The Atlantic, Indian and Pacific Ocean Overturning Circulations.Talley [30] gives an excellent review of our current understandingof the pathways of the overturning circulation in the global oceanbased on estimates of the heat, freshwater and nutrient transports.The schematic in Fig.4 highlights some key results relevant for thisstudy at the expense of ignoring all the complexities of the large-scalethree dimensional circulation.

The overturning circulation is dominated by the intertwined path-ways of abyssal waters, deep waters and intermediate waters illus-trated as ribbons of different colors in Fig.4. We begin our descrip-tion of the overturning from the formation of deep waters throughwintertime convection in the North Atlantic. The North AtlanticDeep Water (NADW) flows southward primarily along deep west-ern boundary currents all the way to the SO where it is brought tothe surface along isopycnals by the Southern Hemisphere westerlies(green ribbon). The core of NADW has densities of 28 kg m−3 andthus upwells in the SO south of the isopycnal 27.9 kg m−3 where thebuoyancy flux is negative. It then flows southward toward Antarc-tica, where it is subject to brine rejection and sinks back into theabyss. These dense waters flow northward as Antarctic Bottom Wa-ter (AABW) at the bottom of the Atlantic, Indian and Pacific Oceans

3We follow [30] in choosing 27.9 kg m−3 as the isopycnal that separates the SO overturningbranches. This is the same density that separates the two branches of the overturning in theSouthern Ocean in [28].

Footline Author PNAS Issue Date Volume Issue Number 3

(blue ribbon). In all three oceans, AABW upwells into the local deepwater, that is, into NADW in the Atlantic (green ribbon), Indian DeepWater in the Indian (IDW, red ribbon) and Pacific Deep Water in thePacific (PDW, red ribbon). The transformation from abyssal to deepwaters is driven by turbulent mixing which diffuses waters acrossdensity surfaces (dashed vertical ribbons). Observations and theory,reviewed in the supplementary material, show that mixing is mostvigorous below 2 km, where internal waves radiated from rough to-pography break and mix [31, 32]. AABW therefore diffuses up to adepth of 2 km in the Indian and Pacific Oceans before returning tothe SO along isopycnals. In the North Atlantic, AABW does not dif-fuse as far up because it has less time to mix into the faster flowingNADW. The IDW and PDW have core densities around 27.9 kg m−3

and outcrop in the SO further to the north than the denser NADW. Theupwelled IDW/PDW in the SO feed both branches of the overturningin the SO: a fraction of the waters upwell north of the buoyancy fluxline and flow across the ACC to join the upper branch, the rest isrecycled back through the lower branch along with NADW. The frac-tion of IDW/PDW that joins the upper branch is a major source ofthe intermediate waters that then feed northward to the NADW for-mation region (orange ribbon), again connecting the upper and lowerbranches.

The key insight of [30] is that turbulent mixing sets the depthto which AABW upwells in the Indian and Pacific Oceans. The27.9 kg m−3 isopycnal that we have chosen as the separatrix3 be-tween the upper and lower SO overturning branches sits close to 2 kmin the Indian and Pacific Oceans, while it is a bit shallower in the At-lantic Ocean for the reasons pointed out above. Abyssal mixing isenhanced below 2 km, the characteristic depth of mid-ocean ridges.In the modern climate this depth is sufficiently shallow that a sub-stantial fraction of IDW and PDW upwell north of the zero buoyancyflux line and thus join the upper branch. The modern overturningcirculation is more akin to a figure eight loop than two separate cells.

The glacial deep ocean stratification and circulationDeep ocean nutrient tracers provide a relatively clear picture of thechanges in water mass distribution between the modern and LGMAtlantic [5, 33–35]. Fig.1 shows that in the modern ocean there isa high δ13C tongue of northern source NADW spreading over the

Present

Longitude

Latitu

de

mm2 s!3

0 50 100 150 200 250 300 350

!75

!70

!65

!60

!55

!50

!45

!40

!0.02

!0.01

0

0.01

0.02

Last Glacial Maximum

Longitude

Latitu

de

mm2 s!3

0 50 100 150 200 250 300 350

!75

!70

!65

!60

!55

!50

!45

!40

!0.02

!0.01

0

0.01

0.02

Fig. 5. The surface buoyancy flux from two CCSM PMIP2 simulations run formodern and LGM climates [36]. The flux is in color with blue representing nega-tive fluxes. The black lines are the 70% and 80% quantiles of annual mean seaice concentration. The gray line is the position of the -2◦C 2 m air temperatureisoline in summer. The area of negative buoyancy flux, the -2◦C isoline and theextent of quasi permanent sea ice shift poleward in excess of 5 latitude degreesat the LGM.

lower δ13C tongue of southern source AABW. At the LGM, thesouthern source water appears to fill the whole ocean volume be-low 2 km, squeezing a much reduced component of northern sourcewaters above 2 km.

