Organic carbon burial in Mediterranean sapropels intensified during Green Sahara Periods since 3.2 Myr ago

Dark organic-rich layers (sapropels) have accumulated in Mediterranean sediments since the Miocene due to deep-sea dysoxia and enhanced carbon burial at times of intensified North African run-off during Green Sahara Periods (GSPs). The existence of orbital precession-dominated Saharan aridity/humidity cycles is well known, but lack of long-term, high-resolution records hinders understanding of their relationship with environmental evolution. Here we present continuous, high-resolution geochemical and environmental magnetic records for the Eastern Mediterranean spanning the past 5.2 million years, which reveal that organic burial intensified 3.2 Myr ago. We deduce that fluvial terrigenous sediment inputs during GSPs doubled abruptly at this time, whereas monsoon run-off intensity remained relatively constant. We hypothesize that increased sediment mobilization resulted from an abrupt non-linear North African landscape response associated with a major increase in arid:humid contrasts between GSPs and intervening dry periods. The timing strongly suggests a link to the onset of intensified northern hemisphere glaciation. Mediterranean sediment records spanning the last five million years suggest an abrupt increase in carbon burial and terrigenous accumulation coincided with enhanced fluvial sediment mobilisation in North Africa during orbitally-driven Green Sahara Periods.

'G reen Sahara Periods' (GSPs) have been a fundamental characteristic of North African climate change for more than 8 Ma ref. 1 Only the most recent GSP during the early-mid Holocene (~11 to 6 thousand years ago, ka; the 'African Humid Period') has been studied in detail at numerous locations [2][3][4] , and it extended to East Africa [5][6][7] . However, there are few continuous, well-dated GSP records that extend beyond 1.5 Ma (or even beyond the last 300 ka), and no high-resolution North African humidity reconstructions encompass the entire Plio-Pleistocene. Mediterranean sediments contain a particularly rich archive of past GSPs. Here, organic-rich layers ('sapropels') form in response to deep-sea anoxia and nutrient inputs when significantly increased run-off enters the basin via the Nile and wider North African margin 8 . These periods correspond to enhanced boreal summer insolation maxima and minima in Earth's orbital precession cycle, resulting in a more northerly and intensified African rain belt 8,9 . Sapropels are a natural testbed for understanding deep-sea redox and carbon burial processes; however, their formation mechanisms are still debated 8 , and there is currently no continuous high-resolution proxy record of Plio-Pleistocene sapropels.
Ba/Al, Ti/Al and planktic δ 18 O in Eastern Mediterranean sediments are useful proxies of past GSPs and sapropels. Ba/Al reliably tracks sapropel intervals because it correlates with original organic carbon burial (C org ) in sapropels and is mostly unaffected by post-depositional redox reactions 10 , while surface water freshening due to monsoon run-off is recorded by strong negative δ 18 O peaks 11,12 . Ti/Al in the open Eastern Mediterranean reflects relative variations in North African aeolian vs riverine inputs to the basin 13 . Aeolian-sourced Ti/Al in the open Eastern Mediterranean is enhanced relative to fluvially-sourced Ti/Al because heavier (Ti-bearing) suspended particles preferentially settle near the Nile fan. Al normalisation (for Ti and Ba) also removes closed-sum effects.
Here, we address the dual need for high-resolution records of North African humidity/aridity changes and sapropel deposition through the entire Plio-Pleistocene, by presenting the first continuous, astronomically dated GSP and sapropel proxy records back to 5.2 Ma from Eastern Mediterranean Ocean Drilling Programme (ODP) Site 967 (Figs. 1-2 and Supplementary  Fig. S1). From the same site, we also present high-resolution records of Saharan dust and riverine inputs over the past 5.2 Ma. Our data reveal a fundamental change in sapropel development, provide much-needed climatic context for understanding hominin evolution and migrations out of Africa, and are essential for modelling the mid-Pliocene-the most recent interval with CO 2 levels approximating modern levels 14 .
