Paleonutrient proxies currently provide the strongest constraints on the past spatial distribution of deep water masses. We review the state of knowledge derived from the trace metal proxy Cd/Ca for the Atlantic Ocean during the Last Glacial Maximum (LGM). We compile published benthic foraminiferal Cd/Ca data, supplemented with new data, to reconstruct meridional Cd sections through the Holocene and LGM Atlantic. Holocene Cd/Ca reflects the modern water masses North Atlantic Deep Water (NADW), Antarctic Bottom Water (AABW), and Antarctic Intermediate Water (AAIW) reasonably well, except for anomalously high values in the intermediate to mid-depth far North Atlantic. LGM Cd/Ca clearly shows that NADW was replaced by the shallower Glacial North Atlantic Intermediate Water (GNAIW), with significant northward expansion of AABW. The boundary level between GNAIW and AABW was very abrupt, occurring between ∼2200 and 2500 m modern water depth. Combined Cd/Ca and Cibicidoidesδ13C data allow us to also calculate the air-sea signature of δ13C (δ13Cas). The LGM Atlantic can be explained entirely by mixing between high-δ13Cas GNAIW and low-δ13Cas AABW. Negative δ13Cas values in glacial AABW were likely caused by poor ventilation during formation, probably associated with extensive sea ice coverage. Overall, our knowledge base would benefit from improved data coverage in the South Atlantic and Southern Ocean, and a better understanding of ancillary effects on the Cd/Ca proxy itself.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Most of what paleoceanographers know about the past distribution of deep water masses is derived from paleonutrient proxies. Such passive tracers offer no direct information about deep water formation rates or fluxes [LeGrand and Wunsch, 1995], but may be used to constrain conceptual and numerical models of the MOC. The most widely applied paleonutrient tracer is benthic foraminiferal δ13C, which is inversely correlated with the labile nutrient phosphate [Kroopnick, 1985; Duplessy et al., 1988; Sarnthein et al., 1994; Curry and Oppo, 2005]. Boyle and colleagues developed benthic foraminiferal Cd/Ca as a deep water tracer at about the same time that δ13C was first applied to the problem, in the early 1980s [Boyle and Keigwin, 1982, 1985/1986]. Cd/Ca was therefore instrumental in codifying the view of a glacial-age shoaling of low-nutrient NADW and northward expansion of nutrient-rich Antarctic Bottom Water (AABW) into the North Atlantic [Boyle and Keigwin, 1987; Oppo and Fairbanks, 1987] (note that we use AABW to include Circumpolar Deep Water (CPDW); the glacial version of AABW/CPDW has also sometimes been referred to as Southern Ocean Water). Benthic foraminiferal Ba/Ca [Lea and Boyle, 1989, 1990; Martin and Lea, 1998] and Zn/Ca [Marchitto et al., 2000, 2002] have since been developed as complementary paleonutrient tracers, though their application has been very limited in comparison to Cd/Ca.
2. Cd/Ca Systematics
Numerous dissolved trace elements in the ocean exhibit vertical profiles that resemble those of major nutrients [Nozaki, 1997]. That is, they are at low concentrations in surface waters and greater concentrations at depth. In some cases, this behavior is linked to the element's importance as a micronutrient. In other cases, nutrient-like behavior may simply result from adsorption onto particulate organic matter in the photic zone and co-remineralization in the deep sea. Regardless of biogeochemical mechanisms, such elements may be useful as nutrient and water mass tracers. Paleoceanographic applications rely on the observation that various elements are accidentally incorporated into benthic foraminiferal CaCO3 during precipitation. In particular, divalent cations are believed to substitute for Ca in the CaCO3 crystal matrix.
