Introduction

The Mid-Pleistocene Transition (1.2–0.7 Ma, MPT) represents a time of planetary transition, from milder 40 thousand-year (kyr) glacial cycles to more extreme 100 kyr glacial cycles. The cause of this shift is debated, but appears to be linked to changes in the carbon cycle (see Herbert (2023) and the references therein). Several lines of evidence suggest that these climate changes may have perturbed the local ecologies of several regions, including Southern Africa. For example, isotopic measurements of faunal remains (e.g., grazer tooth enamel, ostrich egg shells) suggest dietary changes in grazers and increases in relative humidity around 0.9–0.8 Ma1, which, paired with phytolith morphotype analyses from Wonderwerk Cave2,3 and sedimentological data from Mamatwan Mine4, suggests shifts in moisture availability during the MPT. However, the scarcity and coarseness of these records makes it difficult to ascertain whether this is a signal of local variability, a regional trend, or whether a moisture shift is part of an overarching climate regime change.

There are compelling reasons to think that the MPT would affect climate across Southern Africa (SA, Fig. 1a). One prominent hypothesis about the MPT climate shift posits that expanded sea ice near Antarctica strengthened Southern Ocean (SO) stratification and induced ocean circulation changes5,6,7. This would impact global and regional climates, especially in SA, where the surrounding ocean is a confluence of the Benguela Upwelling System (BUS), the Agulhas System, and the Antarctic circumpolar frontal system (AFS). The intensity of the Agulhas Leakage (AL), warm and salty masses of water which escape the retroflection of the Agulhas Current, is hypothesized to be modulated by variations of the westerly wind belt and AFS8. Proxy records associate intervals of reduced AL with northward shifts of the AFS, which occurred over the MPT and during post-MPT glacial periods9,10,11. Northward migrations of the AFS and diminished AL are linked to the intensification of winter rainfall over SA on millennial timescales12,13, however, there is debate over precisely how the westerlies, AFS, and AL influence SA rainfall in the modern14,15,16.

Fig. 1: Regional climatology, site locations, and model results.
Fig. 1: Regional climatology, site locations, and model results.
Full size image

a Modern climatology of the Southern Africa from reanalysis data. The Orange River watershed is mapped using HydroSHEDS data products96. The white star marks the location of Ocean Drilling Program (ODP) Site 1084. Other ODP and International Ocean Discovery Program (IODP) drill sites discussed in the text are marked with gray circles; white circles mark the Mamatwan Mine (M), Wonderwerk Cave (WW), and Gharp Escarpment (GE) archaeological sites discussed in the text. Dashed blue contours over the land denote the extent of the winter, yearly, and summer rainfall zones (WRZ, YRZ, and SRZ, respectively). White arrows highlight the Benguela Current, Agulhas Leakage (AL), Aguhlas Return Current (ARC), and Antarctic Circumpolar Current (ACC). The dashed white contours mark the fronts within the Antarcitc Frontal System (AFS): Subtropical Front (STF), Subantarctic Front (SAF), and Antarctic Polar Front (APF) based on isotherms93,94. The gray contour marks the 35 PSI isohaline and the black contour marks the line of zero Windstress Curl (WSC). b Simulated annual average precipitation for model runs with a northerly placed line of zero WSC minus runs with a southerly placed line of zero WSC. c Simulated latitudinal position of the line of zero WSC for glacials and interglacials before, during, and after the Mid-Pleistocene Transition (MPT), demonstrating that the model produces a northerly displacement of the AFS during the MPT and glacials. The number of models (n) used for each interval is noted.

A more robust understanding of the complex SA hydroclimate system is vital for assessing past and future changes in hydroclimate. Here, we demonstrate that substantial shifts in the leaf wax isotopic signatures of precipitation and vegetation over Southern Africa coincide with significant shifts in the position of the AFS across the MPT. We use HadCM3 simulations to test for a regionally coherent mechanism which reconciles the different perspectives gained from the terrestrial versus marine records (e.g., assess whether a shift in hydroclimate related to large-scale circulation shifts). In particular, we test the hypothesis that northward shifts in the AFS resulted in an expansion of the winter rainfall zone (WRZ) and WRZ-type vegetation (i.e., C3 plants) from 1.5 to 0.9 Ma, and that a sudden withdrawal of the AFS after 0.9 Ma prompted a contraction of WRZ rainfall and vegetation. We conclude by discussing the parallels between the mechanism presented here and that which drove extreme drought in SA in recent years; we raise the question of whether the relatively short-lived, but intense, drought events we see in the modern have the potential to become longer-lived features of the future climatology.

