Marine sediments along the coast of Chile and Peru have been thoroughly studied, as it is a key region in the Southern Hemisphere for studying climate variability related to both atmospheric and oceanographic circulation (e.g., Lamy et al., 1998, 1999, 2002, 2004; Hebbeln et al., 2000, 2002; Mohtadi and Hebbeln, 2004; Stuut and Lamy, 2004; Heusser et al., 2006; Romero et al., 2006; Mohtadi et al., 2008; Muratli et al., 2010a, b; Verleye and Louwye, 2010; Chase et al., 2014). The main circulation patterns include the southern westerly winds (SWWs) and the Antarctic Circumpolar Current (ACC; Fig. 1). The ACC approaches the South American continent from the west and initiates both the Humboldt Current (or Peru–Chile Current) flowing northwards along the continental margin and the Cape Horn Current flowing southwards (e.g., Stuut et al., 2006; Fig. 1). The Humboldt Current flows along the Chilean coast and turns westwards as it approaches the Equator, forming the South Equatorial Current (SEC). Variations in the strength and location of the ACC and southern westerlies are thought to be the main climate controls in this region. Lamy et al. (2002) showed that long-term Holocene trends in temperature and productivity are linked to latitudinal shifts of the ACC and the westerlies. On a millennial to multidecennial scale, paleotemperature variations are strongly connected to millennial-scale climate variations in Antarctica. When the ACC migrates northward, potentially associated with an expansion of Antarctic sea ice, cold subantarctic waters are advected into the Southern Hemisphere midlatitudes, including the Chilean margin (e.g., Lamy et al., 2004; Kaiser et al., 2005). The subtropical high-pressure system is an important climate control north of 33∘ S, which can enhance coastal upwelling (Mohtadi and Hebbeln, 2004).

ODP Site 1234 is located in the southeast Pacific Ocean (36∘13.153′ S, 73∘40.902′ W; 1015 m water depth), ca. 65 km offshore of Concepción, Chile. This site lies in the vicinity of two large Andean river systems: the Río Biobío and Río Itata, which drain basins of 24 000 and 11 200 km2, respectively (Muratli et al., 2010a). The continental margin is under the influence of terrestrial input due to rainfall and land erosion (Lamy et al., 2004), and terrestrial material deposited along the coastal margin derives mainly from two mountain ranges, i.e., the Coast Range and the Andes (Lopez-Escobar et al., 1977; Martin et al., 1999). Moreover, ODP Site 1234 is located within the Peru–Chile upwelling system, which is one of the most important upwelling regions on Earth (Berger et al., 1987) due to the combination of upwelling-favorable winds (Daneri et al., 2000; Carr and Kearns, 2003) and an eastern boundary current (Humboldt Current). The upwelling regime stretches from 5 to 38∘ S, corresponding to ca. 5000 km of coastline. South of 38∘ S, onshore winds suppress upwelling (Strub et al., 1998), but high productivity is sustained by relatively high nutrient supply from the ACC (Hebbeln et al., 2000), which is typically rich in nitrate and limited in micronutrients such as iron (de Baar et al., 1995). Along the Chilean coast, iron availability (as well as other micronutrients) can increase due to fluvial supply as the precipitation of the southern westerlies onshore enhances river runoff (Mohtadi and Hebbeln, 2004).

The core was drilled as part of Ocean Drilling Program (ODP) Leg 202. Sedimentation rates were relatively high (∼80 cm kyr-1 on average; Mix et al., 2003; Heusser et al., 2006). The age model used is from Heusser et al. (2006) and is based upon radiocarbon dates and the correlation of benthic δ18O with deep Atlantic core MD95204 (upper half) and the Vostok ice core chronology (lower half; see Heusser et al., 2006 and references therein). Heusser et al. (2006) switched between the two age models around 69 ka where they overlap, as indicated by the dashed line in Fig. 2a. However, this results in a substantial, and likely unrealistic, dip in the sedimentation rate between ca. 80 and 90 ka in the hiatus between the age models (Fig. 2b). Hence, we chose to use the upper age model (dark blue in Fig. 2), which extends up to ∼83 ka, and linearly interpolate between this age and the age of ∼90 ka constrained by the other age model for the lower half of the core. This depth–age range of linear interpolation between the two age models is highlighted in green in Fig. 2 and results in less abrupt changes in sedimentation rates. The core was sampled with a ∼2 kyr resolution covering the last ∼150 kyr. In total, 74 sediment samples were analyzed for bulk and organic geochemistry.

