The upper branch of the Atlantic Meridional Overturning Circulation predominantly enters the Atlantic Ocean through the southeast, where the subtropical gyre is exposed to the influence of the Agulhas leakage (AL). To understand how the transfer of Indian Ocean waters via the AL affected the upper water column of this region, we have generated new proxy records of planktic foraminifera from a core on the central Walvis Ridge, on the eastern flank of the South Atlantic Gyre (SAG). We analyzed the isotopic composition of subsurface dweller Globigerinoides ruber sensu lato, and thermocline Globorotalia truncatulinoides sinistral, spanning the last five Pleistocene glacial-interglacial (G-IG) cycles. The former displays a response to obliquity, suggesting connection with high latitude forcing, and a warming tendency during each glacial termination, in response to the interhemispheric seesaw. The δ18O difference between the two species, interpreted as a proxy for upper ocean stratification, reveals a remarkably regular sawtooth pattern, bound to G-IG cyclicity. It rises from interglacials until glacial terminations, with fast subsequent decrease, appearing to promptly respond to deglacial peaks of AL. Stratification, however, bears a different structure during the last cycle, being minimal at Last Glacial Maximum, and peaking at Termination I. We suggest this to be the result of the intensified glacial wind field over the SAG and/or of the invasion of the South Atlantic thermocline by Glacial North Atlantic Intermediate Waters. The δ13C time series of the two species have similar G-IG pattern, whereas their difference is higher during interglacials. We propose that this may be the result of the alternation of intermediate water masses in different circulation modes, and of a regionally more efficient biological pump at times of high pCO2.
As the understanding of the oceans circulation evolves, it becomes more apparent that the transfer of Indian Ocean waters into the South Atlantic via the Agulhas leakage (AL) (Figure 1) plays a key role in determining the variability of the Atlantic Meridional Overturning Circulation (AMOC) [Venegas et al., 1998; Biastoch et al., 2008; Garzoli and Matano, 2011]. It is currently agreed that during the last Pleistocene glaciation, the AMOC was more sluggish [Hesse et al., 2011; Negre et al., 2010] or superficial [Lynch-Stieglitz et al., 2007] than at present. Reduction in turnover strength was most severe during the deglaciation [McManus et al., 2004, Gherardi et al., 2005], before punctually recovering its fashion when climate re-emerged to its interglacial mode. It has been suggested that an explanation of this phenomenon, from hampered deglacial to interglacial restored AMOC, involves changes in the AL input: a key point in the upper branch of the global circulation [Knorr and Lohmann, 2003; Peeters et al., 2004; Bard and Rickaby, 2009; Marsh et al., 2007]. To test this hypothesis, research has shown how this southern connection was strengthened and weakened through time, with fauna [Peeters et al., 2004] and sediment [Franzese et al., 2006] from the Indian Ocean leaking into the Southern Atlantic in concomitance with Pleistocene deglaciations. To further our understanding of the repercussions of this contribution, it is necessary to shift the focus to the portion of the South Atlantic downstream from the AL. It has been shown in lagrangian models how the heat and salinity of the South Atlantic, that is relevant to the surface branch of the AMOC, are contained at the depths of the thermocline, in the South Atlantic Central Water, of which probably 90% originates from the Indian Ocean via the AL “warm route” [Donners and Drijfhout, 2004], in contrast to the “cold route” from the Pacific Ocean, via the Drake Passage. This explains the necessity of studying the thermocline at the junction between the South Atlantic Gyre (SAG) system and the AL area, and of verifying the effect of the “warm route” on the structure of the South Atlantic upper ocean. In this context, the following research questions must be addressed. How did the South Atlantic upper ocean stratification change during Pleistocene climate oscillations? Can we recognize a systematic glacial-interglacial (G-IG) pattern of this region's thermocline? And, would that show synchronism with shifts in the amount of Indian Ocean input?
1.1 Foraminifera Proxy Use
The calcite of planktic foraminifera is commonly used to obtain information about water masses in the geological record. Still, a great limitation of analyzing a single species to characterize the past ocean is that the derived picture is necessarily adimensional, i.e., it fails to capture the vertical structure of the upper ocean, and the gradients of physical-chemical properties that drive circulation.
A number of micropaleontological studies have examined the potential of isotopically analyzing more than one species along sedimentary records [e.g., Ganssen and Troelstra, 1987; Wefer et al., 1996; Lea et al., 2000; Spero et al., 2003], but only in a few studies were interspecific signature differences calculated and utilized as evidence of shifts in upwelling regimes [Whitman and Berger, 1992; Mohtadi et al., 2011] or with the explicit aim of inferring the stratification of ancient oceans [e.g., Douglas and Savin, 1978]. The reliability of the latter method has been testified for the Indian [Williams and Healy-Williams, 1980] and for the Atlantic Ocean [Mulitza et al., 1997; Steph et al., 2009; Simstich et al., 2012]; the outcome of such approaches confirms that analysis on pairs of planktic species can function as proxies for the stratification of the upper water masses.
1.2 Oceanographic Setting
In order to address our questions, we opted for a location that is representative of the SAG, and also under the influence of the AL, downstream the “conveyor belt.” We selected a setting not affected by the upwelling associated with the southwest African margin, which inevitably entangles the oceanographic record.