Results and discussion

Leaf-Wax analysis of sediment from ODP 1084

ODP 1084 (25° 30.8345’S, 13° 01.6668E) is located in the central upwelling cell of the BUS (Fig. 1a); age control is derived from shipboard magnetostratigraphy and nannofossil datums17. Terrigenous materials are transported to marine sediments via aeolian processes18 and fluvial transport from the Orange River watershed19. Sediments from the watershed enter the water column at the Orange River delta, approximately 250 km south of ODP 1084; a portion of this sediment is carried northward to ODP 1084 via the Benguela Current20 while the rest is transported southward via a coastal counter current21. The large source area for ODP 1084 means that contributions from a variety of biomes, including the Succulent Karoo, Nama Karoo, Kalahari Savannah, and grasslands, are integrated into a single signal, which ultimately represents regional trends, rather than localized variability.

Paired carbon and hydrogen isotopic analyses of plant leaf waxes are commonly used to infer the relative proportions of C3 to C4 vegetation and the hydrogen isotopic signature of precipitation, respectively, especially in Africa (e.g., Feakins et al.22; Tierney et al.23; Lupien et al.24). Such methodology is appropriate for regions with little or no presence of CAM vegetation, but may not work well in Southwestern Africa, where CAM plants are more prominent25,26; the broad range of δ13Cwax values that CAM plants display make them impossible to differentiate from C3 and C4 using δ13Cwax alone27. Furthermore, there is high uncertainty regarding the apparent fractionation of hydrogen in CAM plants, which is well constrained for C3 and C4 vegetation. To examine the impact of overlooking the potential influence of CAM plants on δDwax, we performed a carbon-correction using a standard methodology to infer the hydrogen isotopic signature of precipitation (see Methods for details) and found that the correction artificially smooths the δDp record (Fig. S5). As such, we utilize the raw, uncorrected δDwax data in our statistical analyses.

While we cannot reliably use δ13Cwax on its own to estimate vegetation changes in this region, we may be able to characterize overall biome transitions by leveraging the ratios between individual n-alkanes. Empirical studies have demonstrated that the ratios between individual n-alkanes are sensitive to environmental changes (e.g., ref. 28), and biomes in Southwestern Africa can be characterized by these ratios29. We use ratios of different n-alkanes (e.g,. Norm31 and Norm33) to roughly track the contribution of the WRZ taxa to our record because these metrics reflect the relative wax contribution of different biomes to the sediment (see methods for more details). Finally, we use simulations from the Hadley Centre Coupled Model 3 (HadCM3) to examine the dynamics driving changes in δDwax. Despite modifications aimed toward more accurately capturing the meridional temperature gradient (MTG)30, HadCM3 remains particularly biased toward warmth over parts of the SO and BUS. Comparing the Preindustrial control run to observational data (Fig. S2), we find the model simulates a weaker temperature gradient over the AFS and weaker westerlies than those observed in the modern. Considering this bias, along with the limitations of a low spatial resolution, we use the model to examine the sensitivity of rainfall to changes in the AFS, rather than treating the model output as an exact picture of MPT climate (see Methods for further analytical and model details). Given the more southerly placement of the SO boundary in the model, we expect rainfall responses to be similarly displaced.

Coordinated changes in Southern African hydroclimate during the MPT

Lomb-Scargle (LS) periodogram analysis on δDwax (Fig. S6) indicates possible low-frequency orbital influence in the 400-kyr and 100-kyr bands from 0.8 Ma to present, though confidence is low (less than 75%). Given the prominence of orbital signals in other African hydroclimate records (e.g., refs. 31,32,33, the lack of orbital variability in our record is probably due to low sampling resolution (>21-kyr between many samples, Fig. S1). Rather than orbital-scale variability, we observe broad trends in the δDwax and δ13Cwax records over the past 3.4 Ma (Fig. 2a, b). We also observe relatively abrupt enrichment in both δDwax and δ13Cwax records over the MPT, particularly at ~0.85–0.75 Ma.