Reconstructions based on a combination of available proxy andnumerical models suggest that global surface temperatures at thepeak of the last glacial period were between 3 and 6 degrees colderthan in pre-industrial times [1], with even more pronounced coolingsouth of 45◦S [4]. Such a cooling is expected to cause a signifi-cant equatorward retreat of the latitudes with above-freezing temper-atures. Summer sea ice tracks the -2◦C isoline of surface tempera-ture (the temperature at which sea water freezes) and thus the surfacecooling is expected to have resulted in an expansion of the area cov-ered by sea ice. Diatom and radiolarian assemblages from sedimentcores confirm that the summer sea ice expanded at the LGM, but thedata are too sparse to determine the overall areal expansion [3].

Numerical simulations show that the sea ice coverage expandsduring glacial climates, following the equatorward shift of the -2◦Csummer air temperature isoline. In Fig.5 we show output fromtwo CCSM PMIP2 (the National Center for Atmospheric ResearchCCSM3 model) simulations run for modern and LGM climates [36].The solid black lines, which mark the quasi-permanent sea ice line4,shifted by at least 5 latitude degrees at the LGM, together with thearea where the air-sea buoyancy flux is negative.

Following the scaling argument given in the previous section, wecan reconstruct the impact of a shift in the quasi permanent sea iceline and the associated zero buoyancy flux line on the ocean strati-fication and overturning. In the SO the isopycnal that separates theupper and lower branches of the overturning has a slope s given byequation (1) and outcrops at the zero buoyancy flux line. The depth ofthis isopycnal, once it reaches the closed oceans basins and becomesflat, is therefore given by,

D = `× s, [2]

where ` is the distance between the quasi-permanent sea ice lineand the northern extent of the ACC (see the schematic in Fig.4). Inthe modern climate s ' 10−3 and the distance between the quasi-permanent sea ice line (65◦S) and the northern boundary of the ACC(45◦S) is close to 20 latitude degrees or close to 2200 km. This givesa depth D ' 2.2 km, which is close to the depth of the isopycnal27.9 kg m−3 in the Pacific and somewhat shallower in the Atlantic.

A 5 degree latitude shift of the quasi-permanent sea ice line isequal to a δ` decrease of close to 500 km. For a constant isopycnalslope s this results into a 500 m shoaling (s × δ`) of the isopyc-nal separating the lower and upper branches of the SO overturning.As the isopycnal shoals above 2 km in the ocean basins, turbulentmixing is no longer able to drive substantial upwelling of AABWacross the isopycnal. Strong turbulent mixing is confined below 2 kmboth in the modern climate and the LGM, because it is generated inproximity of seamounts and mid-ocean ridges which are mostly con-fined below 2 km. As a consequence the deep waters return to theSO below the isopycnal outcropping at the sea ice line, upwell un-der quasi-permanent ice, where the buoyancy flux is negative, andmove toward Antarctica to be transformed back into AABW (blueribbon in Fig.4b). In response the NADW shoals above the expandedabyssal cell to outcrop where the buoyancy flux is positive, and re-turn north as intermediate waters completing a closed loop (green andyellow ribbons in Fig.4b.) Any NADW sinking below 2 km wouldoutcrop under sea ice and be transformed into denser AABW, neverto come back to the North Atlantic. The transformation of watersfrom the upper to the lower cell would continue during a transient,

4In the state estimate for the modern climate, the zero buoyancy flux lines follows the oceanarea covered by ice 70% of the time, while in the simulations it follows more closely the oceanarea covered by ice 80% of the time. This slight difference may reflect deficiencies in the seaice model or the limited temporal coverage of the observations (3 years). Regardless, ourargument is that the buoyancy flux is negative where the ocean is covered by ice most of theyear and it does not matter whether most means 70% or 80%.

4 www.pnas.org/cgi/doi/10.1073/pnas.????????? Footline Author

until the NADW neutral buoyancy level shoaled above 2 km. Theinescapable conclusion is that a 5 latitude degree expansion of quasipermanent sea ice would result in two separate overturning cells ontop of each other. This is the paradigm that is sometimes erroneouslysketched for the modern climate. A more detailed explanation of howthe ocean overturning transitions from a figure eight cell to two sep-arate cells is given in the Supplementary Information.

