The glacial marine isotope stage 14 (MIS 14) appears in many climate records as an unusually warm glacial. During this period an almost monospecific, up to 1.5 m thick, laminated layer of the giant diatom Ethmodiscus rex has been deposited below the South Atlantic Subtropical Gyre. This oligotrophic region is today less favorable for diatom growth with sediments typically consisting of calcareous nannofossil oozes. We have reconstructed temperatures and the stable oxygen isotopic compositions of sea surface and thermocline water (δ18Ow) from planktonic foraminiferal (Globigerinoides ruber and Globorotalia inflata) Mg/Ca and stable oxygen isotopes to test whether perturbations in surface ocean conditions contributed to the deposition of the diatom layer at ∼530 kyr B.P. Temperatures and δ18Ow values reconstructed from this diatom ooze interval are highly variable, with maxima similar to interglacial values. Since the area of the Ethmodiscus oozes resembles the region where Agulhas rings are present, we interpret these hydrographic changes to reflect the varying influence of warm and saline water of Indian Ocean origin that entered the Subtropical Gyre trapped in Agulhas rings. The formation of the Ethmodiscus oozes is associated with a period of maximum Agulhas leakage and a maximum frequency of Agulhas ring formation caused by a termination-type position of the Subtropical Front during the unusual warm MIS 14. The input of silica through the Agulhas rings enabled the shift in primary production from calcareous nannoplankton to diatoms, leading to the deposition of the massive diatom oozes.
 Almost monospecific oozes of the giant diatom Ethmodiscus rex (Rattray) Hendey [Wiseman and Hendey, 1953] have been deposited some 530 kyr ago in the subtropical South Atlantic [Romero and Schmieder, 2006; Schmieder, 2004; Schmieder et al., 2000]. These laminated sediments are 14 to 124 cm thick and have been found between 23° and 34°S in the center of the modern Subtropical Gyre (Figure 1; see auxiliary material), where otherwise calcareous nannofossil oozes primarily constitute the sediment. Today, limitation of silica restricts diatom productivity and deposition in this region. Oligotrophic open ocean conditions kept sedimentation rates always below 1.5 cm/kyr during the last 1.4 Ma [Schmieder et al., 2000]. The diatom horizons represent unique deposition events within the last 1.4 Ma [Gingele and Schmieder, 2001]. They are isochronous between several South Atlantic cores and coeval with shifts in grain size, hiatuses and turbidites in other South Atlantic sediment cores [Schmieder et al., 2000].
 Giant diatoms like Ethmodiscus are known to concentrate along oceanic fronts where they generate massive fluxes to the sediments [Kemp et al., 2006; Yoder et al., 1994]. Like other giant diatom species (e.g., Rhizosolenia sp.) Ethmodiscus is capable of migrating vertically in the water column [Villareal, 1992]. Different scenarios have been proposed to explain the accumulation of the E. rex layers in the oligotrophic South Atlantic. Schmieder et al.  were the first associating them with an open ocean frontal system. Gingele and Schmieder  considered two different scenarios.
 1. An extreme northerly position of the Subtropical Front resulted in a closure of the Agulhas leakage. The stop of inflowing Indian Ocean water allowed nutrient-rich waters of southern origin to penetrate into the South Atlantic Subtropical Gyre providing Ethmodiscus with nutrients.
 2. The deposition of the E. rex oozes was favored by a weak thermohaline overturning circulation during the Mid-Pleistocene Transition interim state which led to stratified conditions. Nutrients accumulated in deep layers where E. rex was able to access them through vertical migration.
Romero and Schmieder  refined the second scenario of Gingele and Schmieder  by arguing that intensified circulation and the depletion of macronutrients at the end of the Mid-Pleistocene Transition caused rapid sedimentation of E. rex fostered by the large size of their frustules. However, the occurrence of E. rex oozes exclusively at the end of the Mid-Pleistocene Transition despite relatively stable conditions for 280 kyr throughout the Mid-Pleistocene Transition interim state [Schmieder et al., 2000], remains enigmatic.
