Our species (Homo sapiens) is thought to have emerged shortly before 300 thousand years ago (ka) in Africa, before dispersing to occupy all the world’s biomes, from deserts to dense tropical rainforest4,5. Although grasslands and coasts have typically been given primacy in studies of the cultural and environmental context for human emergence and spread (for example, refs. 6,7,8), recent evidence has implicated several regions and ecosystems in the earliest prehistory of our species3,9,10. Rainforest habitation in Asia and Oceania is firmly documented as early as 45 ka (refs. 11,12), and perhaps as early as 73 ka (ref. 13). However, the oldest secure, close human associations with such wet tropical forests in Africa do not date beyond around 18 ka (refs. 6,14,15), despite evidence of the widespread presence of Middle Stone Age (MSA) assemblages in regions of present-day African rainforest16,17 (Supplementary Information Section SI-1). Isotopic and zooarchaeological evidence from Panga ya Saidi in Kenya have supported an earlier use of mixed tropical forest and ecotonal environments from at least 77 ka (refs. 18,19), but clear evidence for the dedicated occupation of wet tropical forest remains lacking.
Here we report a suite of analyses from the site of Bété I, located in the Anyama locality of Côte d’Ivoire, West Africa (Fig. 1 and Extended Data Figs. 1–4), that demonstrates a deep-time association between humans and wet tropical forests dating to around 150 ka (Marine Isotope Stage (MIS) 6) (Fig. 2). This association is both geographically and ecologically distinct from contemporaneous sites known from across Africa. The site of Bété I (5.515° N, 4.06° W) is located approximately 20 km north of Abidjan, where an extensive Quaternary sequence is displayed in several deep sedimentary exposures revealed by quarrying activity (that is, Bété I–IV), as first reported in ref. 20. These were resolved into six stratigraphic units (F to A, from the base to the top) and further subdivided into 14 layers (Supplementary Information Section SI-1 and Supplementary Figs. 1–3) during excavations undertaken between 1982 and 1993 by a joint Ivorian–Russian mission that focused on an approximately 14-m step trench, with deeper exposures observed across the quarry site21. These broadly comprise a weathering horizon of the chloritic shale bedrock (Unit F), coarser alluvial deposits attributed to the Continental Terminal (Unit E), finer, gravel-free alluvial terre de barre deposits that show extensive weathering in the lower levels (Units D and C), along with recent subsoil (Unit B) and a topsoil horizon (Unit A). A radiothermoluminescence study initially dated the Unit E deposits to the Early and Middle Pleistocene. Although these ages should be regarded with extreme caution (see Supplementary Information Section SI-1 for a critical evaluation), an estimate of 254 ± 51 ka from Unit D in sediments underlying archaeological horizons provides a tentative terminus post quem for human presence at the site21. Dating of the nearby valley floor deposit has indicated an Early Holocene or terminal Pleistocene incision to establish the modern drainage, probably cutting the terrace deposits (Supplementary Information Section SI-2). Key stone tool assemblages recovered from Unit D include a prominent heavy tool component, such as picks (Supplementary Information Section SI-1 and Supplementary Figs. 1 and 4), alongside small retouched tools (Supplementary Information Section SI-1 and Supplementary Fig. 4). Unit C has assemblages with Levallois reduction and small retouched tools. Unfortunately, the lithic artefact collections were lost during the 2011 civil war.
a, General map showing the African sites dated to MIS 6 (around 130–190 ka). b, Location of Bété I site. c, Sequence at Bété I in 2020 after sampling for geochronology and palaeoecological proxies.
Artefact densities are derived from ref. 21. Numerical age uncertainties are given at 1σ. NC and ND are the total numbers of lithics artefacts found in Units C and D, respectively.
Today, this site lies within the modern distribution of wet–humid West African tropical rainforest, which encompasses a diversity of forest types, including periodically to permanently inundated swamp and riparian forests, as well as evergreen rainforest22,23. Because the site represents the deepest stratified site yet found in Africa’s (present-day) tropical forest regions, we returned to re-investigate it in 2020. The site was unfortunately destroyed later, between 2020 and 2021 (ref. 24), by quarrying activities.
We located the original step trench at Bété I, and cut back and cleaned the uppermost four steps of the original excavation, spanning the top 5.65 m of the sedimentary sequence. Our field records matched the sequence reported in ref. 21, comprising four discrete sedimentary units referred to as Units A–D. A suite of 37 sediment samples were recovered from this sequence for new sedimentological and palaeoecological analyses. The quantified description of the physical characteristics of the deposits matched those from the field records and previous studies, but also highlighted a discrete transition between Units C and D (Fig. 2, Supplementary Information Section SI-2 and Supplementary Fig. 5). The sedimentology supported the interpretation of a low-energy alluvial environment with episodic hiatuses presenting a high-resolution depositional setting for the archaeological assemblages (Supplementary Information Section SI-2 and Supplementary Table 1), with earlier identification of fine knapping debris at the site indicating little likelihood of post-depositional disturbance.