Dissolved Cd has an oceanic distribution very similar to that of the nutrient phosphate [Boyle et al., 1976; Boyle, 1988; Elderfield and Rickaby, 2000] (Figures 1 and 2) . Both are efficiently removed from most surface waters and regenerated at depth, with an intermediate-depth concentration maximum near 1000 m. Cd therefore behaves like a labile nutrient, and its concentration increases about five-fold between the deep North Atlantic and North Pacific. Although Cd has been linked to at least one important algal metalloenzyme, carbonic anhydrase [Cullen et al., 1999; Morel et al., 2003], it is not clear if this use is sufficient to explain its ocean-wide nutrient-like behavior. The global correlation between Cd and phosphate follows a slight curve that may be explained by preferential uptake of Cd relative to phosphate, either actively by phytoplankton or passively by adsorption onto particulate organic matter [Elderfield and Rickaby, 2000] (Figure 1):
where αCd/P is the ratio of Cd/P in particulate organic matter relative to Cd/P in seawater, and averages ∼2 for the global data set. Historically, however, the Cd:phosphate relationship has been described simply by two line segments [Boyle, 1988]:
Benthic foraminiferal Cd/Ca ratios reflect seawater dissolved Cd concentrations, with partition coefficients:
that vary with water depth in calcitic species, from a minimum of 1.3 above 1150 m to a maximum of 2.9 below 3000 m [Boyle, 1992]. The aragonitic benthic foraminifer Hoeglundina elegans has a partition coefficient of 1.0, apparently invariant with water depth [Boyle et al., 1995]. Cd incorporation into benthic foraminifera does not appear to be significantly affected by water temperature [Marchitto, 2004].
3. Challenges and Uncertainties
Several ancillary effects may complicate the reconstruction of past Cd concentrations in seawater. First, many published records rely on combinations of benthic foraminiferal taxa, including epifaunal genera like Cibicidoides and infaunal genera like Uvigerina. Microhabitat is known to affect the δ13C of infaunal taxa because of organic matter degradation within sediment pore waters [Zahn et al., 1986; McCorkle et al., 1997]. Yet Boyle  has shown that there is no consistent Cd/Ca offset between the various calcitic taxa, regardless of microhabitat. Tachikawa and Elderfield  suggest that infaunal species usually record Cd/Ca values similar to Cibicidoides, despite elevated pore water Cd, because they have lower partition coefficients. If true, it would seem to be fortuitous that Cibicidoides and Uvigerina have successfully been used interchangeably in paleoceanographic reconstructions. Our own compilation of published data (section 5.1) indicates that there is no significant bias if infaunal species are included. Nevertheless, improved understanding of possible species-level differences and microhabitat effects is essential.
Benthic foraminiferal trace metal partition coefficients are further affected by seawater saturation state with respect to calcite (ΔCO32−). DCd, DBa, and DZn may be reduced by 50% or more in the severely undersaturated abyssal Indo-Pacific [McCorkle et al., 1995; Marchitto et al., 2000, 2005]. Limited measurements on recently living (Rose Bengal-stained) foraminifera, laboratory dissolution experiments, and theoretical biomineralization considerations implicate reduced trace metal incorporation during growth, rather than post-mortem preferential dissolution [Boyle, 1988; Elderfield et al., 1996; Marchitto et al., 2000]. Benthic foraminifera are believed to be more chemically homogenous than planktonic foraminifera, and therefore not subject to the same dissolution artifacts [Boyle and Rosenthal, 1996]. For Cd the ΔCO32− effect is limited to undersaturated waters, but DBa and DZn appear to be reduced in even mildly supersaturated waters.