In summary, the 5 degree latitude expansion of summer sea ice atthe LGM was accompanied by the splitting of the modern figure eightoverturning circulation in two separate overturning cells as illustratedin Fig.4. As a result waters of Antarctic origin filled all ocean basinsup to 2 km depth, instead of the present 4 km depth. These changesare consistent with the rearrangement of water masses documented inpaleorecords. The δ13C map in Fig.1 shows that the LGM ocean wasfilled with waters of Antarctic origin below 2 km. Profiles of δ18Oconfirm that the temperature and salinities, the two variables that af-fect δ18O, of the deep Atlantic below 2 km reflect their SouthernOcean origin with no signature of a saltier and warmer NADW [37].Atlantic and Pacific deep nutrient distributions were more similar atthe LGM than today suggesting a common origin [38]. The similarityin nutrient distributions above 2 km in the LGM Atlantic and IndianOceans is further evidence of a shallow overturning circulation sepa-rate from abyssal water masses [39].

Estimates of water mass ages also support our simple theory. Inthe present day Indian and Pacific Oceans, there is a mid-depth max-imum in water ages because AABW is younger than the returningIDW/PDW. In the Atlantic Ocean, there is no age maximum, becausethe fast flowing NADW is younger than the slow AABW. Our in-ference that AABW returned south as deep waters below NADW inthe Atlantic at the LGM (blue ribbon in Fig.4b) points to the appear-ance of a mid-depth age maximum in the Atlantic as well. Radio-carbon deep ventilation ages confirm that mid-depth waters in theSouth Atlantic were 2000-3750 years old [40], sandwiched betweenmuch younger waters above [41] and below [42]. A more quantita-tive comparison of our model with radiocarbon profiles is the focusof a companion paper [43].

Our argument so far ignored variations in isopycnal slope be-tween present day and glacial climates. This may seem at odds withthe recent literature on the role of changes in the strength and loca-tion of winds at the LGM [12, 13, 44], which in turn set the isopycnalslope. Equation (2) can be used to estimate the relative importanceof changes in sea ice line versus changes in wind stress on the strat-ification and overturning at the LGM. Changes in the depth of theinterface between the upper and lower branches of the SO overturn-ing, δD, scale as,

δD

D=δs

s+δ`

`=δτ

τ− δK

K+δ`

`, [3]

where we used equation (1) to relate the changes in slope to changesin SO wind stress, δτ , and eddy diffusivity, δK. Paleo reconstruc-tions and atmospheric models suggest that the strength of windschanged by less than 10% at the LGM either due to a few degreespoleward shift of the westerlies or due to a strengthening of the sur-face temperature gradient [45, 46]. Furthermore changes in δK tendto strongly compensate any increase/decrease in the winds [47–49]:geostrophic eddies release the wind energy input in the ocean, so Kincreases/decreases as the winds increase/decrease. The combinedeffect of changes in winds and eddy diffusivities would thus appearto imply a shoaling/deepening of the isopycnal separating the twobranches of the SO overturning closer to 1% than 10%. This is to becontrasted with the 25% change that we estimated from the sea iceline shift (δ`/`=500/2000).

ConclusionsOur analysis suggests that the observed expansion of deep waters ofAntarctic origin at the Last Glacial Maximum (LGM) is dynamicallylinked to the expansion of quasi-permanent (summer) sea ice in the

Southern Hemisphere. The argument is best explained through theschematic shown in Fig.4. The overturning circulation in the South-ern Ocean is dominated by two major branches: an abyssal branchwith waters upwelling under quasi-permanent sea ice and sinking intothe abyss around Antarctica, and a deep branch with waters upwellingnorth of the quasi-permanent ice line and flowing to the north. Theisopycnal that intersects the surface at the quasi-permanent sea iceline separates the two overturning branches along a surface of con-stant slope in the SO and along a flat surface in the ocean basins tothe north (dashed lines in Fig.2 and 4). In the modern climate, thequasi-permanent sea ice edge is close to Antarctica and the isopy-cnal plunges below 2 km depth before reaching the ocean basins.There are many topographic features that reach this depth resultingin strong mixing that drives waters across the dividing isopycnal andconnects the two branches of the overturning circulation in a singlefigure eight overturning cell. At the LGM, the quasi-permanent seaice line extended by at least 500 km according to paleo reconstruc-tions and numerical models. The density budget of the ocean de-mands that the isopycnal separating the two overturning branches ofthe circulation shifted together with the ice line, as shown in Fig.4b,and did not plunge as deep into the ocean basins (the slope of thissurface changes little across different climates.) Once the isopycnalshoaled above 2 km, it did not intersect much topography and expe-rienced little turbulent mixing so that the two overturning branchesbecame two separate closed overturning cells stacked one on top ofthe other. The traditional view of two separate cells, incorrectly usedat times to describe the modern ocean, is only appropriate for theLGM climate.

