Pleistocene climate variability in eastern Africa influenced hominin evolution

Despite more than half a century of hominin fossil discoveries in eastern Africa, the regional environmental context of hominin evolution and dispersal is not well established due to the lack of continuous palaeoenvironmental records from one of the proven habitats of early human populations, particularly for the Pleistocene epoch. Here we present a 620,000-year environmental record from Chew Bahir, southern Ethiopia, which is proximal to key fossil sites. Our record documents the potential influence of different episodes of climatic variability on hominin biological and cultural transformation. The appearance of high anatomical diversity in hominin groups coincides with long-lasting and relatively stable humid conditions from ~620,000 to 275,000 years bp (episodes 1–6), interrupted by several abrupt and extreme hydroclimate perturbations. A pattern of pronounced climatic cyclicity transformed habitats during episodes 7–9 (~275,000–60,000 years bp), a crucial phase encompassing the gradual transition from Acheulean to Middle Stone Age technologies, the emergence of Homo sapiens in eastern Africa and key human social and cultural innovations. Those accumulative innovations plus the alignment of humid pulses between northeastern Africa and the eastern Mediterranean during high-frequency climate oscillations of episodes 10–12 (~60,000–10,000 years bp) could have facilitated the global dispersal of H. sapiens.

. The perennial Segen and Weyto Rivers (Fig. 1b) are the main sources of fluvial inwash from the northeastern and north-western part of the catchment including the high rainfall areas of the 2500-4000 m high southwestern Ethiopian highlands (Fig. 1). The Weyto River forms a deltaic system in the northernmost part of the Chew Bahir basin, with dense vegetation, whereas towards the southern end of the basin the surface is strongly desiccated and can be subject to local aeolian deflation of fine (silty) sediments, remobilization and sedimentation, that play however a minor role compared to the dominating fluvial-deltaic, lacustrine and authigenic processes that control the composition of the deposited sediments 1,6 . Today, rainfall in the area is associated with the seasonal migration of the tropical rain belt, resulting in bimodal rainy seasons in March-May and October-November 7 . However, Chew Bahir is adjacent to regions of unimodal and trimodal rainfall patterns, which may have shifted over the course of long-term climate change 7 . Inter-annual rainfall intensity also strongly depends on Atlantic and Indian Ocean sea-surface temperature (SST) variations associated with the Indian Ocean Dipole and the El Niño-Southern Oscillation 8 . Supplementary Note 2 for radiometric age determination and age-depth modeling. We have developed two independent age models for the composite core HSPDP-CHB14-2 ( Supplementary Fig. S1). For the K/Zr time series discussed here, we applied age model RR2021 9 that is based on multiple independent chronometric techniques using radiocarbon dating of ostracods, optically-stimulated luminescence (OSL) dating of quartz, single-crystal total-fusion (SCTF) 40 Ar/ 39 Ar dating of K-feldspars from tuffaceous zones, and geochemical correlation of the Konso Silver Tuff (Silver Vitric Tuff, SVT) to a visible tephra unit in the core 9,10 . Applying RR2021, the 5 mm spacing of the K/Zr resolution corresponds to a temporal resolution of ~10 years. The 30 ages generated by these chronometric techniques are stratigraphically consistent; Bayesian age-depth modeling incorporating the 14 C, OSL and 40 Ar/ 39 Ar ages, and tephrochronological data, has been used to build an age-depth model for the composite core 9 . The Bayesian age-depth model takes into account the potential for hiatuses and variations in sedimentation rates such as may naturally occur during wet-dry cycles, producing modelled ages with uncertainties throughout the sedimentary sequence. Whilst the mean of the Bayesian age model output effectively linearly interpolates between the sparser data-points in the earliest parts of the core, the uncertainties generated by the model do still cover possible hiatuses and variation in sedimentation rate, averaging across all such possibilities 9 . The 1-sigma absolute uncertainties produced, increase with increasing depth, ranging from <10 ka in the uppermost 50 m (corresponding to ~100 ka BP) and 10-35 ka below 50 m composite depth (~100-620 ka BP). An alternative age model MUBAWA2021 uses the multiband wavelet age modeling (MUBAWA) algorithm that tracks the earth's precession cycle in the wavelet power spectrum of the MSCL-based sediment-color reflectance values 11 . We first use a principal component analysis (PCA) to separate variations of blue-green colors (during wet episodes) to reddish-brown colors (during dry episodes) (stored in the 2nd principal component PC2) from the total reflectance (or brightness) of the sediment (PC1) in sediment-color values. The climatically-controlled color variations recorded in the PC2 show distinct cycles with wavelengths of 10-15 m and of ~40 m in the core, probably a result of the influence of the earth's precession and eccentricity cycles 11 . We then use an adaptive bandpass filter to extract the approximate spatial wavelength range from the sediment color data, which corresponds to the wavelength of our tuning target, the earth's precession cycle according to Laskar et al. (2004) 11,12 . Comparing RR2021 9 and MUBAWA2021 11 reveals that both independent age models agree well with each other, particularly in the deepest parts of the core where they also agree within uncertainties with evidence from direct dating provided by four 40 Ar/ 39 Ar ages (Suppl. Fig. S1). A closer look at the plot of the differences between the modeled ages, reveals that the maximum differences between the two age models accounts for one and a half precession cycles (Suppl. Figs. S1 and S2). These differences between the models are greatest for the upper portion of the core where the direct dating evidence that underpins the Bayesian model RR2021 is the most abundant (e.g. 26 of the 30 radiometric ages from multiple direct dating methods are located within the upper 75m of the core). Below 75 m, only four 40 Ar/ 39 Ar ages are available and the RR2021 model is less well constrained, while the MUBAWA2021 model is calibrated to the precession cycle and therefore provides more detail for this interval. Nevertheless, from ~150 m composite depth through to the base of the core, which spans more than 300 ka duration, the RR2021 and MUBAWA2021 models are in particularly close agreement (Suppl. Fig. S1).
