Since Leakey and colleagues1 (pg. 9) suggested Paranthropus boisei was the “victim” of Homo habilis at Olduvai (Oldupai) Gorge (Tanzania), our ancestors have been implicated in the demise of their sister taxon. Tool-assisted foraging behaviors were traditionally thought to have propelled genus Homo into a broad omnivorous dietary niche, providing an evolutionary edge relative to Paranthropus’ dentognathic procurement strategies across Africa’s savanna habitats2 and resulting in competitive exclusion3. More recent archaeological evidence suggests that both Homo and Paranthropus were plausible inheritors of tool-making behaviors4,5, and stable carbon isotopic (δ13Cenamel) and microwear analyses of fossil specimens have revealed a dynamic and complex evolutionary history of Pleistocene African hominin and non-hominin primate diets (Fig. 1)6,7,8,9,10,11,12,13. Eastern African hominins and Theropithecus, the large-bodied baboon, underwent a dietary transition incorporating more C4 foods in the early Pleistocene6,8,9,12, which have been attributed to behavioral changes in response to complex competitive landscapes12. Distinct from other Pleistocene hominins, P. boisei, the last eastern African paranthropine species, yielded δ13Cenamel and δ44/42Caenamel values indicating a herbivorous13 and primarily C4 diet6 from ~ 2.3 Ma until its last appearance from the fossil record at ~ 1.3 Ma14. Unexpectedly, isotopic evidence for this shift into a C4-plant feeding niche was not mirrored by changes in microwear patterns or dentognathic morphology7,10,15. P. boisei’s diet had comparable mechanical properties to those of C3-C4-mixed feeding hominins such as Australopithecus afarensis7. P. boisei’s use of fallback foods, rather than its dietary staples of C4 plants, may have been the main influence for its distinctive “nut-cracking” form7,10.

Figure 1
figure 1

(A) δ13Cenamel values of EARS hominins and Theropithecus oswaldi (B) duration generalized taxonomic groupings of EARS fossil hominins, (C) δ44/42Caenamel values of EARS hominin and T. oswaldi. See Supplementary Data and Supplementary Information for specimen information, isotopic values, fossil image credits, and references. Shaded area denotes MPT interval (1.3–0.7 Ma).

Wood and Patterson (2020)16 suggested that P. boisei’s 1-myr morphological and dietary stasis signifies that its occupied niche and associated adaptations were “remarkably durable.” Although eastern African hominins may have begun their evolutionary trajectories as C3-C4-mixed-feeding opportunists during the early Pleistocene11, P. boisei evolved to be a C4 plant specialist13 sharing more dietary similarities to T. oswaldi than with members of genus Homo, who remained C3-C4-mixed feeding omnivores (Fig. 1A–C). Thus evidence for dietary niche separation and shared stone tool making abilities cast doubt on competitive exclusion by Homo as a primary cause of P. boisei’s demise.

A prevailing explanation for P. boisei’s extinction is the opposite cause of Homo’s success, that is, its dietary and behavioral inflexibility amidst environmental perturbations17. P. boisei’s extinction, estimated between 1.3 Ma14 and just prior to 1 Ma18, occurs during one of Homo’s significant increases in brain size and its second wave of dispersal out of Africa and into Eurasia19,20. These evolutionary events in the hominin lineage are coincident with the Mid-Pleistocene Transition (MPT; 1.3–0.7 Ma)21, when the Earth’s glacial and interglacial climatic intervals transitioned from an obliquity-paced (41-kyr) cyclicity to an asymmetrical ~ 100 kyr eccentricity pattern of repeated saw-toothed glacial growth and rapid deglaciation22. Global climatic and environmental changes are commonly evoked as major forces of eastern African hominin evolution23. Dietary and technological adaptations in the Plio-Pleistocene have been contextualized in open24 and variable25 landscapes, shaped by the presence of C4 plant communities24. During the MPT interval, current environmental proxy records from the East African Rift System (EARS) show evidence for low percentages of woody cover24 and high abundances of C4-feeders26,27,28,29. This scenario begs the question: why would P. boisei, this “durable” herbivorous C4-feeding hominin, disappear during the dominance of C4 grasslands?

