Biogeochemical evidence for targeted landscape use in ancient foragers of Malawi

Abstract

The Zambezian Biome of southern-central Africa was an important environmental setting for ancient human interactions and biogeographic dispersals. Here, we assess herbivore migratory behavior and habitat selection with a ~ 2,500 km² isoscape of bioavailable strontium (⁸⁷Sr/⁸⁶Sr) for the Kasitu Valley of northern Malawi, and a probabilistic spatial assignment of fossil herbivore enamel from four sites dated to the last 25,000 calibrated years BP. The results show a very large range in bioavailable ⁸⁷Sr/⁸⁶Sr values in the study area (0.7248-0.8521) and reveal that foragers hunted small game locally and procured most large prey in the Afromontane grasslands to the southeast of the study area and along narrow riparian corridors. We found no evidence of migratory behavior in species that exhibit this behavior today. Ancient foragers likely hunted prey that were available year-round, consistent with zooarchaeological and genetic evidence for reduced mobility at the end of the Pleistocene.

Introduction

The Zambezian biome (ZB) is an ecoregion of southern-central Africa, comprising 13% of the African continental land, that is characterized by biotic cohorts assembled on nutrient-poor soils derived from Precambrian geology. These soils support open-canopy woodlands and thickets locally known as miombo, seasonally flooded grasslands and reed swamps (dambos), seasonal gallery forests, and higher-elevation grasslands (generally above ~1600 masl) that contain patches of evergreen-dominated Afromontane forest^1,2. Human foragers inhabited the ZB throughout the Pleistocene, but little is known about land use and subsistence strategies and how they differed compared to more comprehensively studied areas in eastern and southern Africa^3,4. Previous archeological and paleoenvironmental research in Malawi showed an increase in human impact on the environment ~85,000 years (ka) ago, which may have accompanied an increase in regional population sizes⁵. aDNA of Late Pleistocene and early Holocene foragers from across the ZB has shown that foragers participated in mating networks connecting central, eastern and southern Africa between 20 and 80 ka ago, and that sometime after the Last Glacial Maximum (LGM, ending ~19.5 ka ago)⁶ long-distance genetic exchange transitioned to more locally restricted mating interactions at the scale of a few tens of km⁷. Local lifeways are also reflected in a focus on short-distance hunting of small ungulates at Hora 1 (HOR-1) rock shelter⁴. To understand the role of ecology in this major change in population connectivity during the LGM, it is necessary to reconstruct the mobility of people and their prey at a multi-site, landscape level. Understanding the seasonal distribution of prey on the landscape also informs hypotheses of forager socio-territorial organization, where seasonally predictable clustering of resources at key locations may have favored the establishment of more territorial behavior^8,9,10.

Radiogenic strontium isotope ratios (⁸⁷Sr/⁸⁶Sr) in biological hard tissues are a common tool to estimate the provenience and movement history of individuals, since they reflect ⁸⁷Sr/⁸⁶Sr of the substrate with no trophic fractionation^11,12,13. This approach requires an understanding of how environmental ⁸⁷Sr/⁸⁶Sr values vary spatially, which is best achieved with an isoscape, i.e., a predictive model of isotope values at unsampled locations^14,15. Furthermore, when teeth are sampled sequentially along the growth axis, it is possible to obtain intra-annual time series reflecting migratory behavior^16,17. For further details, see Supplementary Text 1. Here, we present a plant-based isoscape (N_plants = 196, plus 25 samples of dust and 12 gastropods) of bioavailable ⁸⁷Sr/⁸⁶Sr for an area of over 2500 km² in the upper part of the Kasitu Valley in northern Malawi (Fig. 1). The area has been under archaeological field investigation by MALAPP (Malawi Ancient Lifeways and Peoples Project) between 2016 and 2023, which has led to the recovery of materials whose provenience can be tracked using a Sr isoscape (Supplementary Text 2), as well as Pleistocene human remains¹⁸ documenting shifts in population connectivity⁷. We applied Empirical Bayesian Kriging¹⁹ to generate the isoscape and used it to create probabilistic assessments of provenience for a sample of archaeological faunal remains (N_teeth = 45, N_samples = 166; Table 1) and test the hypothesis that most prey hunted by humans lived within the confines of our study area during dental development, consistent with the aDNA evidence for reduced mobility and “local” lifeways over the last ~20 ka.