1.2. Marine Isotope Stage 14
 The deposition of the South Atlantic Ethmodiscus oozes took place during MIS 14. An increasing number of marine, terrestrial and ice core records from both hemispheres show that MIS 14 was an unusual warm glacial for the late and mid Pleistocene period. In EPICA Dome C ice cores it is the warmest glacial [Jouzel et al., 2007], shows the lowest glacial aeolian dust flux [EPICA Community Members, 2004], and indicates in four other proxies (non-sea-salt calcium, sea salt sodium, methane, carbon dioxide) the weakest glacial period of the last 800 kyr [Masson-Delmotte et al., 2010]. In the Northern Hemisphere, magnetic susceptibility records from Lake Baikal sediments indicate the presence of only small mountain glaciers and full interglacial conditions for the entire MIS 14 [Prokopenko et al., 2002]. A warm and humid phase has also been recorded in Chinese loess sequences [Kukla, 1987]. At the beginning of the following interglacial (MIS 13), an enormous sapropel layer in the Mediterranean Sea indicates a strong African Monsoon [Rossignol-Strick et al., 1998]. Together with the minimum glacial dust flux to Antarctica during MIS 14 this suggests a global change in atmospheric circulation. Furthermore, a collapse of the West Antarctic Ice Shield during MIS 14 was postulated by Hillenbrand et al.  which would have led to enhanced freshwater input to the Southern Ocean.
1.3. Aim of the Study
 The purpose of this study is to test whether an open ocean front or an alternative phenomenon might have provided the conditions for the deposition of the thick Ethmodiscus oozes. Oceanic fronts are characterized by steep gradients in salinity and temperature. We have reconstructed sea surface and thermocline temperatures, and the oxygen isotopic composition of seawater as proxy for salinity, from paired measurements of Mg/Ca and δ18O in planktonic foraminifers. We show that the unusual climate conditions during MIS 14 led to a change in upper ocean circulation within the South Atlantic Subtropical Gyre providing favorable conditions for the formation of Ethmodiscus oozes.
2. Material and Methods
 We have investigated the E. rex layers from South Atlantic gravity core GeoB 3801-6 (29°30.7′S, 8°18.3′W, 4546 m) from the Angola Basin (Figure 1). The core was collected during Leg M34/3 of RV Meteor [Wefer and Cruise Participants, 1996]. We have investigated the section between 453 and 937 cm core depth (equivalent to 372 and 780 kyr; see section 3). Layers of laminated E. rex oozes are present at 519 to 521 cm and 668 to 792 cm, the latter was investigated in detail.
 Samples were taken every 2 to 5 cm. Two planktonic foraminifer species have been used to reconstruct temperatures and the stable oxygen isotopic composition of seawater (δ18Ow) from the sea surface (Globigerinoides ruber sensu lato (s.l.) white) and the thermocline (Globorotalia inflata) by paired measurements of shell Mg/Ca and stable oxygen isotopes (δ18O). Differentiation between G. ruber morphotypes is based on the work by Wang . Analyses of G. ruber s.l. were performed using tests from the wet sieved size fraction >250 μm while for G. inflata the dry sieved fraction 315–400 μm was used. The size fractions were selected to resemble those in the Mg/Ca temperature calibrations applied in this study [Anand et al., 2003; Groeneveld and Chiessi, 2011]. For G. ruber s.l. the size fraction had to be extended toward larger specimens in order to obtain sufficient foraminiferal tests. Data presented in this study are available from the database PANGAEA (www.pangaea.de).
2.1. Stable Oxygen and Carbon Isotope Measurements
 Depending on the total number and size of available shells, between 3 and 10 visually clean individuals of G. ruber s.l. and G. inflata were selected per sample. Stable isotope measurements were performed with Finnigan MAT251 and Finnigan MAT252 mass spectrometers equipped with an automated carbonate preparation device at the University of Bremen. Results are expressed in ‰ relative to the international VPDB scale (Vienna Pee Dee Belemnite) and are calibrated against the National Bureau of Standards (NBS) standards 18, 19 and 20. The standard deviations of replicate analyses of a laboratory standard (Solnhofen limestone) were lower than 0.06‰ for δ13C and 0.07‰ for δ18O.