The central Walvis Ridge is situated underneath the southeast sector of the SAG, at the interface between the SAG and the turbulent ocean portion that has been referred to as Cape Cauldron [Boebel et al., 2003; Gordon, 2003; Giulivi and Gordon, 2006] (Figure 1). The site is crossed at the surface mostly by SAG waters from SW, advected by the South Atlantic current, which bear the influence of the Southern Ocean, and flow to NE, therefore feeding what is considered the oceanic branch of the Benguela Current [e.g., Garzoli and Gordon, 1996; Garzoli and Matano, 2011]. Circulating within the permanent thermocline are primarily Eastern and Western South Atlantic Central Waters [Poole and Tomczak, 1999]; Gordon et al. , have shown that the majority of thermocline waters in the southeast Atlantic is derived from the Indian Ocean. Additionally, an important source of thermocline water at the central Walvis Ridge are the Agulhas rings that, escaping Cape Cauldron, preferentially travel in close proximity to our study site [e.g., Goni et al., 1997; Dencausse et al., 2010; Boebel et al., 2003; van Sebille et al., 2012]; at this point, rings still contain masses of Indian Ocean origin, characterized by temperature and salinity that exert considerable influence on the mean circulation of the southeast Atlantic [e.g., Matano and Beier, 2003; Garzoli and Matano, 2011]. Waters thus formed constitute the early upper branch of the AMOC.
The mixed layer depth varies seasonally, reaching its maximum in (austral) winter (~100 m) and shoaling in summer to a minimum of ~30 m (Figure 2) (World Ocean Atlas 2009, Loncarini et al. ). The permanent thermocline extends to a depth of ~700 m, and its shape does not undergo relevant seasonal modifications, save for a slight cooling at depths of 100 to 300 m in autumn and winter (Figure 2). The low sedimentation rate that derives from the pelagic setting and the oligotrophic conditions of the SAG, has discouraged the generation of paleoceanographic records with a resolution that permits the observation of suborbital scale variability.
With this study, we intend to fill a gap, regarding the current understanding of suborbital dynamic connections between the AL system and the South Atlantic circulation. We present subsurface and thermocline paleoceanographic records of foraminiferal stable isotopes, and we deliver the first upper ocean stratification record of the east SAG, for the last five Pleistocene climate cycles. We show that the rhythmicity of this oceanographic series is powerful on the 100 thousand years (ka) band, and that it reveals a pattern strictly bound to G-IG successions, allowing us to formulate a number of mechanisms that link SAG stratification to the Agulhas leakage and to changes in AMOC.
Piston core 64PE-174P13 (hereinafter 174P13) is 760 cm long, and was extracted from the central Walvis Ridge (29° 45.71′ S, 2° 24.10′ E) at a depth of 2912 m (Figure 1). Sediments are described as carbonate ooze, reflecting the relatively low pelagic sedimentation regime.
Sampling was conducted at 4 cm intervals (and at 2 cm, dependent upon time resolution). Fossil planktic foraminifera Globigerinoidesruber sensu lato (s.l.) and Globorotaliatruncatulinoides sinistral (sin.) were picked from a narrow size fraction, 250–300 µm, to minimize size-related effects that introduce distortions in the data set. This size fraction is free of preadults and, in the case of G. truncatulinoides sin., of individuals that have undergone gametogenic calcification [Duplessy et al., 1981], which would confound the interpretation of their isotopic signal. Additionally, due to the large size variability of G. truncatulinoides sin. in the record, larger shells of this taxon (350–500 µm ) were selected from a small subset of samples (n = 20), to evaluate the implications of shell size and gametogenic calcite for the δ18O signal. This additional analyses yielded values that differ at times notably from the 250–300 µm fraction, generally displaying higher variability (Figure 3), which we attribute to the wider size range of our additional measurements, and to the higher associated variability in the calcification depth. From these results, however, we cannot ascertain a systematic offset between the two size classes (t-value = 0.338, t-probability = 0.739, DF = 19); the null hypothesis that the composition of large and small sized shells is the same cannot be rejected.
Between 35 and 55 shells for each species were crushed, and a portion of approximately 150 µg of homogenized calcite fragments was used for stable isotope analysis. This approach was adopted to maximize the number of shells involved and therefore the analyses' representativeness of the foraminiferal population. Isotopic measurements were conducted in duplicate, using a Thermo Finnigan Delta Plus mass spectrometer equipped with a Gas Bench II preparation device at the VU University Amsterdam. The instrumental reproducibility was routinely monitored using international calcite standards NBS 19 and IAEA-CO-11 [Ishimura et al., 2008] and yielded an average error of ±0.10 ‰ (1σ) for both oxygen and carbon. A correction was applied for external standard offset, to improve interrun comparability.
Due to the inherent variability typical of fossil foraminifera populations [Killingley et al., 1981], some repeats bore anomalously vast discrepancies (>0.2 ‰); in those cases, triplicate measurements were performed. The standard deviation of the duplicates was less than 0.08 ‰ for both species, and for both oxygen and carbon.
From each of the two curves, we subtracted the record of the average contribution of the global accumulation of ice to the mean δ18O of seawater (δ18Osw), also called ice volume effect (IVE). The latter was calculated using model of Bintanja and van de Wal , forced to match the LR04 global foraminifer δ18O stack [Lisiecki and Raymo, 2005]. Prior to subtraction, we converted the IVE data set to the PDB scale. The propagated error is ±0.12 ‰ for both species. The age scale coherence of the model with that assigned to core 174P13 prevents excessive artefacts generated by time offsets. The δ18O curves freed of the IVE (hereinafter δ18O-IVE) therefore represent estimations of local temperature and salinity changes.
