The Leeuwin Current and the Flinders Current are the main two current systems affecting the ocean region south of Australia (Fig. 1). The East Australian Current, a strong western boundary current (>2 m s-1) transporting tropical heat poleward along eastern Australia, only sporadically affects the southern coast (Bostock et al., 2006). The Leeuwin Current flows southwards along the western Australian shelf break and is characterized as a shallow (upper ∼200 m) coastal current, with low-salinity and nutrient-depleted waters that originate mainly from the Indo-Pacific Warm Pool. It receives further contributions of subtropical waters from the Indian Ocean via the broad equatorward-flowing West Australian Current, which is the eastern branch of the Indian Ocean gyre (Wandres, 2018).

After passing Cape Leeuwin and reaching its highest velocities, the Leeuwin Current turns east into the Great Australian Bight as far as ∼124∘ E (Ridgway and Condie, 2004). At the same time, it becomes saltier, cooler, and denser due to air–sea interactions, subtropical addition, and eddy mixing with Indian Ocean and Southern Ocean waters (see Richardson et al., 2019). Seasonal variations in the Leeuwin Current strength (Ridgway and Condie, 2004; Cirano and Middleton, 2004) reveal that the Leeuwin Current is strongest near the shelf edge in austral winter (June–July), with a maximum poleward geostrophic transport of ∼5 Sv (106 m3 s-1), and weakest in austral summer, with a mean transport of ∼2 Sv (Holloway and Nye, 1985; Rochford, 1986; Feng et al., 2003; Ridgway and Condie, 2004).

Cirano and Middleton (2004) estimated that the contribution of the Leeuwin Current on the total flow along the southern Australian coast diminishes toward the east. Off the eastern Great Australian Bight, the Leeuwin Current only drives ∼15 % of the total flow, while wind forcing (∼47 %) and a pressure gradient term (∼38 %) become more important. Ridgway and Condie (2004) noted that along the western Australian coast, the Leeuwin Current is forced by the alongshore pressure gradient associated with the meridional portion of either less dense and low-salinity water masses from the equatorial Western Pacific Warm Pool or southern-sourced cold, dense, and high-salinity waters, which exceed the equatorward alongshore winds. Along the southern Australian coast, the zonal shelf edge flow is instead forced by the austral winter westerly wind. Ridgway and Condie (2004) suggested that the western coast pressure gradient delivers the Leeuwin Current to the southern coast just in time for the (south)westerly winds to strengthen, thereby maintaining the eastward passage of the current.

The changing atmospheric circulation pattern is closely connected to the Subtropical Ridge, a belt of high-pressure systems (anticyclones) between ∼30 and ∼40∘ S (e.g., Drosdowsky, 2005), which divides the tropical southeasterly circulation (trade winds) from the mid-latitude westerlies. The Subtropical Ridge is shaped by the Indian Ocean Dipole, Southern Annual Mode (which is the zonal mean atmospheric pressure difference between the mid-latitudes (∼40∘ S) and Antarctica (∼65∘ S); Marshall, 2003), and to a lesser degree by ENSO (Cai et al., 2011). During austral autumn/winter (austral spring/summer), it moves north (south), allowing the westerlies to seasonally strengthen (weaken) rainfall in SE Australia (Cai et al., 2011). During El Niño conditions, the Subtropical Ridge is displaced farther equatorward than normal, while during La Niña conditions it is shifted poleward (Drosdowsky, 2003).

Near the eastern edge of the shallow Great Australian Bight shelf, a gravity outflow of warm and high-salinity waters related to intensified surface heating during austral summer spreads across the shelf and continues to flow eastward as the shelf edge South Australian Current (Ridgway and Condie, 2004; Fig. 1). Although relying on different forcing mechanisms, the South Australian Current is widely regarded as the extension of the Leeuwin Current. In the Bass Strait, the Leeuwin Current–South Australian Current system continues south as high-saline and relatively warm Zeehan Current (Ridgway and Condie, 2004; Richardson et al., 2018). South of Australia, the Leeuwin Current System meets the northern boundary of the eastward flowing Antarctic Circumpolar Current (ACC). Below, the deeper (300–400 m) equatorward flow of the Leeuwin Undercurrent is noted (Spooner et al., 2011). During austral summer, when the Leeuwin Current is weak, the equatorward Capes Current establishes at the inner shelf around Cape Leeuwin. Its formation is related to regional upwelling, bringing water masses from the Flinders Current and the lower layers of the Leeuwin Current towards the upper shelf areas (see McClatchie et al., 2006).

The westward-directed Flinders Current is a subsurface northern boundary current along the continental slope of southern Australia (Middleton and Cirano, 2002; Cirano and Middleton, 2004; Fig. 1). Maximum transport is at ∼400–800 m, with velocities of up to 8 cm s-1 (Middleton and Bye, 2007). It originates within the Subantarctic Zone and carries Subantarctic Mode Water (SAMW) and Antarctic Intermediate Water (AAIW) across the STF (McCartney and Donohue, 2007). Southeast of Australia, the Flinders Current is fed and strengthened by the Tasman Outflow, a remnant of the East Australian Current, which injects Pacific waters into the South Australian Basin (Rintoul and Sokolov, 2001) and becomes an important component of the westward flow south of Australia (Speich et al., 2002). The Flinders Current fluctuates in strength on a seasonal timescale (Richardson et al., 2019), with almost doubled transport (∼17 Sv) during austral summer compared to winter (∼8 Sv).

