4.1. South Atlantic G. inflata Apparent Calcification Depth and Mg/Ca-Calcification Temperature Calibration
 The apparent calcification depth of G. inflata from the South Atlantic calculated for nonencrusted specimens from the 315–400 μm fraction is 387 ± 179 m. Plankton tow studies performed off southwest Africa [Lončarić et al., 2006; Wilke et al., 2006] showed that G. inflata occurs over a wide range of water depths with maximum abundance in the thermocline. Based on G. inflata δ18O values Lončarić et al.  showed that calcification also occurs over a large range of water depths with a mean apparent calcification depth of ∼250 m. Elderfield and Ganssen  assigned an apparent calcification depth for G. inflata of 300–400 m, based on core tops from the North Atlantic.
 In contrast to these findings, Cléroux et al.  showed after analyzing a collection of core tops from the North Atlantic that G. inflata mainly records conditions at the base of the seasonal thermocline (<100 m) and only descends deeper in the water column when temperatures at that depth are above 16°C. One explanation for this apparent discrepancy is that G. inflata occupies a different depth habitat in the North Atlantic in comparison with the South Atlantic [Bé and Tolderlund, 1971]. But, this does not explain the differences between the results of Cléroux et al.  and Elderfield and Ganssen  that are both based on North Atlantic samples. A potential explanation for this difference could be a different encrustation state of the analyzed specimens, which was not included in these studies. This apparent disagreement between different studies suggests the importance of defining a clear and narrow state of encrustation of the G. inflata specimens to be used for proxy analyses.
 Elderfield and Ganssen , Anand et al. , and Cléroux et al.  already reported Mg/Ca measurements on recent G. inflata specimens (Figure 3 and Table 2) from the North Atlantic. Our calibration is very similar to the one of Elderfield and Ganssen  and to the one in which the slope was fixed at 0.09 [Anand et al., 2003] extending the Mg/Ca calibration to colder temperatures by ∼5°C. The comparison of our equation with the calibrations of Cléroux et al.  and of Anand et al.  in which the slope was not fixed, however, shows significant dissimilarities. The calibrations from Cléroux et al.  and Anand et al.  not only show a lower temperature dependency than our equation but the absolute Mg/Ca is offset from ours by 0.7–1.0 mmol/mol at a temperature of 15°C. These differences could also be related to the possible existence of different genetic types of G. inflata for the North and the South Atlantic. Different genetic types have been determined for many planktonic foraminiferal species [Darling and Wade, 2008, and references therein]. Although only one genetic type of G. inflata is known yet [de Vargas et al., 1997], different genetic types for another Globorotalia species, Globorotalia truncatulinoides, have been described [de Vargas et al., 2001]. Also, for Neogloboquadrina pachyderma different genetic types for the North and South Atlantic were determined [Darling et al., 2004].
Table 2. Mg/Ca-Temperature Calibration Equations for Several Deep-Dwelling Foraminifera, Source of the Analyzed Foraminifera, Size Fractions, and Temperature Range of the Calibrations
|Species||Curve Code in Figure 3||Source||Size Fraction (μm)||Aa||Ba||r2||Temperature Range (°C)||Reference|
|G. inflata||1||surface samples||n.a.||0.49||0.10||n.a.||7.5–15||Elderfield and Ganssen |
|G. inflata||2||sediment trap||350–500||0.56||0.058||0.55||15–21||Anand et al. |
|G. inflata||3||sediment trap||350–500||0.299||0.09||n.a.||15–21||Anand et al. |
|G. truncatulinoides (dextral)||4||surface samples||n.a.||0.355||0.098||0.92||7–23||McKenna and Prell |
|G. inflata||5||surface samples||355–400||0.71||0.06||0.72||10.5–17.9||Cléroux et al. |
|G. truncatulinoides/G. crassaformis||6||surface samples||355–400||0.84||0.083||0.72||8–15||Regenberg et al. |
|G. inflata||7||surface samples||315–400||0.72b||0.076b||0.81||3.1–16.5||this study|
 Further comparison with Mg/Ca calibration equations for other deep-dwelling planktonic foraminifera, such as G. truncatulinoides and Globorotalia crassaformis, shows a roughly similar picture, although differences are present in absolute values, presumably pointing to interspecies differences and varying states of encrustation (Figure 3) [McKenna and Prell, 2004; Regenberg et al., 2009].