All 74 sediment samples were freeze-dried and homogenized. Small aliquots (ca. 50–100 mg) were used for elemental analysis. For this purpose, all aliquots were acidified with 2 M hydrochloric acid (HCl) to remove all carbonates and rinsed with distilled water to remove salts. Subsequently, the decalcified sediment samples were analyzed for total organic carbon (TOC), total nitrogen (TN), and stable carbon isotope ratios (δ13C) by means of a Thermo Scientific Flash 2000 elemental analyzer coupled to a Thermo Scientific Delta V Advantage isotope ratio mass spectrometer. Total nitrogen (TN) was measured on aliquots that were not acidified. Results are expressed in standard δ notation relative to Vienna Pee Dee Belemnite (VPDB) for δ13C. The precision as determined using laboratory standards calibrated to certified international reference standards was in all cases < 0.2 ‰. TOC mass accumulation rates (ARs; MARTOC) were calculated by multiplication of the sedimentation rate (linear interpolation between depth points of the age–depth model; see above) with an estimated bulk density of 1.6 g cm-3 (Mix et al., 2003) and subsequent multiplication with the TOC percentage.

Benthic stable isotope data were previously published and were generated using standard techniques in laboratories at Oregon State University and the Woods Hole Oceanographic Institution (McManus et al., 1999, 2002, 2003; Heusser et al., 2006). Oxygen isotope ratios in Cibicidoides and Uvigerina were adjusted to each other by 0.64 ‰ (Shackleton, 1974). Additional data on the planktonic species Globigerina bulloides were generated at Lamont Doherty Earth Observatory (LDEO) of Columbia University using a Thermo Delta V Plus gas-source isotope ratio mass spectrometer (IRMS) equipped with a Kiel IV automated individual acid bath sample-preparation device. Although a recent study found no size-related influence on G. bulloides δ18O (Costa et al., 2017), specimens were generally picked from the 250–300 µm size range, with 8–12 individuals selected for analysis. Samples were replicated at ∼5 %–10 % frequency and measured isotope ratios were calibrated to the VPDB isotope scale with NBS-19 and NBS-18. Reproducibility of the in-house standard (1σ) is ±0.06 ‰ (δ18O) and ±0.04 ‰ (δ13C).

The sediment samples (ca. 15 g dry weight) were extracted using accelerated solvent extraction (ASE) using a DIONEX 200 at 100 ∘C, a pressure of 7–8×106 Pa, and a mixture of dichloromethane (DCM) and methanol (MeOH) (9:1; v:v). The total lipid extracts (TLEs) were dried under a stream of nitrogen (N2) using a Caliper TurboVap LV. All TLEs were desulfurized using copper granules activated with 1 M HCL. The copper turnings were added to the TLEs, stirred overnight, subsequently dried over anhydrous sodium sulfate (Na2SO4) in order to remove precipitate and water, and dried down under N2. For quantification purposes, three internal standards were added to the TLEs : 10-nonadecanone (C19:0 ketone) for long-chain alkenones, C22 7,16-diol for LCDs (Rodrigo-Gámiz et al., 2015), and the C46 GDGT for GDGTs (Huguet et al., 2006). The TLEs (aliquots of ∼4.5 mg) were separated into apolar, ketone (containing alkenones), and polar (containing LCDs and GDGTs) fractions by separation over activated (2 h at 150 ∘C) Al2O3 and elution with hexane : DCM (9:1; v:v), hexane : DCM (1:1; v:v), and DCM : MeOH (1:1; v:v), respectively. Polar fractions were split for GDGT (25 %) and LCD (75 %) analysis.