Results and discussion Green Sahara Periods. Ti/Al, Ba/Al, and planktic δ 18 O record GSP timings over the Plio-Pleistocene (Fig. 3). The Ba/Al signal is consistent with a typical sapropel sequence 10,15 , and lower Ti/Al values correspond to Ba/Al spikes, δ 18 O minima and precession minima. The δ 18 O record reflects long-term global sea-level/ice volume changes 16 , with superimposed negative δ 18 O excursions that relate to African run-off into the Eastern Mediterranean and warming within fresher surface-water layers 12,17,18 . African run-off reaching ODP967 would have derived primarily from the Nile 18,19 , which is fed by East African precipitation influenced by both the Indian and African Monsoons and their moisture convergence at the Congo Air Boundary 20 . African run-off also entered the Eastern Mediterranean from the wider North African margin when the African rain belt migrated northward 18,21,22 . This rain belt is associated with the intertropical convergence zone (ITCZ), the West African monsoon, and East African precipitation 20,23 . Thus, ODP967 δ 18 O reflects both West and East African precipitation. While changes in local Eastern Mediterranean precipitation over the entire Plio-Pleistocene are not well-constrained, their impacts on ODP967 δ 18 O are likely to be minimal compared to African run-off because Eastern Mediterranean surface waters are the evaporative source for local precipitation (with little excess δ 18 O fractionation) 8,12 . The similar range of δ 18 O minima at precession minima throughout the record (−1 to −2‰ but reaching −3‰, Fig. 3; note by convention reversed δ 18 O axis) suggests that maximum African (monsoon) run-off into the Eastern Mediterranean reached roughly similar intensities throughout the Plio-Pleistocene. Furthermore, the ODP967 Ba/Al, Ti/Al and δ 18 O records all attest to the continuity of run-off maxima and 'sapropel-like' Pliocene C org fluxes.
The longest continuous (albeit lower-resolution) time-series of Saharan/North African hydroclimate changes extend back tõ 4-4.5 Ma and are based on terrigenous (=non-biogenic) sediment fluxes offshore of Northwest Africa and Arabia 24,25 , where terrigenous components are primarily sourced from aeolian dust 25 . At ODP967, an isothermal remanent magnetisation (IRM) proxy for haematite (IRM 900@120mT ; see Methods) reflects Saharan dust inputs to this site 26 . IRM 900@120mT values remain relatively low until~1.4 Ma, with notable increases at 1.2 and 0.9 Ma, coincident with the Mid-Pleistocene transition (MPT) (Fig. 4e). Increased dust fluxes at the MPT are also observed in a lower resolution ODP967 IRM record extending back to 3 Ma ref. 26,27 , and in dust records from ODP sites 664 (Fig. 4h) and 721/722 (Arabian Sea) 27 . There are also discrepancies among these records, and among terrigenous (dust) records from other ODP sites off Northwest Africa 25,27 (Fig. 1). For example, sites more proximal to the Sahara (e.g. ODP659) have high-amplitude dust fluxes throughout the Plio-Pleistocene 25 , whereas more distal sites (e.g., ODP662/3 and 664) record higher-amplitude dust fluxes from~3 and 2.7 Ma (Fig. 4g, h). At ODP Site 959, Ti/Al values suggest increasing aeolian dust inputs from~3.5 Ma, with potentially a slight increase at 3.2 Ma (Fig. 4i), although this might partly reflect Guinea Current changes 28 . Lower dust inputs tend to coincide with GSPs, but not consistently ( Supplementary Fig. S3); instead, they likely relate to distance from dust source areas and prevailing dust trajectories 25,27 , i.e. factors other than humidity/aridity changes.