Past changes in deep sea saturation state may therefore overprint benthic foraminiferal trace metal records, especially in regions that are poorly saturated today. In the deep Pacific, where glacial-interglacial changes in deep circulation are believed to have been small, ΔCO32− was probably the dominant control on Zn/Ca (and to a lesser extent Cd/Ca) records [Marchitto et al., 2005]. In the North Atlantic, glacial intrusion of low-CO32− AABW may have lowered DZn slightly, but DCd was not likely affected [Marchitto et al., 2002]. It is possible that Cd/Ca records from the deep South Atlantic are overprinted by changes in saturation state, but paleo-CO32− is difficult to estimate. CaCO3 accumulation in a South Atlantic core currently bathed by AABW suggests better preservation during glacial intervals, with highest inferred CO32− levels during glacial terminations [Hodell et al., 2001]. Farther north within the South Atlantic, CO32− changes were likely dominated by the waxing and waning of AABW. Tropical paleo-CO32− records based on planktonic foraminiferal shell weights show dissolution events during Marine Isotope Stage (MIS) 5/4 glacial intensifications [Broecker and Clark, 2003]. However, attempts to use shell weights to reconstruct South Atlantic paleo-CO32− have been hampered by the complicating effects of variable growth conditions and pore water dissolution [Broecker and Clark, 2004]. Paired Cd/Ca and Zn/Ca data could theoretically provide insight because of the different sensitivities of these proxies to ΔCO32− [Marchitto et al., 2000, 2005]. There is also some indication that Cd/Ca in the benthic species Nutallides umbonifera is immune to the saturation effect [Boyle and Rosenthal, 1996; Mackensen, 1997].
An additional challenge arises from within-sample heterogeneity. Boyle  noted that the Cd/Ca scatter between triplicate picks (∼15–20 individuals each) from a sediment sample is more than double that derived from equivalent-sized splits of crushed and homogenized individuals. The pooled standard deviation of the former approach (±0.023 μmol mol−1) is similar to the average within-core scatter of the Cd/Ca measurements compiled here. This error is equivalent to about ±0.08 nmol Cd per kg seawater (at maximum DCd of 2.9), which is relatively large compared to the modern whole-Atlantic gradient of ∼0.5 nmol kg−1 (Figure 2). Hence our reconstructed meridional sections (section 5.1) contain considerable noise, and individual data points must be viewed with caution. Possible causes for heterogeneity include: bioturbational mixing of non-contemporaneous individuals; variable microhabitats, even within the same species; unknown factors such as age or growth rate that may affect Cd uptake; and sedimentary contamination.
Finally, we cannot rule out analytical offsets between the various labs whose work is compiled here. At least three different cleaning protocols are represented: the original oxidative and reductive cleaning method developed for Cd/Ca [Boyle and Keigwin, 1985/1986]; a revised method that reverses the reductive and oxidative steps to better remove CdS overgrowths [Boyle and Rosenthal, 1996]; and a method developed for Ba/Ca that adds a DPTA step to dissolve sedimentary barite [Lea and Boyle, 1990]. We follow Rosenthal et al.  in updating some early measurements [Boyle, 1992] that appear to be affected by CdS contamination. Published analyses were performed primarily by atomic absorption spectrophotometry (AAS) [Boyle and Keigwin, 1985/1986], but also by inductively coupled plasma mass spectrometry (ICP-MS) [Lea and Martin, 1996] and thermal ionization mass spectrometry (TIMS) [Rickaby et al., 2000]. Interlaboratory calibration for Cd/Ca has generally been informal, and there is currently no community-wide standard available for this purpose.
We have compiled published benthic foraminiferal Cd/Ca data from all Atlantic sediment cores that include measurements from the Last Glacial Maximum (LGM, ∼20 kyr BP) (Table 1). The definition of LGM may vary slightly from study to study, but usually corresponds to near-maximum benthic δ18O values [see, e.g., Boyle, 1992]. Some samples may span a few thousand years before and after 20 kyr BP. Combined with bioturbational mixing and the rather low sedimentation rates typical of the deep sea, this means that LGM values may incorporate foraminifera from more extreme intervals like Heinrich event 1, during which the formation of NADW/GNAIW may have ceased [McManus et al., 2004]. Some high-resolution records show very elevated Cd/Ca during Heinrich event 1 [Willamowski and Zahn, 2000; Zahn and Stüber, 2002; Rickaby and Elderfield, 2005] and we avoid those measurements here. Some other cores may be so heavily bioturbated that LGM measurements include foraminifera from the Holocene and/or MIS 3.