The Leeuwin Undercurrent, which is beneath the Leeuwin Current at depths of ∼250–600 m, transports ∼5 Sv of saline (>35.8 psu), oxygen-rich, and nutrient-depleted waters northward as an extension of the Flinders Current (Fig. 1; Thompson, 1984; Smith et al., 1991; Cirano and Middleton; 2004). Both currents are associated with SAMW (Pattiaratchi and Woo, 2009).

We here address three water masses within the uppermost 600 m along the continental slope of southern Australia (Figs. 1 and 2): Subtropical Surface Water (STSW), South Australian Basin Central Water (SABCW), and SAMW. The subtropical warm and saline STSW originates within the surface mixed layer (upper ∼200 m) along Australia's southern margin between 34 and 38∘ S as a result of surface heating and enhanced evaporation (James and Bone, 2011; Fig. 2). STSW constitutes the shallowest water mass along the southern Australian margin and is defined by temperatures >12 ∘C and salinities >35.1 (Richardson et al., 2018). The dissolved oxygen concentration is high (225–250 µmol L-1) and nutrients are low (Richardson et al., 2018). The water mass is additionally fed by low-salinity Tropical Surface Water (TSW) and high-salinity South Indian Central Water (SICW) contributed by the West Australian Current and the South Indian Ocean Current (Cresswell and Peterson, 1993). The maximum depth of the STSW is seasonally dependent: during austral autumn and winter, the Leeuwin Current-transported STSW is thicker (∼300 m in the western and ∼200–250 m in the eastern study area; Richardson et al., 2019) with a rather low vertical temperature gradient in the west (Fig. 2). When the eastward wind stress is strongest and opposing winds cease, it reaches further to the east and may reach the southern tip of Tasmania due to a strong Zeehan Current adjoining the Leeuwin Current (Cresswell, 2000; Feng et al., 2003; Ridgway and Condie, 2004; Ridgway, 2007), which causes warming at depth. During austral summer (November to March), the STSW remains west of ∼140∘ E (Newell, 1961; Vaux and Olsen, 1961; Ridgway, 2007; Richardson et al., 2018). It then is at shallower depths (∼200–250 m in the west and ∼150–50 m in the east; Richardson et al., 2019; Fig. 2), with a well-defined shallow thermocline during times of a weak Leeuwin Current, when opposing winds (blowing from the southwest) are strong (Godfrey and Ridgeway, 1985; Smith et al., 1991; Feng et al., 2003, 2009).

The SABCW, showing a small range in potential density (26.65–26.8 kg m-3), is below the surface mixed layer (Fig. 2b). SABCW is defined by temperatures and salinities of 10–12 ∘C and 34.8–35.1, with a weak dissolved oxygen maximum (>250 µmol L-1; Richardson et al., 2018). Towards the east, the thickness of the SABCW is ∼200 m, while it decreases to ∼100 m in the west (Richardson et al., 2018). The thinning of SABCW towards the west is likely attributed to the presence of near-surface subtropical water in the west (STSW), contributed by the strong eastward-flowing Leeuwin Current. SABCW likely forms south of the STF between 44–46∘ S and 140–145∘ E in winter by convective overturning and subduction of the deep mixed layer (Richardson et al., 2018). The subducted SABCW reaches slope depths of ∼300–500 m at 142∘ E and ∼300–400 m at 130 to 121∘ E. It is transported eastwards towards Tasmania along the STF by zonal flow. The Flinders Current inflow from the southeastern margin then carries SABCW north and west, augmented by the Tasman Outflow and equatorward Sverdrup transport (Schodlok and Tomczak, 1997). Along the southern Australian margin, the boundary between the top surface of SABCW (as part of the Central Water) and the overlying STSW defines the interface between the eastward-directed Leeuwin Current System transporting subtropical waters and the westward flow of the Flinders Current System, which brings subantarctic waters into the region (SABCW coupled to Tasmanian Subantarctic Mode Water (TSAMW) and Tasmanian Intermediate Water (TIW); Fig. 2b; Richardson et al., 2019).

