Understanding changes in eastern African hydroclimate during the Pleistocene is central to investigations of how humans evolved in a variable environment1,2,3,4,5,6,7,8. Over the Pleistocene, eastern African rainfall is thought to have undergone both secular and periodic changes driven by global cooling, evolving tropical sea surface temperature (SST) gradients, low-latitude insolation forcing, and glacial–interglacial cycles3,9,10,11,12,13,14,15,16,17. Each of these forcings has specific implications for the nature and timing of eastern African rainfall changes, which in turn yield predictions for the environmental changes experienced by our hominin ancestors. However, a lack of long datasets capable of resolving orbital cycles (103–105 years) limits our understanding of the relative influences of global climate forcings on the Pleistocene evolution of tropical eastern African rainfall, as well as the effects of paleoenvironmental change on early humans.

Varying seasonal insolation, controlled by the Earth’s orbital precession and eccentricity, causes changes in the differential heating of the African continent and oceans, driving fluctuations in the East African Monsoon strength18,19. 21-kyr cycles in monsoonal rainfall that result from this process are well-documented in eastern African climate records9,11,20,21,22,23,24,25,26, and their varying amplitude has been argued to have played a pivotal role in human evolution6,7,27. Coupled changes in the Earth’s carbon cycle and atmospheric greenhouse gas concentrations, global temperatures, and high-latitude glacial–interglacial cycles are also thought to play a critical role in eastern African climate evolution3,4,28, and long-term variations in these processes may have contributed to the development of bipedalism and other traits29. For instance, soil carbonate isotope (δ18Osc) records indicate gradual drying in northern and tropical Africa30,31, attributed to global cooling and ice-volume growth through the Pleistocene. Records of dust from the eastern Atlantic and the Mediterranean and Arabian Seas suggest transitions from 21- to 41- to 100-kyr periodicity over the Plio-Pleistocene, with shifts toward drier conditions and increased variability starting between 3500 and 2500 ka (onset and gradual intensification of Northern Hemisphere glaciation) and at 1000 ka3,4 (mid-Pleistocene Transition, MPT), matching transitions in the marine oxygen isotopic record of global ice volume32. However, recent accumulation rate corrections33 and time series analyses12 suggest different timings of aridification and a stronger influence of low-latitude insolation. Furthermore, strengthening of zonal SST gradients in the tropical Pacific beginning at ~1700 ka34 is thought to have weakened convection over eastern Africa, contributing to regional drying16. To date, despite the paleoanthropological significance of eastern Africa, the relative importance of low- and high-latitude climate forcings on the region’s rainfall history remain poorly constrained.

The Hominin Sites and Paleolakes Drilling Project (HSPDP) recovered sediment drill-cores that record the environmental history of key hominin fossil locales in Ethiopia and Kenya35,36,37. The cores allow us to develop and compare multiple long, high-resolution records of regional hydroclimate within a set of key time windows to elucidate the forcings and mechanisms of climate change in the region. Here we present a new record of the hydrogen isotopic composition of precipitation (δDprecip) from compound-specific analyses of terrestrial leaf waxes—a novel and powerful proxy for processes related to rainfall38—preserved in middle to late Pleistocene sediments from the Chew Bahir Basin, Ethiopia. This is compared with an existing record of the early Pleistocene from the adjacent Omo-Turkana Basin24 to evaluate changing trends and rhythms in regional hydroclimate, as well as the relative influences of high- and low-latitude forcings during intervals of the early and middle to late Pleistocene.