The stratigraphy for core 174P13 is based on alignment of our G. truncatulinoides sin. record to the LR04 δ18O stack [Lisiecki and Raymo, 2005] (Figures 4 and 5A). This species was chosen for the age control because, calcifying deeper, it is less prone to incorporate in its δ18O distortions due to the higher temperature and salinity variability of the surface (see rest of Methods), which are not negligible in the subtropics.
2.1 Calcification Depth and δ18O Equilibrium
The choice of the symbiont-bearing G. ruber to obtain a subsurface water signal was motivated by the quantity of studies based on stratified net samples, that confirm its photic zone habitat [Deuser et al., 1981; Lonçaric et al., 2006; Peeters and Brummer, 2002], dictated by the light requirement of its symbionts. Successful use in previous paleoceanographic studies [e.g., Thunell et al., 1999; Dekens et al., 2002; Schmidt et al., 2004] attests ecologic and genetic stability across the Pleistocene and grants direct comparability of results.
From stratified tows at a location in proximity of core 174P13, Lonçaric et al.  assign the base of the productive zone for G. ruber at ~125 m, but found highest shell concentrations more shallow, within the seasonal thermocline. Blooms of this species at the central Walvis Ridge occur in (austral) spring and again in winter [Lonçaric et al., 2007].
Modern developments in foraminifera studies render the distinction among subspecies of this taxon obvious and imperative for proxy-based reconstructions, due to genetic differences [Aurahs et al., 2011] that are manifest in non-negligible geochemical offsets between morphotypes [Steinke et al., 2005; Kuroyanagi et al., 2008; Numberger et al., 2009]. We therefore limited our selection to those G. ruber that correspond to the sensu lato morphotype defined by Wang , as it is the most consistently abundant in the core.
For the completion of its reproductive cycle, the species G. truncatulinoides is considered to be dependent on the presence of a deep thermocline [Lohmann and Schweitzer, 1990]. Paleoceanographers have often recommended its use to record “deep” upper ocean signals [e.g., Mulitza et al., 1997; LeGrande et al., 2004; Steph et al., 2009; Cleroux et al., 2007; Ufkes and Kroon, 2012]. Furthermore, the benefit of using G. truncatulinoides for producing a record of thermocline conditions is well supported by observations from core tops extensively covering the Atlantic Ocean, showing how the calcification of this species does not seem to follow isopycnals or to depend directly on temperature [LeGrande et al., 2004]. We therefore assume that its preferential depth in the record did not depart significantly from today's.
Left and right coiling G. truncatulinoides (sinistral and dextral, respectively) correspond to different genotypes [de Vargas et al., 2001; Quillévéré et al., 2011]. These have been shown to have distinct ecological preferences, in particular with respect to depth habitat [Lohmann and Schweitzer, 1990; de Vargas et al., 2001], different seasonality [Lonçaric et al., 2007], and are therefore expected to produce discordant geochemical records. We thus decided to select exclusively specimens of the sinistral variety, as it is the deepest calcifying morphotype [Lohmann and Schweitzer, 1990; Ujiiè et al., 2010] and therefore more apt to capture signals at the base of the thermocline. For this species, at the core location, the base of the productive zone has been established at ~400 m [Lonçaric et al., 2006], and its highest seasonal abundance in the early summer [Lonçaric et al., 2007].
To count on a first-hand estimate of the calcification depth of these taxa specifically for the study area, we carried out isotopic measurements on foraminifera from the core top and compared those to modern local hydrography.
We calculated the isotopic composition of oxygen in sea water (δ18Osw) calibrating in situ measured salinity with δ18O values determined on water samples from station 154P04 [Lonçaric et al., 2006] (n = 8; R2 = 0.96), adjacent to core 174P13:
where the −0.27 ‰ correction is introduced to convert SMOW scale to PDB [Hut, 1987]. We based the calculation of the equilibrium calcification on the empiric relationship of Kim and O'Neil  for inorganic calcite precipitation:
where α is the fractionation factor between calcite and water, and applied its quadratic approximation of Bemis et al. :
where δeq is the isotopic composition of calcite in thermodynamic equilibrium. Solved for δeq, it yields:
Interpolating modern water column hydrography, we can estimate the apparent calcification depth from the core top δ18O of the two species. Applying data from World Ocean Atlas 2009 [Locarnini et al., 2010; Antonov et al., 2010] for the month of major occurrence of each species, calcification depth is 150 ± 43 m for G. ruber s.l. (October), and 525 ± 50 m for G. truncatulinoides sin. (December) (Figure 2). When using to locally measured δ18Osw [Lonçaric et al., 2006], the depth of G. ruber s.l. is 155 ± 26 m, and that of G. truncatulinoides sin. is deeper than 750 m. The standard deviation is calculated based on multiple δ18O measurements in the upper centimeters of the core. Although it is unrealistic to presume that calcite tests are formed at a discrete depth in the ocean, taking into account the vastly incomplete knowledge about the elusive depth at which foraminiferal shells are secreted, we still estimate that our two species record paleoceanographic signals ~350–400 m apart in the water column.
To illustrate the present-day pattern of South Atlantic upper ocean stratification, as close as possible to how it is incorporated in our record, we calculated the difference between the δ18Oeq at 500 m and at 100 m, using temperature and salinity data from the World Ocean Atlas 2009 [Locarnini et al., 2010; Antonov et al., 2010], and applying the aforementioned equilibrium equation, and a South Atlantic salinity-to-δ18O relationship [slope = 0.51, intercept = 17.40, n = 61, LeGrande and Schmidt, 2006]. These values are reported as contoured colors in Figure 1.