The coldest and densest SAMW of the Indian Ocean forms by air–sea interaction and deep winter mixing south of Australia between 40 and 50∘ S (e.g., Wyrtki, 1973; McCartney, 1977; Karstensen and Quadfasel, 2002; Barker, 2004). SAMW is subducted, thereby ventilating the lower thermocline of the Southern Hemisphere subtropical gyres (McCartney et al., 1977; Sprintall and Tomzcak, 1993). The high-nutrient SAMW is defined as a layer of relatively constant density (pycnostad) along the southern Australian continental slope (Richardson et al., 2019; Fig. 2). The pycnostad is clearly defined in the east, notably in summer, but diminishes towards the west (Richardson et al., 2018). The SAMW in this region is located at ∼400–650 m, with temperatures of ∼8–10 ∘C and salinities of 34.6–34.8 (Woo and Pattiaratchi, 2008; Pattiaratchi and Woo, 2009), being therefore fresher than the overlying SABCW and STSW. The top SAMW depth varies seasonally from west to east, as it shallows to ∼350 m during summer and deepens to ∼500 m in winter (Rintoul and Bullister, 1999; Rintoul and England, 2002). In particular, the Tasmanian SAMW (TSAMW) is formed in a clearly defined area at 45–50∘ S and 140–145∘ E (Barker, 2004).

The chronostratigraphy and the paleo-reconstructions were established from isotope-geochemical parameters measured within the calcitic tests of the subtropical shallow-dwelling planktonic foraminiferal species Orbulina universa (Bé and Tolderlund, 1971) and Globigerinoides ruber and the deep-dwelling species Globorotalia truncatulinoides (Lohmann and Schweitzer, 1990). As O. universa preferentially lives in the surface mixed layer and the shallow thermocline, we assigned a calcification depth of ∼30–80 m (see Text S1 in the Supplement). The surface-dwelling G. ruber is the most representative species of warm and annual surface (<50 m) ocean conditions (Anand et al., 2003; Tedesco and Thunell, 2003). For G. truncatulinoides we assume a calcification depth of ∼350–400 m (see Text S1), which corresponds to the base of the summer thermocline (Fig. 2; Locarnini et al., 2019). Most of the G. truncatulinoides specimens in our samples were encrusted (see Text S1).

On average, 10–12 and 30–40 visually clean specimens of O. universa/G. ruber and G. truncatulinoides, respectively, were hand-picked under a binocular microscope from the narrow >315–400 µm size fraction in order to avoid size-related effects on either Mg/Ca or stable isotopes. G. truncatulinoides has no size effect on Mg/Ca (Friedrich et al., 2012), and δ13C and δ18O also show no systematic changes in the selected size fraction (Elderfield et al., 2002). The foraminiferal tests were gently crushed between cleaned glass plates to open the test chambers for efficient cleaning. Over-crushing was avoided to prevent an excessive sample loss during the cleaning procedure. The fragments of the tests were homogenized and split into subsamples for stable isotope (one-third) and trace metal analyses (two-thirds) and transferred into cleaned vials. Chamber fillings (e.g., pyrite, clay) and other contaminant phases (e.g., conglomerates of metal oxides) were thoroughly removed before chemical cleaning and analyses.

Prior to elemental analysis, the foraminiferal samples were cleaned following the protocols of Boyle and Keigwin (1985) and Boyle and Rosenthal (1996). These include oxidative and reductive (with hydrazine) cleaning steps. Elemental analyses were accomplished with a VARIAN 720–ES Axial ICP-OES, a simultaneous, axial-viewing inductively coupled plasma optical emission spectrometer coupled to a VARIAN SP3 sample preparation system at GEOMAR. The analytical quality control included regular analysis of standards and blanks, with results normalized to the ECRM 752–1 standard (3.761 mmol mol-1 Mg/Ca; Greaves et al., 2008) and drift correction. The external reproducibility for the ECRM standard was ±0.01 mmol mol-1 for Mg/Ca (2σ standard deviation). Replicate measurements reveal a reproducibility of ±0.28 mmol mol-1 for G. truncatulinoides (2σ standard deviation).

G. truncatulinoides from core 2614 were only oxidatively cleaned and analyzed on a simultaneous, radially viewing ICP-OES (Ciros CCD SOP, Spectro A.I., Univ. Kiel). A cooled cyclonic spray chamber, in combination with a microconcentric nebulizer (200 µL min-1 sample uptake), was optimized for best analytical precision and minimized uptake of sample solution. Sample introduction was performed via an autosampler (Spectra A.I.). Matrix effects caused by varying concentrations of Ca were cautiously checked and found to be insignificant. Drift of the machine during analytical sessions was negligible (∼0.5 %, as determined by analysis of an internal consistency standard after every five samples; cf. Nürnberg et al., 2008). To account for the different cleaning techniques prior to Mg/Ca analyses, the initial foraminiferal Mg/Ca data of G. truncatulinoides from core 2614 were corrected by 10 % according to Barker et al. (2003). See further details and information on contamination and dissolution issues in the Supplement. In addition, the impact of pH on foraminiferal Mg/Ca is discussed here in detail. In the following, species-specific Mg/Ca ratios are termed Mg/Caruber, Mg/Cauniversa, and Mg/Catruncatulinoides.