The HSPDP core locations lie in the East African Rift System (Fig. 1a), host to many famous hominin fossil sites39,40,41. We generated a new hydroclimate record derived from the hydrogen isotopic composition of terrestrial leaf waxes (δDwax) preserved in paleolake deposits from Chew Bahir, southern Ethiopia (duplicate drill cores HSPDP-CHB14-2A and -2B merged to composite core42,43, hereafter CHB14-2). Coring site CHB14-2 (4° 45′ 40″ N, 36° 46′ 00″ E) is located in the Chew Bahir Basin, just northeast of the Omo-Turkana Basin (Fig. 1b). Today, the southern part of the basin floor is mostly occupied by a saline mudflat. The composite core extends from ~620 ka to present with age constraints based on 40Ar/39Ar dating of tephra, optically stimulated luminescence (OSL), radiocarbon dating, and tephrostratigraphic correlations44. We analyzed waxes spanning the interval from ~250 ka to present-day and synthesized this new dataset with a published record from West Turkana, Kenya24,45 (1900–1400 ka; HSPDP-WTK13-1A, hereafter WTK13) located ~100 km from CHB14-2 (4° 6′ 35″ N, 35° 52′ 18″ E). The age model for WTK13 is based on tephrochronology and magnetostratigraphy and includes very conservative tuning of δDprecip with no impact on the dominance of orbital precession in the spectral properties24 (Fig. S1). Our combined datasets provide a regional hydroclimate record that represents a total span of ~750 kyr during the period 1900 ka to present, with an average sampling resolution of ~3 kyr within each record (Fig. 2).

Figure 1
figure 1

(a) East African Rift System study area map, including HSPDP sites and major rift lakes, generated in Python 3.8; (b) Ethiopian and Kenyan locations of the two paleolake sediment drill cores, WTK13 and CHB14-2, included in this study with Omo-Kibish and Nariokotome Boy hominin sites and the Kokiselei site of the first evidence for Acheulean hand axes48. Map generated in Google Earth Pro 7.3.3.

Figure 2
figure 2

δDprecip records corrected for vegetation, ice volume, and geographic effects (Fig. S4) from CHB14-2 and WTK13 in the context of two million years of zonal mean 20° N June 21st insolation65 (red) and the benthic foraminifera δ18O stack32 (blue). Sampling gaps greater than half of a precession cycle (~10 kyr) are represented with dashed lines and analytical error on δDwax measurements in shading. Age constraints for CHB14-2 and WTK1324,45 with 1σ analytical error depicted along bottom with symbol indicating dating technique (green triangle = 14C; black circle = OSL; red star = 40Ar/39Ar; blue square = magnetostratigraphy).

The combined WTK13 and CHB14-2 data record key intervals when our genus, Homo, was evolving, developing new technologies, and dispersing within and out of Africa46. The Omo-Turkana Basin contains over 100 archaeological sites and 500 fossil finds47, including the earliest and most complete skeletons of H. rudolfensis and H. erectus. The ~1900–1400 ka interval spanned by WTK13 witnessed the development of Acheulean stone tools (earliest evidence for advanced hand axes at ~1760 ka at Kokiselei48, Fig. 1b), the evolution of H. erectus (including the Nariokotome Boy skeleton at ~1600 ka49, Fig. 1b), and what is thought to be the earliest hominin dispersal out of Africa50. The first eastern African evidence of our species, H. sapiens, is dated to ~233 ka at Omo Kibish in the Omo-Turkana Basin51, 100 km northwest of Chew Bahir (Fig. 1b). The past ~250 kyr, recorded in CHB14-2, not only encapsulates human morphological changes, but also social, technological, linguistic, and cultural development, and the dispersal of modern H. sapiens out of Africa42,43,52. These new traits spread to the rest of the world during this interval, and thus, this Turkana-Chew Bahir region may have served as a critical landscape for the development of our ancestors over the Pleistocene. This study, situated within the broader context of the aims of HSPDP (Fig. 1a), provides crucial insight into the nature of environmental change and the potential effects on hominins and other large mammals on the landscape.

Many paleoenvironmental indicators are very sensitive to basin-scale geological processes, limiting the ability for inter-basin comparison. However, δDwax is primarily controlled by δDprecip53, which, in tropical Africa, is dominantly driven by regional atmospheric dynamics that govern rainfall amount54,55. A variety of observational55,56, modeling57, and paleoclimate14,24,58,59,60 studies have revealed δDprecip to be very sensitive to changes in eastern African paleohydrology on orbital timescales. Although we recognize that δDprecip can be influenced by a variety of other processes such as moisture source and transport, and a variety of convective processes including the location of convective cells61, we interpret δDprecip as a qualitative indicator of rainfall amount, consistent with previous studies in the region14,24,54,59,61,62. We directly compare δDprecip between different sedimentary archive sites and time intervals to understand large-scale climate processes.