The Plio-Pleistocene sapropel record. Intriguingly, there are no visibly preserved sapropels in most of the Pliocene portion of ODP967 ref. 29 (Fig. 2) Table S1). The difference among Eastern Mediterranean sites remains unexplained. It is unlikely to relate to Pliocene sedimentation rates, which are similar among the four sites ( Supplementary Fig. S4), but may relate to water depth because Pliocene sapropels are absent from the deepest sites 29 . Spatial offsets in the timing, duration, and intensity of sapropel formation are expected 8,30 , but such a dramatic contrast in Pliocene sapropel preservation among Eastern Mediterranean sites implies basin-wide changes in the balance of deep-water oxygen supply (ventilation) vs demand (C org remineralization and, likely, export).
The ODP967 sapropel-sequence change coincides with a shift to larger amplitude Ba/Al and Ti/Al fluctuations ( Fig. 4b-d), a two-tothreefold increase in terrigenous element concentrations (Supplementary Fig. S5), and a shift in the mean and variance of Ti/Al-based on change-point analysis (Fig. 4c). Change-points at ca 4.4 Ma and ca  1.9-1.2 Ma are fewer/more time-diffuse and do not coincide with major lithological (sapropel) changes, hence we do not focus on these changes. At ODP967, element fluxes reflect sediment density changes associated with sapropel lithology (Supplementary Fig. S7); hence, relative variations in terrigenous vs biogenic element concentrations more accurately indicate terrigenous inputs. Run-off and terrigenous sediment inputs to the Levantine Basin are dominantly sourced from North Africa 19,22,31 , and elevated terrigenous element concentrations at ODP967 tend to be associated with insolation maxima, although not consistently ( Supplementary Fig. S6). Plio-Pleistocene Northern Mediterranean sediment sources to ODP967 are unconstrained and, hence, cannot be excluded, but are likely less significant than African  Fig. 1 for site locations, except ODP Site 1165 which is offshore Prydz Bay, Antarctica). a Tectonic event around Sicily 54 in Chron 2An.2n; b-e ODP967 Ba/Al, Ti/Al, 400-kyr moving standard deviation of Ti/Al (black) and magnetic dust proxy (IRM 900@120mT ) (this study). Change-points in (c) are based on changes in the mean (red) and standard deviation (blue) of the Ti/Al time-series (closed triangles) and Ti/Al residuals (open triangles) after removing low-frequency (140-1200 kyr) variability; f pollen zones from ODP Site 659 ref. 32 (green = more humid; yellow/orange = increasing aridity and humid/arid variability); g, h aeolian dust records from offshore West Africa 24, 25 , i ODP Site 959 Ti/Al (grey) 28 with 11-point running average (red); j tree index based on phytoliths from Baringo Basin core BTB13, with 95% confidence intervals 37 (downward bars limited to the y-axis); k DSDP Site 231 leaf wax δD with inferred shift to more aridity at 3.3-3.0 Ma ref. 38 Table 2), a marked freshwater run-off increase into the Eastern Mediterranean would be registered in ODP967 δ 18 O, irrespective of the freshwater source. However, ODP967 δ 18 O reaches similar minima throughout the last 4.5 Ma (Fig. 3). H 1 also implies a shift to more rainfall after 3.2 Ma, which is inconsistent with pollen data 32 (Fig. 4f) and modelling results 33-35 that indicate a generally greener Pliocene Sahara (i.e. forests and wetlands). Leaf wax δ 13 C 31 at ODP Site 659 ( Fig. 1) is generally more negative in the Pliocene compared with the last glacial cycle, which is consistent with elevated Pliocene North African humidity 36 . Similarly, high-resolution pollen and tree index records from the Baringo Basin reveal a marked shift to drier conditions at 3.2 Ma ref. 37 (Fig. 4j), which is consistent with leaf wax isotope data from DSDP Site 231 in the Gulf of Aden 38 (Fig. 4k). Leaf wax isotope records (δD, δ 13 C) in that study were interpreted in combination with other regional palaeoclimate records and suggest two main shifts to drier conditions: at 5-4.5 and 3.3-3.0 Ma. Hence, available evidence causes us to reject H 