Fig. 2: A compilation of proxy records.
Fig. 2: A compilation of proxy records.
Full size image

Top Right: The full Plio-Pleistocene a δDwax and b δ13Cwax records for Ocean Drilling Program (ODP) Site 1084, error bars indicate ±1σ. Vertical gray bars in both the Top Right and Left panels indicate glacial intervals; the vertical blue bar in each panel highlights the Mid-Pleistocene Transition. Left: A compilation of records from the last 1.5 Ma. c gray bars indicate the latitudinal position the Antarctic Polar Front (and the circumpolar biogenic opal belt) based on the presence of laminated diatom mats (i.e., biogenic opal deposits), dashed lines indicate estimated position of the front50; d the ODP 1081 δDp record62; e, f are the last 1.5 Ma of (a, b), respectively; g alkenone based sea surface temperatures (SST) records from ODP 108461, h Southern Hemisphere SST gradient (extratropical - tropical Atlantic, see Fig. S7 for details); i biogenic SiO2 mass accumulation rates (MAR) from ODP 108417; j alkenone MAR from ODP 108461; k Indian Ocean salinity record52; l G. menardii accumulation rate10); m ice rafted debris records from the Agulhas Ridge7); n the dust MAR record from ODP 109059. Bottom Right: o scatter plot displaying the compiled locations and dates of Southern African hominin fossils or ages of the beds containing said remains (see Data Availability Statement), the hatched bar highlights a relative gap in the fossil record; p, q are the same as (a, b), respectively.

The overarching trends in δDwax and δ13Cwax are likely related to long-term changes in both regional and global SSTs. Varimax factor analysis (see Methods for details) of our records and several regional proxy records (Fig. 3) identifies two primary modes of variability which explain ~60% of the variance in the data. The first mode (Factor 1; Fig. 3a, c), which explains 42.62% of the variance, captures the long-term trends in BUS temperatures, hydroclimate, and vegetation, and dust accumulation; Factor 1 thus links the intensification of the BUS to the overall aridification of the African continent34,35. The secondary mode of varibility (Factor 2; Fig. 3b, d) explains 20.56% of the variance in the data and indicates that changes in the MTG are strongly tied to changes in the nutrient availability within the BUS (ODP 1084 Alkenone MAR) and relatively abrupt variations in δDwax. The abrupt enrichment in δDwax around ~0.9–0.75 Ma is thus likely influenced by similarly stark changes in ocean conditions. The changes in δ13Cwax are probably tied to changing hydroclimate, though δ13Cwax loads less strongly on Factor 2.

Fig. 3: Varimax factor analysis of leaf wax isotopes, BUS sea surface temperatures (SST) and productivity proxies, and Southern Ocean dust flux.
Fig. 3: Varimax factor analysis of leaf wax isotopes, BUS sea surface temperatures (SST) and productivity proxies, and Southern Ocean dust flux.
Full size image

a Timeseries of Factor 1 (black) plotted against the standardized timeseries of the variables with the strongest loadings (ODP 1084 SST and δ13Cwax). b Timeseries of Factor 2 (black) plotted against the standardized timeseries of the variables with the strongest loadings (Southern Hemisphere (SH) SST gradient, alkenone mass accumulation rate (MAR), and 1084 δDwax). c, d Bar plots rank the relative loading of variables on each factor, where the leftmost bar indicates the largest loading and the rightmost bar indicates the smallest loading. Note units are arbitrary and colors match those of the corresponding variable in Fig. 2.

Due to its stratiform character and relatively depleted vapor source (the South Atlantic), precipitation in the WRZ is depleted in deuterium relative that of the summer rainfall zone (SRZ), which is generally convective and utilizes a relatively enriched vapor source (the Indian Ocean)36,37,38. An isotopic gradient exist within the SRZ, as precipitation moves inland from the coast and undergoes fractionation by both the amount and continental effects29,36,38,39; due to its smaller spatial extent and lower overall precipitation volume, the isotopic signature of rainfall in the WRZ is less impacted by these effects38,40. The vegetation of each rainfall zone is distinct: the Fynbos, Succulent and Nama Karoo biomes (mainly C3 and CAM plants) dominate in the WRZ, while grassland and savannah (mostly C4 plants) are more prevalent in the SRZ26. Norm31-Norm33 approaches zero at ~0.9 Ma, which may indicate an increased contribution of waxes from grassland and savannah biomes (Fig. S7). We therefore interpret the abrupt enrichment of the ODP 1084 δDwax and δ13Cwax signals as contraction of the winter rainfall zone and associated vegetation types from ~0.9–0.75 Ma. This interpretation is consistent with multiple terrestrial records from the Kalahari which indicate a substantial hydroclimate and vegetation shift occurred around 0.9–0.8 Ma1,41.