The TOC content varies between 0.4 % and 2.6 %. The TOC content is significantly higher during the interglacial periods (MIS 1, 3, and 5) compared to glacial periods (MIS 2, 4, and 6; two-tailed p<0.001): the average TOC concentration is 1.4 % for the interglacial periods and 0.7 % for the glacial intervals (Fig. 3d), although TOC levels for MIS 3 are quite similar to those of MIS 1 and 2. Similarly, the average TN levels are 0.2 % and 0.1 % during the interglacial and glacial intervals, respectively (Fig. 3d). During Termination 2, both the TOC and TN contents increase rapidly (within < 2 kyr) towards interglacial values, and the highest TOC and TN values are observed around 78 ka (2.6 % and 0.3 %, respectively). The atomic C/N ratio (Fig. 3b) varies from 4.3 to 10.5 and is on average higher during the interglacial periods (7.2) compared to glacial times (6.2). The organic matter δ13C record (δ13COM) also reveals a glacial–interglacial variation (Fig. 3c) corresponding to slightly 13C-enriched values during interglacial times (δ13Caverage=-21.2 ‰) compared to the glacials (δ13Caverage=-21.8 ‰). Although small, these changes are statistically significant (5 % significance level, two-tailed p<0.001).

Using the age model modified from Heusser et al. (2006), we estimated sedimentation rates that varied between ca. 0.2 and 1.7 mm yr-1 (Fig. 3a). Sedimentation rates were highest in MIS 4, during which the sedimentation rate reaches values of around 2 mm yr-1. Sedimentation rates were lowest during the warmest periods (MIS 1 and 5e; ca. 0.3–0.4 mm yr-1). The MARTOC varied between ca. 2 and 32 g m-2 yr-1 and shows a relatively similar pattern as the sedimentation rate, implying that the sedimentation rate strongly controls the AR of TOC (Fig. 3a). Around ca. 100 ka there is a pronounced maximum in the TOC AR, reaching ca. 32 g m-2 yr-1.

In all sediments, long-chain alkenones, GDGTs, and LCDs were present above the quantification limit. The total concentration of crenarchaeol, a specific biomarker for marine Thaumarchaeota (Schouten et al., 2002, 2003), varies between ca. 2 and 16 µg g-1, with highest values during MIS 5, after MIS 5e (Eemian Interglacial; Fig. 4a). Accumulation rates (ARs) range from 1 to 25 mg m-2 yr-1 (Fig. 4a). The summed concentration of the di- and tri-unsaturated C37 alkenones, a productivity proxy for alkenone-producing haptophytes (e.g., Prahl et al., 1988; Prahl and Muehlhausen, 1989; Rostek et al., 1997; Villanueva et al., 1997), varies between ca. 0.5 and 10 µg g-1. Highest concentrations are observed for MIS 5e (126–116 ka; Fig. 4b). Around the boundary of MIS 4 and 5, concentrations decrease from ca. 6 to 2 µg g-1 sediment; however, the alkenone AR does not show this decrease. During the Late Holocene, C37 alkenone concentrations increase again. The alkenone AR record resembles that of the MARTOC with a peak around 100 ka of ca. 10 mg m-2 yr-1 (Fig. 4b). Dinosterol shows the highest abundance during the last interglacial (MIS 5), peaking just after the Eemian Interglacial (between ca. 116 and 110 ka; Fig. 4c) with concentrations of ca. 4 µg g-1. This peak is followed by a gradual decrease towards the LGM with values near 1 µg g-1 sediment. The AR record of dinosterol highly resembles the concentration record but peaks somewhat later, i.e., between ca. 105 and 98 ka. The loliolide concentration shows one maximum during the Late Holocene of ca. 2.3 µg g-1 sediment (Fig. 4d). Its AR varies between ca. 0 and 1 mg m-2 yr-1 throughout the record, with one pronounced peak around 100 ka of around 2 mg m-2 yr-1.