1 . H 2 implies a more erodible North African landscape from 3.2 Ma onward, which in turn suggests more arid or variable climate conditions, or the emergence of new-possibly seasonaldrainage pathways. Equally, a reduction in soil-stabilising vegetation would facilitate sediment erosion through existing channels. Pollen records support a shift to increased aridity and climate variability after 3.2 Ma (see above), while dust records are more equivocal: only two ODP sites (662/3 and 664) record a shift to higher amplitude dust inputs at around 3 Ma (Fig. 4g, h). However, aeolian dust records can reflect other factors (e.g. wind patterns, distance to dust source), which could explain some offsets (Sites 662/3 and 664 are more distal from the Sahara than Site 659). Furthermore, if the amplitude increase in ODP967 Ti/ Al variations (aeolian vs riverine proxy) at 3.2 Ma is Africanclimate driven, then the lack of a coeval shift in ODP967 IRM 900@120mT (aeolian proxy) implies that the Ti/Al change primarily reflects a change in riverine rather than aeolian components. At ODP967, this component is sourced primarily from the Nile 18,19 . While the Sahara has likely existed since the Miocene 39 , the Nile evolved through the Plio-Pleistocene [40][41][42] . The timings of its various development stages are not wellconstrained, but a recent synthesis 41 proposes a Late Pliocene/ early Pleistocene connection of the Blue Nile/Atbara-Tekeze rivers to the palaeo-Nile, and the emergence of the modern Egyptian Nile flood plain and delta (the White Nile joined the main channel within the last 0.5 Ma ref. 41 ). Nile evolution could, therefore, account for a major mid/late Pliocene shift in drainage pathways and suspended sediment.
It is implicit to H 2 (and inferred from our ODP967 δ 18 O record) that freshwater run-off fluxes during GSPs must have been decoupled from long-term changes in suspended sediment loads. Several lines of evidence suggest that this is plausible. First, factors determining river sediment loads over geological time reflect a complex interplay of tectonics (e.g. rock type, relief) and climate (e.g. precipitation, run-off). Peak suspended load does not always correspond to peak run-off, unlike typical annual cycles, so landscape changes driven by tectonic (and climate) evolution can result in more erodible surfaces, and thus increased sediment loads, without an attendant change in local rainfall 43 . Second, ecological modelling and data suggest that African biomes are highly sensitive to small reductions in precipitation 44 . Major biome changes in tropical Africa have been simulated without a total annual precipitation change, simply by altering rainfall seasonality 45 . Moreover, the inclusion of soil feedbacks in GSP simulations can reproduce pollen-inferred vegetation shifts at around 400 mm/yr mean precipitation, relative to~600 mm/yr in the absence of soil feedbacks 46 .
We also cannot ignore the potential effects of large-scale global changes on African climate/ landscape at ca 3.2 Ma. The first major Northern Hemisphere glaciation (based on global benthic δ 18 O) is in stages M2-MG2 at 3.295-3.340 Ma ref. 47 (Figs. 2 and  4n), although the onset was spatially variable and as early as 3.6 Ma ref. 48 . However, the average time-dependent standard deviation of benthic δ 18 O from 25 high-resolution records starts to increase at 3.2 Ma ref. 48 . Stronger high-latitude cooling and intra-and inter-hemispheric SST gradients are observed from 3.3 Ma ref. 49,50 , coeval with a tenfold IRD (ice-rafted debris) flux increase off East Antarctica 51 (Fig. 4m). Central Asian aridification has also been dated to 3.3 Ma, based on halite content, grain size, and magnetic proxies from the Qaidam Basin 52 . All of these developments would have impacted atmospheric dynamics and latitudinal temperature gradients that drive seasonal North African climate variability, so they could relate to either H 1 or H 2 . We have rejected H 1 . Therefore, based on available evidence, we retain H 2 as a possibility.