2.2. Elemental Analyses
 Samples for Mg/Ca measurements were analyzed by inductively coupled plasma optical emission spectrometry (ICP-OES). We typically used 25 to 40 individuals of G. ruber s.l. and 25 to 30 of G. inflata. When samples contained an insufficient number of shells for solution-based analysis, we applied laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). Tests obviously contaminated according to visual inspection were rejected for analysis. Foraminiferal shells were gently crushed and cleaned following the protocol of Barker et al.  and centrifuged for 10 min (6000 rpm) after dissolution to separate insoluble remains before transferring the solution to test tubes. Finally, the samples were diluted with ultrapure water (Seralpur) and analyzed on an ICP-OES (Perkin Elmer Optima 3300RL with auto sampler and ultrasonic nebulizer U-5000 AT) at the Department of Geosciences of the University of Bremen. Each reported Mg/Ca value represents the average of three measurement runs. Stability of the Mg/Ca measurements was monitored by a house standard solution (Mg/Ca = 2.93 mmol/mol) and by analyses of the limestone standard ECRM752-1 with a Mg/Ca of 3.75 mmol/mol [Greaves et al., 2008]. The standard deviations for Mg/Ca of the in-house standard (n = 62) and the ECRM752 (n = 23) were better than 0.09 mmol/mol and 0.06 mmol/mol, respectively. Replicate measurements (n = 5) from the same samples but picked, crushed, cleaned and analyzed during different sessions, showed a mean reproducibility for Mg/Ca of ±0.07 mmol/mol (1σ, equivalent to about ±0.6°C). Al/Ca, Fe/Ca and Mn/Ca were measured to monitor the efficiency of the cleaning [e.g., Barker et al., 2003; Pena et al., 2005; Pena et al., 2008].
 For laser ablation we largely followed the procedure described by Rathmann et al.  and Raitzsch et al. . Up to five individuals of G. ruber s.l. per sample were analyzed. To remove major contaminants from the outside and the pores of the foraminifers, the tests were cleaned for two minutes in an ultrasonic bath filled with deionized water [Raitzsch et al., 2008; Rathmann et al., 2004]. Remaining contaminations were identified by anomalously high Al, Mn, or Zn signals in the time-resolved data, and the respective intervals were excluded from the subsequent evaluation. Measurements were carried out on a NewWave UP 193 Solid State Laser Ablation System with a wavelength of 193 nm connected to a Thermo Finnigan Element 2 ICP-MS at the department of Geosciences, University of Bremen. The tests were ablated by a laser beam with a pulse rate of 5 Hz and a spot size of 35 μm (irradiance: 0.16 to 0.2 GW/cm2). To account for intrashell variations, each test was ablated five times preferably at five different chambers. The data were initially calibrated against the NIST612 glass standard reference material that contains a number of trace elements [Pearce et al., 1997] and which was ablated with 10 Hz and a spot size of 65 μm (irradiance: 0.5 to 1.4 GW/cm2). However, the study by Gao et al.  suggests that the Mg concentration provided in the compilation by Pearce et al.  is not accurate which may explain inaccuracies found when using the NIST612 to calibrate carbonate Mg/Ca [Hathorne et al., 2008]. By cross calibration of the NIST612 and NIST610 glass standard reference materials we found the NIST 612 to overestimate Mg/Ca by 20.5% and have corrected all the affected laser ablation data. Reported Mg/Ca is based on the isotopes 25Mg and 43Ca. 27Al, 55Mn and 64Zn have been examined as indicators for clayey fillings, ferromanganese coatings, and ablation of the sample holder. We used the software package GeoPro from CETAC for evaluation of the raw data. The standard error for Mg/Ca was on average 0.13 mmol/mol per sediment sample.
 Differences between both analytical techniques were detectable but small. Laser ablation (LA-ICP-MS) and solution-based (ICP-OES) measurements on separate subsamples show an increasing difference between both methods with increasing Mg/Ca (slope = 0.92) (Figure 2). Except for two data points, Mg/Ca values of the ICP-OES measurements are systematically lower than those from laser ablation measurements. We attribute this to preferential leaching of Mg during the cleaning process for solution-based analyses [Barker et al., 2003; Rosenthal et al., 2004; Yu et al., 2007], and the non-matrix-matched glass standard used for laser ablation. The maximum difference of 0.4 mmol/mol results in 1.2°C when converted to temperature. For a consistent presentation of Mg/Ca, we corrected the values obtained from laser ablation by multiplying with 0.92.