The main LCDs detected throughout the core are the C28 and C30 1,13-diols, the C28, C30, and C30:1 1,14-diols, and the C30 and C32 1,15-diols. We did not detect the C32 1,14-diol (characteristic for Apedinella radians), but we did identify the C30:1 1,14-diol and C29 12-OH fatty acid, which are typical biomarkers for Proboscia diatoms (Sinninghe Damsté et al., 2003), implying that Proboscia diatoms are most likely the source of the 1,14-diols detected in the sediments. The profile of the summed concentration of 1,14-diols shows four distinct sharp peaks during the MIS 5 (Fig 4f), reaching concentrations up to 1.2 µg g-1. The 1,14-diol AR shows one distinct peak at 94 ka of ca. 1.7 mg m-2 yr-1, whereas throughout the rest of the record the AR varies between 0 and 0.6 mg m-2 yr-1. Figure 4h shows the ARs of the summed 1,13- and 1,15-diols. The AR of the 1,13-diols ranges between ca. 0.1 and 0.8 mg m-2 yr-1 throughout MIS 6 and 5 and peaks (ca. 1.2 mg m-2 yr-1) at the end of MIS 5, revealing the highest ARs during MIS 4, followed by a subsequent gradual decrease towards Holocene values of around 0.2 mg m-2 yr-1. The AR record of the 1,15-diols is highly similar to that of the 1,13-diols for MIS 1 to 4, as well as for MIS 6, but shows higher values during the second half of MIS 5 with maxima around 94, 104, and 122 ka of between ca. 1 and 1.3 mg m-2 yr-1.

We generated a δ18O record of G. bulloides for MIS 5 from ca. 70 to 142 ka, as well as for a short interval within MIS 3 from ca. 36 to 41 ka, which is plotted in Fig. 5b. The δ18O values of G. bulloides varied between ca. 0.3 ‰ and 4.2 ‰ and are most depleted during the Eemian Interglacial around 123 ka (δ18O = 0.3 ‰); enriched values are observed for MIS 6 and 3.

The diol index record shows several maxima, the most evident of which is during MIS 5 (Fig. 4g). During MIS 5e, the diol index reveals two peaks around 126 and 116 ka (up to 0.7), followed by a sharp drop and a subsequent increase around 110 ka, reaching a maximum value at ∼86 ka (up to 0.8). After 86 ka, the index gradually decreases. During MIS 2–4 the diol index varies between ca. 0.1 and 0.5. During MIS 1, the diol index shows a gradual increase from ca. 0.15 to 0.35.

We did not detect the monounsaturated C28 1,14-diol (C28:1 1,14-diol), and therefore this diol could not be included in the calculation of the NDI. In ∼64 % of the surface sediments of the datasets of Rampen et al. (2014b) and de Bar et al. (2016), on which the NDI calibration of Gal et al. (2018) is based, the C28:1 1,14 was also not detected. The NDI record reveals similar variations as the diol index, although more pronounced. During MIS 6 the NDI is close to zero, followed by a small peak at the end of the Eemian Interglacial (NDI = 0.4), coincident with the peak in the diol index at 116 ka. NDI-derived [NO3-] and [PO43-] concentrations around 116 ka are 11 and 0.8 µmol L-1, respectively (Fig. 4h). Similar to the diol index, the ratio then decreases after which the NDI shows a broad peak (∼100–80 ka) with a maximum at 88 ka (NDI = 0.6), which gives concentrations of 20 and 1.4 µmol L-1 when translated to [NO3-] and [PO43-], respectively (Fig. 4h). Between 80 and 0 ka (i.e., MIS 4–1), the NDI does not reveal any distinct maxima, and the ratio varies between ca. 0 and 0.2.

Overall the organic SST proxy records broadly follow the same trend, clearly revealing glacial–interglacial temperature variability (Fig. 5b). Indeed, when we cross-correlate the three proxies, all combinations display significant positive correlations (p<0.001; Fig. 6), albeit with some scatter.

Additionally, the SST trends show good correspondence with the planktonic δ18O record of G. bulloides (Fig. 5c), as well as with the benthic oxygen isotope record for ODP Site 1234 (Heusser et al., 2006; Fig. 5a) and the global stack of benthic δ18O values (e.g., Lisiecki and Raymo, 2005; Fig. 5d). Termination 2 (around 130 ka) is clearly expressed in all three organic proxy SST records, showing a rise in temperature of approximately 4 ∘C within ca. 2 kyr. During MIS 5e, SSTs were between ca. 16 and 18 ∘C based on the three temperature proxies. Then, during MIS 5, the interglacial substages representing alternating cold and warm periods are clearly reflected in all three proxy records. The three records show temperature drops between ca. 4 and 6 ∘C during MIS 5a (ca. 70 ka). During MIS 4, 3, and 2, all SST records show a gradual decrease towards the LGM, during which U37K′ and TEX86H reveal temperatures between 9 and 10 ∘C and an LDI between 7 and 8 ∘C. Around 25 ka, U37K′ and TEX86H temperatures show a similar steady rise to Holocene temperatures of around 16 and 18 ∘C. The LDI reveals an approximate 7 ∘C warming over Termination 1, starting slightly later at around 22 ka. During the last ∼20 kyr, LDI-derived SSTs are ∼3–4 ∘C lower compared to the U37K′- and TEX86H-derived SSTs.