Table 1. Compilation of Published Holocene and LGM Benthic Foraminiferal CdW Dataa
We also compile Holocene data, but only from the cores with LGM measurements. The definition of Holocene is even looser than that of the LGM. In some studies the reported Holocene values are only from core tops, which vary in age due to issues of core recovery and sedimentation. Other studies attempt to report late Holocene means, the true ages of which again vary with core quality.
We include data from various benthic foraminiferal taxa as reported in the literature to allow for maximum spatial coverage. The dominant genera represented are Cibicidoides, Uvigerina, Nutallides, and Hoeglundina. In all cases we convert Cd/Ca to CdW (inferred paleo-seawater Cd concentration, in nmol kg−1) using the depth-dependent calcitic partition coefficients from Boyle  or the aragonitic Hoeglundina partition coefficient from Boyle et al. . CdW is therefore recalculated from studies that use “empirical” or “local” partition coefficients [Bertram et al., 1995; Willamowski and Zahn, 2000]. The average error on individual CdW values, based only on Cd/Ca reproducibility, is estimated to be ±0.08 nmol kg−1 (±1σ).
In addition to published data, we report unpublished data from two Little Bahama Bank cores and new data from seven Mid-Atlantic Ridge cores (Table 2). The Little Bahama Bank measurements (cruise OC205-2) were made by AAS at the Woods Hole Oceanographic Institution following the methods of Marchitto et al.  except that several different taxa were employed (H. elegans, C. kullenbergi, C. wuellerstorfi, and U. peregrina). The new Mid-Atlantic Ridge measurements were made at the University of Colorado on ∼10–15 individuals of C. wuellerstorfi (>250 μm) each, oxidatively and reductively cleaned following the methods of Boyle and Keigwin [1985/6] as modified by Boyle and Rosenthal . LGM intervals were identified using a combination of unpublished radiocarbon dates and percent CaCO3 stratigraphies. Multiple trace metals were measured on a Thermo-Finnigan Element2 sector field single-collector ICP-MS using methods adapted from Rosenthal et al.  [Marchitto, 2006]. Long-term 1σ precision for Cd/Ca is 1.8% across a wide range of Cd/Ca values and sample sizes. Elemental measures of diagenetic (Mn/Ca) and sedimentary (Fe/Ca) contamination were low, ranging from 11 to 68 μmol mol−1 and 4 to 51 μmol mol−1, respectively.
Table 2. New Benthic Foraminiferal CdW Data
Hol. CdW, nmol/kg
LGM CdW, nmol/kg
We have also compiled Cibicidoides δ13C data, where available, from the cores used for Cd/Ca (Table 3). We limit the compilation to Cibicidoides because this epifaunal genus is commonly considered to be the most reliable recorder of seawater δ13C [e.g., Duplessy et al., 1984; Curry and Oppo, 2005]. We combine the δ13C data with CdW to calculate δ13Cas as described in section 5.2.
Table 3. Cibicidoides δ13C With δ13Cas Calculated From CdW Data in Tables 1 and 2a
We present meridional sections of Atlantic CdW and δ13Cas constructed using the program Ocean Data View (R. Schlitzer, Ocean Data View, 2005; http://odv.awi-bremerhaven.de). Our contour plots employ a relatively strong smoothing function (held constant between plots) to minimize the impact of anomalous cores. For CdW, individual values are also plotted for comparison with the smoothed plots. Only one core from the compilation (SO82-05) is not plotted due to its extremely anomalous Holocene Cd/Ca value. We present CdW sections that include all Atlantic data (Figure 3) as well as comparisons of the western and eastern basins (Figure 4). The western and eastern sections each include several cores from the crest of the Mid-Atlantic Ridge and all cores north of 50°N, most of which are from the eastern side of the Mid-Atlantic Ridge. No depth correction is made to the LGM data to account for lowered sea level.