C3 and C4 metabolic processes influence the apparent fractionation between δDwax and δDprecip, but carbon isotopic compositions of the same leaf wax compounds (δ13Cwax; Fig. S2) help estimate vegetation type and correct δDwax to δDprecip (Fig. S3 and S4). While uncertainties exist in the biosynthetic fractionation factor, this correction has minimal influence on the trends and patterns in the precipitation record because the isotopic range in δDprecip is vastly larger than the potential C3–C4 effect. We also correct for geographic differences in δDprecip between WTK13 and CHB14-2 using δDwax and δ13Cwax measurements from late Holocene sediment within each basin to estimate regional δDprecip (Fig. S4). We conduct a series of time series analyses to detect changes in the trends and rhythms of δDprecip and eastern African climate variability.


Leaf wax biomarker record

The hydrogen isotopic composition of long-chain leaf waxes (n-C26, n-C28, and n-C30 alkanoic acids) are strongly correlated in CHB14-2 (C28–C26: r2 = 0.72, n = 100, p << 0.01; C28–C30: r2 = 0.9, n = 117, p << 0.01) demonstrating these compounds were derived from a common source and record similar climate processes. Despite previous work that found that n-C28 may be produced in the lake water column in some lakes63, the strong correlation between long-chain compounds indicates that n-C28 is representative of terrestrial land plants in this basin. As n-C28 is the most abundant long chain n-acid, determined by Average Chain Length (ACL) calculation (28.4), resulting in lower analytical error, we use the hydrogen isotopic ratio of C28 n-acid for all analyses of climate variability for both sites. The Carbon Preference Index (CPI) is a measurement of degradation of the organic compounds in the sediment, where a high even:odd chain length signifies good preservation of alkanoic acids, and a ratio of 1 signifies full degradation64. The CPI in CHB14-2 is acceptable (mean: 2.8; minimum: 1.5), and to further demonstrate the lack of degradation effect on isotope analyses, we compare CPI and δDwax to find an insignificant correlation (r2 = 0.002, n = 125, p > 0.05). In CHB14-2, δDwax ranges from − 164.6 to − 68.7‰.

δ13Cwax averages − 23.8‰ in CHB14-2, and ranges from − 19.9 to − 30.8‰ with one outlier at − 16.8‰ (Fig. S2). The corrected δDprecip record, based on the δ13Cwax data, ranges from − 68.9 to 36.2‰ and closely tracks δDwax (Fig. S3 and S4).

Trend, variability, and spectral properties

Neither of the δDprecip records show significant linear trends towards wetter or drier conditions within the time intervals they span individually or together, nor is there a large difference between the WTK13 and CHB14-2 study intervals (< 2‰ offset in δDprecip; Fig. 2).

Our δDprecip records contain high-amplitude oscillations of up to ~100‰. Lomb-Scargle periodogram analysis demonstrates spectral density at ~21 kyr in the early and middle Pleistocene intervals (1900–1500 ka and 250–130 ka) but no significant spectral properties in the late Pleistocene within the bounds of robust frequency detection (Fig. 3). Gaussian 21-kyr band-pass filtering of δDprecip in the two study intervals supports the spectral analysis findings of strong precession influence in the early and middle Pleistocene, and reveals that this precession-band variation is greatly diminished in the late Pleistocene (Fig. 4). After applying a notch filter to remove variability associated with the ~21 kyr band, we observe gradual D-enrichment from Marine Isotope Stage (MIS) 5 (~125 ka) until the beginning of MIS 2 (~30 ka). This trend coincides with increasing benthic foraminiferal δ18O, suggesting that shifts in the late Pleistocene δDprecip covary with glacial–interglacial cycles (Fig. 4c,d).