 Magnesium-rich manganese coatings might bias Mg/Ca [e.g., Barker et al., 2003; Pena et al., 2005; Pena et al., 2008]. We rejected the results where Mn/Ca exceeded 0.26 mmol/mol. Slightly elevated Mn/Ca values were present below the diatom layer (both species: 0.2 ± 0.05 mmol/mol) (Table 1). For G. ruber s.l. there is no correlation between Mn/Ca and Mg/Ca, but for G. inflata a correlation is present (r2 = 0.77) for the section below the diatom interval (Table 1). However, Mg/Ca in samples with increased Mn/Ca remains within the expected range and varies coherently with glacial-interglacial changes as recorded by benthic δ18O. Therefore, manganese seems to be present in phases that do not influence Mg/Ca.
Table 1. Mean Mn/Ca and Standard Deviations From GeoB 3801-6 for Both Planktonic Foraminifer Species and Both Measurement Techniquesa
Here n represents the number of samples and r2 gives the coefficient of determination of the regression between Mn/Ca and Mg/Ca.
G. ruber s.l.
Above E. rex layer
E. rex layer
Below E. rex layer
Above E. rex layer
E. rex layer
Below E. rex layer
2.3. Calculation of Temperature and Stable Oxygen Isotopic Composition of Sea Water
 We applied the calibrations from Anand et al.  for G. ruber s.l. [Mg/Ca = 0.34 exp (0.102 * temperature)] and from Groeneveld and Chiessi  for G. inflata [Mg/Ca = 0.72 exp (0.076 * temperature)] to calculate temperatures from Mg/Ca.
 We present stable oxygen isotope ratios of seawater (δ18Ow) as measure of relative salinity changes. The estimation of absolute salinities from δ18Ow suffers from large uncertainties in the past relationship between both parameters [Rohling, 2000] and is not further pursued here. We calculated δ18Ow by inserting the temperatures obtained from Mg/Ca (TMg/Ca) and δ18O from foraminiferal calcite (δ18Oc) in the paleotemperature equation from Shackleton , TMg/Ca = 16.9 − 4.38 (δ18Oc − δ18Ow) + 0.10 (δ18Oc − δ18Ow)2. The error in δ18Ow is about ±0.37‰ and primarily stems from a Mg/Ca temperature uncertainty of ±1°C [see Rohling, 2007, and references therein]. We corrected δ18Ow for the effect of global ice volume changes (dg) to estimate the component in δ18Ow that was driven by local hydrographic changes (δ18Owi) [Rostek et al., 1993]. The maximal global glacial-interglacial change of δ18Ow has been estimated as 1.2‰ [Rostek et al., 1993; Wefer et al., 1996]. The correction dg was calculated by normalizing the existing benthic δ18O data from the same core [Bickert and Mackensen, 2004] to this value. Finally, δ18Owi was obtained by subtracting dg from δ18Ow.
 An earlier stratigraphy for core GeoB 3801-6 has been constructed by von Dobeneck and Schmieder  by tuning the susceptibility record to obliquity and precession of Earth's orbit. The tuning resulted in an age of 537 kyr for the top of the Ethmodiscus ooze and a deposition within less than 4 kyr. However, no age tie points exist within the E. rex interval due to low susceptibilities of the sediments. Therefore, we used the benthic δ18O record from Bickert and Mackensen  to develop a new stratigraphy. Age control was established by graphical correlation of prominent minima and maxima in benthic δ18O to the benthic δ18O reference stack from Lisiecki and Raymo  using the software AnalySeries [Paillard et al., 1996]. GeoB 3801-6 and global benthic δ18O agree well with a coefficient of determination of r2 = 0.8. The interval examined here covers MIS 19 to 10. The E. rex layer was deposited before the termination of MIS 14 from 552 to 539 kyr suggesting its formation within 13 kyr. It is characterized by a very high sedimentation rate (9.6 cm/kyr) relative to the core sections above (1.2 cm/kyr) and below (0.6 cm/kyr) (Figure 3). The top age of the layer is well determined by the large δ18O shift related to the termination of MIS 14, but the base of the event is less well constrained due to reduced variance in the isotope record. Low sedimentation rates outside the diatom layer limit the temporal resolution of our age model to 4 to 8 kyr (Figure 3).