5. LGM Atlantic Summary
5.1. Meridional Cd Sections
Major syntheses of benthic foraminiferal Cd/Ca data for the LGM Atlantic have been performed by Boyle , Rosenthal et al. , and Oppo and Horowitz . Of these, only Rosenthal et al.  presented a meridional section of glacial seawater Cd, following the approach that Duplessy et al.  used for δ13C. We repeat that exercise here, incorporating new data that have appeared in the interim. Our LGM composite section contains data from 75 core locations, compared to Rosenthal et al.'s  19. Our main observations are broadly consistent with those of the earlier studies, with significant refinement afforded by the additional data.
Before interpreting the LGM CdW sections, it is instructive to examine the Holocene sections to see how well Holocene foraminifera reproduce the modern Cd distribution. The nutrient pattern of the South Atlantic is captured reasonably well in the composite section (Figures 3a and 3b), with low-Cd NADW sandwiched between high-Cd AABW below and AAIW above. The volume of NADW expands to the north, as expected, with low-nutrient waters filling the entire water column by ∼30°N. However, intermediate to mid-depth cores in the high-latitude North Atlantic tend to exhibit anomalously high Cd/Ca values. This problem is locally pervasive but not ubiquitous, and its cause is not known. Bertram et al.  chose to use elevated empirical partition coefficients in this region based on core top Cd/Ca values, but as discussed above (section 4) we use the global partition coefficients of Boyle . The western Holocene section (Figure 4a) is similar to the composite, but the eastern section (Figure 4b) shows a weaker north-south deep CdW gradient. Reconstructed low to midlatitude deep North Atlantic waters contain more Cd in the east than in the west, while the opposite is true in the deep South Atlantic. Although these observations may be biased by sparse coverage, the North Atlantic data are consistent with the modern distribution of nutrients. Concentrations are lower in the better-ventilated deep western boundary current of NADW and higher in the northeastern basin, where deep recirculation is important for trapping nutrients [Broecker et al., 1991].
The composite LGM section is fundamentally different from the Holocene distribution, with NADW being replaced by low-Cd GNAIW lying above an expanded AABW (Figures 3c and 3d). This pattern is consistent with previous observations from Cd/Ca and δ13C [e.g., Boyle and Keigwin, 1987; Oppo and Fairbanks, 1987; Duplessy et al., 1988; Boyle, 1992; Oppo and Lehman, 1993; Curry and Oppo, 2005]. The boundary level between GNAIW and AABW appears to have been very sharp, occurring between ∼2200 and 2500 m modern depth. The abruptness of this boundary and the dichotomy between the two water masses is well illustrated by the LGM depth profile (Figure 3f). The recent western Atlantic δ13C compilation by Curry and Oppo  also shows the sharpest vertical gradient between ∼2000 and 2500 m. The relatively modest LGM CdW values found in the deepest Atlantic cores may suggest a slight depression of DCd by low-CO32− AABW, but not by enough to obscure the accompanying rise in seawater Cd. The current lack of data poleward of 40°S prevents the accurate characterization of the AABW end-member, but we address this below using paired Cd and δ13C (section 5.2). Data coverage in the intermediate depth South Atlantic is too sparse to infer the glacial presence or absence of AAIW. The deep (>3000 m) east-west CdW contrast during the LGM appears to have been weaker than during the Holocene, though there was perhaps a slightly deeper penetration of GNAIW in the west (Figures 4c and 4d). A reduced deep zonal contrast likely resulted from a combination of the loss of NADW and the spilling of AABW over the Walvis and Mid-Atlantic Ridges into the eastern basins [Beveridge et al., 1995]. Comparison of western [Curry and Oppo, 2005] and eastern [Sarnthein et al., 1994] sections of δ13C at the LGM also suggest relatively little zonal contrast below ∼3000 m.