2.2 Spectral Analysis
The time series produced were analyzed for their spectral components. To verify in which measure the spectra are influenced by the orbital tuning of our record (the L04 stack is in fact orbitally aligned [Lisiecki and Raymo, 2005]), we also computed them after alignment of our δ18O to the planktic chronostratigraphy of Huybers  (H07), which does not rely on orbital assumptions. As a further means of control on the robustness of our analysis with respect to the choice of the age model, we calculated spectra also on the time series with age model obtained through tuning of G. ruber s.l. δ18O to the LR04 stack. The spectral results discussed in this work are valid regardless of the age model.
Core 174P13 spans the past ~460 ka. The δ18O curve of both species predominantly displays the imprint of the last five G-IG cycles (Figure 5A). Spectral analysis of these series reveals identical significant peaks, at 100 and 41 ka (Figure 7A). The 100 ka band of G. truncatulinoides sin. is evidently more pronounced than that of G. ruber s.l.
3.1 Globigerinoides ruber s.l. δ18O-IVE
The IVE-free profile of this species presents variability at the sub-G-IG scale in the order of ~0.7 ‰ (Figure 6B). Underlying this pattern, it is visible a long-term decrease that continues from MIS 12 to MIS 7. No structure is identifiable that replicates across the five G-IG cycles. Despite that, terminations are characterized by the occurrence of low peaks. Such shifts are very sharp at termination (T) V (~1 ‰) and T I (0.5 ‰). They begin earlier and have a more gradual development, prior to T IV, III, and II. In the spectrum of G. ruber s.l. δ18O-IVE, the 41 ka obliquity band is statistically significant (Figure 7B).
3.2 Globorotalia truncatulinoides sin. δ18O-IVE
Maxima of the δ18O-IVE series of this taxon fall at ~440, 360, 270, 145, and 27 ka before present (BP), thus not following the G-IG cycles (Figure 6D). They reveal instead a pointed pulse structure that does not normally appear in oceanographic records, and that is considerably similar to the eccentricity parameter (Figure 6C–D). The spectrum of the δ18O-IVE series shows significant periodicity at ~100 ka, typical of the combination of the orbital components of eccentricity (Figure 7B).
3.3 Δ18O Difference
The difference in oxygen isotopes composition between the two species (from here Δ18O) is on average 0.91 ‰, and ranges between 0.35 and 1.73 ‰ (Figure 5B). This composite signal builds up starting from early interglacial stages, reaching maxima precisely at the subsequent glacial terminations, thus diminishing relatively fast, with minima set just upon establishment of full interstadials. For the secondary climatic cycle represented by the evolution from MIS 4 to 2, this is not equally explicit.
The strongest postdeglaciation Δ18O decrease is observed at early MIS 5 (1 ‰ drop in ~8 ka), followed by MIS 11 (0.7 ‰ in 6 ka), MIS 7 (0.6 ‰ in 10 ka), MIS 9 (0.5 ‰ in 10 ka), and MIS 1 (0.4 % in 5 ka). After minima, values increase gradually, though superimposed events of suborbital scale are visible, during intervals that last from a minimum of ~77 ka, during MIS 11 to 10, to a maximum of ~108 ka during MIS 5 to 2. The duration of the Δ18O buildup is consistent across glacial cycles, each regularly constituting ~88–90 % of the whole duration of the respective G-IG succession. Along MIS 5, values gently increase as in the precedent cycles, but then drop, rise, and fall during MIS 4, MIS 3, and early MIS 2, respectively, steeply increase before T I, and again decrease along MIS 1.
The spectrum of Δ18O is significant on the period of ~100 ka, corresponding to the eccentricity main band (Figure 7C). The alternative age model (based on H07, see Methods) yielded both eccentricity peaks that are more marked, thereby proving that the observed peaks are not artefacts of the orbital tuning of LR04.
3.4 δ13C and Δ13C
The variability in G. ruber s.l. δ13C is larger than that of G. truncatulinoides sin. (Figure 5E), and values of the earlier are almost always higher than the latter's. Besides these differences, the curves of the two species are coupled, and the following considerations hold for both. The heaviest carbon was incorporated in MIS 11 and secondarily in MIS 9. A decreasing trend is indeed visible from MIS 12 to 8 (though G. truncatulinoides sin. records a more accentuated decrease). This is followed by a mildly increasing tendency from MIS 6 to 1. To these trends, a general G-IG pattern is superimposed, with δ13C values being heavier during interglacials and reaching their minima prior to terminations. Exceptionally, from the mid-late MIS 9 to early MIS 8, δ13C maintains the elevated values it had reached at ∼ 325 ka BP.
G-IG patterns are less evident for the Δ13C curve than for Δ18O (Figures 5B and 5F). Still, it is possible to note that the subsurface-to-thermocline difference is more marked in interglacials, and that there is clear alternation of long-sustained Δ13C decreases, from each mid-late interglacial until the following late glacial, and successive more rapid increases which, across terminations, develop into full interglacials.
4.1 Subsurface Signal
Our analysis confirms observations from the southeast Cape Basin, that the coldest sea surface temperature (SST) over the last 460 ka was recorded during late MIS 12 (~435 ka BP), and that the interglacial with longer sustained SST was MIS 11 [Pierre et al., 2001; Dickson et al., 2010] (Figure 6B).