Here we report pedogenic carbonate stable carbon (δ13CPC) and oxygen (δ18OPC) isotopic values (nodules = 53, paired analyses = 95) in the Turkana Basin, northern Kenya, eastern Africa (Fig. 2A–C), from 1.4 to 0.7 Ma, an under-sampled interval for pedogenic carbonates in an otherwise well-studied region of human evolutionary environments (see "Methods", "Supplementary Information"). The Turkana region is the most fossiliferous of the known P. boisei sites18, and Turkana specimens constitute the majority of isotopically analyzed P. boisei enamel samples to date6,8,12,13. We combine our new δ13CPC data to those from other EARS locations preserving the MPT interval, calculate fraction of woody canopy cover (ƒWC)24, and compare the EARS ƒWC record to other environmental, ecological and climatic proxy records from 3 to 0 Ma to examine the contexts of P. boisei’s extinction.

Figure 2
figure 2

(A) Map of Africa with box outline showing (B) map of East African Rift System (EARS) with box outline showing (C) Turkana Basin (base map: Google Maps /TerraMetrics 2021). P. boisei fossil sites18 are denoted with a circle. Sampling locations for EARS ƒWC data that also preserved P. boisei specimens are marked with a circle embedded in a triangle. EARS ƒWC data from sites that did not yield P. boisei fossils are denoted with a triangle. Squares in inset map mark the locations of new δ13CPC and δ18OPC data presented in this study.


In contrast to characterizations of the Turkana Basin during the MPT as extremely grassy and dry23,24,28,29, our new data (n = 53, 95 paired analyses) show excursions to lower δ13CPC and δ18OPC values, consistent with relatively woodier vegetation structures and more humid and/or warmer conditions (Supplementary Fig. S5). Exponentially smoothed ƒWC and δ18OPC values from the Turkana Basin show temporally corresponding peaks of the excursions at 1 Ma (Supplementary Fig. S6). Exponential (Fig. 3A) and Loess (Fig. 3B) smoothing of the compiled EARS ƒWC record from 3 to 0 Ma illustrate the long-term increase in C4 vegetation punctuated by a short-term increase in C3 vegetation beginning at the start of the MPT interval and peaking at 1 Ma. The interpreted pattern from the EARS ƒWC record is substantiated with Bayesian change point analysis detecting two significant changes, which appear to mark the MPT interval (Fig. 3C).

Figure 3
figure 3

EARS ƒWC data fitted with (A) simple exponential smoothed curves (α = 0.1, 0.3, 0.6), (B) loess regressions (3%, 10%, 20%) and (C) Bayesian change point algorithm of a 5-point moving average; posterior probabilities on secondary axis (red line). See Supplementary Data for site locations, δ13CPC values, and references. Shaded area denotes MPT interval (1.3–0.7 Ma).

African basins have idiosyncratic variables (e.g., elevation, topography, temperature, water deficits, tectonism) that influence the distribution of vegetation and local climatic conditions23,30. Individual EARS basins during the time of P. boisei’s evolutionary history show local-scale heterogeneity in vegetation structures (Supplementary Fig. S7). Some EARS sites have low δ13CPC sampling resolutions or may be oversampled in specific time horizons. The Awash (Ethiopia) and Turkana (Kenya) basins have the highest δ13CPC sampling resolutions across the MPT interval (1.3–0.7 Ma)21 with 27 and 53 samples, respectively, and both record significant reductions in C4 vegetation (Supplementary Fig. S7). Olduvai (Oldupai) Gorge and Tugen Hills, yielding small sample sizes from the MPT interval, i.e., 12 and 3, respectively, show persistent grassy vegetation structures (Supplementary Fig. S7). Differences in vegetation structures between the northern and central EARS may be an artifact of low sampling density in the central EARS or true spatial differences within the EARS.


Causes of the EARS C3 excursion

Vegetation structures are dependent on mean annual precipitation (MAP)31, and increasing abundances of C3 vegetation would be predicted with higher rainfall during the MPT. Several mechanisms are proposed to have altered water delivery to Plio-Pleistocene Africa, including eccentricity-modulated precession, glacial forcing of the Inter-Tropical Convergence Zone (ITCZ) position, and tropical sea-surface temperature fluctuations among others33,34,35,36,37,38, yet regional proxy records yield dissimilar evidence for hydroclimate (Fig. 4A–E). Dust flux data from marine cores at sites 721/722 in the Arabian Sea indicate several intervals of increased aridity, including one circa 1 Ma (Fig. 4A)35. In contrast, deep lakes in the EARS are thought to have formed during periods of higher rainfall and forced by 405-kyr eccentricity (insolation) maxima (Fig. 4B)36. During the MPT, Lake Silbo was present in the Turkana Basin39,40, possibly indicating higher rainfall.

