The initiation of the Benguela upwelling has been dated to the late Miocene, but estimates of its sea surface temperature evolution are not available. This study presents data from Ocean Drilling Program (ODP) Site 1085 recovered from the southern Cape Basin. Samples of the middle Miocene to Pliocene were analyzed for alkenone-based (U37K′, SSTUK) and glycerol dialkyl glycerol tetraether (GDGT) based (TEX86, TempTEX) water temperature proxies. In concordance with global cooling during the Miocene, SSTUK and TempTEX exhibit a decline of about 8°C and 16°C, respectively. The temperature trends suggest an inflow of cold Antarctic waters triggered by Antarctic ice sheet expansion and intensification of Southern Hemisphere southeasterly winds. A temperature offset between both proxies developed with the onset of upwelling, which can be explained by differences in habitat: alkenone-producing phytoplankton live in the euphotic zone and record sea surface temperatures, while GDGT-producing Thaumarchaeota are displaced to colder subsurface waters in upwelling-influenced areas and record subsurface water temperatures. We suggest that variations in subsurface water temperatures were driven by advection of cold Antarctic waters and thermocline adjustments that were due to changes in North Atlantic deep water formation. A decline in surface temperatures, an increased offset between temperature proxies, and an increase in primary productivity suggest the establishment of the Benguela upwelling at 10 Ma. During the Messinian Salinity Crisis, between 7 and 5 Ma, surface and subsurface temperature estimates became similar, likely because of a strong reduction in Atlantic overturning circulation, while high total organic carbon contents suggest a “biogenic bloom.” In the Pliocene the offset between the temperature estimates and the cooling trend was reestablished.
 Here we present ocean temperature data from the middle Miocene to Pliocene covering a time span of almost 11 Myr between 13.7 and 2.8 Ma. Our area of investigation is located in the Cape Basin in front of the South African Orange River. We use two different organic geochemical temperature proxies: the alkenone unsaturation index (U37K′, SSTUK) [Brassell et al., 1986] and the TEX86 index using a ratio of glycerol dialkyl glycerol tetraethers (GDGTs) with different numbers of cyclopentane rings, the tetraether index (TempTEX) [Schouten et al., 2002]. While it was originally assumed that both proxies reflect SSTs [Schouten et al., 2002], differences between these two temperature proxies were found that are potentially related to differences in the habitat of the source organisms in specific oceanic areas [e.g., Huguet et al., 2006; Lee et al., 2008]. In this study, we aim to disentangle the additional information that can be obtained from applying two ocean temperature proxies, and we attempt to decipher seawater temperature and upwelling changes in conjunction with global oceanographic events.
2. Benguela Upwelling System and Hydrography of the Cape Basin
 Today's surface water circulation in the southeast Atlantic is dominated by the Benguela Current (BC), which flows northward along the southwest coast of Africa (Figure 1). It originates from the eastward directed cold South Atlantic Current, the northernmost front of the Antarctic Circumpolar Current (ACC), and the warm Agulhas Current (AgC) flowing from the Indian Ocean around the Cape of Good Hope where it retroflects (Figure 1) [Peterson and Stramma, 1991]. The BC divides into the Benguela Ocean Current (BOC) and the Benguela Coastal Current (BCC) around 28°S on its northward flow. The BOC flows northwestward and crosses the Atlantic Ocean as the South Equatorial Current, while the northward directed BCC encounters the southward directed warm Angola Current north of the Walvis Ridge at the Angola-Benguela Front [Peterson and Stramma, 1991].
 The present southeast Atlantic deepwater circulation depends on interactions of the southward flowing, relatively warm, highly saline, oxygen-rich, and nutrient-poor North Atlantic Deep Water (NADW) and the northward flowing, relatively cold, oxygen-poor, and nutrient-rich Antarctic deepwater masses [Berger and Wefer, 1996b; Broecker et al., 1985; Reid, 1989]. In the Cape Basin, NADW occurs between 2500 and 4000 m water depth sandwiched between the cold-water masses of the Antarctic (Figure 1). The hydrography between approximately 2500 and 500 m water depth is generally characterized by Antarctic Intermediate Water (AAIW) and Upper Circumpolar Deep Water (UCDW) (Figure 1). South Atlantic surface waters down to approximately 400 m are warm and salty (15°C to 23°C and 35.4 to 36.0 psu) compared to AAIW located underneath (6°C to 16°C and 34.5 to 35.5 psu) [Schneider et al., 2003], which leads to a strong permanent thermocline. Along the East Atlantic margin the cold thermocline waters extend to the bottom of the continental shelf and are uplifted to the surface in the upwelling cells along the African coast [Andrews and Hutchings, 1980].