Antarctic front migration mechanism

Assessment of regional proxy records presents a compelling mechanism for this shift in seasonality. Around 1.5 Ma, Southern Hemisphere extratropical temperatures cooled rapidly, falling below preindustrial levels (Fig. S742,43). This cooling resulted in a strengthening of the MTG (Figs. 2h, S6) and allowed for an expansion of sea ice (Fig. 2m7), which prompted a reorganization of ocean circulation6,7,44,45. In the surface waters, a steeper temperature gradient over the SO shifted the AFS northward; the presence of foraminifera generally found in Antarctic and sub-Antarctic waters (e.g., refs. 11,46,47.), along with records of opal deposition48,49,50 and primary productivity11,51, indicate a more northerly presence of the AFS prior to ∼1 Ma. The northward presence of the AFS may have also created a choke-point at the Agulhas Plateau which limited AL (Fig. 2l10,) and, combined with lower sea levels and weakened Indonesian Throughflow, caused a build-up of warmer, saline waters in the western Indian Ocean (Fig. 2k52,). The strengthened MTG, expansion of sea ice, and limited AL probably impacted deep water circulation as well: foraminiferal trace elements and authigenic Nd isotope proxies indicate increased carbon storage and weaker Atlantic Meridional Overturning Circulation over the MPT, culminating in a meridional overturning minimum at ~0.9 Ma6,44.

Changes of the MTG influence the width (and intensity) of atmospheric Hadley circulation, with models and modern observations demonstrating that a weaker gradient drives an expansion of the Hadley cell53,54 and a poleward propagation of the westerly storm track53,55,56,57,58. By this same mechanism, a steepening of the MTG from ~1.5 to 1.1 Ma and from ~1 to 0.8 would cause a contraction of the Hadley cell and an equatorward propagation of the westerly storm track. As winter storms travel further north, the WRZ would expand and the proportion of winter to summer rain would increase; an increase in isotopically depleted winter rains would contribute to the gradual depletion of deuterium in δDwax prior to 0.9 Ma38.

As mentioned previously, the model’s resolution limitations and temperature bias result in a more poleward-displaced SO position in PI simulations compared to observations, and any rainfall response to changes in the SO is expected to be similarly displaced. The HadCM3 simulations feature northward shifts of the subtropical front and a strengthening of the southern westerlies which result in both a zonal increase and intensification of winter rainfall over the waters south of Africa and parts of SA during the MPT (Fig. 1b, c). The model results thus demonstrate the sensitivity of winter rainfall to dynamical changes in the westerlies and SO. Further work examining this response in models of higher resolution and with the ability to trace water isotopologues is needed.

Frontal system retreat

From ~1.1 to 1 Ma and ~0.8–0.75 Ma, the Southern Hemisphere MTG rapidly weakened (Figs. 2h, S7). The first weakening, from ~1.1 to 1 Ma, occured as tropical SSTs rapidly decreased, is probably due to an increase of equatorial upwelling as the Hadley Circulation intensified59,60,61. Despite a weaker MTG, the proxy evidence indicates that the AFS remained in a northerly position from ~1.1 to 1 Ma.

From ~0.8 to 0.75 Ma, SSTs in both the tropics and Southern Hemisphere Atlantic extratropics increased - the weakening of the MTG appears to have been driven by differences in the rates of warming, as the extratropics appear to have warmed more rapidly than the tropics (Fig. S7). During this interval, sea ice retreated7 and AL resumed10. The coincident decrease and increase in the respective accumulation rates of biogenic silica and alkenones at ODP 1084 suggests the withdrawal of polar nutrients from the BUS17,49,60 as the AFS retreated.