Deep ocean tracer distributions are consistent with Antarctic ori-gin waters filling all oceans below 2 km at the LGM and only below4 km in the modern climate (see Fig.1). Our analysis suggests thatthis expansion tracks the appearance of a closed abyssal overturningcell coming to the surface only around Antarctica under sea ice. Thisvoluminous water mass was likely very salty, consistent with porousfluid measurements [50], as it experienced strong brine rejection un-der the sea ice as discussed in the Supplementary Information.

The dynamical model presented here provides solid underpin-ning for quantifying the role of ocean circulation on the glacial car-bon budget. The ocean-atmosphere partitioning of carbon is set bya balance between biological carbon export from the surface-ocean,the "biological pump", and the return of carbon to the surface bythe ocean’s overturning circulation, the "physical pump" [6–8]. Ourwork shows that the primary changes in the glacial physical pumpconsisted of three effects. The LGM waters filling all deep oceansbelow 2 km were at freezing temperature, thus increasing the seawa-ter solubility of CO2. The LGM deep waters leaked less carbon thantoday, because they did not experience much mixing with the over-lying waters of North Atlantic origin [16] and they outcropped onlyunder quasi-permanent sea ice, thus experiencing little surface out-gassing of CO2 [15]. Finally the fourfold expansion of poorly venti-lated waters at the LGM (the ocean waters below 2 km occupy fourtimes the volume of waters below 4 km) increased the ocean storagecapacity of carbon at depth [51]. An assessment of the atmosphericCO2 drawdown in response to these physical changes would requirea quantification of the associated changes in the biological pump andit is the subject of ongoing work. The main contribution of the workpresented here is to constrain the major changes in the physical pumpat the LGM. Arguably the lack of constraints on both the physicaland biological pump has stymied the study of glacial-interglacial at-mospheric CO2 fluctuations by allowing the exploration of too manyscenarios not relevant for the LGM climate.

It is important to emphasize that we proposed a diagnostic re-construction that connects sea ice expansion, ocean overturning cir-culation, and deep water masses. This is different from proposing aprognostic theory for how the ocean transitioned from the modern toa glacial climate. The sea ice expansion could have been triggered bya temperature drop in the Northern or in the Southern Hemisphere.

Footline Author PNAS Issue Date Volume Issue Number 5

The power of our argument is to show that disparate observations ofthe LGM climate can be brought together into a unified frameworkthrough ocean dynamics. Future work will need to assess to what ex-tent the zonally averaged approach introduced in this study capturesthe major features of the three dimensional ocean circulation at theLGM.

References1. T Schneider von Deimling, A Ganopolski, H Held, and S Rahmstorf.

How cold was the last glacial maximum? Geophys. Res. Lett., 33(14),2006.

2. E Monnin, A Indermuhle, A Dallenbach, J Fluckiger, B Stauffer, T. FStocker, D Raynaud, and J. M Barnola. Atmospheric CO2 concentra-tions over the last glacial termination. Science, 291, 2001.

3. R Gersonde, X Crosta, A Abelmann, and L Armand. Sea-surface temper-ature and sea ice distribution of the southern ocean at the EPILOG lastglacial maximum–a circum-antarctic view based on siliceous microfossilrecords. Quatern. Science Rev., 24(7-9):869 – 896, 2005.

4. MARGO Project Members. Constraints on the magnitude and patternsof ocean cooling at the last glacial maximum. Nature Geoscience, 2:127– 132, 2009.

5. W. B Curry and D. W Oppo. Glacial water mass geometry and the distri-bution of δ13C of

∑CO2 in the western Atlantic Ocean. Paleoceanogr.,

20(1):PA1017, 2005.6. F Knox and M. B McElroy. Changes in atmospheric co2: Influence of

the marine biota at high latitude. J. Geophys. Res.: Atmospheres, 89:4629–4637, 1984.

7. J. L Sarmiento and J. R Toggweiler. A new model for the role of theoceans in determining atmospheric pCO2. Nature, 308:621–624, 1984.

8. U Siegenthaler and T Wenk. Rapid atmospheric co2 variations and oceancirculation. Nature, 308:624–626, 1984.

9. V Brovkin, A Ganopolski, D Archer, and S Rahmstorf. Lowering ofglacial atmospheric CO2 in response to changes in oceanic circulationand marine biogeochemistry. Paleoceanogr., 22:PA4202, 2007.

10. M. P Hain, D. M Sigman, and G. H Haug. Antarctic stratification, NorthAtlantic Intermediate Water formation, and subantarctic nutrient draw-down during the last ice age: Diagnosis and synthesis in a geochemicalbox model. Global Biogeochemical Cycles, 24:47–55, 2010.