Strait of Sicily reconfiguration. Considering H 3 and H 4 ( Table 2), Mediterranean tectonics could alter basin and sill depths, which would in turn affect deep-water ventilation. Tectonics within the wider Eastern Mediterranean catchment could also affect terrigenous fluxes to the seafloor, by rerouting transportation pathways and/or increasing continental erosion. The Mediterranean Basin has been tectonically active since reaching its modern configuration in the Oligocene-Miocene 53 . Detailed palaeomagnetic work in Sicily suggests a rapid (80,000-100,000 yr) differential clockwise rotation at 3.21 Ma, near the C2An.2n-C2An.2r boundary 54 (Fig. 4a). The rotation compressed the Sicilian fold-and-thrust belt and its foreland [54][55][56] and accelerated the Tyrrhenian Sea opening between 3.5 and 2 Ma ref. 54 , and is consistent with the evidence of middle Pliocene tectonics in the Sicilian Strait 56 . The timing corresponds with the Trubi-Narbone Formation boundary and may correspond to a rotational phase in northern Italy and a tectonic event north of Crete 54 .
The close timing between lithological changes at ODP967 (and likely also at Site 964 ref. 29 ) and inferred tectonic adjustment of the Sicilian Strait is tantalising. In the modern Mediterranean, Eastern Mediterranean deep-water flushing over the Sicily sill depends on Bernoulli aspiration (akin to 'pulling' waters out of the eastern basin) and Eastern Mediterranean deep-water formation 'pushing' deep waters out 8 . At the present sill depth Deep-sea ventilation versus carbon export. Reduced deep Eastern Mediterranean ventilation during sapropel/monsoon periods is forced by surface buoyancy gain (i.e. freshening), yet the relationship between sapropel C org content and degree of buoyancy forcing is complex 8 . Sapropel C org is typically of marine origin, but the type, mechanism, and amount of primary/export production is uncertain; however, consensus suggests a welldeveloped deep chlorophyll maximum for most sapropels 8 . If increased terrigenous fluxes from 3.2 Ma onward (e.g. hypothesis H 2 ) brought more labile terrestrial C org and nutrients to the Eastern Mediterranean, they may have fuelled higher primary productivity and C org export rates relative to the Pliocene, leading to more intensely developed sapropels. Evidence to support/refute this is equivocal or non-existent. Riverine nutrient inputs remain unquantified even for the most recent sapropel 57 . Pliocene sapropels in astronomically dated sections from Sicily and Cyprus contain lower C org concentrations than typical Pleistocene sapropels 58,59 , but sapropels in outcrops typically contain less C org and total N than sapropels in offshore cores 60 . There is general consensus among earlier studies 60-62 that Pliocene and Pleistocene sapropels are similar, and that Pliocene sapropels can contain high C org (≤30%) ref. 62 , but the down-core depth intervals studied in these examples correspond to ages younger than 3.1 Ma. Meanwhile, similar early/mid-Pliocene and Pleistocene abundances of Florisphaera profunda-indicative of a deep chlorophyll maximum-are observed in sapropels from outcrops in Cyprus 59 and at ODP967 ref. 63 , implying similarly elevated Pliocene and Pleistocene primary productivity during sapropel deposition. Nonetheless, hypothesis H 2 cannot account for Eastern Mediterranean ventilation changes after 3.2 Ma, so improved sapropel preservation after 3.2 Ma must be solely due to increased C org export. In contrast, Sicilian Strait shoaling (H 3-4 ) could, in principle, induce an increased tendency toward Eastern Mediterranean deep-water stagnation, but-as yet-evidence remains circumstantial and sparse. Major benthic foraminiferal turnovers are documented