5.2. Air-Sea δ13C
The distributions of Cd and the δ13C of dissolved inorganic carbon (DIC) are influenced by similar biological cycling, but δ13C is also strongly impacted by air-sea exchange. If biological cycling were the only process acting on δ13C, it would decrease by about 1.1‰ for each 1 μmol kg−1 increase in dissolved phosphate, and a modern oceanic range of ∼3.3‰ would be possible [Broecker and Maier-Reimer, 1992]. However, air-sea exchange of CO2 can isotopically enrich the DIC pool (increase δ13C), with greater enrichment in cases of (1) greater exchange due to long air-sea contact time or high winds; (2) exchange at colder temperatures; or (3) net CO2 efflux out of the ocean into the atmosphere [Mook et al., 1974; Lynch-Stieglitz et al., 1995]. The net effect of the various air-sea exchange processes spans a modern surface ocean range of ∼2‰.
where 2.7 is an arbitrary constant introduced to bring deep Pacific and Indian Ocean waters close to zero. δ13Cas is a conservative tracer of deep waters because once a parcel of water leaves the surface ocean, its δ13Cas signature is only altered by mixing [Lynch-Stieglitz and Fairbanks, 1994]. δ13Cas can therefore be used to separate mixing and biogeochemical aging processes. Modern NADW is characterized by negative δ13Cas values, while AABW and AAIW have positive signatures. AAIW is especially enriched in δ13Cas due to formation from surface waters that are partially equilibrated with the atmosphere at cold temperatures [Charles and Fairbanks, 1990; Charles et al., 1993].
Lynch-Stieglitz et al.  noted that during the LGM, organic matter δ13C was ∼2‰ higher [Rau et al., 1991] and mean ocean δ13C was ∼0.3‰ lower [Duplessy et al., 1988], so the relationship between δ13C and phosphate was slightly different. Taking this into account and assuming unchanged oceanic Cd and phosphate inventories, they derived the following equation for δ13Cas in the LGM deep ocean (P > 1.34 μmol kg−1):
Two additional equations can be derived for waters with (P < 1.34 μmol kg−1):
We use Cibicidoidesδ13C (where available) from our compiled Atlantic Cd/Ca sites to calculate δ13Cas during the Holocene and LGM, using equations (6)–(9) (Table 3). It must be noted that the resulting Holocene meridional section of δ13Cas bears relatively little resemblance to its modern distribution (Figure 5a). Although NADW below ∼2000 m displays negative values and AAIW appears to be positive (both as expected), upper North Atlantic values are generally too positive and there is no sign of a positive AABW. The upper North Atlantic anomaly is related, at least in part, to the anomalously high Holocene Cd/Ca values found in this region, as noted above. The paucity of paired Cd:δ13C data in the deep South Atlantic may hinder the proper reconstruction of AABW δ13Cas.
Given the problems with the Holocene reconstruction, the LGM δ13Cas section (Figure 5b) should be viewed with some caution. Nevertheless, some interesting patterns emerge. As in previous reconstructions [Lynch-Stieglitz and Fairbanks, 1994; Oppo and Horowitz, 2000] GNAIW is clearly characterized by positive δ13Cas values, compared to ∼−0.4 to −0.6‰ for modern NADW [Charles et al., 1993; Lynch-Stieglitz et al., 1995]. The deep South Atlantic, in contrast, had a negative δ13Cas signature instead of the slightly positive values of modern AABW. It has long been noted that South Atlantic Cd/Ca shows little to no glacial-interglacial change while δ13C was much lower during the LGM, resulting in low δ13Cas [Boyle, 1992; Boyle and Rosenthal, 1996; Rosenthal et al., 1997]. Although some very low Southern Ocean δ13C values have been questioned because of microhabitat effects caused by high fluxes of algal detritus [Mackensen et al., 1993], the spatial uniformity of Southern Ocean δ13C during the LGM argues for a real drop in deep water values [Ninnemann and Charles, 2002]. As noted above, deep South Atlantic Cd/Ca may hypothetically have been affected by decreased seawater CO32− during the LGM, but the magnitude of this effect is difficult to estimate.