3.1. Stable Oxygen Isotopes
δ18O from both species shows glacial-interglacial variability. Values for G. ruber (G. inflata) range from ∼0.2‰ (1.7‰) during interglacials to ∼1.2‰ (2.7‰) during glacials (Figures 4 and 5). However, amplitudes in δ18O for G. ruber s.l. white between the top of the diatom layer and the top of the investigated core section are about 0.5‰ smaller than those below the diatom layer down to the bottom of the studied core section. This change in amplitudes is not evident in δ18O from G. inflata. Within the diatom layer G. inflata shows pronounced δ18O variability. On average, δ18O values remain on an intermediate level for G. ruber s.l. (∼0.7‰) and on a glacial level for G. inflata (∼2.5‰) (Figures 4 and 5).
3.2. Mg/Ca and Temperatures
 For both species Mg/Ca and reconstructed temperatures correlate with δ18O showing similar changes between glacials and interglacials (Figures 4 and 5). For the surface-dwelling G. ruber Mg/Ca is 2.5 mmol/mol (equivalent to ∼20°C) on average over the entire record with an average glacial value of 2.2 mmol/mol (18°C), and an average interglacial value of 2.9 mmol/mol (21°C). Values for the thermocline-dwelling G. inflata are 1.3 mmol/mol on average (7°C), 1 mmol/mol (4°C) for glacials and 1.7 mmol/mol (11°C) for interglacials.
 Within the Ethmodiscus ooze average Mg/Ca is on glacial level for G. inflata (1 mmol/mol), but relatively high for G. ruber (2.5 mmol/mol). Variability is high for both species and by far exceeds that for glacials other than MIS 14 (G. ruber s.l.: 1.7 to 3.1 mmol/mol; G. inflata: 0.58 to 1.54 mmol/mol). Within the diatom ooze Mg/Ca repeatedly changes between values that would be typical for glacial and interglacial conditions. The equivalent temperature amplitudes are ∼4°C for the sea surface (G. ruber) and ∼8°C for the thermocline (G. inflata). Where warm temperatures occur in glacials other than MIS 14 they can be considered part of the inception or termination. The recurrent rapid switch between very cool and very warm temperatures (and vice versa) is an exclusive feature of MIS 14.
3.3. Stable Oxygen Isotopic Composition of Sea Water
 General trends in the δ18Owi records from both planktonic species are similar and parallel the temperature records (Figure 5). δ18Owi values for the sea surface are on average 1.0‰ both within and outside the diatom layer. Average thermocline δ18Owi values are about −0.6‰ outside and −1.3‰ within the diatom layer. Within the latter, δ18Owi is highly variable ranging from about 0.2‰ to 1.5‰ for G. ruber s.l. and about −2.0‰ to 0.2‰ for G. inflata. Akin to the variations in Mg/Ca and reconstructed temperatures, the δ18Owi amplitude within the diatom layer approaches glacial to interglacial variations. δ18Owi values at the thermocline tend to be lower during glacial compared to interglacial times.
4.1. Reliability of Reconstructed Temperatures
 The modern lysocline in the South Atlantic is located at a water depth of approximately 4000 m [Volbers and Henrich, 2004]. Bickert and Wefer  have shown that during glacials, at least back to MIS 10 (∼360 kyr), the northward spread of corrosive Antarctic Bottom Water caused a lysocline shoaling. For the last glacial maximum South Atlantic lysocline depths of 3800 to 3200 m have been reconstructed [Bickert and Wefer, 1996; Volbers and Henrich, 2004]. Site GeoB 3801-6 (∼4500 m depth) is located below the lysocline today, and was most likely situated below the lysocline throughout the time interval investigated here.