 Upwelling along the southwest African coast occurs in small upwelling cells in response to southeast trade wind–induced Ekman transport. The strongest perennial upwelling within the Cape Basin takes place in the northern part of the BUS (19°S–28°S) between Lüderitz and the Walvis Ridge and extends 130–230 km offshore (Figure 1, gray shaded areas) [Shannon and Nelson, 1996]. Less intense and smaller upwelling cells are located south of Lüderitz in the southern part of the BUS (28°S–34°S) [Shannon and Nelson, 1996]. Within these cells the upwelling of cold and nutrient-rich Antarctic waters from 200 to 500 m water depth induces perennial high productivity [Hart and Currie, 1960; Lutjeharms and Meeuwis, 1987; Lutjeharms and Stockton, 1987]. Intensified southeast trade winds during austral summer (December to April) lead to enhanced upwelling, and upwelling filaments may extend up to 600 km offshore [Lutjeharms and Stockton, 1987; Summerhayes et al., 1995]. Secondary upwelling may occur at a shelf-break thermal front, sometimes extending 1300 km offshore [Summerhayes et al., 1995].
3. Material and Methods
3.1. Study Area and Samples
 We analyzed samples from Ocean Drilling Program (ODP) Hole 1085 A that was drilled in 1997 during ODP Leg 175 on the southwest African continental margin (core position 29°22.5′S, 13°59.4′E, 1713 m water depth; Figure 1 [Wefer et al., 1998]). The site is situated in the mid-Cape Basin south of the modern upwelling center. Today, the site is influenced by filaments of seasonal upwelling [Lutjeharms and Stockton, 1987]. Annual mean SSTs are around 17.9°C, while warmer temperatures occur during the austral summer (∼20.1°C, January to March) and colder during the austral winter (∼15.9°C, July to September) [Locarnini et al., 2010]. Seasonal temperature changes have an impact down to a depth of ∼125 m [Locarnini et al., 2010]. A temperature depth profile of nearby GeoB Site 8337 indicates a surface mixed layer of 40 m water depth [Lee et al., 2008]. In surface waters, concentrations of silica and phosphate as well as primary productivity are relatively low compared to the northern part of the upwelling area [Berger et al., 2002; Mollenhauer et al., 2002].
 ODP Site 1085 is located approximately 300 km away from the river mouth of the perennial Orange River at the margin of today's river plume, which carries suspended terrigenous material. The shelf in this area is relatively wide (180 km) [Shannon and Nelson, 1996; Wefer et al., 1998]. The Orange River discharges on average 0.02 Gt yr−1 organic carbon [Compton et al., 2009]. The sediment load is transported to the shelf and slope and distributed southward by nearshore south flowing countercurrents [Bremner et al., 1990; Summerhayes et al., 1995].
 A high-resolution age model based on oxygen isotope ratios (δ18O) of benthic foraminifera and X-ray fluorescence scanning (Fe and Ca) provides a detailed chronology for the late Miocene to early Pliocene (13.9 to 4.7 Ma) [Vidal et al., 2002; Westerhold et al., 2005]. Ages of Pliocene and Pleistocene samples were determined by using an age model based on biostratigraphic and paleomagnetic events [Berger et al., 2002]. Sixty-six samples of ∼10 mL sediment were taken between 134 and 592 m composite depth dated from 2.8 to 13.7 Ma. For comparison three samples were selected from the late Pleistocene (14.8, 32.0, and 69.3 ka; see Table 1).
 Before sampling at the IODP Core Repository in Bremen, cores of ODP Leg 175 Site 1085 were stored at 4°C since their recovery. Sediment samples were freeze-dried and ground and homogenized by mortar and pestle. Total organic carbon (TOC) contents were determined by decarbonating a sediment aliquot of ∼25 mg using 6 N hydrochloric acid before combustion at 1050°C in a Heraeus CHN-O-rapid elemental analyzer. The relative precision of the measurements is based on duplicates as well as on the control analysis of a lab internal reference sediment sample.