The glacial expansion of AABW allows for better δ13Cas characterization of this water mass than in the Holocene section, yet we still have no data from the Southern Ocean poleward of 40°S. However, we can use a δ13C:CdW cross-plot to extrapolate to an AABW end-member estimate for the LGM (Figure 6). The δ13C:CdW plot allows for the separation of biogeochemical aging (along lines of constant δ13Cas) from mixing of water masses with different δ13Cas signatures (across lines of constant δ13Cas). The LGM Atlantic data can largely be explained by mixing between high-δ13Cas GNAIW and low-δ13Cas AABW. Deep North Atlantic data fall between these two end-members, suggesting some mixing across the bathyal nutricline at ∼2200 to 2500 m. Ninnemann and Charles  reconstructed uniform LGM AABW values of ∼−0.8‰ in the Atlantic sector of the Southern Ocean. Extrapolating the mixing line to this value suggests that the AABW end-member had a Cd concentration of roughly 0.6 to 0.8 nmol kg−1 and a δ13Cas signature of perhaps −0.5 to −1‰. The implication of this analysis is that much of the δ13C depletion in glacial AABW was due to altered air-sea processes. Widespread microhabitat [Mackensen et al., 1993] or CO32− effects [McCorkle et al., 1995] are not required to explain the distribution of δ13C:CdW data in Figure 6.
Although there are no LGM δ13Cas data from the modern core region of AAIW, we note that upper South Atlantic values are generally lower than in GNAIW (Figures 5b and 6). This pattern could be explained by simple mixing between GNAIW and AABW, but we cannot rule out the existence of a negative-δ13Cas AAIW during the LGM.
Most theories put forward to explain the deep water configuration of the LGM Atlantic have relied on changes in buoyancy forcing in the high latitude North Atlantic. That is, fresher and therefore less dense surface waters in the North Atlantic reduced the formation rate and/or penetration depth of NADW/GNAIW [e.g., Weyl, 1968; Broecker et al., 1985; Boyle and Keigwin, 1987]. Ocean circulation models suggest that reduced sea surface salinities and greater sea ice cover in the glacial North Atlantic could cause not only a shallower mode of deep water production, but also a shift in its production site to south of the Nordic Seas [Rahmstorf, 1994, 2002; Ganopolski et al., 1998]. Such a change in deep convection sites would reduce ocean heat transport to higher latitudes, even if the rate of NADW/GNAIW production was unchanged. Alternatively, reduced production rates in the north may have been balanced by increased rates of AABW production in the south [Broecker, 1998]. Although the question of rate cannot be addressed with paleonutrient data alone [LeGrand and Wunsch, 1995], the sharp mid-depth nutricline reconstructed here argues for a significantly altered mode of circulation during the LGM.
It is also possible that surface processes in the Southern Ocean were mainly responsible for the LGM circulation pattern. Longstanding theories of the modern MOC require that deep waters return to the surface by vertical mixing through the thermocline [Stommel, 1958; Stommel and Arons, 1960]. The magnitude of this mixing, which derives its energy from winds and tides, may ultimately determine the global rate of deep water formation [Munk and Wunsch, 1998]. Toggweiler and Samuels [1995, 1998] argue that much of the modern mixing occurs not across the thermocline but in the Southern Ocean, where circumpolar westerlies blowing over the latitude band of Drake Passage force northward Ekman drift that is balanced by deep upwelling. This upwelling, combined with adequately dense surface waters in the North Atlantic, may set the southward flux of NADW. Today net precipitation over the Southern Ocean allows for the buoyant conversion of NADW into AAIW. Keeling and Stephens  hypothesized that during the LGM, deep water temperatures approached the freezing point, so exposure to the Southern Ocean surface resulted in extensive sea ice cover. This sea ice shielded deep waters from precipitation, retarding the buoyant conversion of NADW. Salt became concentrated in AABW and AAIW such that both theoretically became denser than NADW/GNAIW. The upper depth limit of AABW would then correspond to the lower limit of AAIW, which is set by the depth of the seafloor ridges crossing the latitudes of Drake Passage; below this depth topographic boundaries would cause southward geostrophic recirculation of AAIW. Warren  estimated the relevant ridge depth to lie between ∼1500 and 2500 m, which spans the GNAIW-AABW Cd nutricline (∼2200–2500 m) reconstructed here.