between 3.6 and 2.6 Ma in the Punta Piccola section 64 (Southern Italy) and deep ODP sites in the Western and Eastern Mediterranean basins 65 , which indicate increasing instability in bottom-water conditions. However, the main change is focussed at~2.7 Ma, some 500,000 years after the events discussed here. Coupled Mg/Ca-δ 18 O and εNd trends suggest that Mediterranean Outflow Water (MOW) intensified between 3.5 and 3.3 Ma ref. 66 , but this pre-dates our observed geochemical shift. Nevertheless, seismic data and core sediments from the Gulf of Cadiz (Fig. 1) indicate a hiatus at ca 3.2-3.0 Ma and subsequent contourite deposition, which is attributed to MOW intensification in response to Strait of Gibraltar tectonics 67 . The timing is based on preliminary biostratigraphic results and might be older (~3.4-3.3 Ma) according to a revised mid-late Pliocene chronology for Site U1387 ref. 68 . Regardless, deep-water exchange through the Sicilian Strait, rather than the Strait of Gibraltar, is the critical factor for deep ventilation of the Eastern basin 8 . Contributions to Pliocene buoyancy forcing from local precipitation/evaporation are unknown, but we assume that they were negligible relative to African run-off forcing. Likewise, elevated SSTs cannot solely account for buoyancy forcing, given that they only partially contribute to water-column stratification during the Last Interglacial sapropel formation 17,18 , and mostly in response to salinity-driven stratification.

Conclusions
We have evaluated different hypotheses (climatic vs tectonic) to explain a marked shift in Eastern Mediterranean sediments 3.2 Ma ago. We find that sill tectonics (H 3,4 ) could account for reduced ventilation after 3.2 Ma (which best explains subsequently increased ODP967 Ba and S concentrations) and may coincide with inferred tectonic activity in the Strait of Gibraltar. However, hypothesis H 3 cannot satisfactorily account for a doubling (or more) of terrigenous element concentrations at ODP967 after 3.2 Ma. Hypothesis H 4 might account for this if the terrigenous influx was locally sourced (i.e. Eastern Mediterranean borderlands), but circumstantial evidence is lacking, and-more importantly-Pleistocene terrigenous sediments at ODP967 predominantly originated from North Africa 19 . Conversely, the non-linear response of riverine terrigenous loads to African climate/landscape evolution (hypothesis H 2 ) better accounts for increased ODP967 terrigenous element concentrations and Ti/Al fluxes after 3.2 Ma. H 2 can also be directly linked to midlate Pliocene global climate evolution. The 3.2 Ma shift could represent a critical North African landscape transition in response to a global climate state shift to icehouse conditions 69 . Increased geochemical signal variance is common prior to critical transitions in Eastern Mediterranean sediments 70 , and the biggest systematic Ti/Al variance increase is from 3.4 to 3.2 Ma (Fig. 4d). African pollen evidence attests to a more arid Pleistocene compared to the Pliocene 32,71 , and a significantly more erodible landscape must exist before a stepwise amplitude increase in fluvial suspended sediment can occur. Furthermore, North African desert-soil albedo and vegetation feedbacks strongly amplify rainfall variability 46,72,73 . The 3.2 Ma increase in ODP967 Ti/Al and fluvial-sourced terrigenous elements might, therefore, signify the onset of modern-day North African aridity. Further quantitative studies (e.g. weathering proxies) are needed to confirm this. Nevertheless, we cannot (yet) fully reject H 4 . It is conceivable that both H 2 and H 4 are valid and were coupled to major global climatic and tectonic changes; future model-data comparisons should test this possibility.