 To aliquots of 6 to 13 g of sediment samples (Table 1) known amounts of n-nonadecan-2-one were added as internal standards. The sediments were extracted by Dionex™ accelerated solvent extraction (ASE) using dichloromethane/methanol (9:1, v/v; three times 5 min., 70 bars, 100°C). The total lipid extracts were separated by Al2O3 column chromatography using hexane/dichloromethane (9:1, v/v) to elute an apolar fraction (hydrocarbons) and dichloromethane/methanol (1:1, v/v) to elute a polar fraction. The latter fraction, containing GDGTs for the TEX86 and BIT analyses as well as alkenones for the U37K′ analyses, was dried under a continuous nitrogen stream, ultrasonically dissolved in a hexane/isopropanol (99:1, v/v) mixture to a concentration of ∼2 mg mL−1, and filtered through a 0.4 μm pore size PTFE filter. The fractions for GDGT analyses were analyzed in duplicate using an Agilent HP1200 high-performance liquid chromatography (HPLC) system coupled to an Agilent 6120 mass spectrometer equipped with an atmospheric pressure chemical ionization (APCI-MS) ion source according to the method described by Hopmans et al.  and refined as reported by Schouten et al. [2007b]. The TEX86 index was calculated using the peak areas of ions for different GDGTs. The TEX86 index is defined as
where GDGT-1, GDGT-2, and GDGT-3 indicate GDGTs containing 1, 2, and 3 cyclopentane moieties, respectively, and Cren' is the crenarchaeol regionisomer [after Kim et al., 2010]. Temperatures were calculated using the calibration for subtropical and tropical oceans (GDGT index 2) by Kim et al. :
The error of the calibration equation is ±2.5°C [Kim et al., 2010]. The BIT index was calculated using the peak areas of ions for different GDGTs and the following formula [after Hopmans et al., 2004]:
where I, II, and III indicate GDGTs without cyclic components in the structure and IV indicates crenarchaeol. Standard deviations of GDGT indices reported are calculated from results of duplicate analyses. A consistently analyzed lab internal standard sediment (core-catcher sediment from the continental slope off Namibia: GeoB 1712–4, 23.26°S 12.81°E, 998 m water depth, 1029 cm core length; recovered during R/V Meteor cruise M20/2) [Schulz et al., 1992] revealed a standard deviation for analyses of TempTEX of 3.54% or ±0.51°C using the calibration of Kim et al. . Reported standard deviation errors of some data exhibited large deviations (up to 2.9°C; Table 1) that were due to the low content of organics, especially of GDGTs in individual sediment samples. However, TempTEX data with large analytical errors generally fit into the trend depicted by adjacent more reliable TempTEX data (see Table 1 and Figure 2). We interpret all data only outside their analytical errors.
 The remaining polar fraction was reduced to dryness and saponified using 0.5 M KOH in methanol (1 mL, 85°C, 2 h). The products were extracted with n-hexane after the addition of distilled water. After this clean-up step the polar compounds were converted to their trimethylsilyl ether derivatives before analysis in duplicates by gas chromatography with flame ionization detection (GC-FID) for U37K′ calculation. The U37K′ is defined as
The error of the calibration equation is ±1°C [Müller et al., 1998]. A consistently analyzed lab internal standard sediment (GeoB 1712-4) revealed a standard deviation for analyses of SSTUK of 2.21% or ±0.38°C by using the calibration of Müller et al. .
 A complete set of analytical data of this study is available through the PANGAEA database (http://www.pangaea.de/).
 TOC, SSTUK, and TempTEX of this study are displayed in Figure 2, whereas numerical data are given in Table 1. The TOC content record ranges between 0.1% and 1.8% and exhibits a gradual increase from 0.2% at approximately 12 Ma to 0.7% at 10 Ma. These relatively low values are followed by a slight overall decrease to ∼0.4% at 6.9 Ma and a subsequent increase to the highest values of ∼1.8% in Pliocene sediments. TOC values of Pleistocene sediment samples (∼1.2%) are slightly lower than those of the Pliocene (Figure 2).
 SSTs estimated by U37K′ exhibit a monotonous decrease of about 8°C from maximum values of 27.5°C between 13.7 and 11.6 Ma to ∼19°C at 2.8 Ma. TempTEX values decline steeply from 30.5°C at 13.7 Ma to ∼19°C at 9.5 Ma and then remain rather constant until 6.5 Ma. Between 6.5 and 5.0 Ma the TempTEX values fluctuate between 25°C and 17°C, and afterward the values exhibit a gradual decline from ∼17°C to 15°C at 2.8 Ma. The overall amplitude of TempTEX data amounts to about 16°C. Pleistocene water temperature estimates of both proxies indicate temperatures of around 17.5°C to 18.0°C (Figure 2 and Table 1). In general, TempTEX fluctuations have larger amplitudes than those of the SSTUK and mostly show an offset of around 4°C up to 7°C to colder temperatures. Between 6.5 and 5 Ma and during the Pleistocene, SSTUK and TempTEX are more similar (Figure 2).
5.1. Significance of Proxies
 An organic deep-sea sediment proxy record of a time span of over millions of years may have been affected by evolutionary changes in source organisms and/or microbial, chemical, or oceanic processes like degradation, resuspension, transport, or dissolution. Therefore, we first discuss the reliability and limits of our data.
