Site 1168 (42∘36.5809′ S, 144∘24.7620′ E; 2463 m modern water depth; Fig. 1) is located on the continental slope of the western Tasmanian continental margin. The site sits on the northern edge of the Subtropical Convergence Zone, which separates warm, saline subtropical waters from comparably cold and fresh subantarctic water masses (Heath et al., 1985; Exon et al., 2001), with a modern SST ranging from 13–17 ∘C (winter–summer). During the Neogene, the location of Site 1168 tectonically drifted along with Tasmania and Australia from 52∘ S at 23 Ma to its modern position at 42∘ S (Van Hinsbergen et al., 2015). During this northward tectonic drift, the southern margin of Australia was continuously bathed by the eastward-flowing proto-Leeuwin Current (McGowran et al., 2004; Hoem et al., 2021). Hence, Site 1168 is well suited to the study of the Neogene evolution of the ACC and the STF.

The post-cruise bio-magnetostratigraphic age model includes nannofossil, planktonic foraminifer, diatom, radiolarian and dinocyst biostratigraphy, with constraints from magnetostratigraphy and stable isotope data (Stickley et al., 2004). Here, we recalibrated these data to the Geological Time Scale 2020 (Gradstein et al., 2020) by using state-of-the-art biostratigraphic constraints from Nannotax and Foramtax as well as updated diatom biostratigraphic constraints (Cody et al., 2008). We then fitted a loess smooth curve through these data, whereby we assigned a 10-fold weight to magnetostratigraphic and benthic δ18O tie points. We interpolated this loess curve to obtain ages for the samples. We derive average sediment accumulation rates of 1.8 cm kyr-1 for the top 360 m (22–0 Ma) and 7.1 cm kyr-1 between 360 and 462 m below seafloor (m b.s.f.; 23–22 Ma; Fig. 2).

A total of 883.5 m of sediment was recovered from Site1168 Hole A, dating back to the late Eocene to modern (Exon et al., 2001). The Neogene interval is represented in the upper 413 m. Between 260–413 m b.s.f. (early to mid-Miocene; 23.0–15.6 Ma), sediments are comprised of clay-bearing nannofossil chalk with a gradual decrease of non-carbonate minerals (Robert, 2004). The upper 260 m (mid-Miocene to modern; 15.6–0 Ma) contains calcareous biogenic oozes, with 85–97 wt % calcium carbonate (Exon et al., 2001); a sharp decrease in detrital clay content occurs at the boundary between these lithologic units (Robert, 2004).

Lipid biomarkers were extracted from 433 powdered and freeze-dried samples with a Milestone ETHOS X microwave system using dichloromethane / methanol (DCM / MeOH) 9:1 (v/v). Activated Al2O3 columns were used for the separation of the total lipid extract into three fractions, using the solvent mixtures hexane / DCM 9:1 (v/v), hexane / DCM 1:1 (v/v) and DCM / MeOH 1:1 (v/v) for apolar, ketone and polar fractions, respectively. Polar fractions were filtered using a 0.45 µm polytetrafluorethylene filter and analysed using an Agilent 1260 Infinity series HPLC system coupled to an Agilent 6130 single quadrupole mass spectrometer, following the instrumental and analytical setup described in Hopmans et al. (2016). A total of 99 ng of C46 standard was added to the polar fraction in order to quantify the absolute concentration of GDGTs. The ketone fractions of 163 out of 433 samples were dissolved in ethyl acetate and analysed on a gas chromatograph (GC) coupled to a flame ionisation detector (GC-FID, Hewlett Packard 6890 series) equipped with a CP-Sil 5 fused-silica capillary column (25 m × 0.32 mm; film thickness 0.12 µm) and a 0.53 mm precolumn. Samples were injected on-column at 70 ∘C with helium as a carrier gas at a constant pressure of 100 kPa. The oven programme was as follows: 70 ∘C for 1 min, then it was ramped to 130 ∘C at 20 ∘C min-1, then to 320 ∘C at 4 ∘C min-1, and then it was held to be isothermal for 10 min. Di- and tri-unsaturated C37 alkenones were identified based on retention time. Data are stored at Zenodo (Bijl et al., 2022).