Figure 3
figure 3

Lomb-Scargle spectral analyses for unevenly sampled data of δDprecip from the early (1900–1500 ka), middle (250–130 ka), and late (130–0 ka) Pleistocene. Precession-band 19- and 23-kyr periodicities lie above the 95% confidence line (dashed grey) in the early and middle Pleistocene. Frequency distribution is plotted from ½× the Nyquist frequency as the high-frequency cutoff to 1/3 of the total length of interval as the low-frequency cutoff, thus the differing x-axes of the three windows depend on the resolution and length of the specific interval.

Figure 4
figure 4

Gaussian 21-kyr ± 5-kyr band-pass (b) and notch (c) filtering of the δDprecip study intervals truncated to 1870–1500 ka and 250–30 ka to omit low sampling resolution sections. June 21st zonal mean 20° N insolation65 (a) plotted and highlighted in light red demonstrate similarity with high- and low-amplitude variability packets in gaussian band-pass filtered δDprecip. Our selection of June 21st insolation at 20°N is based on observations from latest Pleistocene and Holocene records demonstrating the sensitivity of eastern African precipitation to this date and latitude14,61,84,88. We note that the chronologies for the CHB14-2 and WTK13 records are too imprecise to determine the phase of the response of δDprecip to orbital forcing; however, the choice of latitude and season does not influence our spectral analyses nor other results. Benthic foraminifera δ18O stack32 (d) plotted with recent interglacial MIS’s and highlighted in light blue to demonstrate similarily with late Pleistocene notch-filtered (precession-band periodicities removed) δDprecip. Means were removed in both band-pass- and notch-filtered data to feature changes in variability.


Our δDprecip records indicate eastern African rainfall experienced high-amplitude, orbitally-driven wet/dry cycles during long intervals of the early, middle, and late Pleistocene. Variability in the early Pleistocene 1900–1400 ka and middle Pleistocene (230–150 ka) intervals is dominated by orbital precession, with strong 21-kyr cycles in δDprecip (Figs. 3, 4), as well as 100-kyr eccentricity-band amplitude modulation (Fig. S5). Ice volume and associated global climate processes varied primarily at the 41-kyr period during the early Pleistocene and had a saw-tooth pattern and 100-kyr periodicity in the middle Pleistocene32, yet we see no robust signal of obliquity in the early Pleistocene (Fig. 3) nor visual similarity between δDprecip and ice volume through most of the record (Fig. 2). Instead, eastern African rainfall varied primarily at a 21-kyr precession rhythm (Fig. 3) with modulation of that variability by eccentricity into high- and low-amplitude packets (Fig. S5), in sync with low-latitude summer insolation forcing65 during the early to middle Pleistocene.

We observe no difference in mean values of δDprecip between the WTK13 and CHB14-2 records, suggesting remarkable long-term stability in eastern African rainfall during the Pleistocene. The similar lack of trend in the eastern Africa soil carbonate δ18O compilation31 suggests that the Omo-Turkana and Chew Bahir Basins, despite their aridity relative to surrounding basins, capture regional paleoclimate changes, especially because of the large-scale integrative nature of the leaf wax biomarker proxy. The long-term hydroclimate stability occurs despite evidence for regional C4 grassland expansion66,67,68,69,70,71,72, supporting recent work suggesting that declining atmospheric CO2, rather than hydroclimate, plays a dominant role in C4 grass expansion in Africa73,74.

Orbital-scale vegetation change, though, covaries with hydroclimate variations in intervals throughout the Quaternary24,59,75, and we observe substantial changes in the amplitude of orbital-scale variability within each of our records. Band-pass filtering of the precession signal in our δDprecip records isolates packets of high-amplitude variability that generally align with high orbital eccentricity and intervals with the strongest seasonal insolation forcing (Fig. 4 and S5). Although not every high eccentricity interval produces high-amplitude δDprecip oscillation (i.e., 1900–1800 ka), this result further suggests a dominant role for precession-driven seasonal insolation change in controlling eastern African rainfall during the early and middle Pleistocene.