4.2. Potential Bias Caused by Different Size Fractions and States of Encrustation
 As G. inflata calcifies over a large depth range, Mg/Ca represents an average signal over this depth range. As the specimens descend through the water column they also acquire a calcite crust recording lower temperatures than the primary calcite. Hence, larger specimens are expected to contain a larger portion of calcite crust and lower Mg/Ca. Hathorne et al.  showed for G. inflata specimens from a North Atlantic sediment trap that Mg/Ca of the primary calcite is 2–3 times higher than the calcite crust. Cléroux et al. , on the other hand, analyzed two different size fractions of G. inflata, 250–315 μm and 355–400 μm, indicating that no significant difference in Mg/Ca was present between both size fractions.
 In this study we extended the range of size fractions to detect potential biases in Mg/Ca, though always selecting specimens with the same state of encrustation (defined as nonencrusted) as used for the calibration. Additionally, we also included samples with heavily encrusted, shiny specimens. An increase in size fraction is systematically related to a decrease in Mg/Ca (Figure 5). Mg/Ca in specimens <250 μm is warmer (2.0°C on average) than the nonencrusted 315–400 μm specimens. Mg/Ca from the 250–315 μm fraction (+0.7°C) is most similar to the 315–400 μm fraction, which is in agreement with Cléroux et al. . Lowest Mg/Ca is recorded by the largest specimens (>400 μm) and heavily encrusted specimens, which deviate 2.2°C and 4.4°C, respectively, from the fraction used for our calibration equation (315–400 μm). This experiment suggests that with increasing size the proportion of calcite crust increases, which was also shown for several other planktonic foraminifer species [Caron et al., 1990] indicating that the specimens calcified a larger fraction of the total test mass deeper in the water column. Therefore, their average chemical signature represents deeper conditions in the water column if compared to smaller, nonencrusted specimens. We state that careful and consistent selection concerning size and state of encrustation of specimens of G. inflata is essential for a reliable reconstruction of paleotemperatures.
Figure 5. G. inflata Mg/Ca and temperatures for different size fractions (<250 μm, 250–315 μm, 315–400 μm, and >400 μm) of nonencrusted specimens and for encrusted specimens (315–400 μm) of gravity core PS2495-3, with the dashed line representing the downcore record determined on the 315–400 μm fraction. Analyses on more than one size fraction/encrustation stage were performed for the Holocene, MIS2, MIS5, and MIS6. The vertical black bar depicts the error (±1σ) associated with our Mg/Ca-calcification temperature calibration equation.
Download figure to PowerPoint
 It is important to note that even though the temperature bias is calculated using our new Mg/Ca-calcification temperature calibration, that equation is not necessarily applicable to the other size fractions and encrusted specimens. Calcite crust was suggested to have a different temperature dependency than primary calcite [Bentov and Erez, 2006] and, therefore, when different portions of calcite crust are present also different temperature dependencies would apply.
4.3. Downcore Record of PS2495-3: A Test Case Application
 In order to illustrate the application of our Mg/Ca-calcification temperature calibration of G. inflata as a recorder of permanent thermocline temperatures we established a downcore Mg/Ca record spanning the last two glacial-interglacial cycles for core PS2495-3 in the central South Atlantic (Figure 1a). This site has been the focus of several paleoceanographic studies [Mackensen et al., 2001; Gersonde et al., 2003, 2004] (Figure 4) providing a well-dated stratigraphic framework.
 Presently, core PS2495-3 is located within the SAZ. Data from the World Ocean Atlas show that there is a temperature gradient between the SAZ and the PFZ of ∼5°C at 350–400 m water depth [Locarnini et al., 2006]. The reconstructed temperature change at core PS2495-3 over Termination I is ∼8°C (Figure 4). This value is similar to the one reconstructed for the sea surface based on foraminifera transfer functions [Gersonde et al., 2004]. Temperature reconstructions for midlatitude South Atlantic sites not under the influence of migrating oceanic fronts show changes in SST over Termination I of 2–4°C [Gersonde et al., 2003]. This suggests that an oceanic front migrated over our site during the Termination resulting in an additional 4–6°C temperature change. Therefore, our Mg/Ca temperature reconstruction shows that core PS2495-3 was located within the PFZ before Termination I and due to the southward migration of the SAF became under the influence of the SAZ at the end of Termination I.