We follow the approach by Sluijs et al. (2020) and Bijl et al. (2021) to assess non-temperature factors in relation to the relative distribution of isoGDGTs and thus the TEX86 value they represent. Briefly, this involves checking the weighted average of cyclopentane moieties of isoGDGTs compared to modern values (with the Ring Index – RI; Zhang et al., 2016), overprints from methanotrophic archaea (with the Methane Index – MI; Zhang et al., 2011; Weijers et al., 2011) and methanogens (with the GDGT-0 / Cren ratio; Blaga et al., 2009), as well as contributions from deep-dwelling archaea (with the GDGT-2 / GDGT-3 ratio; Taylor et al., 2013) or terrestrial GDGTs (with the BIT index; Hopmans et al., 2004; Weijers et al., 2006). The BIT index is determined by the ratio of branched GDGTs (brGDGTs) produced by terrestrial bacteria and marine-originated crenarchaeol. However, recent studies have proved that brGDGTs can be produced in situ in marine environments (Peterse et al., 2009; Sinninghe Damsté, 2016; Dearing Crampton-Flood et al., 2019). Thus, the sources of brGDGTs are assessed using the weighted number of cyclopentane moieties in tetramethylated branched GDGTs (#ringstetra), where a value of > 0.7 is assumed to indicate a marine rather than a terrestrial source of these compounds (Sinninghe Damsté, 2016). In order to assess the influence of potential algae distribution on the U37k′-based SST reconstruction, we explored the ratio between C37 and C38 (C37 / C38; Rosell-Melé et al., 1994) and the ratio between all C37 and the ethyl C38 alkenones (C37 / C38Et; Zheng et al., 2019).

Numerous calibrations have been implemented to translate TEX86 into sea surface temperature (e.g. Schouten et al., 2002; Kim et al., 2010; Tierney and Tingley, 2014). However, improved understanding of archaea ecology questions the validity of TEX86 as a true proxy for the ocean mixed-layer temperature. This is especially due to the variable export production zone depth (50–200 m) of marine Thaumarchaeota. Fortunately, this can be revealed by the GDGT-2 / GDGT-3 ratio, which suggests that many modern core top samples actually receive contributions from deep-dwelling archaea (Van der Weijst et al., 2022). Thus, the ambient temperature of Thaumarchaeota, which determines the cyclisation of GDGTs, is not strictly sea surface temperature. Even though the GDGTs may, to a variable extent, derive from around the thermocline, it was shown that SST has a strong relationship with surface temperature (Van der Weijst et al., 2022). Subsurface calibrations (Tierney and Tingley, 2014; Kim et al., 2015; Ho and Laepple, 2016) use variable methods to integrate temperature over depth; these methods still induce uncertainty with regard to their reliability. Nonetheless, even though a perfect calibration does not exist yet, TEX86 is still a valuable proxy that reflects the temperature of a relatively stable layer of the ocean (e.g. Kim et al., 2016; Hurley et al., 2018) and provides a robust ocean temperature change in both trend and variability (Van der Weijst et al., 2022) because the relationship between the TEX86 in sediments and surface temperature actually derives from the strong relationship between subsurface and surface temperature. In particular, when TEX86 is used along with other temperature proxies (e.g. Super et al., 2020; Leutert et al., 2020), such as the one we use here (U37k′), both proxies together can provide better constrains on SST reconstructions.

Here, we apply the spatial linear Bayesian calibration BAYSPAR (Tierney and Tingley, 2014, 2015) to translate TEX86 values into temperatures using both surface (0–20 m) and depth-integrated temperature (0–200 m) calibrations (prior mean of 20 ∘C; prior standard deviation of 20 ∘C). We applied the U37k′ paleothermometer based on alkenones as an independent additional paleothermometer. U37k′ index values were calculated following Prahl and Wakeham (1987) and were converted to SST using the BAYSPLINE calibration of Tierney and Tingley (2018) (prior standard deviation of 10 ∘C). We choose the Bayesian calibrations for both proxies for the consideration of both high-temperature applicability and spatial characteristics. These two proxies, together with the TEX86-related overprint indices, are combined to determine the sea surface temperature (SST) change.