Methods
Bulk element geochemistry. We combine published x-ray fluorescence (XRF) core-scanner data from ODP Site 967 spanning 0-3 Ma ref. 15 with new XRF corescanner data for the 3-5 Ma interval. Scanning was performed on archive core sections at MARUM University of Bremen, on an Avaatech XRF core scanner. Core sections were covered with 4 mm-thick Ultralene film and measured at 50 and 30 kV with 0.55 mA current and with a Cu and Pd-thick filter, respectively, and at 10 kV with 0.035 mA current (no filter); count time for all runs was 7 s. Element 'counts' for the entire 0-5 Ma interval were converted into element concentrations by multivariate log-ratio calibration 74 , using published 35,75 and new wavelength dispersive (WD)-XRF reference element concentrations (Supplementary Figs. S8, S9). For these, 42 bulk sediment samples were chosen to cover a range of lithologies based on the XRF scan, then 1 cm 3 dried ground sample was mixed with a lithium tetraborate/lithium metaborate flux and fused into 39 mm diameter beads. Major element abundances were analysed by WD-XRF using a Bruker S8 Tiger™ spectrometer at Geoscience Australia. Loss on ignition (LOI) was measured by gravimetry after combustion at 1000°C. One in every ten samples was duplicated along with multiple analyses of three international standards (NCS DC70306, MAG-1, ML-2) and an internal basalt standard (WG1). Quantification limits for all major element oxides are <0.2% and reproducibility is within 1%.
The upper 40 m (0-1.3 Ma) of ODP967 contains cm-thick turbidites, which are not always visible, and which typically occur within sapropels 76 . Sudden 'extreme' enrichments in Ti, Si, Zr, and to a lesser extent Fe and K, and corresponding Ca depletions, may therefore indicate a turbidite 75 . Anomalous K spikes (relative to background values) are apparent at 2.3 and 2.9 Ma (Supplementary Fig. S4), but elsewhere any potential turbidite signal is obfuscated by the dominant sapropel/ marl signal.
ODP967 chronology. Published age-depth tie-points for the 0-3 Ma interval are based on tuning a principal-component derived sapropel proxy to precession 15 . Here, we extend the age model to 5 Ma by tuning the Ba/Ti record from XRF core-scanning to precession minima with zero phase lag ( Supplementary Fig. S1). Phase lags between a sapropel mid-point and nearest precession minimum/insolation maximum can be up to 3 ky but are not systematic and are likely limited to major deglaciations 12 , hence (minor) lags would only be expected after the MPT.
Environmental magnetism. ODP967 U-channel samples were sliced at 1-cm intervals into non-magnetic plastic cubes and measured on a 2-G Enterprises cryogenic magnetometer at the Australian National University (ANU). IRM 900@120mT was obtained after imparting a 900 mT induction in a direct current field, followed by alternating field (AF) demagnetisation in a peak 120 mT field.
Stable oxygen isotopes (δ 18 O). Bulk sediment samples were washed through 63 μm sieves with DI water and the residue was oven-dried at 45°C for 24 h. Globigerinoides ruber (white) specimens were picked from the >300 μm size fraction, adhering to 'sensu stricto' morphotypes and size range (±25 μm). Picked tests (typically 10-20, depending on abundance) were gently crushed and cleaned by briefly ultrasonicating in methanol, then air-dried after decanting the fine suspended matter. Samples were analysed at ANU using a Thermo Fisher Scientific Delta Advantage mass spectrometer coupled to a Kiel IV carbonate device for sample digestion. Isotope data were normalised to the Vienna Peedee Belemnite (VPDB) scale using NBS-19 or IAEA-603, and NBS-18. External reproducibility (1σ) was always better than 0.08‰.
Bernoulli aspiration depth (d). We use values from ref. 8 in the equation of ref. 77 . Data analysis. Change-points were estimated using the MATLAB built-in function 'findchangepts', following ref. 78 . Change-points were determined based on the mean and standard deviation of the raw geochemical records, and a maximum of n = 3 change-points was stipulated as a parsimonious balance between constraining output to the most significant changes while allowing a >1 outcome. The moving standard deviation of the ODP967 Ti/Al record was calculated using the MATLAB built-in function 'movstd'.

Data availability
ODP Site 967 data from this study are available from Panagea (www.pangaea.de) under 'Plio-Pleistocene scanning XRF, stable isotope and environmental magnetic data from ODP Site 967' and are also available as online Supplementary Data accompanying this article.