As the AFS and winter storm track migrated southward, the extent of the WRZ would be limited; the abrupt enrichment in δDwax at ODP 1084 and ODP 108162 is consistent with an increase in the proportion of summer rainfall38. The stark fluctuation in δ13Cwax values over this interval indicate a vegetation shift over the leaf wax source area, and our inference that wax contribution from grassland and savannah biomes increased at ~0.8 Ma is consistent with an expansion of the SRZ. Improved sampling resolution during the Late Pleistocene is needed to further examine the impacts of glacial/interglacial variability on δDwax and δ13Cwax after the MPT.

Comparison to terrestrial and tropical records

The interpretation of the ODP 1084 leaf wax records presented here agrees with multiple terrestrial records from the Kalahari which indicate a substantial hydroclimate shift occurred around 0.9–0.8 Ma1,4,41,63. For example, sedimentalogical evidence from Mamatwan Mine, ~1000 km inland from ODP 1084, points to the presence of a shallow water body from ~1.5 to 1 Ma4,63. The remains from lechwe antelope (Kobus leche), which require standing bodies of water, have been found at Wonderwerk Cave1 and regions of the Ghaap Escarpment64, indicating that similar shallow water bodies were present throughout the southern Kalahari during this interval. Just as the extension of the WRZ, and adjacent yearly rainfall zone, would help sustain these shallow waters, the contraction of these zones could help explain their disappearance after 0.9 Ma.

The mechanism presented here also supports tropical vegetation reconstructions for the MPT. Leaf wax records from ODP 1077, near the equator, indicate an increase (decrease) in tropical savannah (C3 rainforest taxa) in the Congo Basin from ~0.95–0.85 Ma, and a reverse of this trend after ~0.85 Ma, which the authors link to tropical SSTs65. East African soil carbonates demonstrate large inter-site variability; most sites in East Africa display a broad trend toward tropical C4 savannah over the last ~4 Ma, however, the records from several locations feature substantial C3 incursions (e.g., Maslin et al.66, Lüdecke67, Quinn et al.68). The shifting MTG and Hadley circulation would have impacted the placement of the Intertropical Convergence Zone, water availability, and subsequent ecosystems variation in the tropics.

Implications for hominin evolution

Elucidation of past hydroclimate change may be useful to contextualize hominin evolution in SA. Hominin individuals living in this region would have experienced abrupt, but extensive landscape change alongside dwindling freshwater resources and a coincident vegetation shift. At the same time, there is a relative reduction in artifact densities in the corresponding strata of Wonderwerk Cave2,69, as well as the last appearance of Paranthropus boisei in the SA archaeological record (e.g., Herries 2022). However, more rigorous work examining the impact of precipitation and landscape change on Paranthropus is necessary.

Implications for future climate

A deeper understanding of the mechanism presented in this study is vital for understanding and predicting modern and future climate events. From 2015 to 2018, the Day-Zero Drought devastated water supplies across southwestern Africa70,71,72. Intense scrutiny of this drought has pointed to an expansion of the Hadley cell and subsequent poleward displacement of the mid-latitude westerlies as the primary driver57,73,74,75. At the synoptic scale, the Hadley expansion may be increasing the frequency of South Atlantic Anticyclone ridging high-pressure systems57,75,76, ultimately resulting in a delayed and shortened wet-season75,76. Outside of SA, the response of other Mediterranean-type climates to changes in Hadley circulation is varied and subject to regionally specific drivers77. Compilation of hydrogen isotopic records of precipitation from Chile and South Australia (the other Southern Hemisphere Mediterranean-type climate zones) are thus needed to interrogate the local hydroclimate response to changes of the Hadley circulation and AFS over both modern and paleo-timescales.

The results presented in this paper bolster the view that long-term shifts in the Southern Hemisphere storm tracks may ultimately be responsible for hydroclimate shifts in SA. Given the similarities of these past mechanisms to those observed in the modern, further work is needed to examine whether the relatively short-lived drought events we see in the modern have the potential to become longer-lived features of the future climatology. While we cannot prevent the next extreme drought from occurring, an enhanced understanding of the underlying mechanics of regional hydroclimate may provide additional insight which improves our projections of regional climate, aids mitigation efforts, and strengthens the resilience of SA communities to the effects of anthropogenic climate change.