11. D. M Sigman, M. A Altabet, R Francois, D. C McCorkle, and J.-F Gail-lard. The isotopic composition of diatom-bound nitrogen in southernocean sediments. Paleoceanogr., 14(2):118–134, 1999.

12. R. F Anderson, S Ali, L. I Bradtmiller, S. H. H Nielsen, M. Q Fleisher,B. E Anderson, and L. H Burckle. Wind-driven upwelling in the SouthernOcean and the deglacial rise in atmospheric CO2. Science, 323:1443–1448, 2009.

13. J. R Toggweiler. Shifting westerlies. Science, 323:1434–1435, 2009.14. A. M de Boer, D. M Sigman, J. R Toggweiler, and J. L Russell. Effect

of global ocean temperature change on deep ocean ventilation. Paleo-ceanogr., 22(2):PA2210, 2007.

15. B Stephens and R. F Keeling. The influence of Antarctic sea ice onglacial-interglacial CO2 variations. Nature, 404:171–174, 2000.

16. D. C Lund, J. F Adkins, and R Ferrari. Abyssal Atlantic circulation dur-ing the Last Glacial Maximum: Constraining the ratio between transportand vertical mixing. Paleoceanogr., 26, 2011.

17. D. M Sigman, M. P Hain, and G. H Haug. The polar ocean and glacialcycles in atmospheric CO2 concentration. Nature, 466:47–55, 2010.

18. W. H Munk. Abyssal recipes. Deep-Sea Res., 13:707–730, 1966.19. J. R Toggweiler and B Samuels. On the ocean’s large-scale circulation

near the limit of no vertical mixing. J. Phys. Oceanogr., 28:1832–1852,1998.

20. A Gnanadesikan. A simple predictive model for the structure of theoceanic pycnocline. Science, 283:2077–2079, 1999.

21. M Nikurashin and G Vallis. A theory of deep stratification and overturn-ing circulation in the ocean. J. Phys. Oceanogr., 41:485–502, 2011.

22. D. R Jackett and T. J McDougall. A neutral density variable for theworld’s oceans. J. Phys. Oceanogr., 27(9):237–263, 1997.

23. V. V Gouretski and K. P Koltermann. WOCE global hydrographic cli-matology. Technical report, Berichte des Bundesamtes fur Seeschifffahrtund Hydrographie, 2004.

24. J Pedlosky. Geophysical Fluid Dynamics. Springer, 1987.25. J Marshall and K Speer. Closure of the meridional overturning circulation

through Southern Ocean upwelling. Nature Geosci., 5:171–180, 2012.

26. M. R Mazloff, P Heimbach, and C Wunsch. An eddy-permitting southernocean state estimate. J. Phys. Oceanogr., 40:880–899, 2010.

27. A Ganachaud and C Wunsch. Improved estimates of global ocean circu-lation, heat transport and mixing from hydrographic data. Nature, 408:453–457, 2000.

28. R Lumpkin and K Speer. Global ocean meridional overturning. J. Phys.Oceanogr., 37:2550–2562, 2007.

29. L. D Talley, J. L Reid, and P. E Robbins. Data-based meridional over-turning streamfunctions for the global ocean. J. Climate, 16:3213–3226,2003.

30. L. D Talley. Closure of the global overturning circulation through the In-dian, Pacific and Southern Oceans: schematics and transports. Oceanog-raphy, 26:80–97, 2013.

31. L. C St. Laurent, H. L Simmons, and S. R Jayne. Estimating tidally drivenmixing in the deep ocean. Geophys. Res. Lett., 29:21–1–21–4, 2002.

32. M Nikurashin and R Ferrari. Overturning circulation driven by breakinginternal waves in the deep ocean. Geophys. Res. Lett., 40:3133–3137,2013.

33. E. A Boyle and L. D Keigwin. Deep circulation of the North Atlanticover the last 200,000 years: geochemical evidence. Science, 218:784–787, 1982.

34. W. B Curry and G. P Lohmann. Carbon isotopic changes in benthicforaminifera from the western South Atlantic: reconstruction of glacialabyssal circulation patterns. Quatern. Res., 18:218–235, 1982.

35. J. C Duplessy, N. J Shackleton, R. G Fairbanks, L Labeyrie, D Oppo,and N Kallel. Deepwater source variations during the last climatic cycleand their impact on the global deepwater circulation. Paleoceanogr., 3:343–360, 1988.

36. B. L Otto-Bliesner, C. D Hewitt, T. M Marchitto, E Brady, A Abe-Ouchi,M Crucifix, S Murakami, and S. L Weber. Last Glacial Maximum oceanthermohaline circulation: PMIP2 model intercomparisons and data con-straints. Geophyisical Research Letters, 34:L12706, 2007.