Fig. 1: Maps of the study area.

Left: digital elevation model of the study area showing the excavated sites, the three main rivers, and the sampling locations, color-coded by ⁸⁷Sr/⁸⁶Sr of the samples collected at each location. Color-coding of the samples follows natural breaks (jenks⁵⁷) in the data. Note the transect of samples extending east from Hora Mountain to the Kasitu River. Right: map of Africa with Malawi indicated, showing the extent of the Zambezian biome (green shading). The study location is located in the north of the country.

Table 1 Summary of the enamel samples analyzed in this study

We defined our study area as the landscape within a radius of ~25 km from Hora Mountain and encompassing all the archaeological sites investigated by MALAPP (Fig. 1). Such a radius should reasonably encompass most of the daily foraging activities of the people inhabiting these sites, based on ethnographic data on modern foragers from the ZB^20,21. The overall landscape is dominated by Precambrian to Lower Paleozoic Basement Complex bedrock granitic gneiss inselbergs, as well as biotite and muscovite schists, and gneiss outcrops, interspersed between broad flat valleys filled with young (Quaternary) sandy sediments. The valleys also include alluvial and floodplain terrace dambos supporting “rivers of grass” with patches of reeds and sedges^2,22 (Fig. 2). The southeastern quarter of our study area is occupied by a mosaic of Afromontane vegetation with grassland being the dominant component¹ (Supplementary Fig. 1). The rest of the study area is mostly miombo woodland, which loses most of its understory grasses during the dry season (Supplementary Fig. 1).

Fig. 2: Types of habitats available within the study area during the dry season.

A Open woodland; B dambo in the Kasitu Valley; C dambo in Vwaza marsh (close but outside of the study area); D Mountain grassland in the Nyika plateau. Credit for photos (A, C, D) to Jessica Thompson; credit for photo (B) to Chelsea Smith.

Results

Our plant sample records a very large range and variation of ⁸⁷Sr/⁸⁶Sr among collection sites (Fig. 3A and Supplementary Data). The range of ⁸⁷Sr/⁸⁶Sr in non-riparian plant samples is wider (0.7248–0.8521; Δ = 0.1273, x̄ = 0.7669) than that of riparian (including dambo) plant samples (0.7383–0.7860; Δ = 0.0477, x̄ = 0.7571). Land snail shell ⁸⁷Sr/⁸⁶Sr ranges from 0.7183 to 0.7861 (Δ = 0.0678). Although gastropods were collected opportunistically, they sample a variety of localities (Supplementary Data). Intra-plant ⁸⁷Sr/⁸⁶Sr variability (leaves 1 m above ground—leaves 2 m above ground; n = 34 pairs; Fig. 3B) averages 0.0007 ± 0.0011. Intra-locality ⁸⁷Sr/⁸⁶Sr variability (mixed grass—leaf average; n = 30 pairs; Fig. 3 C) averages 0.0029 ± 0.0033. ⁸⁷Sr/⁸⁶Sr differences between leaves and the dust deposited on them are high and average 0.0104 ± 0.0086 (n = 25 pairs; Fig. 3D). In 22 of 25 cases, dust had lower ⁸⁷Sr/⁸⁶Sr than the leaves it had coated.

Fig. 3: ⁸⁷Sr/⁸⁶Sr values in environmental samples.

A Distribution of plant ⁸⁷Sr/⁸⁶Sr values across bins 0.01 wide. Color-coding highlights natural breaks (jenks) in the data. Note that some natural breaks occur within bins. B Relationship between ⁸⁷Sr/⁸⁶Sr values in leaves collected 1 m above ground vs 2 m above ground, within individual trees. C Relationship between ⁸⁷Sr/⁸⁶Sr values in mixed grasses vs leaves collected at the same locality. D Relationship between ⁸⁷Sr/⁸⁶Sr values in leaves vs the dust deposited on them. Red dashed lines in plots (B, C, D) represent perfect correspondence (slope = 1) between the pairs of samples.