Our findings are supported by records that indicate a dominant role for orbital precession in controlling African climate history, particularly in subtropical and northern Africa9,20,22,23,24,25,26,76. For instance, sapropel records from the Mediterranean indicate precessional insolation forcing has been a dominant driver of northeast African rainfall throughout the Plio-Pleistocene9. Our results are also consistent with some paleoclimate model simulations17, though others predict a stronger role for atmospheric greenhouse gases in eastern equatorial Africa76 than suggested by our records. Synchronized pulses of deep lakes in multiple East African Rift basins have been suggested to occur during intervals of high eccentricity5. Our δDprecip records indicate that high eccentricity intervals were times of much wetter, as well as much drier, conditions (Fig. S5), and the alternation between extreme endmembers suggested by our data could drive selection for generalist or adaptable traits in early humans27,77,78.

Despite the dominant role of orbital precession in our records, our δDprecip suggests global climate conditions became increasingly influential on tropical African rainfall between the middle and late Pleistocene after the last interglacial at ~130 ka. After removing precessional periodicity from our data, we observe a trend toward drier conditions from MIS 5e (when ice volume levels were similar to the Pliocene32) until the Last Glacial Maximum (LGM; Fig. 4). Previous work has documented strong influences of ice volume on eastern African climate during the latest Pleistocene, such as drying over most of the region during the LGM43,79. A ~210 kyr-long δDwax record from the Gulf of Aden also documents strong precession-band rainfall variations during MIS 5, 6, and 7 superimposed on alternating humid and arid conditions that track ice volume14. This mixture of signals of insolation and ice volume in the Gulf of Aden potentially results from its more northern location or the larger area of leaf wax supply to this marine record. However, a dust record from the Mediterranean, which is thought to record Northeast African monsoon strength, also demonstrates precession-band fluctuations throughout the last 3000 kyr until a large, 100-kyr, sawtooth-shaped excursion begins in MIS 5e12,76.

Climate model simulations suggest strong atmospheric teleconnections between eastern African rainfall and the northern high latitudes80. One potential mechanism for the influence of late Pleistocene glacial–interglacial cycling in tropical Africa could be that cooling in the northern high latitudes is advected by the westerlies into Eurasia, which enhances the boreal winter Arabian anticyclone80. Northerly winds originating from this circulation advect cool and dry air over eastern Africa, suppressing boreal fall and winter rainfall. These simulations rely on freshwater hosing to cool the northern high latitudes and are therefore not directly analogous to the Northern Hemisphere glaciation cycles. However, these simulations demonstrate an atmospheric mechanism linking eastern African rainfall and northern high latitude climate via Eurasia that could apply on longer timescales.

Our δDprecip data suggest that low-latitude insolation forcing controls much of the long-term variability in eastern African rainfall, including during the middle Pleistocene when ice volume changes were large. However, ice volume fluctuations leave distinct signals from 130 ka to the present (Fig. 4) and there is also a stark lack of similarity between δDprecip and precession (Fig. 3) and eccentricity modulation (Fig. S5) during this time. We suggest that this arises due in part to the relative strengths of high- and low-latitude forcings. High-amplitude seasonal insolation forcing under high orbital eccentricity causes strong, periodic changes in eastern African rainfall17. However, when ice volume fluctuations strengthen and insolation forcing weakens, such as occurred from ~130 ka to the present, ice volume changes can emerge as a strong influence on eastern African hydroclimate. The shift from insolation-driven to ice volume-driven fluctuation at ~130 ka in our record suggests a nonlinear sensitivity of eastern African rainfall to seasonal insolation forcing and to high-latitude-driven climate change at this orbital time scale. This varying sensitivity to forcings of variable amplitude may reconcile the large number of records that document eastern African aridity during the LGM43,58,79,81,82,83,84,85,86, when ice volume changes were large and eccentricity was particularly low, against the longer Pleistocene records that show a dominant control of orbital precession on eastern African rainfall. This hypothesis may further explain the absence of 41-kyr cycles in African rainfall during the early Pleistocene, as ice volume changes were generally small compared to those during the late Pleistocene. Climate modeling experiments have suggested threshold responses of tropical climate to Northern Hemisphere ice volume changes, due to shifts in the position of westerly jets and their ability to perturb the tropical atmospheric circulation87. Additionally, threshold-like responses of African hydroclimate to insolation have been documented61,87 and attributed to various processes, including feedbacks involving vegetation, soil moisture, and SST43,58,88. The interaction of these nonlinear responses to high- and low-latitude climate drivers may have triggered shifts in sensitivity, depending on the relative strengths of each forcing.