Figure 4
figure 4

(A) Arabian Sea dust flux (%) from core sites 721/722 with proposed periods of increased aridity (orange bars)35 and heightened environmental variability25 (purple bars). (B) 100/405-kyr eccentricity record and proposed EARS lake phases36 (blue bars). (C) plant wax δDppt values from Indian Ocean core sites 235/24138, (D) EARS ƒWC record (exponentially smoothed, α = 0.1) (E) Turkana Basin δ18OPC values. See Supplementary Data and Supplementary Information for site locations, δ18OPC values, and references. Shaded area denotes MPT interval (1.3–0.7 Ma).

Recent studies have suggested that hydroclimate may not be the primary cause of changes to African vegetation structures38,41,42. Plant wax δD values from ODP cores 235/241 in the Indian Ocean off of eastern Africa indicate no directional trend in calculated precipitation δD over the last 10 million years, and the MPT interval in those cores, specifically, does not appear to have experienced a significant shift in regional paleohydrology yielding intermediate values within the 3–0 Ma study interval (Fig. 4C)38. However, the sampling density (n = 3) of ODP 235/241 through the MPT is of particularly low resolution. If orbital climate forcing36 was the primary influence of the EARS C3 excursion, we would predict multiple C3 excursions coinciding with insolation maxima, which is not the case (Fig. 4D). The Turkana Basin δ18OPC record, a proxy of rainfall source, rainfall amounts, and/or temperature, shows a long-term trend consistent with more arid and/or cooler conditions, followed by a return to warmer and/or wetter conditions during the MPT (Fig. 4E). Additional rainfall proxy records from the EARS, able to distinguish rainfall amount from sources of rain42,43, are warranted to further test for links between hydroclimate and vegetation structure.

Vegetation structures are impacted by herbivore communities31; thus major changes in animal community compositions may have influenced the EARS C3 excursion (Fig. 5A–D). Significant declines in eastern African large-bodied carnivore speciosity from ~ 4–1 Ma (Fig. 5B)44 and megaherbivore diversity from ~ 6 to 0 Ma (Fig. 5D)41 correspond to the long-term EARS C4 trend (Fig. 5A). C3-browsing by fewer megaherbivores could have resulted in an increase in woody cover. However, after the C3 excursion, the EARS ƒWC record returns to previous percentages, maintaining the long-term C4 trend, which is not predicted as megaherbivores continued to decline. A recent compilation of the number of EARS grazer species from ~ 7 to 0 Ma45 parallels the spread of C4 grasslands as previous shown23,26,28; however, there is a major decline in the number of EARS grazer species, specifically non-ruminant grazers, beginning at ~ 1 Ma and coincident with the peak of the EARS C3 excursion (Fig. 5C). Faith and colleagues45 proposed that non-ruminant grazers were outcompeted by ruminant grazers and mixed feeders due to habitat loss during aridity pulses beginning with the MPT. The Turkana Basin δ18OPC record is not consistent with increased aridity during the MPT, but the EARS ƒWC record indicates that C4 grasslands contracted significantly, which may have influenced the decline in non-ruminant grazers.

Figure 5
figure 5

(A) EARS ƒWC record (exponentially smoothed, α = 0.1) compared to EARS fossil faunal abundance data: (B) carnivore fraction (medians);44 (C) number of grazing species (residuals);45 (D) megaherbivore diversity (residuals)41. Shaded area denotes MPT interval (1.3–0.7 Ma).

The long-term increase in eastern African C4 grasslands and its impact on faunal communities has been associated to concurrently decreasing pCO238,41. Global climate and vegetation models46,47 predict the cause-effect relationship between higher pCO2 and destabilized C4 vegetation48; moreover, woody thickening is proposed as a consequence of rising pCO246,47. Modeled and proxy pCO2 records, showing discrepancies in estimations across the MPT (Fig. 6A–E), fuel various hypotheses about the role of pCO2 in Earth’s climatic reorganization21,22,49,50,51. The EARS ƒWC record (Fig. 6G) is consistent with the Chinese Loess Plateau paleosol pCO2 record49 (Fig. 6E) and one of the pCO2 models51 (Fig. 6A) indicating relatively higher pCO2 confined to the MPT interval. The clumped-isotope paleo-thermometer record, derived from paleosols in the Nachukui Formation at Turkana, shows higher temperatures during the MPT (Fig. 6F)52, which are predicted with increasing pCO253. The Turkana Basin δ18OPC excursion circa 1 Ma is also consistent with higher temperatures (Fig. 4E).