To avoid the potential circularity of using an orbitally-tuned age model to investigate a potential proxy for climate, particularly when a robust age model based on multiple direct dating techniques exists, the Bayesian model RR2021 9 is used to provide the chronology for the K/Zr record presented in this study. Exploring the detailed differences between the XRF-based (K/Zr) record adjusted to the RR2021 9 and MUBAWA2021 11 age models (Suppl. Fig. S2 illustrates the previously noted offsets, but shows that the general long-term trends are in good agreement, and the differences between them lie within their uncertainties. Because of the uncertainties in our chronology, and especially because of greater uncertainties in the comparative archives, our interpretation and comparisons of transitions in this record with those in other climate archives is undertaken with great caution. Supplementary Note 3 for Forcing. Forcing and SST variations: A second modifier of hydroclimatic change in Chew Bahir is sea-surface temperature (SST) variations of the surrounding oceans with particularly the Indian Ocean as the main humidity source for eastern Africa 7,13,14,15 . In addition to the insolation effect on air mass transport controlled by the land-sea temperature gradients, higher SSTs can enhance temperaturedependent evaporation over the ocean and therewith strongly influence the amount of water in the troposphere. A weakening of low-level winds along the eastern coast of Africa and across the Arabian Sea during long-term insolation driven phases of warmer ocean temperatures, had dampening effects on upwelling, once again enhancing convergence over the western equatorial Indian Ocean 16 . Thus the CHB long-term wet-dry oscillations show generally a strong similarity to SST fluctuations recorded in ODP 722 17 , that reflect the intensity of coastal upwelling in the Arabian Sea, and thus the strength of the summer monsoon 18,19 , suggesting that amplified wet phases in southern Ethiopia might have been caused by enhanced moisture advection and a stronger summer monsoon in phase with orbital forcing (Suppl. Fig. S3).
Forcing and glacial boundary conditions: In addition to insolation forcing and SST changes of the neighboring Indian Ocean, atmospheric CO 2 variability has also been proposed as a driver of tropical climate change throughout the geological past 20 . Our comparison of the Antarctic EPICA CO 2 21 record with CHB moisture fluctuations suggests, however, that greenhouse gases only played a minor role for the moisture budget of CHB with possible exceptions during the eccentricity minimum phases between ~430-360 ka and ~120-0 ka, respectively. Here, the lowered insolation levels appear to coincide with distinct moisture phases at the transition into interglacial conditions (Suppl. Fig. S3). The forcing of large ice sheets however could have played a role for precipitation in eastern Africa during muted orbital forcing, when under glacial conditions and lowered greenhouse gas concentrations convection was generally reduced in the near-equatorial zones and westerly controls weakened monsoons and stormtracks 22,23 . This is in agreement with model simulations considering orbital, greenhouse gas, and ice sheet forcing of the moisture variability over northern Africa and the Arabian peninsula suggesting that during interglacial conditions the variations between orbitally forced wet/dry extremes has seen some of the wettest phases due to the increase of NH monsoons in addition to increased winter rains from the Mediterranean basin. On the other hand a decreased NH seasonality coincided with weakened monsoons and precipitation decrease from the Mediterranean storm tracks 22 . This is particularly well expressed at ~430 ka and the transition into MIS 11, and could potentially relate to the simultaneous Mid-Bruhnes Event (MBE). Although the origin of the MBE is not completely understood, it was associated with persistently enhanced interglacial CO 2 levels during the late mid-Pleistocene 24 . In turn, increased CO 2 levels might have resulted in enhanced annual and seasonal rainfall rates, and increased extreme wet period frequency as predicted for eastern African climate under current climate change.
Supplementary Note 4 on the Categorization of dimensions in cultural innovation. Important developments in cultural and technological innovation have been proposed to be bundled in eight grades that describe the expansion of cultural capacity (EECC model) by Haidle et al. (2015) 25 (Fig. 3). Each onset of these phases is represented by "minimum ages", i.e. the oldest certain evidence. Stage 6, from <200 ka on, encompasses the development of critical cognitive abilities that enable humans to plan and execute technological augmentations such as composite tools as found in hafted tools and compound adhesives [25][26][27][28][29] . It has been argued that this key set of innovation also represents an important expansion of cultural capacity 27 . The onset of Stage 7, from 64 ka on, is marked by a notable advance in complementary cultural capacity, that entails the development and use of a set of cultural modules as an acting entity with two or more interdependent and exchangeable parts. An example would be the emergence of bow-and-arrow, needle-and-thread, screw-and screwdriver, key-and-lock 26,30 . Earliest evidence for the emergence of a notional cultural capacity, Stage 8, appears around ~40 ka 31 . The development of a notional cultural capacity contains that notional concepts are mentally constructed and socially shared entities and relationships that can be represented in a) the signification of objects/signs (e.g. cross, crescent, and Star of David as symbols of religions), b) systems of ideas (e.g. myths, religious beliefs, philosophical questions, constitutions of states), c) normative definitions (e.g. metric and value systems), or d) virtual beings (e.g. angels) and characters (e.g. protecting capacities of an amulet) 25 .
The transport of raw material over longer distances can be used as a key indicator for the formation of larger social networks. As early as ~320 ka ago, obsidian has been shown to be transported and exchanged in Olorgesailie (Kenya) over a distance of ~50 km, whereas evidence from Baringo (Kenya) ~200 ka ago indicates already a much larger network radius of 166 km [31][32][33] .