Both orbital-scale variability and secular trends in eastern African climate have been postulated as drivers of hominin evolution and dispersal3,7,42,43,77,89. Our proxy records indicate that orbital-scale variability (up to 100‰ in a single precession cycle) is much larger than the long-term mean change occurring since ~2000 ka. Extremely high-amplitude fluctuations occurred in the region during critical times of early hominin evolution in eastern Africa and potentially promoted an environment that favored behavioral and morphological plasticity or adaptability in our ancestors8,27.


Geochemical analyses

We analyzed the isotopic composition of terrestrial leaf wax biomarkers preserved in sediment from composite core HSPDP-CHB14-2 (hereafter termed CHB14-242) archived at the National Lacustrine Core Repository. Plants produce epicuticular waxes to shield leaf surfaces from evaporation and physical damage90. These waxes may be ablated and transported by eolian and fluvial processes to lakes, where they are preserved in sediment over geological time. The waxes include long-chain n-alkanoic acids, which we use to reconstruct water isotope compositions. Lipid extraction, purification, and isotopic analytical procedures91 were performed at Brown University. Lipids were extracted from freeze-dried and homogenized sediment using a DIONEX Accelerated Solvent Extractor 350 with dichloromethane:methanol (9:1). The total lipid extract was separated into neutral and acid fractions via aminopropylsilyl gel column with dichloromethane:isopropanol (2:1) and ether:acetic acid (24:1). The acid fraction was then methylated using acidified methanol, and the resulting fatty acid methyl esters (FAMEs) were purified using a silica gel column. Relative concentrations of the FAME chain lengths were quantified using an Agilent 6890 gas chromatograph (GC) equipped with a HP1-MS column (30 m × 0.25 mm × 0.25 µm) and flame ionization detector (FID).

Hydrogen isotopes (δDwax) were measured using an Agilent 6890 GC, equipped with HP1-MS column (30 m × 0.32 mm × 0.25 µm), coupled to a Thermo Delta Plus XL isotope ratio mass spectrometer (IRMS) with a reactor temperature of 1445 °C, although some of the samples from the CHB14-2 core were analyzed with a Thermo Delta V Plus IRMS using the same conditions. On both instruments, D/H ratios were measured in triplicate using H2 as an internal standard with He as the carrier gas, and corrected using a known FAMEs lab standard. Carbon isotopes (δ13Cwax) from CHB14-2 and the late Holocene analogues were measured at Brown University with these same procedures on the Thermo Delta V Plus GC-IRMS with a reactor temperature of 1100 °C. Isotope ratios were corrected for the added methyl group (δDMeOH =  − 123.7‰ and δ13CMeOH =  − 36.62‰). We report δDwax relative to Vienna Standard Mean Ocean Water (VSMOW) and δ13Cwax relative to Pee Dee Belemnite (PDB) in per mil (‰) notation.

We successfully analyzed 125 samples (out of 143 samples) for δDwax and 92 samples for δ13Cwax from the CHB14-2 composite core. The sediment samples integrate up to 4 cm (~80 years44) and have a mean temporal resolution of ~1.75 kyr since 250 ka. Hydrogen isotopic analyses of the FAMEs standard had a standard deviation (1σ) of 3.2‰ and the H3 factor was 1.76 ppm/nA. For hydrogen, 56 samples were run in triplicate (average 1σ = 1.5), 20 in duplicate (average difference = 2.3‰), and 49 as single injections due to limited concentration. For carbon, all samples were measured in duplicate, with an average FAMEs standard 1σ of 0.25 and average intra-sample difference of 0.14‰. Five samples were removed from further analysis because they lie between two ages that constrain a potential sedimentary hiatus or dramatic reduction in sediment accumulation rate around the LGM from ~30–10.5 ka42,43,44.