The chronology of the Bété I site was obtained using a combination of single-aliquot (SA) and single-grain (SG) optically stimulated luminescence (OSL) and multiple centre (MC) electron spin resonance (ESR) dating, both applied to quartz grains extracted from the sediments (Supplementary Information Section SI-3). In total, eight SA and SG-OSL and five MC-ESR ages (all reported with 1σ error) were calculated for various samples from Units C and D (Fig. 2 and Supplementary Information Section SI-3). The ages are, overall, stratigraphically consistent from the base to the top of the Bété I sequence, and range from the late Middle Pleistocene to the Pleistocene/Holocene transition, contributing to the establishment of a coherent age–depth model for the in situ stone tool assemblages from Units D and C. Our critical evaluation of the combined OSL-ESR dataset indicated that: (1) the SG-OSL results can be regarded as the most reliable estimates of the true burial age of the deposits; (2) the SG-OSL chronology is supported, in most samples, by the semi-independent age control provided by the SA-OSL and MC-ESR results; and (3) among the various sets of ESR ages obtained through the MC approach, those derived from the titanium centre of quartz (Ti–H) signal are regarded as providing a closer estimate to the true burial age. A complete discussion of the dating results is presented in Supplementary Information Section SI-3 (see also Supplementary Figs. 6–14 and Supplementary Tables 2–11) and Extended Data Figs. 5 and 6. At the bottom of the sequence, the SG-OSL age of the deepest sample, ANY20-09, was 166 ± 14 ka (placing it in MIS 6, 130–190 ka, ref. 25), providing a maximum age constraint for the deposits. The chronology of the lithic assemblages from Unit D, with the larger tool component, is bracketed by SG-OSL ages of 146 ± 9 ka (ANY20-08) and 55 ± 3 ka (ANY20-05). By comparison, the Ti–H ESR ages are older, but consistent at 1σ, because of their large associated uncertainties (Supplementary Information Section SI-3). In particular, sample ANY20-08 is associated with the deepest location of lithic artefacts in Unit D, and provides an age of around 150 ka (in MIS 6) for the earliest evidence of human presence at the site (Supplementary Information Section SI-3).
The ages of 35 ± 3 ka (SG-OSL) and 44 ± 19 ka (ESR) from the transition from Units D to C are relatively coherent, and agree at 1σ, indicating the end of Unit D deposition towards the end of MIS 3. Further up in the sequence, two SG-OSL ages from Unit C constrain the typical MSA artefacts to between 20 ± 1 ka (ANY20-03) and 12 ± 1 ka (ANY20-02), placing this unit in MIS 2.
The delta 13 carbon (δ13C) measurements of the bulk soil organic matter (SOM) from the sediments taken through the Bété I sequence are shown in Fig. 2, Supplementary Table 10 and Extended Data Fig. 7. The values range from −25.4 to −27.6‰ (−26.6 ± 0.6‰, n = 35) and overlap with the bulk δ13C values measured from contemporary African rainforest contexts26, corrected for the Suess effect27,28,29,30, of −24.8 to −34.5‰ (−28.1 ± 1.9‰, n = 24). The SOM δ13C is usually assumed to represent standing plant biomass and plant remains introduced by either humans (for example, carried for bedding or other uses) or nature (for example, from wind and water transport)31, with an increase of +1 to +3 per mil in SOM δ13C values relative to the contributing plant matter being the result of microbial activity32. With this in mind, the Bété I bulk δ13C values primarily indicate C3 biomass, with an increase in values recorded from sample 31 (Unit D3) to sample 29 (Unit D2), from 4.2 to 3.8 m deep, and δ13C fluctuations from sample 21 (Unit D1, at 2.2 m deep) upwards with a steady trend towards increasing values towards the top of the sequence (Unit A). To discern the exact drivers of these trends, we combined this isotopic analysis with biomarker analyses (leaf-wax) and the examination of phytoliths and pollen from the sequence.