The shift of ~0.5 ‰ in δ18O-IVE prior to T I, if interpreted solely in terms of temperature change, is fully compatible with the regional SST warming of 2 °C inferred for the last deglaciation from faunal assemblages [Niebler et al., 2003], and from a recent integrated model-data study [Annan and Hargreaves, 2013]. The only other relatively fast subsurface warming/freshening happened before T V (~1 ‰). The similarity in upper ocean hydrographic changes between these two transitions generally echoes the parallelism previously highlighted between the two following interglacials (MIS 11 and 1, respectively), which bore similar orbital configuration [e.g., Berger and Loutre, 2003; de Abreu et al., 2005]. Minima in our record appear aligned to maxima in the tilt of the Earth's axis (obliquity), in particular across terminations (Figures 6A and 6B). The spectrum of G. ruber s.l. δ18O-IVE corroborates this notation, with the ~41 ka band emerging as significant (Figure 7B). Obliquity has recently been suggested as a plausible forcing for T I and II [Drysdale et al., 2009] and for the entire Pleistocene [Huybers, 2007]. It determines the total combined radiation received by the high latitudes of the two hemispheres, and it turned out to be an important component of Antarctic paleo records [Jouzel et al., 2007]. This orbital parameter was shown to pace temperature and salinity changes in the Agulhas Current [Caley et al., 2011], and it is a component of the release of AL in the east Cape Basin [Peeters et al., 2004]. We show that it also influences the subsurface at the central Walvis Ridge, mediated by meridional shifts of the frontal systems at higher latitudes. Such shifts have been postulated for G-IG time scales to result from the intensity of the latitudinal insolation gradient [Hays et al., 1976b; Bard and Rickaby, 2009], in turn paced by obliquity. The associated contraction and expansion of Antarctic sea ice can regulate the (sub)tropical SST, as modeled by Lee and Poulsen .
The warm/fresh excursions from G. ruber s.l. at terminations were also recorded by winter SST maxima from transfer functions in the Namibian upwelling, 10°E of our site [Chen et al., 2002], even though we detect close structural resemblance to their series only for T V and MIS 3 to 1. Whereas from the record of Chen et al., , it could be incautious to extract SSTs from faunal abundances, due to the implications that the possible transport of G. ruber by the AL has on the transfer function, our reconstruction provides a geochemical hint for southeast Atlantic upper ocean warming. Mix et al.  showed how surface warming of the southeast tropical Atlantic corresponded to the last deglacial interval, laying down the first formulation for what became later known as the bipolar seesaw concept (e.g., Broecker ). This is ultimately linked to the strong reduction of the AMOC, brought about by melt water discharge in the North Atlantic [Vellinga and Wood, 2002], as can be seen from the first derivative of the LR04 benthic stack (Figure 5D), seen primarily as a proxy for the rate of ice volume change. We hence suggest that a response of the southeast Atlantic to this mechanism was a constant mode of G-IG climate change of the last 460 ka. Further, our subsurface record also supports the notion of increased efficiency of Indian to Atlantic heat transport at glacial terminations [Knorr and Lohmann, 2003; Peeters et al., 2004].
4.2 Thermocline Signal
The high δ18O-IVE, likely pointing at remarkably low temperatures at MIS 12, and the sustained climatic optimum of MIS 11 interglacial, from the G. ruber s.l. δ18O-IVE and from other studies [Hodell et al., 2000; Dickson et al., 2010], are even more manifest in the G. truncatulinoides sin. δ18O-IVE (Figure 6D), meaning that the intensity of MIS 12 glacial was (regionally) more pronounced in thermocline waters than at the subsurface. This is in agreement with the lowest values of AL fauna from Peeters et al. , which indicates minimal connection between Indian and South Atlantic Oceans, and thus suggests a potential response of our record to the AL.
The varying position of maxima and minima of the δ18O-IVE of this species, with respect to G-IG cycles, leads us to deduct that Pleistocene climate shifts do not form an exhaustive explanation for variations in water properties of the east SAG thermocline.
We note that, while tropical intermediate depth warming during likely weakened AMOC (i.e., terminations) was reported for the low-latitude tropics during the last two glacials [Ruhlemann et al., 2004; Lopes dos Santos et al., 2010] and for a more coastal record in the southeast Atlantic during MIS 12 [Dickson et al., 2010], our G. truncatulinoides sin. does not clearly reflect this mechanism, and its extension to the southeast Atlantic gyre for the last five deglaciations is not possible. Our findings in this sense seem in line with the temperature insensitivity to changes in the AMOC recently found for the tropical Atlantic thermocline [Huang et al., 2012].
Examining the relationship between thermocline δ18O-IVE signal and eccentricity (Figures 6C and 6D), we propose a mechanism of eccentricity-paced regulation of the southeast Atlantic properties of thermocline waters, through changes in the tropical seasonal contrast. As proposed by Berger et al. , there is a strong 100 ka cycle in the equatorial seasonality that emerges from the difference of the maximum minus minimum insolation record. This modulation of seasonality is noticeable in the tropics, and we show how it seems to have a profound impact on thermocline heat content over the last 460 ka.
4.3 Upper Ocean Stratification
Each species' isotopic composition is a result of the effects of ice volume, temperature, and local seawater anomaly, ultimately attributable to salinity [Emiliani, 1955]. Because the IVE is the same in two coeval populations, the magnitude of Δ18O is governed by the difference between the temperatures recorded by the respective species, plus the difference between the δ18O of seawater (dependent on salinity).