 During the middle to late Miocene an effect of low TOC preservation is evident for ODP Site 1085, possibly because of oxic bottom water conditions [Diester-Haass et al., 2004; Dupont et al., 2011]. Although TOC is a first-order proxy for primary productivity being low during times of low productivity, preservation effects may also be reflected in TOC values, especially during the middle Miocene period (Figure 2). Organic preservation is better after 11 Ma and probably excellent between 9 and 8 Ma as well as between 6 and 5 Ma, as indicated by the high abundance of palynomorphs that are sensitive to oxygenic degradation [Dupont et al., 2011]. Hence, middle Miocene temperature data might be biased by low preservation. Although surface productivity (upper continental slope off high upwelling areas) might be less, the preservation of organic matter is enhanced in slope depocenters because of the deposition of resuspended material [Inthorn et al., 2006; Mollenhauer et al., 2002]. Sediments of the shelf eroded by rapid sea level changes and by dissipation of wave and tidal energy are transported laterally in southward flowing nepheloid layers, aligned by bottom ocean currents, and deposited at the upper continental slope [Compton and Wiltshire, 2009; Compton et al., 2009; Mollenhauer et al., 2002, 2007]. On the margin, deposition was continuous since the middle Miocene [Weigelt and Uenzelmann-Neben, 2004]. On the shelf, major composite erosional unconformities occurred in the middle Miocene to Pliocene/Pleistocene [Compton et al., 2004]. Lateral transport of organic material may have biased our water temperature estimations, but age estimations resulting in offsets between organic temperature proxies are effective only on short time scales [Mollenhauer et al., 2007]. Hence, lateral transport is present but is negligible for our trend observations over millions of years.
 The calibration of Müller et al.  for the U37K′ index used in this study is based on recent sediments containing alkenones mainly produced by the haptophyte Emiliania huxleyi. This coccolithophorid species became the dominant alkenone-producing organism during the last 70–80 kyr and first appeared 268 kyr ago [Thierstein et al., 1977]. Alkenones in sediments dating back to the Eocene (55.8 to 33.9 Ma) are attributed to extinct genera of the family Gephyrocapsaceae, possible ancestors of Emiliania [Marlowe et al., 1990]. Several studies confirmed that these species produce alkenones without significant differences in SST dependence on Emiliania [e.g., Conte et al., 1995; McClymont et al., 2005; Müller et al., 1997; Villanueva et al., 2002]. We therefore assume that the U37K′ calibration by Müller et al.  is applicable to Miocene samples.
 The U37K′ calibration was shown to be linear over a temperature range between 0°C and 27°C [Müller et al., 1998]. Above 27°C, as found in tropical or subtropical oceans, alkenone-producing haptophytes synthesize only trace amounts of triunsaturated alkenones and the U37K′ becomes insensitive to temperature. Obviously, the sensitivity limit is reached in our data during most of the middle Miocene, in which the SSTUK curve flattens and reaches its maximum value of 27.5°C (Table 1 and Figure 2).
 Regardless of the calibration accuracy, oxic bottom water conditions affect the diunsaturated and triunsaturated species to different extents; the triunsaturated alkenone may be degraded faster, resulting in a relative increase of U37K′ that would translate into a shift of SSTUK to higher values [Gong and Hollander, 1999; Hoefs et al., 1998; Huguet et al., 2009; Kim et al., 2009; Rontani et al., 2005, 2008; Teece et al., 1998]. Other studies, however, showed that the U37K′ index seems to be unaffected or that differential degradation has only a minor effect on the estimated temperature (up to 1.2°C), despite a strong decrease in the amount of alkenones [Huguet et al., 2009; Kim et al., 2009; Prahl et al., 1989]. Oxic degradation is obvious in ODP Site 1085 sediments of the middle Miocene, where the U37K′ reached its calibration limit (Figure 2). The SSTUK data prior to 11 Ma are, therefore, not usable to estimate temperatures. The preservation of organic matter increased after 11 Ma, when we presume the effects of postdepositional oxidation to be minor.
 The TEX86 is defined as the abundance ratio of GDGTs with different numbers of cyclopentane rings, synthesized by picoplankton of the phyllum Thaumarchaeota, which was previously thought to be related to the lineage Crenarchaeota [Schleper and Nicol, 2010; Schouten et al., 2002; Wuchter et al., 2004]. Some species of the Thaumarchaeota are extremophiles (e.g., hyperthermophilic organisms) and can grow at temperatures higher than 60°C [Blöchl et al., 1998; Karner et al., 2001]. Marine (pelagic) species are thought to have adjusted their membrane lipids to colder environments by introducing crenarchaeol, a cyclohexane containing GDGT [Sinninghe Damsté et al., 2002]. Hence, TEX86 is not limited by maximum surface temperatures as U37K′ is. Instead, it exhibits a nonlinear behavior for water temperatures below 5°C [Kim et al., 2008, 2010], which are not relevant in our study (Table 1). The TEX86 calibration of recent core top sediments has been applied for water temperature reconstructions of the past 120 Myr [e.g., Dumitrescu et al., 2006; Pearson et al., 2007; Schouten et al., 2007a; Sluijs et al., 2006], and no deviations are known resulting from evolutionary effects of the source organisms. Studies on the effect of differential degradation of GDGTs deposited under oxic or anoxic sedimentary conditions are inconclusive, suggesting either a temperature bias of up to 6°C or no significant change [Huguet et al., 2008, 2009; Kim et al., 2009; Schouten et al., 2004]. On the other hand, a radiocarbon study by Mollenhauer et al.  reveals large differences between radiocarbon contents of alkenones and crenarchaeol, suggesting selective preservation of alkenones, while crenarchaeol might be degraded more rapidly in oxygenated water masses. However, since 11 Ma, oxic degradation should have had relatively small effects on organic molecules used to estimate water temperatures.