Our new GDGTs data are derived from 412.7 to 0 m b.s.f. (22.6–0 Ma). The concentrations of all GDGTs are consistently high (∼ 50 µg g-1 sediment for total isoGDGTs; see Fig. S1 in the Supplement) in the early Miocene and show a normal relative distribution, except for the interval around the MCO (287–256 m b.s.f.; Fig. S2). The isoGDGT concentration drops to 5 µg g-1 sediment at the onset of the MCO and remains stable until 7 Ma. In the MCO interval, of all GDGTs, crenarchaeol (cren) and crenarchaeol (cren') decrease most strongly (to 1/400; GDGT-3 decreases to 1/200; GDGT-2 and GDGT-1 decrease to 1/100; GDGT-0 decreases to 1/25; Fig. S1). As cren is not in the TEX86 index, its anomalous trends in abundance do not directly affect TEX86 values but do affect the GDGT indices and ratios that have cren in the denominator. As a result of the extra decline in cren, GDGT-0 / cren, MI and GDGT-2 / cren all yield abnormally high values in the MCO interval (Fig. 3). The GDGT-2/3 ratio gradually increases from 5 to 8 throughout the record, with transient peaks in the early Miocene and MCO. Cut-off values for this ratio vary between 3 and 10 among users and sites (e.g. Leutert et al., 2020; Bijl et al., 2021; Van der Weijst et al., 2022). Hurley et al. (2018) demonstrated that the GDGT-2 / GDGT-3 values rapidly rise from 3–5 in the surface mixed layer (upper 150 m) to 20–25 at the thermocline depth (see also Basse et al., 2014; Hernández-Sánchez et al., 2014; Kim et al., 2016; Van der Weijst et al., 2022). In any case, sediments with GDGT-2 / GDGT-3 values > 3 might, to some extent, be biased towards deeper waters and thus to lower temperatures. The ΔRing Index varied from -2.2 to 1 in the whole record, and 152 data points fell outside the 95 % confidence interval of the RI-TEX86 array (Fig. S3).

BIT index values show a large range of variation, between 0.1 and 0.9, and show a prominent peak (∼ 0.9) during the MCO and consistent low values (∼ 0.1) in the Pliocene, indicating a potentially large contribution of GDGTs from land (Fig. 3). However, the #ringstetra values are highly varied throughout time but are consistently elevated between 17 and 7 Ma, from 0.3 to more than 1.0, suggesting that brGDGTs have an in situ marine origin. In a ternary diagram of the tetra-, penta- and hexa-methylated brGDGTs, Site 1168 samples generally plot offset to the global soil cluster (Fig. S4), which also supports a non-soil origin. This would imply that the BIT index cannot be interpreted as an indicator of the input of terrestrial matter at this site.

TEX86 values of the early Miocene were around 0.65 but fluctuated, then they reached 0.8 at 16 Ma in the MCO interval, although these values are probably affected by non-thermal overprints. There is an abrupt decline in TEX86 values after the MCO and then a long-term decrease to 0.4 until 5 Ma. TEX86 values increased to 0.6 in the early Pliocene, then decreased to 0.36 at 1.35 Ma and eventually increased to 0.52 in the youngest sediment (Fig. 3). SSTs derived from the TEX86 are around 25 ∘C in the early Miocene section. TEX86 values at the peak MCO would equate to SSTs of 34 ∘C, followed by first a rapid cooling at around 14.5 Ma and then a more gradual down to 7 ∘C towards the end of the Miocene (Fig. 4). After an ephemeral warming to 20 ∘C in the early Pliocene, SST decreased to 6 ∘C in the mid-Pleistocene and then recovered to the modern level around 17 ∘C. Throughout the record, the difference between temperatures derived from surface and depth-integrated subsurface calibrations is small (< 2 ∘C).