Figure 6
figure 6

(A) Modeled global pCO2;51 (B) modeled global pCO2;50 estimated pCO2 (ppm) derived from compiled (C) phytoplankton-alkenone records23, (D) compiled δ11B values23, (E) Chinese Loess Plateau (CLP) paleosols51, (F) paleotemperature estimates based on clumped isotope analysis of EARS pedogenic carbonates52, (G) EARS ƒWC record (exponentially smoothed, α = 0.1). Shaded area denotes MPT interval (1.3–0.7 Ma).

We interpret that the EARS C3 excursion was primarily forced by a transient increase in pCO2, potentially accompanied by an increase in temperature. There is debate, however, about the primary drivers of African vegetation change38,54, and vegetation proxy records from various African regions record dissimilar trends across the MPT. Plant wax δ13C data from ODP core 1077 in the Lower Congo Basin indicated an increase in C4 vegetation, which was interpreted as a response to an increased aridity circa 900 Ka34. Arabian Sea plant wax δ13C data (core sites 721/722) yielded evidence for a long-term increase in C4 vegetation but no significant change during the MPT interval38. Lake Malawi plant wax δ13C record, from the southern EARS (Malawi Rift), shows persistent C3 vegetation throughout the Plio-Pleistocene and across the MPT interval55. Our analysis of individual basins demonstrates that the C3 excursion occurred in the northern EARS and may not have been a significant event in the central EARS. However, we emphasize that sampling resolution of some of these records may not be sufficiently high to resolve vegetation changes within the MPT interval.

Finding discrepancies between the EARS ƒWC record and other regional vegetation datasets supports interpretations that marine sediments to the north and west of Africa may not capture the complete range of paleoenvironmental conditions within the EARS56,57. In contrast to the limited spatial averaging of pedogenic carbonates (see Supplementary Information), marine core records of terrestrial vegetation represent integrated signals without specific provenance for large regions, for example, aeolian transport from southern Africa in the case of ODP core 107734. Site- and region-specific controls on vegetation may differentially respond to changes in ice volume, sea surface temperatures, and pCO238,57. Moreover, model data and lake core studies demonstrate that rift basins respond differently to global environmental change such as during the Last Glacial Maximum58,59. Therefore, we suggest that a transient rise in global pCO2 causing the EARS C3 excursion does not necessitate synchronous declines in C4 vegetation in other African regions.

A new behavioral and ecological scenario for the extinction of P. boisei

P. boisei’s extinction occurred during a significant contraction of C4 grasslands within the EARS, and specifically in one of its known habitats, the Turkana Basin. Admittedly, ƒWC estimates do not fully characterize the diversity and complexities of African vegetation communities during the Pleistocene, but rather provide evidence for relative changes in the dominant (C3 vs. C4) vegetation structures. The decline in C4 grasslands likely resulted in the loss of P. boisei’s exploited C4 plant foods (Fig. 7A), but the identities of those C4 plant foods remain unknown. Microwear evidence suggest that foods items may not have involved hard components7,10 but rather “novel mechanical challenges” entailing masticating C4 grasses and sedges for long periods of time12. Faunal-based studies have emphasized that EARS environments during human evolution were “non-analogous” to modern faunal community structures26,45,60. In a common thread, paleovegetation communities also evolved through time, thus limiting the use of modern EARS environments and current vegetation proxy methods to identify specific elements of vegetation communities as well as particular plant species consumed by Pleistocene hominins. Moreover, dietary reconstructions of eastern African hominins and non-hominin primates with δ13Cenamel and δ44/42Caenamel data pose issues of isotopic equifinality61 where many types of food combinations may result in comparable isotopic values.

Figure 7
figure 7

Proposed ecological and behavioral influences of P. boisei’s extinction. Food resources represented by (A) EARS ƒWC record (exponentially smoothed, α = 0.1); dietary competition shown with (B) speciosity of EARS grazers45 and (C) T. oswaldi δ13Cenamel values; dietary niche shown with (D) P. boisei δ13Cenamel and (E) δ44/42Caenamel values; estimated time of extinction denoted by (F) P. boisei’s duration18. See Supplementary Data and Supplementary Information for isotopic values, fossil image credit, and references. Shaded area denotes MPT interval (1.3–0.7 Ma).