Summary

This study highlights substantial shifts in rainfall seasonality and vegetation cover which are associated with migrations of the AFS over the MPT. We use HadCM3 simulations to further demonstrate that SA precipitation is sensitive to variations in the westerlies and AFS; this work would benefit from additional analyses using isotope-enabled simulations with higher spatial resolution. Our records supplement the existing terrestrial records of hydroclimate change in SA. We present a mechanism of hydroclimate change during the MPT which is analogous to that of the Day-Zero Drought, suggesting that these impactful drought events we see in the modern record have the potential to become persistent features of the region’s climatology. Continued investigation of this mechanism in the past and present is therefore needed to improve our projections of, and preparations for, future extreme climate events.

Methods

Sample preparation and analytical methods

Samples were freeze-dried and homogenized, and total lipids (TLEs) were extracted and purified via column chromatography. Concentrations of C29 and C31 alkanes were quantified using a Trace 1310 GC-FID, and their hydrogen and carbon isotopic com- positions were measured using a Thermo Scientific Trace 1310 GC and a PTV injector connected via an IsoLink II pyrolysis furnace to a Delta V Plus isotope ratio mass spectrometer (GC-IRMS). H2 and CO2 gases calibrated to an n-alkane standard (A7 mix provided by Arndt Schimmelmann at Indiana University) provided references for each analysis and were used to monitor drift. H3 factor measured daily over the course of the hydrogen analyses averaged 4.703 ppm/V over a range of 1–8 V. Samples were run in triplicate for δD to obtain a precision better than 3‰ (1-σ), and duplicate or triplicate for δ13C to obtain a precision better than 0.2‰ (1-σ).

We calculated a mean carbon preference index >1, indicating terrestrial rather than petrogenic inputs78, and an mean average chain length (ACL) of 29.35 ± 0.089 (1-σ). Because concentrations of the C29 and C31 alkanes were similar (within 2.5 ng/g of one another), and because previous studies from SWA have focused on the C31 alkane as a proxy for precipitation (e.g., refs. 35,62), we report isotopic results from this chain length.

Inferring δD of precipitation from leaf waxes

The carbon isotopic measurements of 70 samples were paired with 106 hydrogen isotopic measurements of the C31 alkane (δ13Cwax and δDwax, respectively) to infer the hydrogen isotopic signature of precipitation, δDp. δDwax is offset from the hydrogen isotopic signature of the environmental source waters, generally assumed to be precipitation; this apparent fractionation, εp−w, varies systematically across plant clades79,80, with graminoids generally having a larger εp−w than those of eudicots79,81.

Within a Bayesian framework, we use δ13Cwax and empirically derived constraints on εp−w from modern grasses and eudicots (Table S123,) to infer the proportion of graminoids in our sample and weigh our εp−w accordingly. The resulting εp−w accounts for the variable fractionation between graminoids and eudicots (Eq. 1).

$$\varepsilon=({f}_{C4})\cdot {\varepsilon }_{C4}+(1-{f}_{C4})\cdot {\varepsilon }_{C3}$$
(1)

We then use our weighted εp−w to infer δDp (Eq. 2).

$$\delta D(precip)=\frac{1000+\delta {D}_{wax}}{(\varepsilon /1000)+1}-1000$$
(2)

This approach has been widely applied to infer δDp from leaf wax δDwax in a variety of paleoclimatic studies (e.g., Tierney et al.23, Bhattacharya et al.82, Windler et al.83).

n-Alkane ratio calculation

Herrmann et al.29 measured n-alkanes in soils, sediments, and river deposits from several biomes across Southern Africa and calculated the ratio of n-alkane homologues C29 and C31 and C29 and C33 as

$$Norm31={C}_{31}/({C}_{29}+{C}_{31})$$
(3)

and

$$Norm33={C}_{33}/({C}_{29}+{C}_{33})$$
(4)

where Cx is the concentration of the n-alkane with x carbon atoms.

Their results showed that in samples from the Grassland and Savannah biomes (which are mainly SRZ flora), Norm31 ≈ Norm33, but for samples from the Fynbos, Succulent Karoo, and Nama Karoo biomes (primarily in the WRZ), Norm31 > Norm33.

Based on their results, we tentatively infer the relative contributions of SRZ flora vs WRZ flora by taking the difference of Norm31 and Norm33. If Norm31-Norm33 ≈ 0, we infer that wax contributions mostly came from the Grassland or Savannah biomes. If Norm31-Norm33 > 0, we infer that wax contributions include more contributions from the Fynbos, Succulent, or Nama Karoo biomes.