Our new alkenone data are derived from 363 to 0 m b.s.f. (21.9–0 Ma). Existing U37k′ data from this site were published in Guitián and Stoll (2021). C38 alkenones are mostly at or below detection limit in the sediments older than 8 Ma. In the younger sediments, we find four clear peaks that represent C38 alkenones. Here, C37 / C38 fluctuates between 0.9 and 1.2, while C37 / C38Et varies between 1.2 and 1.4 (Fig. S5). The U37k′ index record varies between 0.43 and 0.93 and generally follows the trends of TEX86, except during the MCO and the early Pleistocene (Fig. 4). Early Miocene sediments have an average U37k′ value of ∼ 0.8, in agreement with previously published data from the same site (Guitián and Stoll, 2021). In the MCO interval, U37k′ rose to 0.93. None of the analysed sediments in the MCO interval have saturated U37k′ index values. Subsequently, U37k′ gradually drops to 0.43 at the end of the Miocene and recovers to ∼ 0.6 in the Pliocene. U37k′-based SSTs reveal similar temperatures to TEX86-based subsurface temperature except in the MCO interval (Fig. 4). In the early Miocene, SSTU37k′ yields ∼ 24 ∘C on average, then increases to ∼ 28 ∘C in the MCO. Subsequently, SSTU37k′ cools down to 10 ∘C at 5 Ma and increases to 16 ∘C in the Pliocene. SSTU37k′ shows a similar trend to that of SSTTEX86 in the Pleistocene and varies between 17 and 7.5 ∘C. However, the difference between SSTU37k′ and SSTTEX86 increases from the onset of the Pleistocene due to the plunge of SSTTEX86.

In all intervals of the records besides the MCO, temperature estimates derived from the U37k′ and TEX86 have similar trends and absolute temperatures, demonstrating that both proxies represent the same water layer. The large U37k′ / TEX86 discrepancy during the MCO suggests that one of the paleotemperature proxies is affected by non-thermal overprints (Figs. 4, 5). Given the many indices that signal anomalous isoGDGT distributions in the MCO interval, it is likely that TEX86 has a non-pelagic GDGT assemblage and thus reflects an unreliable SST. However, the high GDGT-0 / cren, MI, GDGT-2 / cren and BIT index values in the MCO interval are all caused by the anomalously low contribution of cren (i.e. the denominator) and not that of the elevated relative abundances of the signalling compound for that overprint (the nominator). Moreover, the low total organic carbon (TOC) wt % (< 0.5 %; Exon et al., 2001) in the MCO sediments does not support the existence of any cold seeps, anaerobic oxidation of methane or methane hydrate production despite the high GDGT-0 / cren, MI and GDGT-2 / cren values (Fig. 3). If it is indeed the excess relative decrease in cren concentration that causes the high BIT, GDGT-0 / cren, MI and GDGT-2 / cren and the low RI because all have cren involved in their equation – the question is to what extent this affects TEX86 values. In any case, the indices may not necessarily reflect the overprints that they are usually associated with at this site. However, since TEX86 and U37k′ disagree in the same interval as the anomalous GDGT compositions, and to be conservative, we discard the TEX86-based SSTs with high confounding-factor values in the MCO interval at this site.

The anomalously low relative abundance of cren could be explained by different preservation efficiency and/or degradation rates for distinct GDGTs. The interval in which GDGT concentrations decrease (17.34–16.85 Ma, 294.66–274.28 m b.s.f.) is concomitant to an interval of decreased terrigenous clay and quartz and increased calcium carbonate content (Fig. 5; Robert, 2004). The loss of terrigenous clay can lead to reduced preservation of organic matter (Ransom et al., 1998; Wu et al., 2019) because enhanced pore-water flow in the overlying sediment enhances oxygen exposure time (Huguet et al., 2008; Schouten et al., 2013). It was shown that isoGDGTs with more cyclopentane moieties are less resistant to oxidation than those with less cyclopentanes (Ding et al., 2013). As a result, degradation processes would result in lower TEX86 values and an underestimation of SSTs. However, in the MCO interval, TEX86 was very high and led to SST reconstructions that were much higher than SSTUK37. While selective degradation could be an explanation for the relatively excessive loss of cren, it cannot explain the anomalously high TEX86-based SSTs in that interval. On the other hand, preferential degradation of alkenones could have biased U37k′ at this site as well, notably in the form of a warm bias in SST (Freitas et al., 2017). Yet, in the MCO interval, we do not observe anomalous warmth, and the index is not yet saturated. Therefore, the selective degradation in alkenones cannot explain the SST difference between SSTs derived from U37k′ and TEX86.