Of the 37 palaeoenvironmental samples, 31 had sufficient lipid material for plant wax biomarker analysis through gas chromatography mass spectrometry (GCMS). Even-numbered, mid-chain-length (C22–C24) n-alkanoic acids (as fatty acid methyl esters (FAMEs)) dominated the biomarker distributions, indicating high input from submerged or emergent plant wax sources for most samples33,34. The average chain length (ACL20–34) ranged from 23.9 to 27.3 (25.2 ± 0.84, n = 31) (Supplementary Table 13), which is typically lower than previous reports of modern ACLs from African terrestrial plants35,36,37,38,39. The aquatic plant ratio (Paq C22+24) ranged from 0.41 to 0.86 (0.66 ± 0.09, n = 31), and the submerged/terrestrial ratio (STR) of the C24 FAME (STR24) ranged from 0.15 to 0.44 (0.25 ± 0.07, n = 31) (Fig. 2 and Supplementary Table 13). Both proxies indicate that abundant wetland-adapted species, either fully submerged or emergent plants, were principal contributors of plant wax biomarkers to the site (Supplementary Information Section SI-4 and Supplementary Figs. 15 and 16). The STR values, however, indicate that terrestrial plants also provided plant wax biomarkers to the Bété I sediments. There also seems to be a changing pattern in the relationship between the abundance of the C24 FAME and the bulk sedimentary δ13C. For instance, below 2.0 m (from the lower section of Unit D3 to the middle of Unit D1), the δ13C co-varies in the same direction as both the STR and Paq (Spearman’s correlation STR24 rs = 0.248, P = 0.394; Paq rs = 0.597, P = 0.024). However, above 2.0 m (from the middle of Unit D1 to the middle of Unit A), there is an anti-phased correlation between the C24 FAME and the bulk sedimentary δ13C (Spearman’s correlation rs = −0.377, P = 0.136). That is, as the abundance of C24 increases relative to the other FAMEs, the δ13C decreases. Additionally, as both the STR and Paq values increase, indicating greater input from submerged/aquatic plant biomarkers, the bulk sedimentary δ13C shifts lower (Spearman’s correlation STR24 rs = −0.441, P = 0.077; Paq rs = −0.240, P = 0.353). It is possible that, as local forest characteristics changed, such as with forest succession, the bulk isotope signal changed in accordance with the abundance of terrestrial or aquatic plants.
Nine samples were analysed for phytoliths and pollen to establish the preservation and representation of microbotanical evidence, selected to match peak artefact densities, correlate to human habitation of the site, and provide further insights into the trends observed in the biochemical data. The deposition and preservation of these two types of plant microfossil is inversely correlated, such that samples with a low phytolith influx (Supplementary Information Section SI-4 and Supplementary Fig. 17) tend to have a higher influx of pollen (Supplementary Information Section SI-4 and Supplementary Tables 15 and 16), but every sample yielded plant microfossils.
The phytolith assemblages produced by nine samples (samples 10, 11, 21, 22, 26, 27, 30, 31 and 36) (Supplementary Table 14) established their preservation. Morphotype identification demonstrated that all samples were dominated by arboreal phytolith morphotypes (Extended Data Fig. 8), with 82–96% of these morphotypes representing trees/shrubs, and therefore being indicative of a C3-dominated biomass in all cases (Supplementary Fig. 18). All of the arboreal phytolith types identified in the nine sediment samples are produced in most dicotyledons, so they could not be attributed to the genus or species level. Arecaceae phytoliths (palms) made up between 2 and 7% of the identified assemblages in seven of the nine samples, reaching a maximum in the uppermost sample (sample 10). Grass phytoliths made up between 1 and 18% of the assemblage, with the highest values coming from the lowermost sample (sample 36). Grasses peak again in sample 27 (16%) and fell to their lowest percentages in the uppermost unit (sample 10, 3%, and sample 11, 1%). It was not possible to directly compare the number of phytoliths produced by monocots and dicots because plants have different abilities to produce phytoliths—in particular, the Poaceae produce more than other monocotyledons40. As monocotyledons can produce up to 20 times more phytoliths than the dicots41, the dominance of arboreal phytoliths in all samples (more than 81%) is a clear signal of local forest cover with relatively few grasses or palms.
The pollen assemblages (Supplementary Figs. 19–26, Supplementary Tables 16 and 17 and Extended Data Fig. 9) in these samples were dominated by dicot pollen types (70–80%), followed by grasses (10–20%) and palms (5–10%). The pollen types were attributable to species, in some cases (Elaeis guineensis Jacq.), and to the genus (Hunteria) or family (Poaceae) level in others. There was a consistent presence of pollen types typical of wet–humid West African rainforests, riparian forests and swamp forests. Early riparian forest succession is signalled by the co-occurrence of E. guineensis (oil palm), dense shrubs/trees belonging to the genera Alchornea and Macaranga, and gap-colonizing trees, such as Anthocleista (Supplementary Fig. 22 and Supplementary Table 17). Canarium schweinfurthii and Pentadesma are both large trees frequently found in the later stages of forest succession in seasonally inundated conditions near rivers and lakes. These pollen types were more common in Unit D (specifically in the middle of D2 and the top of D3) and were absent from Unit C, being replaced by Uapaca, which is a diverse genus with many riparian-affiliated species. Unit C also yielded taxa typical of riparian forests (Parinari sp.), as well as types widespread in the Guineo–Congolian forest zone (Rauvolfia sp.). The Unit D samples contained more Hunteria pollen, probably attributable to Hunteria umbellata, which is common in wet–moist West African tropical rainforests23, particularly in forests adjacent to waterways42. The presence of anther fragments from E. guineensis and Hunteria (Fig. 3) in sample 30 from the top of Unit D3 presents strong evidence for whole flowers/anthers falling directly from the plant and being incorporated into the sedimentary matrix, supporting the local rainforest signal.