 A bias affecting TEX86-based water temperature estimates has been observed in sediments rich in terrestrial organic matter. The ratio of soil-derived GDGTs versus marine crenarchaeol, the BIT index, is used as a proxy of riverine-transported terrestrial organic matter. In theory, BIT = 0 is expected for open ocean conditions, and BIT = 1 is expected for pure terrestrial soils [Hopmans et al., 2004]. Weijers et al. [2006, 2007] reported that river runoff of terrestrial soils may contain GDGTs used for the TEX86 parameter and that large amounts of terrestrial organic matter associated with high BIT values may shift the TEX86 values toward warmer temperatures. Sediment samples of this study are potentially influenced by river runoff from the adjacent Orange River. High BIT values of around 0.88 were found for the middle Miocene (Figure 2b and Table 1) and relatively low values of around 0.13 afterward (Table 1). A TempTEX deviation of +1°C may be reached at a BIT index of 0.2–0.3 [Weijers et al., 2006]. The degree of biasing depends on the GDGT composition in river discharged soil material [Weijers et al., 2006]. The latter, however, is unknown. As the bias is higher for relatively low water temperatures and less so for warm waters [Weijers et al., 2006] but TempTEX estimates with low BIT values at that time indicate that values >25°C and U37K′ values reach their calibration limit close to 28°C, we infer that the overestimation of TempTEX is likely minor. This is supported by the general resemblance of the TempTEX record with deep water cooling (Figure 2a). As BIT values are low during the later periods, the rest of the TempTEX record is not affected by Orange River discharge. In Figure 2 we marked TempTEX values (pale gray) that are associated with BIT values exceeding 0.3. Excluding these potentially biased data points does, however, not change the general trend of steep TempTEX decline during the middle to late Miocene.
5.1.3. Differences Between SSTUK and TempTEX
 We observe an offset between both water temperature proxies between 11 and 7 Ma as well as during the Pliocene (Figure 2), which for the above mentioned reasons we consider not to be related to differential preservation and systematic degradation of individual lipids. It is commonly assumed that alkenone- and GDGT-based temperature estimates represent water temperatures of the upper parts of the water column, because both indices correlate well with mean annual surface temperatures [e.g., Herbert, 2001; Kim et al., 2008; Prahl and Wakeham, 1987; Prahl et al., 2000; Schouten et al., 2002]. Several studies, however, found offsets between SSTUK and TempTEX estimates [e.g., Huguet et al., 2006; Lee et al., 2008; Leider et al., 2010; Lopes dos Santos et al., 2010]. These studies attributed the findings to different seasonal production and/or depth habitats. Alkenones are produced in the euphotic zone by haptophyte phytoplankton species, which require sunlight for photosynthesis. They may be outcompeted, e.g., by diatoms during strong upwelling of silicate-rich deep waters [Mitchell-Innes and Winter, 1987]. Over Site 1085, however, elevated surface productivity is supported only by filaments laterally derived from coastal perennial upwelling [Lutjeharms and Stockton, 1987]. The global core top calibration of U37K′ [Müller et al., 1998] contains samples from the BUS area showing no deviation from the mean annual SSTs.
 SSTUK data from surface sediments of the Cape Basin as well as SSTs derived from satellite data (SSTsatellite) [Lee et al., 2008] agree well, whereas TempTEX data are biased to colder temperatures (Figure 3a). The offset between SSTUK and TempTEX data is smaller at greater water depth or at a greater distance from the upwelling near the coast. Sites within a strong upwelling exhibit the largest offset between both temperature proxies. TOC data of the same core top sediments [Inthorn et al., 2006] correlate to the temperature offset (ΔTemp, Figure 3b). This suggests that ΔTemp depends on the degree of marine primary productivity in the surface waters. We calculated a ΔTemp of our data based on reliable temperature estimates without SSTUK close to the calibration limit and TempTEX accompanied by BIT > 0.3 (Figure 2c) to avoid any potentially biased temperature estimates. Comparing the Miocene with Pliocene records of ΔTemp and TOC (Figure 2d), however, no correlation is observed despite good preservation of organic matter since 11 Ma. The highest ΔTemp is indeed observed during the first increase of TOC at around 11 Ma, but the high TOC values around 6 Ma are not associated to a high ΔTemp and, instead, correlate to similar SSTUK and TempTEX values. Elevated primary productivity alone can thus not explain the temperature offset. Apparently, elevated surface water primary productivity only causes displacement of Thaumarchaeota to subsurface water layers, but temperature variations recorded in the subsurface layers are primarily driven by other processes. Interestingly, TempTEX values in our study never decrease below 15°C, which is the present temperature at the base of the surface water layer extending to around 400 m water depth [Schneider et al., 2003]. We thus infer that TempTEX in the Cape Basin records subsurface water temperatures below the surface mixed layer but above the permanent thermocline. In a recent study by Lopes dos Santos et al.  in the eastern tropical Atlantic a similar conclusion was drawn that TempTEX records thermocline temperatures and connections were made between thermocline adjustments and the strength of the meridional overturning circulation.