P. boisei likely competed directly with the EARS grazing species for C4 plant foods (Fig. 7B). We suggest that P. boisei and several other non-ruminant grazers were outcompeted by ruminant grazers for declining C4 plant foods during the MPT interval. Of course, P. boisei was not an herbivorous ungulate but rather a large-bodied, bipedal, encephalized, and likely stone tool using hominin15,18,62, thus its life history strategies, social organization, reproductive rates, activity times, caloric and nutritional requirements, home and day ranges, and cognitive abilities were likely more similar to other hominins and non-hominin primates than to most of the ungulates occupying the EARS C4 biome. T. oswaldi was likely in direct competition with P. boisei for C4 plant foods throughout the Pleistocene (Fig. 7C). δ44/42Caenamel values, however, suggest that T. oswaldi engaged in omnivory throughout its evolutionary history, providing some dietary niche separation from the herbivorous P. boisei but also a competitive advantage, as C4 plant foods contracted.

Although much evidence indicates dietary niche partitioning between P. boisei and members of Homo, some dietary competition may have occurred. Patterson and colleagues (2019)63 suggested that in the Koobi Fora region of the Turkana Basin after 1.65 Ma, Homo sp. δ13Cenamel values slightly converge with those of P. boisei, and H. erectus was by and large the C4 interloper (Fig. 1). The post-1.65 Ma change in Homo sp. δ13CEC values was not detected in other sympatric mammals from the Koobi Fora region, and local vegetation structures appear stable; consequently, as these authors63 suggest, it seems likely that a behavioral change rather than a change in resource base underpinned H. erectus’ dietary shift. H. erectus and other members of Homo display wide ranges of δ44/42Caenamel values after 1.65 Ma suggestive of omnivory (Fig. 1C). Ungulate C4-grazer meat and marrow may have been the primary source of H. erectus’ C4 diet13,63, supporting the interpretation of niche separation rather than direct competition between the hominin sister taxa even after 1.65 Ma. But notably a few Homo specimens approach P. boisei’s δ13Cenamel and δ44/42Caenamel values13, which could be interpreted to indicate that segments of the omnivorous Homo populations exploited some resources within P. boisei’s dietary niche12.

Similar biogeographic distributions and habitat preferences of members of Homo and P. boisei62 implies competition for non-food resources including but not limited to feeding territories, sleeping sites, and potable water. Evidence for the origins5,64 and evolution65,66 of Early Stone Age technologies has shifted the discussion of presence vs. absence of tool making abilities by P. boisei and other non-Homo hominin species further toward the cognitive and social learning capacities required for habitual and advanced stone tool making67. The large-brained and -bodied H. erectus remains the most likely maker of advanced tools during the MPT interval62 and may have outcompeted P. boisei for non-food resources and some C4 plant foods with those tools.

In summary, EARS environments experienced a significant reduction in C4 grasslands during the MPT interval potentially forced by an increase in pCO2 and associated with a rise in temperature. The EARS C3 excursion, peaking circa 1 Ma, escalated dietary competition amongst the abundant C4-feeders, which influenced the decline of non-ruminant grazers. Dietary niche separation amongst the EARS hominins may have served as a strategy to reduce competition in the C3-C4-mixed feeding niche during the early Pleistocene. However, with C4 plant food loss, P. boisei’s inability to return to its ancestral C3-C4-mixed diet due to competitive exclusion by H. erectus and/or its own behavioral inflexibility likely played a role in its extinction (Fig. 7D–F).


Our study site is located in the Lake Turkana Basin, which is part of the northern Kenyan rift in the eastern branch of the EARS (Fig. 2A–C). On the northwest side of the basin, the Nariokotome Member is the uppermost unit of the Nachukui Formation and attains a thickness of ~ 60 m69. Previous interpretations suggested the Nariokotome Member was accumulated through the period of 1.30–0.75 Ma70. Near the Nariokotome Catholic mission, we studied this member’s outcrops and sampled pedogenic nodules along two NW trending transects (Supplementary Fig. S1, Supplementary Table T1). Two representative composite stratigraphic sections, measuring 50–55 m thick, were recorded (Supplementary Fig. S6). Walking along marker horizons (e.g., stromatolites layers) or using a transit to level the relative positions of marker horizons facilitated correlations between successive outcrops (Supplementary Fig. S2). Sedimentary strata of these outcrops have dips of about 3–5° into the west or are nearly flat lying. The strata comprise rounded volcanic-clast gravels, quartzo-feldspathic sands of varying grain sizes, and mudstones. Locally, the mudstones preserve carbonate nodules and slickensided fractures that indicate paleosols. Occasionally interbedded with these detrital clastic sediments are thinner units of stromatolite-encrusted gravels, mollusk sandstones, and tuff layers.