The average bulk organic carbon δ13C over the last 150 kyr is -21.4 (±0.6, SD) ‰, and the atomic C/N ratio displays an average value of 6.8 (±1.3), both implying a dominant marine organic carbon source (Bordovskiy, 1965; Emerson and Hedges, 1988; Meyers, 1997). Today, the amount of sedimentary organic carbon is a good indicator of export production along the Chilean margin (Hebbeln et al., 2000). Hence, higher TOC levels during the interglacial intervals in the ODP 1234 record may suggest enhanced marine productivity during these intervals for this site. However, enhanced interglacial productivity is in contrast with previous studies that state that coastal productivity on the Peru margin was highest during the LGM and diminished during the Holocene (e.g., Thomas et al., 1994; Marchant et al., 1999; Thomas, 1999; Hebbeln et al., 2000, 2002; Lamy et al., 2002, 2004; Romero and Hebbeln, 2003; Mohtadi and Hebbeln, 2004). These studies suggest that during the LGM the ACC migrated northward, supplying nutrients (particularly nitrate and phosphate; Levitus et al., 1994), together with a northward shift of the southern westerly belt as the main precipitation source onshore, resulting in enhanced micronutrient supply via continental runoff. This combined effect would have stimulated productivity along the Chilean coast during the LGM. Upon deglaciation the climate zones propagated southward, resulting in a lowering of productivity, as indicated by pollen, sedimentological, and continental studies (e.g., Heusser, 1990; Lamy et al., 1998, 1999, 2001, 2004; Brathauer and Abelmann, 1999; Haberle and Bennett, 2004; Stuut and Lamy, 2004; Kaiser et al., 2005; Heusser et al., 2006). These studies suggest a 5 to 6∘ northward movement of the southern westerly belt during the LGM.

In contrast to the hypothesis of stimulated productivity during the LGM and lower productivity during the Holocene, two studies suggested low paleoproductivity during the LGM for core sites at the same latitude as that of our core (35–36∘ S) (Romero et al., 2006; Mohtadi et al., 2008). Mohtadi et al. (2008) hypothesized that if the climate zones shifted 5 to 6∘ northward during the LGM, the SWW would be just above the core site (35–36∘ S), blowing directly onshore and thereby preventing coastal upwelling. In turn, during the Early Holocene the subtropical high pressure would have become the dominant atmospheric player according to Romero et al. (2006), favoring upwelling. However, Muratli et al. (2010b), who reconstructed paleoproductivity over the last ∼30 kyr for ODP 1234, suggested that the rise in TOC after the LGM is probably not the result of increased productivity, but of a lower oxygen availability due to decreased Antarctic Intermediate Water ventilation and thus increased preservation. The authors based these suggestions on the fact that neither opal concentrations nor opal or TOC MARs increased simultaneously with TOC concentrations. Chase et al. (2014) support this hypothesis based on the Th-normalized organic carbon fluxes. Our (unnormalized) MARTOC supports these findings as well since although TOC levels show a steady rise after the LGM, we do not observe this increase in the accumulation of organic carbon, and the MARTOC is higher during the LGM compared to the Holocene, suggesting higher glacial productivity. In summary, our findings seem consistent with previous studies suggesting that higher TOC levels during the Holocene likely do not result from increased productivity, but rather from increased preservation.