5.2. Development of the Benguela Upwelling System (BUS)
5.2.1. Onset and Intensification of Upwelling
 The high-resolution stable oxygen isotope record (δ18O) of benthic foraminifera from ODP Site 1085 (Figure 2a) indicates Antarctic ice growth and global cooling, until 10 Ma, that is associated with the cooling of bottom waters at ODP Site 1085 [Westerhold et al., 2005]. Both late Miocene and Pliocene temperature data sets of this study show a trend to cooler temperatures over 11 Myr (Figure 2b). In conjunction with the benthic δ18O data, the temperature signals represent a record of increasing impact of cold waters of Antarctic origin on surface and subsurface waters of the southeast Atlantic Ocean. TempTEX values agree better with benthic oxygen isotopic values compared with SSTUK values. It seems that the TempTEX record is more strongly affected by changes in temperatures of southern sourced cold water. The SSTUK development, on the other hand, indicates a more gradual cooling in the upper water layers at ODP Site 1085 consisting of ocean surface waters mixed with waters of filaments derived from coastal upwelling areas.
 Large temperature shifts between 3°C and 10°C since the middle Miocene climatic optimum (16 Ma) were found by several studies outside of the BUS [Billups and Schrag, 2002; Browning et al., 2006; Flower and Kennett, 1994; Miller et al., 1991; Shevenell et al., 2004; Williams et al., 2005], while the largest differences between late Miocene and modern water temperatures occurred at low latitudes of the Southern Hemisphere [Williams et al., 2005]. In the eastern Pacific, nearly 50% of the diatom assemblages were replaced by cold-water species during a major middle Miocene cooling from 14.9 to 12.4 Ma, reflecting increased upwelling and cooling of surface waters [Barron and Baldauf, 1990]. Increased Antarctic glaciations promoting intermediate and bottom water formation [Diekmann et al., 2003] and intensification of the Hadley circulation [Flohn, 1978], combined with an intensified Southern Hemisphere southeast wind system [Roters and Henrich, 2010], are inferred causes. Both mechanisms intensified upwelling at the west coasts of the Southern Hemisphere continents. Unfortunately, the lack of reliable temperature estimations for the middle Miocene does not allow dating the start of the cooling trend in the Benguela system, which might have started in the Cape Basin during the early to middle Miocene [cf. Compton et al., 2004].
 Between 11 and 9 Ma the offset between SSTUK and TempTEX increased steeply to a maximum of 7°C at 9.5 Ma and declined and leveled off to 4°C on average afterward (Figure 2c). We interpret this as increased inflow of cold and nutrient-rich intermediate Antarctic waters to the Cape Basin. Between 11 and 10 Ma, the increase of TOC values indicated the start of enhanced marine primary productivity (Figure 2d), which is substantiated by TOC data and foraminifera counts by Diester-Haass et al. . During the TOC maximum between 9.5 and 9 Ma the largest difference between both temperature proxies of up to 7°C is found. We interpret this as the result of cold Antarctic intermediate water advection being overlain by cool surface water filaments originating at the upwelling centers along the coast of the southern BUS. At ODP Site 1085 water of upwelling filaments would have caused changes in surface productivity and the TEX source organisms would have been displaced to subsurface water layers by a gradual increase of the primary productivity in the surface waters. A downward shift of the GDGT source organisms would have resulted in an increasing offset between the two temperature estimates from 11 to 9 Ma (Figures 2b and 2c). By 9 Ma, the upwelling system seems to have been established, resulting in an average temperature difference of ∼4°C between surface and subsurface waters over Site 1085.
5.2.2. Changes in Atlantic Ocean Circulation
 Progressive intensification of upwelling of cold and nutrient-rich Antarctic waters from late Miocene to Pliocene times is reflected by an overall increase in TOC as well as a parallel cooling in SSTUK and TempTEX since 10 Ma, except during the period between 6.5 and 5.0 Ma (Figure 2). At 7 Ma, TempTEX values converge with SSTUK values, and similar or even higher water temperatures were recorded until 5 Ma, indicating warming of subsurface waters (Figure 2b). From the previous interpretation, similar surface and subsurface temperatures could indicate a strong decline in upwelling, which is, however, inconsistent with the relatively high and rising TOC values in combination with high accumulation rates of benthic foraminifera, suggesting high marine primary productivity at ODP Site 1085 during this period [Diester-Haass et al., 2002, 2004] (Figure 2). The high marine productivity at ODP Site 1085 is not a local phenomenon, but is part of global phenomenon [Diekmann et al., 2003; Diester-Haass et al., 2002], which is called the “biogenic bloom” event [e.g., Dickens and Owen, 1999; Farrell et al., 1995]. It has been suggested that changes in intermediate waters caused fundamental changes in the nutrient cycling during this event.