37. I. N McCave, L Carter, and I. R Hall. Glacial-interglacial changes in wa-ter mass structure and flow in the sw pacific ocean. Quat. Sc. Rev., 27:1886–1908, 2008.

38. E. A Boyle. Cadmium: chemical tracer of deepwater paleoceanography.Paleoceanogr., 3:471–489, 1988.

39. E. A Boyle. Characteristics of the deep ocean carbon system during thepast 150,000 years:

∑CO2 distributions, deep water flow patterns, and

abrupt climate change. PNAS, 94:8300–8307, 1997.40. L Skinner, S Fallon, C Waelbroeck, E Michel, and S Barker. Ventila-

tion of the deep Southern Ocean and deglacial CO2 rise. Science, 328:1147–1151, 2010.

41. S Barker, G Knorr, M. J Vautravers, P Diaz, and L Skinner. Extremedeepening of the Atlantic circulation during deglaciation. Nature Geo-science, 291:567–571, 2010.

42. A Burke and L. F Robinson. The Southern OceanÕs Role in CarbonExchange During the Last Deglaciation. Science, 335:557–561, 2011.

43. A Burke, A Stewart, J Adkins, R Ferrari, M Jansen, and A Thompson.Changes in water mass ages at the Last Glacial Maximum. Paleoceanogr.,2014.

44. J. R Toggweiler, J. Russell, and S. R Carson. Midlatitude westerlies, at-mospheric CO2, and climate change during the ice ages. Paleoceanogr.,21, 2006.

45. K. E Kohfeld, R. M Graham, A. M De Boer, L. C Sime, E. W Wolff,C Le Quere, and L Bopp. Southern Hemisphere westerly wind changesduring the Last Glacial Maximum: Paleo-data synthesis. Quatern. Sci-ence Rev., 68:76–95, 2013.

46. L. C Sime, K. E Kohfeld, C. L Quere, E. W Wolff, A. M de Boer, R. MGraham, and L Bopp. Southern Hemisphere westerly wind changes dur-ing the last glacial maximum: model-data comparison. Quatern. ScienceRev., 64:104 – 120, 2013.

47. R Hallberg and A Gnanadesikan. The role of eddies in determining thestructure and response of the wind-driven southern hemisphere overturn-ing: Results from the Modeling Eddies in the Southern Ocean (MESO)project. J. Phys. Oceanogr., 36:2232–2252, 2006.

48. R Abernathey, J Marshall, and D Ferreira. The Dependence of SouthernOcean Meridional Overturning on Wind Stress. J. Phys. Oceanogr., 41:2261–2278, 2011.

49. D. R Munday, H. L Johnson, and D. P Marshall. Eddy saturation of equi-librated circumpolar currents. J. Phys. Oceanogr., 43(3):507–532, 2012.

50. J. F Adkins and D. P Schrag. Reconstructing Last Glacial Maximum bot-tom water salinities from deep-sea sediment pore fluid profiles. EPSL,216:2341–2354, 2003.

6 www.pnas.org/cgi/doi/10.1073/pnas.????????? Footline Author

51. L Skinner. Glacial - interglacial atmospheric CO2 change: A possiblestanding volume effect on deep-ocean carbon sequestration. Climates of

the Past, 5:537–550, 2009.

Footline Author PNAS Issue Date Volume Issue Number 7

-6000

-5000

-4000

-3000

-2000

-1000

0

)m( htpe

D

4.0

.0 4

.0 8

8.08.0

8.0

2.1

-6000

-5000

-4000

-3000

-2000

-1000

0

)m( htpe

D

-60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70

Latitude

8.0-

0- 4.

0

0

4.0

8.02.1

6.1

Fig. 1.

Latitude

Dep

th [m

]

−80 −60 −40 −20 0 20 40 60 80−4000

−3000

−2000

−1000

0

Neu

tral

den

sity

[kg/

m3 ]

25262727.727.827.92828.128.228.3

Fig. 2.

Longitude

Latitu

de

mm2 s!3

0 50 100 150 200 250 300 350

!75

!70

!65

!60

!55

!50

!45

!40

!0.02

!0.01

0

0.01

0.02

Fig. 3.

8 www.pnas.org/cgi/doi/10.1073/pnas.????????? Footline Author

80S

l1

0km

2km

4km

0km

2km

4km

50S 80N

Last Glacial Maximuml2

ICE

ICE

Modern

Fig. 4.