Mg/Cauniversa values were converted into sea surface temperatures (SSTMg/Ca) using the species-specific paleotemperature calibration of Hathorne et al. (2003): Mg/Ca=0.95⋅exp(0.086⋅T). This calibration function is based on a North Atlantic core-top calibration study and provides reliable SSTMg/Ca estimates (Figs. S8 and S9 in the Supplement) with an error (standard deviation 2σ) of ±0.2 units of ln(Mg/Ca), which is equivalent to ±1.1 ∘C. The calibration provides a mean Holocene (<10 ka BP) SSTMg/Ca estimate of ∼20.5 ∘C in the eastern core 2609, which exceeds the modern annual SST conditions by ∼3–5 ∘C (cf. Fig. 4c). In the western core 2614, the SSTMg/Ca estimate of ∼19.6 ∘C (Fig. 4c) is in broad agreement with the modern austral summer SST range at 30–80 m water depth in the upper thermocline or mixed layer (see further discussion below; cf. Figs. S8 and S9). In the case of G. ruber, we refrained from converting the Mg/Caruber ratios into temperatures due to reasons discussed in the Supplement.

The Mg/Catruncatulinoides values were converted into thermocline temperatures (TTMg/Ca) using the deep-dweller calibration equation of Regenberg et al. (2009): Mg/Ca=0.84⋅exp(0.083⋅T). This calibration provides core-top TTMg/Ca estimates (on average ∼10–12 ∘C; Fig. 5), which agree with the modern annual thermocline temperatures (∼9–12 ∘C) at the preferred depth of G. truncatulinoides (∼9–12 ∘C; Fig. 2). The error (standard deviation 2σ) of the calibration is ±1.0 ∘C. The TTMg/Ca estimates from other existing paleotemperature calibrations specific to G. truncatulinoides are discussed in the Supplement (Figs. S8 and S9).

For the vertical gradient calculation, we used evenly sampled (200 years apart) and linearly interpolated datasets using the software AnalySeries (Paillard et al., 1996)partly because foraminiferal specimens were too rare, not allowing for combined isotope and trace element analyses throughout the entire records, and partly because data were missing in one record or another. In particular for core location 2614, negative vertical ΔTSST-TT values were interpreted in a way that thermocline temperature came close or even became similar to sea surface temperatures (cf. Fig. 6a). Even though the calibrations were carefully chosen, there remains considerable uncertainty in the absolute temperature values over time. First, calibrations should ideally be region specific to allow for the best reconstructions. None of the calibrations applied, however, were developed for the region south of Australia. Second, the range in downcore temperature amplitudes highly depends on the applied calibration. The less exponential the calibration, the larger the downcore amplitude variations. These imponderabilities cannot be solved in this context.

Measurements of stable oxygen (δ18O) and carbon isotopes (δ13C) on foraminiferal test fragments were performed at GEOMAR on a Thermo Scientific MAT 253 mass spectrometer with an automated Kiel IV carbonate preparation device. The isotope values were calibrated versus the NBS 19 (National Bureau of Standards) carbonate standard and the in-house carbonate standard “Standard Bremen” (Solnhofen limestone). Isotope values in δ notation are reported (in ‰) relative to the VPDB (Vienna Peedee Belemnite) scale. The long-term analytical precision is ±0.06 ‰ for δ18O and ±0.05 ‰ for δ13C (1–σ value). Replicate measurements were not done due to the low numbers of specimens found. A previous study on the same device revealed a δ18OVPDB reproducibility of ±0.14 ‰ from 148 replicate measurements of G. truncatulinoides (Nürnberg et al., 2021). In the following, species-specific δ18O values are termed δ18Oruber, δ18Ouniversa, and δ18Otrunca.

Commonly, modern δ18Osw and salinity are linearly correlated in the upper ocean. Unfortunately, the sparse database of modern δ18Osw south of Australia does not allow for accurate description of the relationship (see Schmidt et al., 1999). Past local salinity variations at the sea surface and thermocline depths were approximated from δ18Osw derived from combined δ18O and SSTMg/Ca respective TTMg/Ca measured on the surface and thermocline-dwelling foraminiferal species (e.g., Nürnberg et al., 2008; 2015; 2021). First, the temperature effect was removed from the initial foraminiferal δ18O by using the temperature versus δ18Ocalcite equation of Bemis et al. (1998) for planktonic foraminifera: δ18Osw=0.27+((T-16.5+4.8⋅δ18Oforam)/4.8). By applying the correction of 0.27 ‰ (Hut, 1987), we converted from calcite on the VPDB scale to water on the Vienna Standard Mean Ocean Water (VSMOW) scale. Second, we calculated the regional ice-volume-corrected δ18Osw record (δ18Osw-ivc) by accounting for changes in global δ18Osw that were due to continental ice volume variability. Here, we applied the Grant et al. (2012) relative sea level reconstruction to approximate variations in the global ice volume because it provides a high temporal resolution during MIS 3 and times of rapid Dansgaard–Oeschger climate variability (Fig. 4a).