To interpret the Δ18O, we intend to deal with the complications that commonly affect the use of foraminiferal isotope records. By excluding some of them, we wish to assign a clearer paleoceanographic meaning to the isotopic difference. In the first place, close inspection of assemblages from the 63–125 µm fraction reveals that the shells of delicate juvenile specimens are pristine throughout the core, hence dismissing the possibility that taxon-specific dissolution played a role in scaling the Δ18O (and the Δ13C), as was to expect from the results of Howard and Prell , which constrain the glacial Cape Basin lysocline shoaling. Even though the two species have different seasonal occurrence [Lonçaric et al., 2007], a change in the strength of the annual temperature gradient cannot account for a significant portion of the observed changes in Δ18O. In fact, fauna-based reconstructions of Niebler et al.  show that seasonal SST differences in the southeast sector of the South Atlantic remained reasonably stable from the Last Glacial Maximum (LGM) to present. Nevertheless, there is to date no reconstruction of this pattern for older G-IG cycles. Another eventuality that requires attention is that the depths, to which the Δ18O is referred, might not be stable over the last 0.5 million years. As mentioned before, we are keen on interpreting the record of G. truncatulinoides sin. in terms of changes in water mass properties at its presently observed calcification depths, neglecting distortions to the signal enforced by relevant habitat shift, that probably did not occur [LeGrande et al., 2004]. As for the changes in the depth of G. ruber s.l., other studies have assessed its habitat stability for the last G-IG cycles [Spero et al., 2003]. Moreover, the interpretation of the observed increase in Δ18O (Figure 5B) as reflecting habitat migration of G. ruber s.l., pursuing warmer conditions in more shallow layers as glacial temperatures deteriorate, does not seem plausible for three reasons. First, is counter-intuitive that upward migration went as far as to reach levels warmer than at climate optima; second, we examined distribution maps for this species at our location (Arthur et al., manuscript in preparation), and found no significant difference between its present and LGM abundance, meaning that, for either climatic state, G. ruber s.l. did not undergo stress levels that warrant meaningful habitat migration; last, because this interpretation does not conciliate with the timing of SST changes recorded elsewhere in the southeast Atlantic [Chen et al., 2002; Peeters et al., 2004].
On another note, to understand the influence of size-related variations in the δ18O of G. truncatulinoides sin. on the Δ18O (see Methods), we show that the Δ18O calculated between the larger shells of this species and G. ruber s.l. indeed has a less clear pattern (Figure 3). This demonstrates that, by opting for a narrow size fraction of G. truncatulinoides sin., we likely targeted a more defined gradient than we would have by picking larger individuals, whose habitat depth is considered less stable, and that have occasionally developed secondary calcite crusts [Mulitza et al., 1997].
Taking into account the aforementioned, we interpret the evolution of the Δ18O as a proxy for the stratification of the permanent thermocline. Further, being foraminiferal δ18O, as density, directly proportional to salinity and inversely to temperature [Lynch-Stieglitz et al., 1999], we take Δ18O as a qualitative estimation of density stratification.
Values of Δ18O are quite low, when compared to what found between surface and thermocline dwellers, in G-IG records from other tropical locations [Wefer et al., 1996; Spero et al., 2003], confirming that along our record, the southeast Atlantic maintained a deep thermocline, as it is suggested by the constant presence of G. truncatulinoides sin. [Lohmann and Schweitzer, 1990].
Glacial values are statistically different from interglacial ones, with high significance (p <0.001). The sawtooth pattern reveals that stratification was lowest (i.e., deepest thermocline) in interglacials and maximal (thermocline most shallow) at terminations. Note that only the LGM deviates from this scheme, being characterized by a minimum in stratification, and alluding to the exceptional spatial/temporal structure of the last transition. Two different mechanisms must therefore account for changes in the last glacial cycle and in the four preceding ones. We first proceed by proposing a potential explanation for the former.
Several independent proxy [Stuut et al., 2002; Moreno et al., 1999; Lambert et al., 2008; Bard and Rickaby, 2009] and model [Clauzet et al., 2007] reconstructions have pointed out that southern hemisphere easterlies and westerlies blew more intensely during the LGM, as also reflected in stronger upwelling east of our site [Chen et al., 2002]. Intensified zonal winds imply enhanced Ekman pumping in the subtropical gyre, at the junction of the two winds systems. This mechanism appears to be reflected in our record, where stratification is minimal at the LGM, when Ekman-induced downwelling was supposedly at its peak, and then swiftly increased before T I, when winds strength waned.
To assist the interpretation of the older G-IG cycles, we recur to comparison with modern observations. As can be seen in the South Atlantic map in Figure 1, between 20°S and 50°S, the present-day δ18Oeq difference between 500 and 100 m (see Methods) varies largely with latitude, with values decreasing southwards, while north of 20°S, the gradient rather follows the direction of the subsurface South Equatorial current. In this view, we can infer, for periods of Δ18O buildup: (1) a southeastern expansion of the higher stratification zone towards the core site, with subsequent rapid retreat; (2) a gradual rearrangement of the stratification pattern in the South Atlantic basin, with displacement of highly stratified thermocline to the southeast; or a combination of the two. Either of the envisioned scenarios implies substantial alteration of the southern hemisphere AMOC.