 Some foraminifer shells from the studied core section show signs of dissolution (e.g., etched surfaces and fragmentation). For geochemical analysis we have used only the specimens that showed the best preservation. However, the microscopic inspection used by us would not identify dissolution of the inner shell, and we cannot rule out the possibility that affected shells have been included in the geochemical analyses.
 Dissolution preferentially affects the Mg rich parts of a shell [Dekens et al., 2002; Rosenthal and Lohmann, 2002] leading to an underestimation of reconstructed temperatures. In G. inflata the Mg/Ca of newly formed calcite decreases during ontogeny, resulting from the increase in habitat depth with associated cooler water temperatures, and the secretion of an outer crust consisting of Mg-depleted calcite [Groeneveld and Chiessi, 2011]. In this crust Mg/Ca is 2–3 times lower than in the primary shell [Hathorne et al., 2009]. In contrast, G. ruber calcifies within the surface mixed layer [Anand et al., 2003] and lacks an outer crust, and hence is expected to display a smaller degree of ontogenetic variability in shell Mg/Ca. Even though G. ruber is more susceptible to dissolution than G. inflata [Berger, 1970] the impact on shell Mg/Ca is probably smaller.
 Mg/Ca from G. ruber from Core 3801-6 yields the expected range of sea surface temperatures. Interglacial temperatures (∼21°C) are in line with modern instrumental observations (22°C [Locarnini et al., 2006]), and temperatures for glacials other than MIS 14 (18°C) are similar to the estimates for the last glacial maximum for this core derived from foraminiferal transfer functions [Niebler et al., 2003]. For the diatom layer, the warm excursions are consistent with the temperature estimate of about 20°C for core GeoB 3813-3 (Figure 1) represented by an alkenone-derived datum [Gingele and Schmieder, 2001]. Thermocline temperatures based on G. inflata (∼11°C for interglacials, ∼2°C for glacials) tend to be somewhat too low compared with modern instrumental (14°C [Locarnini et al., 2006]) and modeled last glacial maximum estimates (6–7°C [Paul and Schäfer-Neth, 2003]). Within the diatom layer reconstructed thermocline temperatures occasionally fall below 0°C, which appears unrealistic.
 Taken together, reconstructed sea surface temperatures are not or only marginally affected by dissolution of foraminiferal shells. The low thermocline temperatures are probably due to preferential dissolution of the primary shell of G. inflata compared with the more resistant crust; nevertheless, the qualitative pattern of temperature changes between glacials and interglacials is well reflected in our record.
4.2. South Atlantic Ethmodiscus Oozes
 The Ethmodiscus ooze was deposited at an unusually high sedimentation rate (>9 cm/kyr; Figure 3). High planktonic foraminiferal δ13C indicates contemporaneously enhanced primary productivity (Figure 6). Due to the preferential uptake of 12C during photosynthesis 13C is enriched in seawater leading to higher δ13C in planktonic foraminiferal carbonate [Curry et al., 1988]. For the nearby site GeoB 3813-3 in the Brazil Basin Gingele and Schmieder  found increased bulk sediment Ba/Al ratios within the diatom layer, providing additional evidence for increased primary productivity. These authors also excluded the accumulation of diatom shells through focusing, since Site GeoB 3806-1 is sheltered from major bottom currents.
 Hypotheses on the origin of the subtropical South Atlantic E. rex layers discuss a weak thermohaline circulation, more stratified conditions, and subsequent deep enrichment of nutrients which were accessible for E. rex through vertical migration [Gingele and Schmieder, 2001; Romero and Schmieder, 2006]. However, at least from a North Atlantic perspective, the δ18Obulk record from IODP Site U1308 (Figure 7) provides no indication for Heinrich-like events during MIS 14 that would be expected if the thermohaline circulation was weaker [Hodell et al., 2008].
 The high variability in surface and thermocline temperatures at site GeoB 3801-6 during MIS 14 would be theoretically consistent with the repeated passage of an ocean front, which could explain the occurrence of the giant diatom ooze [Kemp et al., 2006]. However, no open ocean front exists within the South Atlantic Subtropical Gyre today [Stramma and England, 1999]. For the last glacial maximum, model studies and proxy data show that the gyre was compressed and shifted toward the equator by 3° to 6° [Clauzet et al., 2007; Trend-Staid and Prell, 2002], which is insufficient for the Subtropical Front to reach the core position. It is therefore highly unlikely to assume the presence of an oceanic front within the stable regime of the Subtropical Gyre during MIS 14.