 An explanation for the detected subsurface warming between 6.5 and 5.0 Ma may be related to variability in the strength of North Atlantic deepwater formation on the temperatures of the intermediate waters of the South Atlantic. The existence of Northern Component Deep Water since the late Miocene is assumed by many authors [Berger and Wefer, 1996a; Diester-Haass et al., 2004; Frank et al., 2002; Robert et al., 2005; Woodruff and Savin, 1989]. Deep water exchange between the Pacific Ocean and the Atlantic Ocean was restricted long before the final closure of the Central American Seaway in the Pliocene and might have stimulated North Atlantic deepwater production since the late Miocene [Molnar, 2008; Schneider and Schmittner, 2006]. Lee et al.  document that, at present, when the NADW formation is strong, SSTUK and TempTEX show an offset in the Cape Basin. Conversely, the reduction of subsurface heat export from the South Atlantic and deepening of the thermocline that are due to a reduction in the rate of NADW formation would decrease the offset between both temperature proxies. A subsurface warming and a downward mixing of heat from the South Atlantic thermocline has been observed during times of reduced NADW formation [Huang et al., 2000; Rühlemann et al., 2004]. Models indicate that the warming of the South Atlantic resulting from reduced deepwater formation is most pronounced in the subsurface waters extending to 30°S [e.g., Dahl et al., 2005; Haarsma et al., 2008; Huang et al., 2000; Rühlemann et al., 2004; Vellinga and Wood, 2002]. While the continuous cooling of surface waters in combination with high surface productivity suggests enhanced upwelling, subsurface water temperatures might have increased as a consequence of reduced NADW formation. Hence, the convergence of both temperature proxies between 6.5 and 5 Ma would be the result of cooler surface temperatures and warmer subsurface ones. Similarly, Lopes dos Santos et al.  attributed changes in the offset between SSTUK and TempTEX in the eastern equatorial Atlantic to variations in the strength of the Atlantic overturning circulation.
 The dating of events related to the Messinian Salinity Crisis corresponds to the period for which we observe convergence of surface and subsurface water temperature estimates (6.5–5.0 Ma, Figure 2b). The isolation of the Mediterranean Sea and the cessation of the inflow of salty Mediterranean water into the Atlantic Ocean was originally dated between 5.96 and 5.33 Ma [e.g., Hilgen et al., 2007; Krijgsman et al., 1999; Roveri et al., 2008]. However, van Assen et al.  found evidence that the input of Atlantic waters into the Mediterranean Sea became restricted much earlier, already by 6.84 Ma. Decrease of the saltwater input into the Atlantic would have decreased deepwater formation and thus weakened Atlantic overturning circulation [e.g., Barreiro et al., 2008]. We therefore interpret the strong subsurface warming observed in TempTEX between 6.5 and 5 Ma, indicating a downward mixing of heat from the thermocline, as an effect of reduced NADW formation in association with the reduced salt import into the Atlantic Ocean during the Messinian Salinity Crisis.
 Also for the Pleistocene samples, no offset between both temperature proxies (SSTUK and TempTEX) exists, while TOC is at the same relatively high level as during the late Pliocene (Figures 2b, 2c, and 2d). All three Pleistocene samples used for comparison are from the last glacial age when NADW formation was weak, South Atlantic subsurface waters were relatively warm, and the thermocline shifted downward [cf. Lopes dos Santos et al., 2010; Rühlemann et al., 2004]. This observation is consistent with our proposed scenario that differences between in the Benguela current system reflect the strength of NADW formation.
5.2.3. Intensification of Northern Hemisphere Glaciation
 Since 5 Ma the Northern Hemisphere converted into permanently glaciated conditions with fluctuating ice sheets, which is linked to further uplift of the Isthmus of Panama and, finally, to the closure of the Central American Seaway at around 2.7 to 2.6 Ma [Haug and Tiedemann, 1998; Larsen et al., 1994; Molnar, 2008; Zachos et al., 2001]. The closure of the Seaway caused intensification of the NADW formation that resulted in enhanced heat export from the South Atlantic [Haug and Tiedemann, 1998; Klocker et al., 2005] as well as amplification of Milankovitch cycles starting around 3 Ma [Fedorov et al., 2006]. Additionally, strengthening of global major currents led to a northward shift of the Antarctic polar front and the Subtropical Convergence Zone in the South Atlantic, which might have resulted in an increase in velocity of the BUS [Diester-Haass, 1988; Diester-Haass and Rothe, 1987], changes in zonal temperature gradients, and major changes in midlatitudinal SSTs [Brierley and Fedorov, 2010]. The ocean heat fluxes probably responded to changes in atmospheric winds [Barreiro et al., 2008; Boccaletti et al., 2005]. Intensification of trade winds are therefore expected to initiate a positive feedback and to enhance coastal upwelling intensity [Marlow et al., 2000]. In the southeast Atlantic, evidence for trade wind intensification is provided by the parallel decrease of both SSTUK and TempTEX values from approximately 5 to 2.8 Ma (Figure 2b), as well as by increasing TOC values, which point to further enhancement of marine productivity (Figure 2d) [Diester-Haass et al., 2004].