The propagated 2σ error in δ18Osw-ivc is ±1.16 ‰ for G. truncatulinoides (see Reißig et al., 2019), and hence it is larger than for shallow dwellers (±0.4 ‰ for G. ruber; e.g., Bahr et al., 2013; Schmidt and Lynch-Stieglitz, 2011). The overall Holocene (<10.5 ka BP) δ18Osw-ivc amplitude of ∼1 ‰ calculated for O. universa and G. truncatulinoides corresponds to the modern surface δ18Osw variability of ∼-0.5 to 0.5 ‰ for close-to-coast regions south of Australia (Schmidt et al., 1999). The calculated late Holocene (<5 ka BP) surface δ18Osw-ivc (O. universa) values of 1.2–2 ‰, however, are heavier than the δ18Osw values reported by Richardson et al. (2019) for surface waters (STSW >0.05 ‰). In addition, the calculated late Holocene (<5 ka BP) subsurface δ18Osw-ivc (G. truncatulinoides) values of 0.2 ‰–0.3 ‰ appear heavier than the reported δ18Osw value for TSAMW (-0.1 ‰ to -0.25 ‰; Richardson et al., 2019). In spite of the potential errors in our δ18Osw-ivc calculations, which are related to (i) the large ecological and hydrographical variability and (ii) the comparatively large uncertainty of the Mg/Ca temperature calibrations applied, we note that the relative difference between the isotopically heavy STSW and the light TSAMW is well reflected in the calculated sea surface and thermocline δ18Osw-ivc values. The δ18Osw-ivc values were not converted into salinity units, as it is not apparent that the modern linear relationship between δ18Osw and salinity held through time due to changes in the ocean circulation and freshwater budget (e.g., Caley and Roche, 2015). We therefore interpret the downcore δ18Osw-ivc records as relative variations in salinity.

All raw analytical data of cores 2416 and 2409 versus core depth are presented in the Supplement (Figs. S6 and S7). Over the last 60 kyr, the SSTMg/Ca development in the western and eastern study areas differ substantially. In the western area south of Cape Pasley (core 2614), the MIS 3 (∼57–29 ka BP; Lisiecki and Raymo, 2005) is characterized by long-term sea surface warming by ∼4 ∘C on average from ∼17 to 21 ∘C until ∼37 ka BP (Fig. 4c). This warming trend is underlain by large-amplitude variations in SSTMg/Ca of up to 4–5 ∘C, ranging between ∼15 and 22 ∘C. The sea surface warming pulses are commonly accompanied by changes to more saline conditions (high δ18Osw-ivc-values; Fig. 4d). Most of the short-term changes to warm and saline sea surface conditions appear at the Antarctic Warming Events 3 and Antarctic Isotope Maxima (AIM) 12 and 8 and during Northern Hemisphere cool periods. These glacial MIS 3 warming pulses compare to and even exceed the modern SST conditions. After ∼37 ka, the SSTMg/Ca decline continuously, accompanied by short-term and high-amplitude warming events rather similar to those events observed during the early MIS3.

The subsequent MIS 2 (∼29–14 ka BP; Lisiecki and Raymo, 2005) shows rather low SSTMg/Ca of ∼14–17 ∘C and fresh conditions specifically at the beginning of MIS 2. While O. universa specimens are missing during the remaining MIS2, the highly variable Mg/CaG. ruber data during MIS 2 imply similarly variable SSTMg/Ca conditions to those during MIS 3 (see Fig. S8).

During the last deglaciation (∼18–12 ka BP), the SSTMg/Ca gradually increase from ∼15 to 20 ∘C, with intermittent prominent high-amplitude SSTMg/Ca variations and maxima of up to ∼22 ∘C. Similarly, salinity conditions vary considerably (δ18Osw-ivc=1.5±0.5 ‰), with δ18Osw-ivc values mostly exceeding the modern values (>0.05 ‰; Richardson et al., 2019) and pointing to rather saline conditions during times of sea surface warming. The high-amplitude SSTMg/Ca variations of ∼4 ∘C during the Holocene (<10 ka BP) are close to the modern austral summer SST conditions, but during the late Holocene in particular they exhibit a slight cooling and freshening trend.

In the eastern study area (core 2609) northwest of King Island, the SSTMg/Ca ranges between ∼16 and ∼20 ∘C during MIS 3 (Fig. 4c). This is at the upper limit of the modern SST range in this area, which is overall cooler than the western study area. Only temporally does SSTMg/Ca come close to the core 2614 SST conditions. SSTMg/Ca amplitudes are approximately half the amplitude observed in the western core 2614. The δ18Osw-ivc variations are rather comparable to those of core 2614, pointing to commonly more saline sea surface conditions than today (Fig. 4d). Notably, the prominent AIM-related sea surface warming pulses observed in the western core 2614 and the synchronous changes to saline conditions are not seen in core 2609.

During the Last Glacial Maximum (LGM; between ∼24 and 18 ka BP), the SSTMg/Ca decline to on average ∼11–16 ∘C, clearly cooler by ∼2 ∘C than modern austral winter conditions, and temporally reach values of even <12 ∘C. The δ18Osw-ivc values of 0.5 ‰–1.5 ‰ gradually approach the modern values, pointing to fresher conditions when sea surface is cooling. During the deglaciation, the core 2609 SSTMg/Ca increases gradually by >5 ∘C, with increasingly saline sea surface conditions. Conditions became relatively similar in both the eastern and western study areas despite remaining more variable in the west (Fig. 4c).