Interestingly, during MIS 3, 4, and 5 we observe even higher MARTOC values, reaching a pronounced maximum around 100 ka during MIS 5, i.e., during an interglacial period. Hence, this record suggests that the general productivity over the last ∼150 kyr was highest around 100 ka. This interpretation is supported by most of the AR records of the individual lipid biomarkers peaking around 100 ka. The C37 alkenone, crenarchaeol, and dinosterol ARs are all at their maximum between ca. 105 and 95 ka (Fig. 4b, a, and c, respectively), simultaneous with maximal MARTOC (Fig. 4j). The AR of loliolide (indicating diatom abundance) peaks at 100 ka (Fig. 4d), coincident with a maximum in total diatoms and Chaetoceros diatom counts (Mix et al., 2003; Fig. 4c) and concurrent with the peak in MARTOC. Chaetoceros diatoms are generally associated with upwelling (e.g., Abrantes, 1988; Abrantes and Moita, 1999), and also for the Peru–Chile upwelling system, the diatom genus Chaetoceros is thought to dominate the diatom community during upwelling conditions (e.g., Anabalón et al., 2007; Schrader and Sorknes, 1991; Romero et al., 2001; Vargas et al., 2004; Abrantes et al., 2007; González et al., 2007; Sanchez et al., 2012). Therefore, the high abundance of Chaetoceros diatoms at ∼100 ka suggests that there was a peak in upwelling intensity around this time. This upwelling would have introduced macronutrients from deeper colder waters into the euphotic zone, stimulating several phytoplankton communities, including haptophytes, dinoflagellates, and diatoms. During MIS 4 we also observe a peak in MARTOC (Fig. 3a) but there is no corresponding peak in Chaetoceros abundance (Fig. 4i). However, the high MARTOC during MIS 4 is linked to high sedimentation rates (Fig. 3c), whereas the peak in MARTOC during MIS 5 is not, confirming that the MARTOC maximum during MIS 5 indeed likely resulted from increased primary productivity, whereas during MIS 4 the high MARTOC did not.

Interestingly, peaks in the diol index and 1,14-diol concentrations and accumulation rates occur before and after (around 116 and 88 ka; Fig. 4a and j), but not during, the time interval with enhanced upwelling (i.e., around 100 ka; purple band in Fig. 3). Contreras et al. (2010) also reported high abundances of the C28 1,14-diol around 120 ka near the coast of Peru (11∘ S). This suggests that Proboscia diatoms were more abundant before and after this period of intense upwelling and is in agreement with observations that in the present-day Chile–Peru upwelling region, Proboscia alata is more dominant when upwelling is less intense (Tarazona et al., 2003; Herrera and Escribano, 2006). Although the recent timescale is quite different from our long time record, which represents an integrated signal of several hundreds to thousands of years, it suggests that over the time period 120–80 kyr, on average, upwelling became stronger, reaching a maximum at 100 ka before subsequently decreasing. Shortly before and after the period of maximum upwelling, the conditions (averaged over multiple years) were apparently optimal for Proboscia diatoms. As previously suggested by Rampen et al. (2014b), this indicates that the diol index should perhaps be considered as a specific indicator for Proboscia productivity, rather than upwelling strength generally, as the environmental conditions determining Proboscia abundance likely differ from region to region.

Recently, the NDI was introduced as a quantitative paleo-nutrient proxy. The diol index, the 1,14-diol AR, and the NDI are low during the time of most intense upwelling (around 100 ka), suggesting a minimum in annual mean nitrate and phosphate concentrations, which is highly unlikely. The NDI is based on the saturated and monounsaturated C28 1,14-diol relative to other diols. Whereas we observe a strong correlation between the ARs of the monounsaturated and saturated C30 1,14-diol (R2=0.70), the correlations between the C28 1,14 and C30 1,14-diol (R2=0.32) and the C28 1,14 and C30:1 1,14-diol (R2=0.36) are weak, indicating different source organisms for the C28 and C30 1,14-diols, in agreement with previous studies (Rampen et al., 2014b; de Bar et al., 2016; Gal et al., 2018). In fact, Sinninghe Damsté et al. (2003) found that 98 % of the diols in P. alata consisted of the saturated and monounsaturated C28 1,14-diol, whereas P. indica produces similar amounts of the C30 and C28 1,14-diol and P. inermis mainly produced the C28 1,14-diol (Rampen et al., 2007). P. alata is the Proboscia species detected along the modern coast of Chile and Peru (Tarazona et al., 2003; Herrera and Escribano, 2006), and therefore it is likely that P. alata is the predominant producer of the C28 1,1,4-diol detected in the ODP 1234 sediments. Consequently, the NDI record likely reflects variations in the abundance of P. alata over the last 150 kyr in the Chilean margin, whereas the diol index (which also includes the C30 1,14-diol) more likely reflects the abundance of multiple species of Proboscia.