The extremely high (up to 100) GDGT-2 / GDGT-3 ratio found in the MCO is unprecedented in both paleo and modern ocean records (e.g. Taylor et al., 2013; Hernández-Sánchez et al., 2014; Hurley et al., 2018; Besseling et al., 2019; Bijl et al., 2021; Van der Weijst et al., 2022). Perhaps the isoGDGTs found in the MCO interval were produced by other archaeal communities, e.g. Marine Group II and/or III (Besseling et al., 2019), and the ratio could have been further increased through selective degradation of GDGT-3 over GDGT-2. Given the temperature trend in the MCO interval, these GDGT producers may have also responded to water temperature, although indirectly or with a different dependency. In the early and late Miocene, GDGT-2 / GDGT-3 values are still relatively high (> 5), indicating some isoGDGT input from deep-water sources. The overall input of deeper-dwelling GDGTs may bias the reconstruction of absolute SSTs from TEX86, but because the GDGT2/3 ratio in our post-MMCT interval is stable and without a trend, the trend of TEX86 should not be influenced (Ho and Laepple, 2016; Leutert et al., 2020; Van der Weijst et al., 2022), and the amplitude is well constrained by U37k′. Thus, for the late Miocene and Pliocene, the sediments are not discarded despite high GDGT-2 / GDGT-3 values. Despite nearly half of the data points falling out of the 95 % confidence interval of the TEX86 Ring Index, we decided to not discard those because most of the sediments with abnormal Ring Index values are caused by the highly reduced contribution of cren.

C37 / C38 and C37 / C38Et ratios do not show any profound change; thus, we infer that alkenone composition is not affected by algae composition changes, and thus, U37k′ represents sea surface temperature. While TEX86-derived SST is in a consistently similar range or above U37k′-derived SST throughout most of the record, with the onset of the Pleistocene, SSTTEX86 drops significantly when compared to SSTs derived from alkenones. However, the difference is around 3 ∘C, which is still within the calibration errors. Nevertheless, such difference is mainly due to the use of BAYSPAR calibration. SST results using BAYSPAR calibration are barely different from TEX86H based on the temperature output when TEX86 > 0.5. However, when TEX86 is smaller than 0.5, namely in the Pleistocene, TEX86H-based SST is well in line with SSTU37k′ (Fig. S6).

Overall, considering the small difference (∼ 2 ∘C) between surface and subsurface calibration of TEX86 and SSTU37k′, the relatively large calibration error of proxies, and the similar extent of variability and the high GDGT-2 / GDGT-3 ratio throughout the study interval, we deem it that both proxies mainly reflect temperature of the surface layer, with TEX86 integrating a deeper component. Hence, we claim that our temperature record is a sea surface temperature (SST). However, we focus on the U37k′ record when we interpret the record during the MCO.

The new SST record for Site 1168 shows in broad lines a similar trend to that of the global compilation of the benthic foraminiferal oxygen isotope stack (δ18Obf); however, there are some interesting deviations (Fig. 6). The mid-to-late Miocene interval contains a remarkable ∼ 10 ∘C gradual SST cooling that is much less prominent in δ18Obf (Fig. 6). Other than the late Miocene cooling, we have found that subtropical SSTs fluctuated around 26 ∘C in the early Miocene, which is similar to those in the Oligocene (Fig. 4; Guitián and Stoll, 2021; Hoem et al., 2022). The SST record from > 23 Ma is elaborately discussed in Hoem et al. (2022), so we will focus on the SST trends in the interval < 23 Ma. SSTs were slightly elevated in the MCO and rapidly cooled by 5 ∘C across the MMCT. Pliocene and Pleistocene SSTs at Site 1168 have larger variability than those in the late Miocene. However, the variability remains in the Pliocene and Pleistocene, which deviates from the δ18Obf feature.

The northward movement of the site, from ∼ 52∘ S in the early Miocene to ∼ 42∘ S at present, may have dampened the amplitude of Neogene long-term cooling to an unknown extent. However, during the Oligocene, ocean conditions at this site also barely changed despite the northward drift, likely because the ocean currents migrated northwards along with the tectonic drift of Australia (Hoem et al., 2021). Similarly, for the same reason, the northward tectonic drift of Australia during the Neogene, in other words the latitude change, may have had little effect on the temperature evolution at this site, thus records the temperature resulting in both global climate and water mass change at Site 1168. The synchronous tectonic drift of other mid-latitude sites warrants the conclusion about the latitudinal temperature gradient drawn from the comparison (Fig. 6a). With the consistency of both paleotemperature proxy results in consideration, we will further discuss the SST evolution per time interval, focusing on variability within the record, comparison to the δ18Obf as a representation of deep-sea temperature and global ice volume trends, and comparison to other SST records in the region to reconstruct latitudinal SST gradients.