Relying on modern ocean observations, the Agulhas Current is clearly identifiable on the basis of its stratification values, markedly lower compared to the surrounding region (Figure 1). This feature is explicit also from observational studies [Lutjeharms, 2006] and is furthermore seen in Agulhas rings entering the South Atlantic [Arhan et al., 2011; Souza et al., 2011], where the anticyclonic rotation depresses the isopycnals. Although the mesoscale processes involved in the encounter of Indian Ocean masses with the South Atlantic are characterized by high complexity [e.g., Biastoch et al., 2008], we formulate a connection of the changes in the stratification of the SAG with the AL. To secure the synchronicity of our record with that of the AL fauna [Peeters et al., 2004], we tuned the δ18O of the latter to the LR04 stack. We propose a cause-effect relationship between the release of poorly stratified waters from the AL at glacial terminations, and the relatively fast decrease of stratification in our SAG record (Figures 5B and 5C). The conceptual succession of events is hypothesized as follows: when the AL “warm route” becomes weaker or, alternatively, loses its characteristic deep thermocline (i.e., it becomes more stratified), stratification at our site progressively builds up, until terminations. At that point, the powerful release of AL distributes into the South Atlantic poorly stratified waters characterized by lower thermocline density [Arhan et al., 2011], thereby making our Δ18O values drop along an interval of 6–10 ka . In other words, by evaluating the fingerprint of thermocline water masses from modern observations, we are able to propose a mechanism of the effect of the AL in the upper South Atlantic Ocean hydrography.
In turn, the AL fauna appears tightly coupled to the rate of ice volume change (Figures 5C and 5D). Remarkably, maxima in the rate of ice volume change, indicating melt water input at terminations, coincide both with SAG stratification drops and with AL peaks. This suggests that fresh water discharge in the North Atlantic caused a southern displacement of southern hemisphere fronts [Timmermann et al., 2007] visible in increased Indian to Atlantic communication south of Africa, coherent with the ideas of, e.g., Hays et al. [1976b] and Bard and Rickaby , thus assigning emphasis to the role of the interocean exchange in the termination of glacial periods.
4.4 Periodicity of Stratification
The spectral peak of the Δ18O at 100 ka compares with the G-IG cycle, but also with orbital eccentricity (Figure 7C). It is renown that eccentricity effect on insolation is too mild to noticeably force the paleoclimate record (the “100 ka problem” [Hays et al., 1976a]). Still, it is important to keep in mind that the eccentricity cycle functions as a modulator of precession amplitude.
Special interest in the detection of typical G-IG 100 ka periodicity in the Δ18O curve arises from acknowledging that, as we have explained, this signal is per definition independent of IVE, which commonly introduces the G-IG pace into foraminifera δ18O records. In this light, we detect in the Δ18O climatic evolution evidently tied to G-IG cycles, but that is solely explained by changes in water column structure.
Investigating on which of the individual δ18O curves that compose the Δ18O is accountable for this pacing, we underline that neither exhibits univocal connection to G-IG cycles. In G. ruber s.l. δ18O-IVE, the 100 ka band is not present at all, while obliquity is significant (Figure 7B). Even though the spectral analysis of G. truncatulinoides sin. δ18O-IVE reveals very strong periodicity at ~100 ka (even stronger when tuned to the H07 chronostratigraphy), the structure of this series does not coincide with G-IG cycles as defined by the δ18O stacks (Figure 6D). Considering the aforementioned, the powerful G-IG periodicity of our Δ18O record is attributable to the thermocline signal. Further research on thermocline water masses, covering several G-IG cycles, is required to disentangle the forcing of orbital parameters on upper ocean stratification at the subtropical latitudes.
4.5 Mechanisms Determining the 13C Gradient
The higher interglacial δ13C values, and their low glacial counterparts (Figure 5E), are consistent in magnitude with the carbonate ion effect described by Spero et al. , and with the 40 µmol/kg G-IG shift in [CO32−] suggested by these authors, and with the recent finding that Southern Ocean alkalinity was higher during glacials than interglacials, over the time span of core 174P13 [Rickaby et al., 2010]. The decreasing trend of MIS 12 to 8, also visible in benthic δ13C from the southeast Cape Basin [Pierre et al., 2001], and in the Pacific Ocean [e.g., Mix et al., 1995], corresponds to the mid-Brunhes climatic event [e.g., Jansen et al., 1986]. The heaviest values of both species at MIS 11 are in agreement with planktic records in the Southern Ocean [Hodell et al., 2000]. Notably, in late MIS 9 and early MIS 8, δ13C maintains the high values it had reached at ∼ 325 ka BP. This plateau is observable also at similar longitude at 42.5°S [Hodell et al., 2000] and therefore probably indicates long sustained productivity at the junction of Southern and Atlantic Oceans. Further, the absence of resemblance of our Δ13C curve (Figure 5F) to that of site 1087 [Pierre et al., 2001] tells that our results are not extendable to the upwelling area further to the east, but seem representative of the low productivity region trapped in the SAG.
Several mechanisms might have concurred to determine the discrepancies in the δ13C of our two species, and the resulting Δ13C. In the first place, the mentioned carbonate ion dynamics vary with water depth, and their effect on foraminifera δ13C was shown to be species specific, likely due mainly to the activity of symbionts [Spero et al., 1997]. To our knowledge, there is no such published study dealing with G. truncatulinoides, and exploration of the differential effect is thus so far precluded.