 At ODP Site 1084 offshore Lüderitz, which today is situated under the highest productive cell of the BUS around 5° of latitude north of ODP Site 1085 (Figure 1), Marlow et al.  found SSTUK values of approximately 26°C to 27°C from 4.6 to 3.2 Ma decreasing to fluctuating values between 12°C and 18°C for the late Pleistocene (Figure 2b). In contrast, our results imply cooler SSTUK of around 21°C until about 3.9 Ma followed by a gradual decrease to 19°C until 2.8 Ma and by a shift to about 16°C and 19°C in the Pleistocene (Figure 2b), similar to values found by Lee et al. . We interpret this as a northward shift of the area of strongest upwelling to a position offshore Lüderitz during the Pliocene, which is suggested by the further cooling of SSTUK after 3 Ma. This implies that the strengthening of upwelling intensity started in the southern BUS and progressed northward later on. Possibly a northward shift of strong upwelling areas occurred as a result of further expansion of polar ice sheets and corresponding frontal systems.
6. Summary and Conclusions
 Two organic molecular proxies for ocean temperatures were used in conjunction with TOC content and published data from ODP Site 1085 to identify the history of the BUS in the southeast Atlantic Ocean from Miocene to Pliocene times. In concordance with global cooling from the Miocene climatic optimum to the Pliocene, both, SSTUK and TempTEX temperature proxies exhibit a cooling trend between 13.7 and 2.8 Ma with a significant overall shift of 8°C and 16°C, respectively. Temperature trends depicted by both proxies differ in rate and timing, which may be related to different habitats of their source organisms. The SSTUK data reflect the warmer surface water temperatures of the euphotic zone, while TempTEX data likely reflect colder subsurface waters. Increased primary productivity presumably leads to migration of the GDGTs source organisms (TempTEX) to subsurface waters below the surface mixed layer.
 TempTEX data exhibit a similar temporal trend as δ18O of benthic foraminifera, likely indicating advection of cold Antarctic intermediate waters, the latter influencing subsurface water temperatures. SSTUK data record a more gradual decrease in surface temperatures. We suggest that the intensified Benguela Upwelling during the Miocene led to a gradual decrease in sea surface temperatures over ODP Site 1085 by mixing with cold waters from filaments originating from coastal upwelling. Rapid cooling of subsurface waters and an increase in marine productivity after 11 Ma point to an intensified upwelling in the southern Benguela system, while a significant offset between both temperature proxies developed at 10 Ma.
 During the period of the Mediterranean Salinity Crisis (6.8 to 5.3 Ma) a rapid warming of subsurface waters to temperatures similar to or exceeding those of the surface waters suggests deepening of the thermocline as the result of decreased Atlantic overturning circulation that was due to a decrease in or cessation of salt input from the Mediterranean Sea into the Atlantic Ocean.
 Parallel cooling of surface and subsurface waters resumed in the Pliocene, when the Northern Hemisphere glaciations intensified.
 We thank Martin Butzin (University of Bremen, Germany), Torsten Bickert (MARUM, University of Bremen, Germany), Gregor Knorr (Alfred Wegener Institute, Bremerhaven, Germany), Stephan Mulitza (MARUM, University of Bremen, Germany) and Gerrit Lohmann (Alfred Wegener Institute, Bremerhaven, Germany) for helpful advice and constructive discussion, as well as Ralph Kreutz and Hella Buschhoff (both from University of Bremen, Germany) and Elke Joost, Katharina Siedenberg, and Abhinav Gogoi (all MARUM, University of Bremen, Germany) for analytical support. John Compton (University of Cape Town, South Africa), Jaap S. Sinninghe Damsté (Royal Netherlands Institute for Sea Research, Netherlands), and an anonymous referee are gratefully acknowledged for constructive reviews. Samples were taken at the ODP repository in Bremen. Data are archived in PANGAEA (http://www.pangaea.de/). This study was financially supported by the Deutsche Forschungsgemeinschaft (DFG, Bonn, Germany) within the research unit “Understanding Cenozoic Climate Cooling: The Role of the Hydrology Cycle, the Carbon Cycle, and Vegetation Changes” (FOR 1070), grant SCHE 903/6.