Glacial pore water δ18O measurements, combined with benthic foraminiferal δ18O, suggest that the deep ocean was more homogenously cold than today, very close to the freezing point of seawater [Adkins et al., 2002; Schrag et al., 2002]. Pore water chlorinity further suggests that glacial AABW within the Southern Ocean was much saltier and therefore much denser than deep waters in the North Atlantic [Adkins et al., 2002]. Such a reversal of the meridional salinity gradient (modern AABW is fresher than NADW) could be explained by glacial AABW formation by brine rejection beneath extensive sea ice [Adkins et al., 2002]. This scenario is broadly consistent with the Keeling and Stephens  hypothesis of filling the deep Atlantic with dense Southern Ocean waters. However, it is not clear how well the pore water evidence for dense AABW coincides with the Cd and δ13C reconstructions of widespread AABW: deep (4584 m) and mid-depth (2184 m) North Atlantic sites appear to have been equally less dense than AABW, with no clear evidence of a mixing gradient. Additional Atlantic pore water sites may be required to elucidate the glacial spreading of salty AABW.
Benthic foraminiferal Cd/Ca and δ13C both clearly support an LGM expansion of high-nutrient AABW and a replacement of low-nutrient NADW by the shallower GNAIW. Cd/Ca suggests that the main mixing boundary between AABW and GNAIW occurred abruptly between ∼2200 and 2500 m modern water depth. Cd concentration differences between the western and eastern basins of the deep Atlantic may have been diminished during the LGM due to the spilling of AABW over mid-ocean ridges. Combined Cd and δ13C data indicate that the entire glacial Atlantic can be characterized by mixing between high-δ13Cas GNAIW and low-δ13Cas AABW. Waters in the upper South Atlantic may alternately reflect mixing between GNAIW and a relatively low-δ13Cas AAIW. Although we cannot determine whether conditions in the North Atlantic or Southern Ocean were more important for setting the LGM circulation pattern, it is apparent that both GNAIW and AABW formed under different conditions than their modern counterparts. In particular, low δ13Cas suggests that AABW was more poorly ventilated than today, likely contributing to decreased atmospheric CO2.
Cd/Ca data coverage is sparsest in the tropical South Atlantic and the Southern Ocean, with no LGM values poleward of 40°S. Glacial Cd/Ca data from the Southern Ocean are needed to better define the end-member composition of AABW and to determine the status of AAIW. Paired Cd and δ13C data from the upper South Atlantic would be valuable for determining the air-sea conditions that AAIW may have formed under. Improved understanding of benthic foraminiferal trace metal incorporation and potential artifacts is also warranted, especially in the high latitude North Atlantic where correspondence between core top Cd/Ca and modern seawater Cd is often poor. Laboratory culture work could be particularly valuable in this regard. Finally, numerical modeling of paleonutrient distributions is essential for exploring the dynamic and climatic implications of the data.
This paper evolved from an invited presentation by Marchitto at the March 2005 SCOR/IMAGES Workshop on Past Ocean Circulation in Atlanta, organized by J. Lynch-Stieglitz and C. Kissel. We thank E. Clark and C. Wolak for laboratory assistance; R. Came, H. Elderfield, Y. Rosenthal, and R. Zahn for suggestions regarding compiled data; W. Curry and U. Ninnemann for insightful discussions; and C. Charles and an anonymous reviewer for comments that significantly improved this manuscript. Oceanus cores are curated at WHOI with support from NSF, ONR, and USGS, and Robert Conrad and Vema cores are curated at LDEO with support from NSF and ONR. This work was supported by NSF grant OCE-0221979 to Broecker, S. Hemming, J. Lynch-Stieglitz, and Marchitto.