On the northeast side of the basin, the Koobi Fora Formation’s Chari Member (1.38–0.75 Ma) was examined because it is nearly time equivalent with the Nariokotome Member. We compiled fieldwork observations and samples from the Chari Member outcrops exposed near the town of Ileret (Supplementary Fig. S3). Sediments, chronostratigraphic constraints, and interpretations of the depositional environments for the Ileret outcrops have been described in detail elsewhere39. We sampled pedogenic carbonates from sedimentary strata documented by the section PNG-0439 The PNG-04 section is redrawn in Supplementary Figure S4 and relevant latitude and longitude data are listed in Supplementary Table T1. At PNG-04, an unconformity occurs ~ 7 m up from the base of the Chari Tuff. Samples were derived from stratigraphic levels above and below the unconformity, measured relative to the dated tuff units. Lithostratigraphic thicknesses, sedimentological data, and bedding attitudes were collected at the outcrops using standard field procedures and measuring instruments.

Pedogenic carbonate nodules were extracted from all preserved carbonate nodule-bearing paleosols throughout each of three outcrop sections in the Turkana Basin as the stratigraphic distribution of relevant geological materials dictated. Paleosols were identified from the presence of vertic features and slickensides. Pedogenic carbonate nodules were sampled at levels > 30 cm below the contact with overlying stratum and ~ 50 cm deep into the outcrop. Ages were determined through linear scaling between the stratigraphic levels of the radioisotopic dates of the Lower Nariokotome Tuff (1.30 Ma) and the Silbo Tuff (0.75 Ma) and the Chari Tuff (1.38 Ma) and the Silbo Tuff70. Scaled ages were calculated from the reported sedimentation rate70. The large age gap between samples Ileret 514–1 and 520–2 (Supplementary Fig. S5; Supplementary Data) is due to a ~ 500 kyr unconformity in the lower Chari Member (Supplementary Fig. S4)39.

Pedogenic nodules were cross-sectioned to expose the inner surface. Carbonate powders were eroded with a hand-held rotary tool (Foredom Series) affixed with a 0.5 mm carbide bit. We avoided sparry calcite and collected micrite from the nodules. Ninety-five δ13C analyses of extracted powders from fifty-three pedogenic carbonate (PC) nodules were conducted on a FISIONS Mass Spectrometer in the Department of Earth and Planetary Sciences at Rutgers University. Samples were reacted at 90 °C in 100% phosphoric acid for 13 min. δ13CPC and δ18OPC values are reported in the standard per mil (‰) notation: = (Rsample/Rstandard – 1)*1000, relative to Vienna-Pee Dee Belemnite (V-PDB) using the laboratory standard NBS-19. Analytical error is < 0.05‰.

We utilized δ13CPC data from published sources and this study (n = 53) to characterize vegetation structures in each of the EARS basins dated to 3–0 Ma, which spans the evolutionary history of P. boisei (Supplementary Data). Published data were taken from the compilation of Levin (2015)23 and also from Quinn and others (2013)71, Patterson and colleagues (2019)63, and Potts et al. (2018)72. We then compiled δ13CPC values from EARS sampling locations that included data spanning the MPT interval to gauge relative changes in vegetation structures from 3 to 0 Ma, which included Awash, Turkana, Olduvai Gorge, and Tugen Hills. Due to potentially different rainfall sources across different EARS basins73, we restricted time-series analysis to δ18OPC data from the Turkana Basin and Lower Omo Valley.

After Cerling and others24 we subtracted 14‰ from the δ13CPC values to convert to the isotopic equivalent of organic carbon (δ13Com) and used the equation: ƒWC = {sin[−1.06688 – 0.08538(δ13Com)]}2 to generate estimates of fraction woody canopy cover (ƒWC) for classification into UNESCO categories of African vegetation (see Supplementary Information). Eastern African savanna plant communities demonstrate a wide range of δ13CPC values, and due to differential paleosol deposition and preservation, δ13CPC data points are not evenly distributed through time. In order to assess trends in the central tendency of vegetation structures through time from the EARS ƒWC datasets, we performed simple exponential smoothing (α = 0.1, 0.3, 0.6), Loess regressions (3%, 10%, 20%, 30%), and a Bayesian change point algorithm of a 5-point running mean (see Supplementary Information).