4.3. Agulhas Rings
 An alternative way to create rapid changes in the upper ocean hydrography is the passage of Agulhas rings. The water masses within the rings are primarily of Indian Ocean origin, but contain admixed Southern Ocean waters [Arhan et al., 1999; Garzoli et al., 1999; Lutjeharms and van Ballegooyen, 1988; van Aken et al., 2003]. The mixed layer of Agulhas rings is warm, saline and silica-rich, and can be clearly distinguished from the cooler and fresher waters of the Subtropical Gyre [Gordon, 1985; van Aken et al., 2003]. We suggest that the sudden changes in temperature and δ18Owi, reconstructed from G. ruber s.l. and G. inflata, were caused by changes in the frequency of Agulhas rings passing over Site GeoB 3801-6. Individual rings would hardly be identifiable in the reconstructed upper ocean hydrography, but long-term changes in the frequency of rings would, where a particularly high frequency would lead to warmer average temperatures and salinities. We note that the areas of Agulhas ring paths and Ethmodiscus ooze deposition match quite well [Byrne et al., 1995] (Figure 8).
 The formation of Agulhas rings at the Agulhas retroflection is the main mechanism for interocean heat exchange between the Indian and the Atlantic oceans [Lutjeharms, 1996]. After their formation south of Africa, the majority of these rings drifts off into the South Atlantic traveling through the Subtropical Gyre as far as the western boundary of the Atlantic (40°W; Figure 8) [Byrne et al., 1995; Lutjeharms, 1996]. Their pathway is located between the Benguela Current to the north and the Subtropical Front to the south [Byrne et al., 1995]. GeoB 3801-6 is located in the center of this pathway and as such ideally positioned to record increasing intensity/frequency of Agulhas Rings (Figure 8). Of the five gravity cores containing Ethmodiscus oozes [Schmieder, 2004; Schmieder et al., 2000], GeoB 3801-6 is closest to the Agulhas retroflection. The influence of Agulhas rings has been probably more intense at site GeoB 3801 than at the others. This is also reflected in the thickness of the Ethmodiscus layer in other South Atlantic GeoB cores which is becoming thinner toward the edge of the Agulhas ring path (Figure 8). Additionally, core GeoB 3801-6 is located near the Mid Atlantic Ridge, a topographic feature which causes the Agulhas rings to slow down [Byrne et al., 1995]. This might have prolonged their presence in this area and further enhanced the deposition of E. rex. Schmieder et al.  and Schmieder  reported from other cores of the subtropical South Atlantic shifts in grain size, turbidites and hiatuses that are coeval with the E. rex oozes (Figure 8). These cores are mainly located near the margins of the region influenced by Agulhas rings, where average giant diatom productivity was probably insufficient for the generation of distinguishable layers.
 While leakage was generally higher under interglacial than under full glacial conditions, the maxima occurred during the “second halves of the glacial terminations” [Peeters et al., 2004]. MIS 14 differs significantly from typical glacials in having been weak and warm in both hemispheres, and having been characterized by low global ice volume, almost interglacial conditions at Lake Baikal, weak trade winds and low dust transport to Antarctica (Figure 7) [e.g., Jouzel et al., 2007; Lambert et al., 2008; Lisiecki and Raymo, 2005; Prokopenko et al., 2002]. This setting resembles that of terminations, but was of much longer duration.