Previous studies showed that P. alata preferentially blooms over other diatoms when nutrients are high but silica concentrations are low (as P. alata needs little Si to build its frustule) (Haake et al., 1993; Sakka et al., 1999; Koning et al., 2001; Smith, 2001). In fact, although P. alata is often observed in high nutrient and/or upwelling regions (Hernández-Becerril, 1995; Lange et al., 1998; Koning et al., 2001; Smith, 2001), the conditions in which the species is abundant are often described as post-bloom, stratification, early upwelling season and/or the oceanic side of the upwelling front (e.g., Hart, 1942; Takahashi et al., 1994; Katsuki et al., 2003; Moita et al., 2003; Tarazona et al., 2003; Herrera and Escribano, 2006; Sukhanova et al., 2006; see references in Table 1 of Rampen et al., 2014b). Moreover, sediment trap studies from the Arabian Sea showed that the maximum flux of Proboscia lipids was at the start of the upwelling season (Prahl et al., 2000; Wakeham et al., 2002; Sinninghe Damsté et al., 2003; Rampen et al., 2007). When silicate concentrations increase (i.e., during upwelling), P. alata is likely outcompeted by more heavily silicified diatoms (such as Chaetoceros; e.g., Riegman et al., 1996). This suggests that the C28 1,14-diol likely reflects early or post-upwelling nutrient conditions. Hence, the dip in reconstructed [PO43-] and [NO3-] concentrations at ∼100 ka is not realistic and likely due to the low abundance of P. alata at this time of intense upwelling. In other regions, the C28 1,14-diol might be produced by other Proboscia species (e.g., P. inermis) and the NDI might thus not reflect solely P. alata productivity. In summary, we suggest that the NDI at the Chilean margin likely reflects Proboscia productivity and may therefore not be suitable as paleo-nutrient tracer.

The three organic proxy-based SST records broadly follow the trend of the planktonic oxygen isotope record for MIS 5, as well as the benthic oxygen isotope record for ODP Site 1234 (Heusser et al., 2006; Fig 5), and the global stack of benthic δ18O (e.g., Lisiecki and Raymo, 2005) for the entire record, suggesting that the evolution of SST at our study location generally follows global climate patterns. The resolution of our record is not high enough to recognize possible millennial-scale SST variations related to climatic events such as the Antarctic Cold Reversal. Nevertheless, the overall SST patterns are similar to other Southern Hemisphere records (e.g., Kaiser et al., 2005; Kaiser and Lamy, 2010; Caniupán et al., 2011; Lopes dos Santos et al., 2013) and Antarctic ice core stable isotope records (e.g., Blunier and Brook, 2001). The three temperature proxies are all significantly positively correlated and present a coherent view of regional climate variability (Fig. 6).

In principle, we would expect similar reconstructed temperatures from the LDI and U37K′ index, since both proxies are based on biomarkers produced by photosynthetic algae (Prahl and Wakeham, 1987; Rampen et al., 2012). Indeed, Kim et al. (2002) showed that for surface sediments off Chile, U37K′-derived temperatures strongly correlated with annual mean temperatures of the sea surface mixed layer. The linear regression for the U37K′ vs. the LDI is close to the 1:1 line, suggesting the LDI is reflecting SST. Interestingly, the LDI reveals a greater amplitude over the record compared to the U37K′ and TEX86H, with a maximum temperature difference between the coldest (LGM) and warmest (Eemian Interglacial) temperature of ca. 10 ∘C, whereas this is ca. 7 and 8 ∘C for the U37K′ and TEX86H, respectively. Other LDI and U37K′ applications on glacial–interglacial timescales also show somewhat greater amplitudes for the LDI compared to the U37K′ (Rampen et al., 2012; Lopes dos Santos et al., 2013; Rodrigo-Gámiz et al., 2014; Jonas et al., 2017). This might potentially suggest that the surface sediment calibration of the LDI requires some modification, such as a lower slope. Further analysis of surface sediments should reveal this.