Present

Longitude

Latitu

de

mm2 s!3

0 50 100 150 200 250 300 350

!75

!70

!65

!60

!55

!50

!45

!40

!0.02

!0.01

0

0.01

0.02

Last Glacial Maximum

Longitude

Latitu

de

mm2 s!3

0 50 100 150 200 250 300 350

!75

!70

!65

!60

!55

!50

!45

!40

!0.02

!0.01

0

0.01

0.02

Fig. 5.

Footline Author PNAS Issue Date Volume Issue Number 9

Supplementary Information for “An ocean tale oftwo climates: Modern and Last GlacialMaximum”Raffaele Ferrari ∗, Malte Jansen †, Jess F. Adkins ‡, Andrea Burke ‡ , Andrew L. Stewart ‡ , and Andrew F. Thompson ‡

∗Massachusetts Institute of Technology, Cambridge, MA,†Geophysical Fluid Dynamics Laboratory, Princeton, NJ, and ‡California Institute of Technology, Pasadena,CA

Submitted to Proceedings of the National Academy of Sciences of the United States of America

Vertical profile of mixing in the deep oceanVertical mixing is typically quantified in terms of a vertical diffu-sivity κ, which represents the rate at which mixing would spread apatch of tracer over time. Direct estimates show that vertical dif-fusivities are strongly enhanced a few hundred meters above roughbottom topography [e.g. 1]. Direct estimates suggest that the verti-cal diffusivities averaged over ocean basins are of O(10−4) m2 s−1

at 4 km, the average ocean depth, and decay to O(10−5) m2 s−1 at2 km, the typical height of oceanic ridges [e.g. 2, 3]. Similar resultsare found from inverse models that infer vertical diffusivities fromtracer distributions [4, 5] and from theoretical calculations of the in-ternal wave generation rates [6]. A review of the extensive literatureconcludes that diffusivities of O(10−4) m2 s−1 drive an overturn-ing of O(10) Sv, as observed in the real ocean, while diffusivities ofO(10−5) m2 s−1 could only return a very weak flow of O(1) Sv [1].

Numerical models of the Last Glacial MaximumOur claim that the rearrangement in deep water masses observed atthe LGM is linked through simple dynamics to the expansion of seaice around Antarctica may seem at odds with reports that numeri-cal models of the LGM struggle to reproduce this rearrangement [7].While the CCSM PMIP2 model used to generate Fig.5 shows an ex-pansion of the abyssal water mass, other models do not. We believethat the range in model responses stems from various inconsistenciesin the physics. Most models do not include a profile of vertical mix-ing that decays with height above the bottom, thereby missing a keydynamical aspect of the LGM oceans. Models differ in their treat-ment of atmospheric radiation and sea ice with some models captur-ing the expansion in sea ice better than others. The representation ofgeostrophic eddies is deficient, because it overestimates changes inisopycnal slope due to wind changes at the LGM.

Based on our study, we believe that LGM simulations wouldlikely become more consistent if all models represented the key dy-namical features of the LGM oceans, namely the vertical profile ofvertical mixing, an appropriate representation of geostrophic eddies,and consistent sea ice physics. We plan to run such a model to testto what extent our zonally averaged view of the changes in the oceancirculation at the LGM will stand the test of a three dimensional cir-culation.

Shoaling of North Atlantic Deep Water at the Last Glacial

MaximumThe shoaling of NADW is best understood if one considers the tran-sient response to an expansion of SO sea ice. Any water sinking in theNorth Atlantic below the isopycnal separating the two cells would betrapped in the abyssal cell, because mixing would no longer be ableto lift it high enough in the water column to outcrop around Antarc-tica in areas experiencing a positive buoyancy flux. The waters would

instead experience a negative buoyancy flux, become denser and sinkback into the abyss. This transient conversion of NADW into denserAABW would continue until the whole abyss is filled with watersdenser than the ones formed in the North Atlantic. At this point wa-ters sinking in the North Atlantic will start flowing above the abyssaldense cell and the equilibrium solution sketched in Fig.4 would beachieved.

We sketched the solution with two overturning cells relevant forthe LGM in Fig.4b. However a second solution is possible [8]. Ifthe Northern Hemisphere air-sea fluxes generated waters lighter thanany of the waters found at the surface in the Southern Ocean, the up-per overturning cells would disappear and a pool of stagnant waterswould develop in the North Atlantic, a solution possibly relevant forHeinrich events. The circulation in the upper cell is, however, incon-sequential for the abyssal cell, which is the focus of this study.

The deep salinity budget at the Last Glacial MaximumIn the present day ocean, North Atlantic Deep Water comes to the sur-face under sea ice in the Southern Ocean with temperatures between2-3◦C. These warm waters drive strong melting of the ice shelves re-sulting in a freshening of the surface waters visible in Fig. 1 as a lightdensity tongue right at the sea surface around Antarctica. This fresh-water offsets much of the increase in salinity driven by the formationof new ice and brine rejection. This results in weak salinity variationsand as a result temperature dominates the density stratification in thepresent Southern Ocean.