 We propose the following scenario for the South Atlantic for MIS 14: Since global climate was intermediate between full glacial and interglacial conditions, we assume that the Subtropical Front was located intermediately between its typical glacial and interglacial positions. This led to an Agulhas leakage that was not only active, but as intense as during advanced terminations. The geometry of the African coast would have favored the formation of Agulhas rings [Zharkov and Nof, 2008] and the position of the Subtropical Front effectively facilitated the separation of newly formed Agulhas rings by wedges of cold Subantarctic Surface Water [Lutjeharms and van Ballegooyen, 1988]. Furthermore, enough space between the Subtropical Front and South Africa was available to allow for frequent Agulhas ring formation. These conditions may have resulted in a maximum Agulhas leakage for at least 13 kyr, long enough to form the thick Ethmodiscus oozes. In contrast, during terminations the duration of maximum Agulhas leakage was too short to allow for the accumulation of distinct diatom layers, while during interglacials leakage was too weak to form Agulhas rings in adequate frequency. The enhanced leakage and frequent formation of Agulhas rings resulted in the local maxima in reconstructed temperatures and salinities recorded at site GeoB 3801. Although average conditions were favorable for ring formation, a high degree of variability in ring occurrence and/or their average pathway is suggested by the frequent and pronounced hydrographic changes at the core locations.
4.5. Change From Carbonate to Opal Primary Productivity
 A characteristic feature associated with the Ethmodiscus oozes is the shift from carbonate production dominated by coccolithophores to opal production dominated by diatoms. This shift is interesting because the modern surface and subsurface waters of the Subtropical Gyre are characterized by a low silicate concentration (Figure 9) rarely exceeding 2.5 μmol/l [Garcia et al., 2006] and are unsuitable for high diatom productivity. Experiments have shown that diatoms require Si(OH)4 concentrations above 2 μmol/l to outcompete coccolithophores and other nannoplankton [Egge and Aksnes, 1992]. In the source region of Agulhas rings silicate concentrations are higher than those of the Subtropical Gyre and may exceed 6 μmol/l (Figure 9). Agulhas rings are thus relatively silica-rich, for example, in the surface mixed layer of ring Astrid an average silica concentration between 2 and 4 μmol/kg was measured [van Aken et al., 2003]. We suggest that during MIS 14, the numerous Agulhas rings advected silica to the Subtropical Gyre. This would not only explain the shift toward diatom-based primary productivity, but also the dominance of Ethmodiscus within the diatom community. The transition between an Agulhas ring and the surrounding water with its strong hydrographic gradients and outcropping of isotherms and isohalines at the sea surface [van Aken et al., 2003] is comparable to the conditions at open ocean fronts. Convergent flow at this ring front concentrates organisms that are able to control their vertical movement [Olson and Backus, 1985], which has been observed repeatedly in or next to anticyclonic warm-core eddies [Wilson and Qiu, 2008, and references therein]. These conditions might have provided the environment for buoyant Ethmodiscus to become dominant and to form the thick diatom deposits during MIS 14.
 Repeated upper ocean temperature and salinity changes between values typical for glacials and interglacials have been reconstructed from a thick layer of Ethmodiscus giant diatom ooze deposited during a 13 kyr time interval within MIS 14. This diatom ooze occurs within an area that today is influenced by Agulhas rings.
 The upper water masses within Agulhas rings are warmer, saltier, and enriched in silica compared with the ambient Subtropical Gyre waters. The transition between rings and the surrounding water provides conditions comparable to those at oceanic fronts. Both the front-like conditions and the silica supply favored the growth of Ethmodiscus and the accumulation of the thick E. rex oozes. The reconstructed hydrographic variability most likely reflects the changing influence of Agulhas rings which in turn is linked to a variable, but generally high (compared with today) ring formation during MIS 14. This glacial was unusually warm, and we hypothesize that the Subtropical Front was located between its glacial and interglacial positions, creating a situation analogous to the later stages of typical late Pleistocene terminations when Agulhas leakage peaks. Particularly strong leakage and ring formation would explain the MIS 14 temperature and salinity maxima in our record.
 We thank A. Klügel, M. Kölling, S. Pape, and M. Raitzsch for technical support in the laboratory; M. Segl and B. Meyer-Schack for carrying out the isotope measurements; and S. Kemle-von Mücke and S. Renken for sample preparation. J. Pätzold, A. Paul, F. Schmieder, M. Schulz, T. von Dobeneck, G. Wefer, and K. Zonneveld are thanked for fruitful discussions. The paper benefited from detailed comments of two anonymous reviewers. This work was supported by the Bremen International Graduate School for Marine Sciences (GLOMAR) that is funded by the German Research Foundation (DFG) within the frame of the Excellence Initiative by the German federal and state governments to promote science and research at German universities.