For the TEX86, it has been shown to potentially reflect subsurface rather than surface water temperatures (Huguet et al., 2007; Kim et al., 2010, 2015; Schouten et al., 2013; Chen et al., 2014) due to the production of isoprenoid GDGTs below the surface mixed layer. Overall, the TEX86H record agrees reasonably well with the U37K′ record for ODP 1234, suggesting that it mainly reflects SST. Also, Kaiser et al. (2015), who established a regional TEX86H calibration, suggested that this proxy mainly reflects SST. However, all cross-correlations still reveal relatively large scatter, reflecting the multiple different constraints on the respective proxies, as each one is affected by different parameters and can reflect different seasonal temperatures. Nevertheless, for the largest part of the record, the absolute temperature differences between the three temperature proxies is smaller than the maximal possible discrepancy that can be explained by the combined calibration errors (calibration errors of U37K′, TEX86H, and LDI are 1.5, 2.5, and 2.0 ∘C, respectively; Müller et al., 1998; Kim et al., 2010; Rampen et al., 2012, respectively).

The only period during which the offsets are larger than the calibration errors is the last ∼25 kyr, i.e., from the LGM to the Holocene, and the relatively brief interval between 52 and 56 ka. During these time spans, the LDI-derived temperatures are ca. 3 to 6 ∘C lower than those derived from U37K′ or TEX86H. This offset could potentially be related to terrestrial input; however, we have assessed the relative contribution of terrestrially derived organic carbon by means of the BIT index (Hopmans et al., 2004) and the fractional abundance of the C32 1,15-diol (de Bar et al., 2016; Lattaud et al., 2017a, b), which were both always < 0.2, implying that there was no substantial riverine input from terrigenous organic matter. Alternatively, the LDI might be compromised by a high input of 1,14-diols (Rodrigo-Gámiz et al., 2015); however, we do not observe a correlation between the LDI (or the SST offset between the LDI and U37K′ or TEX86H) and the fractional abundance of the summed 1,14-diols. Potentially, the production of the 1,13- and 1,15-diols shifted to a colder season due to regional climatic change associated with the southward migration of the ACC and SWW upon the deglacial warming. Finally, we observe an early onset of the deglacial warming for the LDI compared to the U37K′ and TEX86H (ca. 6 kyr; Fig. 5b), the reasons for which are elusive.

We can compare our SST records of ODP Site 1234 with those of other SST records along the coast of South America, in particular those generated by the U37K′, which has been used the most often (Fig. 7). Our U37K′ temperature record agrees well with other records in the vicinity of ODP 1234, both in terms of absolute temperature and in glacial–interglacial temperature amplitude. Our U37K′ SST record shows an approximate 7 ∘C warming over Termination 1, which is in agreement with other past SST records for the central and southern Chilean margin (Kim et al., 2002; Lamy et al., 2002, 2004; Kaiser et al., 2005; Romero et al., 2006; Kaiser and Lamy, 2010). Although the timing of deglacial warming associated with Termination 2 is comparable to other records, the last deglaciation (Termination 1) reveals an early start (around 26 ka) at Site 1234 compared to neighboring sites. The TEX86H record reveals this same timing, whereas the LDI suggests a timing more comparable to other records in the vicinity (around 20 ka). Regional differences in the timing of deglacial warming have previously been related to the southward shift of the SWW, which directly and indirectly influences local paleoproductivity and upwelling intensity, thereby potentially leaving site-dependent, unique signatures in the SST records (Mohtadi et al., 2008). This may partially explain the discrepancies that we observe between our U37K′ record and the other records. For instance, site GeoB 3327-5 is located at the northern extent of the ACC and thus largely influenced by its latitudinal movement, which is closely related to the migration of the SWW (Ho et al., 2012), whereas for ODP 1234 this influence might be less. Furthermore, the combined regional records show an increase in the magnitude of U37K′ temperature variations over glacial–interglacial timescales towards the south (Fig. 7), which agrees with the idea of greater glacial–interglacial temperature differences at high latitudes compared to low latitudes (e.g., CLIMAP, 1976; Mohtadi and Hebbeln, 2004; Mohtadi et al., 2008).