The Holocene SSTMg/Ca in core 2609 increases to ∼19–22 ∘C, seemingly warmer and more saline (δ18Osw-ivc=1.6‰–2.4 ‰) than modern austral summer conditions and those conditions at the western site 2614. This disparity will be discussed further below. We note, however, that the youngest samples in both cores provide rather similar SSTMg/Ca and salinity conditions when relying on the G. ruber proxy data (cf. Fig. S9: SSTMg/Ca in both cores is 16–18 ∘C, which reflects modern conditions at depths <50 m fairly well). We also note that the youngest O. universa-derived SSTMg/Ca estimate from core 2609 matches the SST long-chain diol index (SSTLDI) estimate of ∼22 ∘C from nearby core MD03-2607 (Lopes dos Santos et al., 2013; Fig. 4c). The SSTLDI estimates are based on long-chain diols, and LDI-inferred temperatures supposedly reflect SSTs of the warmest month (Lopes dos Santos et al., 2013).

We hence hypothesize that the O. universa SSTMg/Ca signal is seasonally biased towards the austral summer season. We note also that the entire core 2609 SSTMg/Ca record matches the SSTLDI record from nearby core MD03-2607 reasonably well, with similar absolute temperature estimates (∼11–24 ∘C) and particularly similar deglacial amplitudes of up to 7 ∘C (Fig. 4c). Both the SSTLDI and SSTMg/Ca estimates are warmer by ∼4 ∘C than the alkenone-based SSTUK′37 estimate from cores MD03-2607 (Lopes dos Santos et al., 2012) and MD03-2611 (Calvo et al., 2007; 36∘44′ S, 136∘33′ E; Fig. 1), likely due to the fact that SSTUK′37 reflects the cooler early spring conditions.

All raw analytical data of cores 2614 and 2609 versus core depth are presented in the Supplement (Figs. S6 and S7). Over the last 60 kyr, the development at thermocline depth in the western study area south of Cape Pasley (core 2614) differs substantially from the eastern area, with prominent and rapid high-amplitude changes in TTMg/Ca and the according δ18Osw-ivc in the western area. The proxy records from the eastern core 2609 instead appear rather muted, cooler, and fresher (Fig. 5c and d).

During MIS 3, the TTMg/Ca in western core 2614 ranges between 10 and 21 ∘C, revealing a long-term cooling trend from ∼20 ∘C at 60 ka BP on average to ∼11 ∘C at ∼23 ka BP (Fig. 5c). This cooling trend is accompanied by high-amplitude TTMg/Ca variations even exceeding 5 ∘C. The TTMg/Ca and thermocline depth δ18Osw-ivc minima correspond to the modern TT (9–11 ∘C; cf. Fig. 2) and δ18Osw ranges at core location 2614 (Richardson et al., 2019), while distinct warming pulses at thermocline depth and saline conditions exceed modern conditions by up to ∼10 ∘C and ∼2‰, respectively (Fig. 5c and d). Although some of these TTMg/Ca warming pulses are only represented by single Mg/Ca data points (due to rare foraminiferal sample material), we assess them as robust as the peaks are mostly supported by several δ18OG.truncatulinoides and δ13CG.truncatulinoides excursions to light values (Fig. 5b).

In the eastern core 2609, the MIS3 TTMg/Ca ranges between ∼7 and 11 ∘C, which is cooler by a maximum of 2 ∘C than the modern TT range of 9–11 ∘C (cf. Fig. 2). The thermocline depth δ18Osw-ivc values (-0.5 ‰ to ∼0.5 ‰) are mostly equal to or more positive than the modern value (Richardson et al., 2019; cf. Fig. 5d) but remain clearly fresher by up to 2 ‰ and less variable than at the western core (0 ‰–2 ‰). During MIS 2 and during the LGM in particular, the conditions at thermocline depth at core 2609 are cooler than modern temperatures by ∼2 ∘C, while remaining fresher and lower in amplitude compared to the clearly more variable and warmer thermocline conditions at core 2614 (Fig. 5c and d). The western location rather exhibits short-term TTMg/Ca variations between ∼8 and ∼13 ∘C, which is close to the modern TT in the region. Relative salinity varied correspondingly (0.5 ‰–1.5 ‰).

In the western study area, the deglaciation is characterized by rapid and prominent changes in thermocline conditions (Fig. 5c). Increases in TTMg/Ca by up to 10 to a maximum of 20 ∘C and in δ18Osw-ivc by up to 2.5 ‰ in amplitude occur during the early Heinrich Stadial 1, the early Bølling/Allerød, and the Preboreal. In contrast, the deglacial change in the eastern study area lags behind the western development and is less prominent, with TTMg/Ca rising from 7 to 12 ∘C in line with the Southern Hemisphere deglacial climate change as reflected in the EDML δ18O record (EPICA Community Members, 2006; Fig. 5f).