Using this index, we infer more contribution from the Fynbos, Succulent Karoo, and/or Nama Karoo biomes 1.5–1 Ma, and decreasing contribution from these biomes after 1 Ma. The agreement of this inference with evidence of vegetation changes from the Kalahari interior (e.g., phytolith morphotpe analyses2,3, dietary changes in grazers1), lends credence to this index. Further work needs to be done to examine whether this index is viable outside of southwestern Africa.

Varimax factor analysis

Varimax Factor Analysis (VFA) is a type of matrix decomposition similar to Principal Component Analysis (PCA); the goal of both techniques is to reduce the number of variables while retaining as much of the original information as possible84,85. However, VFA assumes there is an unobserved variable (or latent factor) which exerts a causal influence upon the observed variables85 -this assumption ultimately allows for a more simple interpretation of the underlying structure of the correlations between measured variables.

We performed this analysis using the scikit-learn python toolkit86. The dataset was first standardized and normalized using the scikit-learn pre-processing tools, then the VFA was performed using the appropriate functions from the scikit-learn decomposition module. The results are listed in Table S2.

Model description and results

We utilize a suite of simulations from the Hadley Centre Coupled Model 3 (HadCM3). HadCM3 is a coupled climate model comprised of a 3-dimensional dynamical atmo- sphere component with a resolution of 2.5° × 3.75° and 19 vertical levels, and an ocean model with a resolution of 1.25° × 1.25° and 20 vertical levels (see Valdes et al.87 for a detailed overview). These simulations cover the last 3.6 Ma and were run every 4 ka using a configuration of HadCM3 which includes a tiling of surface types within MOSES2.1 d88 and additional atmospheric and vegetation moisture stress param- eterizations to better simulate a ‘Green Sahara’ during the mid-Holocene89. The model also includes several parameterizations to better resolve the ‘Cold-Pole Paradox’30: major parameterizations include modified cloud condensation nuclei density and cloud droplet effective radius, which better aligned the simulated meridional temper- ature gradient with proxy observations; adjustments to the ocean model to removed nonlinear instabilities; and the addition of sea ice, which was calculated on a zero-layer model and allowed for partial sea-ice coverage.

Despite these modifications, HadCM3 remains particularly biased toward warmth over parts of the Southern Ocean - comparison to COBE SST data90 demonstrates that HadCM3 simulates a substantially weaker temperature gradient over the AFS. Considering this bias, along with the limitations of a low spatial resolution, we use the model to examine whether a northward shift of the Southern Ocean is simulated during the MPT and whether this northward movement is associated with any change in the WRZ, rather than treating the model output as an exact picture of MPT climate.

We select 39 simulations from six categories: Pre-MPT Interglacials and Pre-MPT Glacials, MPT Interglacials and MPT Glacials, and Post-MPT Interglacials and Post- MPT Glacials. Each category is comprised of select model runs from within the specified interval and a complete list of these simulations can be found in Table S3.

We assess whether the selected model simulations capture shifts in SA precipitation that are consistent with δDp across the MPT. We focus on zonal winds and precipitation, and we use the line of zero WSC (WSC = 0)91,92 as indicator of the northern edge of the AFS (i.e., the STF). We considered using the position of the 11degC isotherm as a marker of the AFS91,93,94), however Southern Ocean temperature biases in the model87 mean the isotherm may not actually correspond to the location of the front.

The average latitudinal position of WSC = 0 across all model runs is 49.48°S - any model run with a WSC = 0 placement north of 49.48°S was classified as northward-shifted and those with a WSC = 0 placement south of 49.48°S was considered southward-shifted (Fig. 1b). Most Pre- and Post-MPT glacal runs were classified as northward-shifted, as are most MPT glacial and interglacial runs. Pre- and Post-MPT interglacial runs were largely categorized as southward-shifted.

To evaluate the dynamical and hydroclimate responses to simulated shifts in the STF, we subtract the average precipitation and wind fields of the southward-shifted models from the northward-shifted models (Fig. 1c). As expected, the westerly wind belt is shifted northward and strengthened in northward-shifted runs. These model runs simulate less terrestrial rainfall than the southward-shifted runs, particularly over the SRZ. However, the WRZ does expand zonally and experiences a slight increase in rainfall, and precipitation south of Africa increases substantially.