 A marked difference between our record and the SST reconstructions from Gersonde et al.  is that Mg/Ca temperatures show a clearly warmer MIS3 in comparison with MIS2 and MIS4 (Figure 4). The modern gradient of ∼4°C between the sea surface and the permanent thermocline at our core site [Locarnini et al., 2006], which was also found for the Holocene, MIS2, and MIS4, is absent during MIS3. As our reconstruction of G. inflata apparent calcification depth is constant throughout the South Atlantic it seems unlikely that this can be explained by a change in habitat depth. A meridional vertical profile of the water column shows that midlatitudinal fronts are present down to a water depth of 400–500 m with the boundary between warmer and colder waters deepening toward the north resulting in the modern temperature gradient (Figure 1b). But when the surface and the permanent thermocline were bathed in the same water mass the water column would have been less stratified. This could possibly explain the similar temperatures reconstructed for the sea surface and for the permanent thermocline during MIS3. We suggest that during MIS2 and MIS4 the permanent thermocline at our site was under the influence of the PFZ, and the surface under influence of the SAZ, whereas both were bathed by the SAZ during MIS3.
 Reconstructed permanent thermocline temperatures for MIS5 and MIS6 are significantly warmer (∼4°C) than those reconstructed for the Holocene and MIS2, respectively (Figure 4). For MIS5 and MIS6 reconstructed temperatures based on Mg/Ca approach those reconstructed for the sea surface based on foraminifera transfer functions (Figure 4) [Gersonde et al., 2004], possibly suggesting a less stratified water column. We hypothesize that the warmer temperatures recorded for the permanent thermocline at core PS2495-3 during MIS5 are related to a stronger influence of the Subtropical Zone (STZ) if compared to the Holocene. Likewise, the PFZ would not have extended as far north during MIS6 if compared to its northernmost extension during MIS2, leaving the permanent thermocline at core PS2495-3 under the influence of significantly warmer waters of the SAZ. The warmer conditions during MIS5 and MIS6 in the permanent thermocline therefore seem to point to a more southward position of the SAZ in the permanent thermocline rather than large changes at the surface.
 The reconstructed temperatures for MIS2 and MIS4 appear close to or even lower than the lowest temperature tolerated by G. inflata of ∼3°C (Figure 5) [Bé and Hutson, 1977], and could have been caused by dissolution. Dissolution of biogenic carbonate in the water column or at the sediment-water interface preferentially dissolves higher-Mg portions of foraminiferal calcite [e.g., Brown and Elderfield, 1996]. As dissolution predominantly occurs in water masses undersaturated with respect to CO32− [Dekens et al., 2002; Regenberg et al., 2006; Mekik et al., 2007], deeper core locations are more easily affected by dissolution than shallower locations. At present core PS2495-3 is located in a water depth of only 3134 m and bathed by noncorrosive North Atlantic Deep Water and preservation is good. This is supported by Mg/Ca for G. inflata from a core top transect down to a water depth of 4000 m at the Rio Grande Rise which did not show any influence of dissolution [Mekik et al., 2010]. But, we cannot exclude that during glacial time periods more corrosive Antarctic Bottom Water had some influence at the site. Reconstruction of the calcite lysocline based on ultastructural investigations of the planktonic foraminifer G. bulloides showed increased influence of Antarctic water masses throughout the South Atlantic during glacial periods [Volbers and Henrich, 2004]. This led to a general shoaling of the calcite lysocline toward ∼3000 m water depth possibly causing some dissolution in our samples, and biasing Mg/Ca toward lower values. The correction of the Mg/Ca for dissolution would lead to an increase of 1–2°C [Dekens et al., 2002; Regenberg et al., 2006]. Thus, even corrected temperatures would still be significantly colder than MIS3.
 An alternative bias on the lowest temperatures of our downcore record could be related to the so-called cold-end effect of our calibration curve. This is a common feature of all Mg/Ca-temperature calibrations, both for planktonic and benthic foraminifera [Martin and Lea, 2002; Meland et al., 2006; Raitzsch et al., 2008]. Because small changes in Mg/Ca lead to large changes in temperature, slight differences in laboratory methods can have a significant effect on the reconstructed temperatures [Rosenthal et al., 2004]. However, considering the procedure at our laboratory as well as the occurrence of adjacent samples with low Mg/Ca during both MIS2 and MIS4, we consider that the cold-end effect is probably not significant and the reconstructed pattern of temperature change is therefore most likely representative.