The Holocene is characterized in both regions by subtle variations in TTMg/Ca and corresponding δ18Osw-ivc. The western core shows higher TTMg/Ca (∼12–14 ∘C and warmer-than-modern conditions) than the eastern core (∼10–12 ∘C, rather similar to modern conditions at thermocline depth), while the salinity (0 ‰ to 0.5 ‰) in both areas appears rather similar and close to the modern values (which is 34.8–35.1 in the western core and 34.7–34.9 in the eastern core; Fig. 5c and d).

We interpret the SST and surface δ18Osw-ivc data derived from O. universa in terms of changes in the surface mixed layer, which is dominated by STSW (contributions of Leeuwin Current-transported TSW and South Indian Ocean Current-transported SICW) at the western core location and by the South Australian Current (SAC) in the eastern study area (Fig. 1). The thermocline-dwelling G. truncatulinoides proxy data instead reveal changes in the underlying Central Water, which comprises SABCW and Tasman Subantarctic Mode Water (TSAMW). The boundary between STSW and Central Water defines the interface between the eastward-directed Leeuwin Current System and the westward flow of the Flinders Current System (see Fig. 2b; see Sect. 2.2).

To assess the dynamics of the Leeuwin Current-transported STSW and its interaction with both the surface SAC and the underlying SABCW/TSAMW south of Australia through time, we calculated the vertical temperature gradients at both core locations (see Sect. 3.3). The vertical temperature gradient (ΔTSST-TT) provides insight into the thermocline depth, with small (large) ΔTSST-TT pointing to a shallow (steep) thermal gradient and a deep (shallow) thermocline with accompanying strong (weak) vertical mixing. In conjunction with the lateral gradients at both sea surface (ΔSSTwest-east) and thermocline depths (ΔTTwest-east; Fig. 6a and b), which define the regional differences at the two depth levels, we derive insight into how the Leeuwin Current System developed spatially in relation to the Flinders Current System during different climate regimes. The similarity (R=0.87) between the ΔTTwest-east record (Fig. 6b) and the TTMg/Ca record of the western core 2614 (Fig. 5c) pinpoints that it is the thermocline changes in the western area that are crucial to the oceanographic setting south of Australia and best reflect the relative presence of the different water masses.

Palynological studies on our western core 2614 record a substantial decline of the dung fungus Sporormiella, a proxy for herbivore biomass, which was taken as evidence for the prominent Australian megafaunal population collapse from ∼45 to 43.1 ka BP (van der Kaars et al., 2017; Fig. 8a). Climate change likely played a significant role in most of the disappearance events of the continent's megafauna during the Pleistocene, while human involvement appears likely for the last megafaunal population collapse after ∼45 ka BP in particular, but this is still debated (Wroe et al., 2013).

A new chronology constrains the early dispersal of modern humans out of Africa across southern Asia into “Sahul” (northern Australia and New Guinea connected by a land bridge at times of glacially lowered sea level; see Saltré et al., 2016) to ∼65–50 ka BP (Clarkson et al., 2017; Tobler et al., 2017). The further settlement comprised a single, rapid (within a few thousand years; Tobler et al., 2017) migration along the eastern and western coasts, with Aboriginal Australians reaching the south of Australia by ∼49–45 ka BP. It is also clear that humans were present in Tasmania by ∼39 ka BP (Allen and O'Connell, 2014) and in the arid center of Australia by ∼35 ka BP (Smith, 2013). This places the initial human colonization of Australia clearly before the continent-wide extinction of the megafauna (cf. Saltré et al., 2016). Rule et al. (2012) and van der Kaars et al. (2017) claimed that human arrival causing overhunting, vegetation change due to landscape burning, or a combination thereof was the primary extinction cause rather than climate change. Brook and Johnson (2006) showed with model simulations that species with low population growth rates, such as large-bodied mammals in Australia, might have been easily exterminated by even small groups of hunter-gatherers using stone-based tools. Saltré et al. (2016) also hypothesized that climate change was not responsible for late Quaternary (last 120 kyr) megafauna extinctions in Australia, as they appeared to be independent of climate aridity and variability.

Our record of detrended ΔTTwest-east, which approximates the strength of the Leeuwin Current, provides additional views on these issues. It shows a robust covariance on millennial to centennial timescales from ∼60 to 20 ka BP to a charcoal composite record reflecting biomass burning in the Australasian region (Mooney et al., 2010; Fig. 8b). Commonly, fewer fires appeared during periods associated with an intensified Leeuwin Current, a southward-located STF and Subtropical Ridge, and wetter conditions in the Australasian region at times of a weakened ACC (cf. Fig. 8b and c). The consistent timing of changes in both ocean dynamics and biomass burning over such a long period even prior to the arrival of humans in Australia (see Singh et al., 1981) suggests a strong coupling between climate-modulated changes related to the Leeuwin Current and changes in terrestrial vegetation productivity and distribution. This might have been an important factor for controlling Australasian fire regimes (Mooney et al., 2010). We hence argue that it is the joint interplay between natural ocean and climate variability, vegetational response, and human interference that caused the Australian megafaunal extinction.