Isotopic corrections

A series of corrections to δDwax were performed to convert values to δDprecip (Fig. S4). Once all corrections were made, one outlier (outside 3 standard deviation units) was removed from the WTK13 record.

Vegetation correction

C3 trees and C4 grasses fractionate hydrogen to different degrees during leaf wax synthesis due to differing metabolic pathways and plant physiologies. This causes different apparent fractionations between leaf waxes and precipitation (εwax-P), which can affect paleoclimate records based on δDwax if vegetation changes38. We calculated a ‘vegetation correction’ based upon δ13Cwax values (Fig. S2) to correct δDwax for these differences91. We use δ13Cwax endmember values for C3 and C4 plant types previously described from a Omo-Turkana Basin outcrop92, in which the δ13C of n-C30 acids is − 32.9‰ for the C3 endmember and the δ13C of n-C30 acid is − 19.0‰ for the C4 end member. We adjust these values to account for observed differences between n-C30 and n-C28 acids93, thereby using − 32.15‰ and − 20.63‰ as the C3 and C4 endmembers. Samples with δ13Cwax values more enriched than this C4 endmember value were treated as 100% C4. After applying this C3/C4 mixing model to our δ13Cwax data, we then applied εwax-P values of − 112.8‰ and − 124.5‰ for C3 and C4 vegetation with a 25‰ correction for C27 n-alkane to C28 n-acid38,91,94 to correct for ‘vegetation effects’ on δDwax and estimate δDprecip (Fig. S3).

Because not all δDwax measurements have a corresponding δ13Cwax measurement, typically due to concentration limitations, we used AnalySeries95 to mathematically resample the δ13Cwax data to δDwax resolution to obtain a δDprecip record with the same resolution as δDwax. In Fig. S2 we demonstrate that this does not have a meaningful impact on our results as the corrections are much smaller than the hydroclimate signals in δDwax and δDprecip. We show the CHB14-2 δDprecip record with and without the additional resampled δ13Cwax corrections to demonstrate that the difference between the δDwax and the empirically derived δDprecip is negligible.

Ice volume correction

We use the benthic δ18O stack32 to estimate past ocean water isotopes to correct the δDprecip for different source water δD91. Age uncertainty in our records and in the LR04 stack limits our ability to precisely align the two, so we average the stack δ18O in each study interval, anomalize that value to late Holocene, and convert it to δD based on the meteoric water line. We then apply this anomaly to each study interval to obtain an ice volume-corrected signal of δDprecip (Fig. S4).

Geographic correction

δDwax and δ13Cwax measurements of late Holocene analogue sediment (Table S1) lets us obtain δDprecip measurements from both sites. One sample from the Chew Bahir Basin96 and 12 averaged samples from the Omo-Turkana Basin85 were used to represent the late Holocene (last 5 kyr) leaf wax isotope signature of each region (Table S1). Our late Holocene analogue measurements of δDprecip are similar to modeled precipitation isotope data97, indicating that we have appropriately captured the differences between study sites. We anomalized the Chew Bahir measurements to Turkana δDprecip. This “geographic” correction (12‰) was then added to the mean of the CHB14-2 record (Fig. S4) to produce the fully corrected eastern African δDprecip Pleistocene record (Fig. 2).

Time series analyses

We analyzed the linear trends within the WTK13 and CHB14-2 records, as well as throughout the entire 1900 kyr interval. Comparisons between δDprecip and insolation were performed using June 21st insolation at 20° N, which is based on observations from late Pleistocene and Holocene records demonstrating the sensitivity of eastern African precipitation to this date and latitude 14,60,83,87. We also performed Lomb-Scargle analysis of δDprecip to study spectral density of unevenly spaced data with the plomb function in MATLAB98,99. This method was applied to the two study intervals, 1900–1500 ka and 250–30 ka, which exclude low-resolution intervals. We then used the frequency of the densest spectral peak from each interval (early Pleistocene, 22 kyr; middle to late Pleistocene, 25 kyr; each with bandwidth of ± 5 kyr) to inform gaussian band-pass and notch filtering exercises, which were performed using the time series analysis program AnalySeries version 2.0.895.