We therefore discuss two causes that likely had large impact on the Δ13C. It is currently understood that, in the framework of the reorganization of water masses in the glacial Atlantic, Antarctic Intermediate Waters (AAIW) were supplanted, at their present depth, by Glacial North Atlantic Intermediate Waters (GNAIW) [Curry and Oppo, 2005; Lynch-Stieglitz et al., 2007], with heavier δ13C. Evidence of isotopic signature changes at the latitudes and depth of core 174P13 is only available from benthic results in the west of the basin, basically indicating that the GNAIW did not reach 30°S at the Brazil margin [Curry and Oppo, 2005]. In the absence of indications for the eastern longitudes, it is difficult to draw the geometry of the southern invasion of GNAIW. Nevertheless, it is plausible to postulate that similar water masses played a role also during older glacials. If indeed, such water invested the southeast Atlantic, it would have been sensed by our G. truncatulinoides sin. in the low part of its calcification depth range [Mulitza et al., 1998a], thereby recording a signal enriched in 13C relative to the subsurface. This is sustained by the lowest values of Δ13C reached during the five glacials embraced by 174P13. In addition, this hypothesis of southern extension of GNAIW seems coherent with a more uniform upper water column, and a minimized density stratification seen in our Δ18O for the LGM, whereas it does not seem to have prevailed in dictating the Δ18O stratification in older glacials.
Still, the graduality of the Δ13C increases across terminations requires another explanation than merely the GNAIW to AAIW shift, which must have occurred faster. The higher variability of G. ruber s.l. δ13C with respect to G. truncatulinoides sin. (Figure 5E), similar to what reported for the north edge of the SAG (core GeoB 1413 [Wefer et al., 1996]), possibly indicates that this species captures the productivity-induced effects on seawater 13C more efficiently. Likewise, the constantly higher subsurface values, compared to the thermocline, reflect the production-respiration budget, which differs in the two layers, and the photosynthetic activity of symbionts hosted by G. ruber s.l. [Spero, 1998]. Generally, heavier δ13C corresponds to enhanced phytoplankton production. Because of these dynamics, the surface to intermediate 13C gradient seems to function as a proxy for the vertical nutrient gradient in the eastern tropical Atlantic [Mulitza et al., 1998b]. Given the decreasing gradient of the ratio between production and respiration, from the photic zone to the less productive thermocline, we see how the Δ13C might also provide a qualitative measure of the intensity of the “biological pump”. It appears therefore that this process worked best during climate optima, generally losing efficiency as glacials progressed into their coldest phase. This allows the hypothesis that at times of higher pCO2 [Luthi et al., 2008], this region of the ocean functioned more efficiently as atmospheric carbon sink.
Unfortunately, a complete clarification of the predominant causes of the Δ13C shifts is still impeded by the not yet resolved geometry of the GNADW southward extension, not to mention that of similar north-source water masses in older glacials. Also necessary are more strict constrains on past thermodynamic processes determining the 13C of intermediate waters upon interaction with the atmosphere [Charles et al., 1993].
5 Summary and Conclusions
We produced O and C isotope records of subsurface and thermocline dwelling foraminifera, covering the last five G-IG cycles, for a sector of the ocean that links the AL with the AMOC, previously undocumented at these time scales.
Subsurface low δ18O-IVE values indicate warming/freshening and appear paced by obliquity, suggesting response to high latitude dynamics. Their coincidence with glacial terminations testifies a response of this region to the bipolar seesaw, and is compatible with at SAG warming resulting from increased AL input.
In contrast, the SAG thermocline δ18O does not parallel G-IG successions, whereas it is rather anchored to eccentricity cycles.
MIS 12 appears the most extreme glacial of the last five, with exceptionally heavy thermocline δ18O values. The SAG seems to record the minimal transfer of warm Indian Ocean waters seen from the east Cape Basin.
The upper ocean Δ18O gradient is a proxy for density stratification, revealing a sawtooth pattern that is strictly bound to G-IG cycles, as also evident in the frequency spectrum. From MIS 12 to 5, stratification was minimal during interglacial optima, increased until the following termination, and then dropped relatively fast. We propose that the gradual increase in stratification indicates a progressive loss of Indian Ocean input to the South Atlantic from the AL and that the quick drop, well synchronized with intense AL release at terminations, depicts a SAG thermocline readily replenished by poorly stratified waters from the Agulhas Current.
Exceptionally, stratification was minimal during LGM. Given the availability of paleo records for this period, we are able to provide potential mechanisms for this observation, namely enhanced Ekman downwelling forced by the intensified wind field over the SAG, and/or southward invasion of GNAIW. Both mechanisms imply profoundly altered LGM South Atlantic circulation.
The Δ13C pattern seems to reflect the substitution of GNAIW to AAIW, repeatedly over the last five glacials, and/or a more efficient biological pump at interglacials, lending support to the idea of increased surface productivity at times of high pCO2.
Our study depicts the southeast Atlantic as a sensitive area for dynamics of G-IG transitions and provides a mechanism for the connection between SAG and AL. We underline that to reconstruct past inter-basins communication, there is critical need for more records that, reaching below the surface ocean, should target thermocline stratification in key areas, such as the oceanographic bottlenecks of the AL, the Drake Passage and the North Brazil Current, and in other sectors of the SAG.
The work described in this paper has received funding from the European Community's Seventh Framework Program FP7/2007-2013 – Marie-Curie ITN, grant agreement 238512, GATEWAYS project. We thank Geert-Jan Brummer for making the core available, and Ralph Schneider for his advice on the drilling location. We acknowledge an anonymous reviewer and editor Christopher Charles for their contributions, which greatly improved previous versions of the manuscript.