At the Last Glacial Maximum, the deep waters upwelling to thesurface around Antarctica were a close to the freezing point and re-sulted in little melting of ice shelves [9]. The waters that flew pole-ward under the permanent sea ice were therefore exposed to signifi-cant brine rejection, resulting in a strong increase in their salinities.We can derive a scaling law for the the salinity changes experiencedby the waters as they flowed under ice.

In steady state the meridional salinity variations of waters flow-ing under sea ice can be computed from the simple formula [10]:

ψSy ≈ F .

Reserved for Publication Footnotes

www.pnas.org/cgi/doi/10.1073/pnas.0709640104 PNAS Issue Date Volume Issue Number 1–3

Here ψ is the mass transport in the surface layer of the ocean exposedto the salinity flux F and Sy is the meridional salinity gradient. Thisequation has been obtained averaging the equation for the conserva-tion of salinity over the depth of the ocean mixed layer (the upperlayer of the ocean where temperature and salinity are vertically wellmixed) and along a latitude circle. If we further integrate the equationin latitude between the Antarctic continent and the permanent sea iceline, we obtain that the salinity change ∆S of waters at the permanentsea ice edge and those around the Antarctic continent scales as,

∆S ≈ F × `/ψ

where ` is the meridional extent of sea ice.The CCSM PMIP2 model used to generate Fig.5 simulates a

mass transport of ψ = 20× 106 m3s−1 in the LGM mixed layer un-der sea ice, close to the present day value. The area averaged salinityflux is instead much larger than today as a result of the expansionof the permanent sea ice and the reduction of ice melting by warmocean waters and F × ` = 2 × 107 m3psu s−1. According to ourscaling law, these values imply a salinity difference ∆S ≈ 1psu be-tween the deep waters that upwell at the sea ice edge and the abyssalwaters that sink into the abyss around Antarctica. This is seen in themodel salinity profiles and It is consistent with pore fluid estimatesof LGM salinity, which suggest that the abyssal ocean was at leastone psi saltier than overlying waters. This large salinity differencedominated the density stratification in the Southern Ocean. A ten de-gree temperature difference would be required to match the impacton density of a one psu salinity difference. And the deep LGM oceanwas everywhere within a few degrees of freezing. Based on modeloutput and observations, we can then conclude that the isolated over-

turning cell spanning the deep ocean at the LGM and depicted in theschematic in Fig. ?? was associated with a strong salinity gradient.

References1. C Wunsch and R Ferrari. Vertical mixing, energy, and the general circu-

lation of the oceans. Ann. Rev. of Fluid Mech., 36:281–314, 2004.2. L St. Laurent and H Simmons. Estimates of power consumed by mixing

in the ocean interior. J. Climate, 19:4877–4890, 2006.3. L. C St. Laurent, H. L Simmons, and S. R Jayne. Estimating tidally driven

mixing in the deep ocean. Geophys. Res. Lett., 29:21–1–21–4, 2002.4. A Ganachaud. Large-scale mass transports, water mass formation, and

diffusivities estimated from world ocean circulation experiment (WOCE)hydrographic data. J. Geophys. Res.: Oceans, 108, 2003.

5. R Lumpkin and K Speer. Global ocean meridional overturning. J. Phys.Oceanogr., 37:2550–2562, 2007.

6. M Nikurashin and R Ferrari. Overturning circulation driven by breakinginternal waves in the deep ocean. Geophys. Res. Lett., 40:3133–3137,2013.

7. B. L Otto-Bliesner, C. D Hewitt, T. M Marchitto, E Brady, A Abe-Ouchi,M Crucifix, S Murakami, and S. L Weber. Last Glacial Maximum oceanthermohaline circulation: PMIP2 model intercomparisons and data con-straints. Geophyisical Research Letters, 34:L12706, 2007.

8. C. L Wolfe and P Cessi. The adiabatic pole-to-pole overturning circula-tion. J. Phys. Oceanogr., 41:1795–1810, 2011.

9. M. D. M. J. F. A. D Menemenlis and M. P Schodlok. The role of oceancooling in setting glacial southern source bottom water salinity. Pale-ooceanogr., 27:PA3207, 2012.

10. J Marshall and T Radko. Residual-mean solutions for the Antarctic Cir-cumpolar Current and its associated overturning circulation. J. Phys.Oceanogr., 33:2341–2354, 2003.

2 www.pnas.org/cgi/doi/10.1073/pnas.0709640104 Footline Author