Figure 8a shows the Sporormiella record of western core 2614 (van der Kaars et al., 2017) in comparison to our detrended record of Leeuwin Current variability. It is evident that before ∼45–43.1 ka BP the Sporormiella abundances were highly variable, placing Sporormiella abundance maxima (>10 %–13 %) into times of extensive thermocline expansion and the strong southward transfer of tropical heat via the Leeuwin Current (warming phases 7, 6, and 5; Fig. 8a and b). This is when Antarctica cooled (WAIS Divide Project Members, 2013) and the ACC weakened, likely in response to sea ice expansion (Wu et al., 2021; Fig. 8c and d). Sporormiella minima (<∼8 %) instead consistently occurred during times of a shallow thermocline and a weakened Leeuwin Current, with percentages becoming lower stepwise during Antarctic warm periods A3 (∼7 %) and AIM12 (∼6 %) until they reach their lowest values (∼2 %) during AIM11 at ∼45–43.1 ka BP (Fig. 8a).

The successive decline of Sporormiella during Antarctic warm periods and its rapid recuperation in between during times of Antarctic cooling, sea ice expansion, and ACC slowdown is mirrored in the decline of the Australian megafaunal bird Genyornis newtoni. From widespread eggshell fragments of Genyornis exhibiting diagnostic burn patterns, Miller et al. (2016) concluded that humans depredating and cooking eggs significantly reduced the reproductive success of Genyornis. They dated the egg predation and the related extinction of Genyornis to ∼47 ka BP, although they admitted that an age range from ∼54 to 43 ka BP could not confidently be excluded (Fig. 8a). This places the given extinction dates of Genyornis into the periods of prominent declines in Sporormiella abundances (A3, AIM12, AIM11) and hence into periods of a weak Leeuwin Current system, while in the warming Southern Ocean (WAIS Divide Project Members, 2013) sea ice extension shrank and the ACC strengthened (Wu et al., 2021; Fig. 8c).

The tight coupling between oceanographic changes and changes in the Australian megafauna brings ocean dynamics as an important player into the game. We hypothesize that the apparent rapid variations in the ocean–climate system from ∼60 to ∼43 ka BP with an overall tendency towards a weakening of the Leeuwin Current and the equatorward migration of the Southern Hemisphere frontal system (Fig. 8) must have caused considerable climatic and ecosystem response in Australia, with negative aftereffects on the continent's megafauna. A recuperation of the megafauna, however, is documented (and expected) by the increasing Sporormiella abundances during each of the short time periods 7, 6, and 5 of an intensified southward transfer of tropical heat via the Leeuwin Current and poleward shift of the Subtropical Ridge (Fig. 8b), even though human impact should have persisted or even raised during this period.

The final extinction phase defined to ∼45–43.1 ka BP on the basis of the core 2614 Sporormiella record (van der Kaars et al., 2017) and supported by other studies (e.g., Miller et al., 2005, 2016; Rule et al., 2012) appeared synchronous to the significant decline in the core 2614 thermocline temperature, salinity, and depth; the reduction of ΔTTwest-east by more than 10 ∘C; and the clearly warmer and more saline sea surface conditions in the western study area, while the eastern sea surface remained cool and fresh (Figs. 5 and 6). This all points to the drastic weakening and shoaling of the Leeuwin Current (analogous to the modern austral summer conditions) with the STF being pushed to the north and a larger impact of the glacial Southern Ocean via an enhanced Flinders Current. The significant reorganization of the ocean circulation south of Australia at ∼45–43.1 ka BP is accompanied by a transient change in climate and vegetation in Australia. Bowler et al. (2012) described a drying trend in SE Australia (Willandra Lakes) since ∼45 ka, synchronous with the weakening of the Australian monsoon (Johnson et al., 1999) and also visible in the Mooney et al. (2010) charcoal record (Fig. 8b). The dietary δ13C change in the Australian emu Dromaius novaehollandiae population at that time (Fig. 8a) also points to the reorganization of vegetation communities across the Australian semiarid zone (Miller et al., 2005). The abrupt decline in C4 plants between ∼44 and 42 ka BP observed in core MD03-2607, however, was interpreted by Lopes dos Santos et al. (2013) not in terms of climate change but in terms of a large ecological change, primarily caused by the absence of the megafaunal browsers due to extinction. The extinction left increased C3 vegetation biomass in the landscape, which would have fostered fires, eventually aided by human activities (Lopes dos Santos et al., 2013).

We hypothesize, alternatively, that the centennial-scale severe change in the ocean–climate system beginning at ∼45 ka BP must have had aftereffects on the continental environment. We argue that the megafauna, which might have been significantly decimated by human activity at that point, likely did not keep track with the rapidly increasing ecological stress and was no longer able to adapt to the changing conditions related to the weakening of the Leeuwin Current. Humans might indeed have effectively contributed to the extinction of the Australian megafauna as previously suggested (e.g., Rule et al., 2012; Miller et al., 2016; van der Kaars et al., 2017), but the ocean–climate dynamics provide an important prerequisite and amplifying factor until a tipping point was reached, after which faunal recuperation no longer happened.