Examples of anther fragments (sample 30) from pollen taxa typical of rainforest (Hunteria) and flooded forest (E. guineensis) and phytoliths (respectively from samples 20, 30 and 22) preserved in Unit D of the Bété I sequence. Scale bar, 10 µm.
The SG-OSL dates of around 20–12 ka for the MSA assemblage in Unit C and 150–55 ka for the assemblages differentiated by a large tool component in Unit D document persistent human occupation of the Anyama area after the Middle Pleistocene. However, it is the lithic assemblages in Unit D that are of particular interest, as they are associated with a wet and forested environment. The palaeoenvironmental proxies from Unit D show no evidence for open and dry grassland, sparse savannah, or wooded savanna vegetation cover, despite the tendency for grasses to be over-represented in microfossil records. The sedimentary, biomarker and microfossil results are remarkably consistent, showing evidence for alluvial deposition in a tropical forest environment and consisting of riparian, swamp and rainforest taxa. Currently, the assemblages of Bété I from Unit D dating from MIS 6 are the oldest found outside the sahelian and sudanian savannah biomes of West Africa (Fig. 4). On a broader scale, the Unit D lithic assemblages are contemporaneous with other MIS 6 MSA stone tool assemblages found in other African sites located in different ecoregions, such as savannahs, or close to modern coasts (n = 16) (Fig. 1 and Supplementary Table 18 in Supplementary Information Section SI-5 and references therein) in West Africa (Bargny 1 and Bargny 3), in northern Africa (Bizmoune, Ifri n’Ammar, Wadi Lazalim (site 16/29), Bîr Tirfawi, Taramsa 1), in eastern Africa (Sai Island 8-B-11, EDAR 135, Abdur) and in South Africa (Amanzi Springs, Florisbad, Pinnacle Point 13B, Wonderwerk, Border Cave, Bundu Farm).
Insolation curve (dark blue) from ref. 46. Inter-regional African humidity curve (purple) from ref. 47. Benthic curve (light brown) built from ref. 48. Synthesis of dated occupations from West Africa from refs. 48,49,50 (for more details, see Supplementary Information Section SI-5). The Later Stone Age sites are not included. Error bars associated with OSL and ESR ages represent ± 1σ uncertainties.
Several independent lines of evidence have confirmed the association between humans and tropical wet broadleaf forest at Anyama, starting at least 150 ka. The consistent forest signal over time also indicates that this area of West Africa possibly acted as a rainforest refugium during arid periods (Supplementary Information Section SI-4). This corroborates projections of Middle and Late Pleistocene vegetation showing reduced, but persistent, rainforest cover at lower latitudes43. These data confirm a deep-time connection between human evolution and wet tropical forests, and highlight the importance of Africa’s many biomes and diverse ecoregions in this process9,44. The assemblages from Unit C, featuring Levallois flakes and points, side and end retouched pieces, add to the emerging evidence of a chronologically persistent MSA in West Africa towards the end of the Late Pleistocene that is probably a key regional characteristic7,45.
The assemblage in Unit D, featuring large tools alongside a small tool component, may support long-held views that the diverse heavy-duty tool assemblages seen in Central and West Africa are convergent adaptive solutions to tropical forest habitation6,16. The combination of an MIS 6 age with the ecological context of the Unit D assemblage is without precedent elsewhere in Africa. As a result, the archaeological classification of the Unit D assemblage warrants circumspection and further study (Supplementary Information Section SI-1). This is particularly the case because West Africa remains under-researched compared to other regions of the continent, and its archaeological sequence and specific regional characteristics are yet to be fully understood (Fig. 2 and Supplementary Information Section SI-1). Most importantly, however, our results confirm a deep-time connection between human evolution and tropical forest biomes, opening up a new chapter in the human past in which our species occupied dense, wet tropical forests much earlier than widely thought. This association confirms the predictions of the pan-African model of human evolution, and highlights the importance of Africa’s many regions and ecosystems in this process9.