Concentrations of trace elements that might suggest diagenetic alteration (rubidium, uranium, vanadium, and rare earth elements)²³ were higher than expected based on modern mammals in 20 enamel samples, but none of the ⁸⁷Sr/⁸⁶Sr ratios are outliers relative to other samples in the same tooth or in other teeth of the same taxon (Supplementary text 8). Therefore, all measurements are included in the analysis. Our enamel samples span ⁸⁷Sr/⁸⁶Sr from 0.7299 to 0.7813 (Δ = 0.0514; see Supplementary Data). Different taxa yield different ranges of ⁸⁷Sr/⁸⁶Sr ratios (Fig. 4). Zebras have the lowest ratios (0.7299–0.7654). Bovids have intermediate ratios (0.7390–0.7697), when excluding an individual reduncin (puku, ⁸⁷Sr/⁸⁶Sr = 0.7813) with a very high ratio. The average ⁸⁷Sr/⁸⁶Sr ratio for zebra (0.7482 ± 0.0103, 1 sigma) is not significantly lower than that of bovids (0.7570 ± 0.0005) (Wilcoxon rank sum test [two-tailed], W = 138, p = 0.1). Bushpigs (0.7509–0.7568) also have intermediate ratios, while warthogs (0.7587–0.7770) and rhinos (0.7473–0.7729) have higher ratios, except for one warthog with low ratios (0.7346-0.7393). Warthogs and rhinos are not statistically significantly different from zebras and bovids. Diachronically, average zebra ⁸⁷Sr/⁸⁶Sr is significantly higher 25–15 ka (0.7497 ± 0.0108) than 14–10 ka (0.7361 ± 0.0004) (W = 271, p = 0.02). The average zebra ⁸⁷Sr/⁸⁶Sr 6.5–0 ka (0.7468 ± 0.0025) is indistinguishable from the 25–15 ka average, but significantly different from the 14–10 ka average (W = 0, p < 0.001). Average small bovid ⁸⁷Sr/⁸⁶Sr decreases between the older time bins (25–15 ka: 0.7560 ± 0.0042; 10– 8 ka: 0.7615 ± 0.0055) and 6.5–0 ka (0.7500 ± 0.0062), but the difference is statistically significant only between the two Holocene bins (W = 15, p = 0.03). When pooling the data by archaeological site (Supplementary Fig. 2), the HOR-1 samples yield a significantly higher average (0.7588 ± 0.0069, N_samples = 105) compared to the MAZ-1 samples (0.7437 ± 0.0089, N_samples = 55) (W = 5355.5, p < 0.001). The sample sizes from HOR-5 and KAD-1 are too small for this comparison.

Fig. 4: Summary of the enamel ⁸⁷Sr/⁸⁶Sr results by taxon (N_samples/N_teeth) and time-bin.

Note that the two Hippotragus sp. teeth sample the same individual. Silhouettes from phylopic.org.

Figures 5 and 6 show ⁸⁷Sr/⁸⁶Sr variation within individual sequentially sampled zebra Equus quagga and alcelaphin teeth, plotted against δ¹⁸O series from the same teeth as a proxy for season of formation (high δ¹⁸O values representing the dry season and low δ¹⁸O values representing the wet season; data from Bertacchi et al., in review). In most individuals, variation is small compared to the range of values measured in plant samples (Table 2), which is expected as plants sample point locations while mobile animals average across their range. Intra-tooth variability in ⁸⁷Sr/⁸⁶Sr in several individuals follows a quasi-sinusoidal pattern consistent with directional movement over several months (e.g., specimens #12711, #46459, #101126) but with very small overall change. Zebra #67935 stands out as the specimen with the largest intra-tooth variation (0.0121) and the pattern of change in ⁸⁷Sr/⁸⁶Sr is mirrored by changes in δ¹⁸O values, with a few mm of “temporal lag” between the two^cf 24. This individual has higher ⁸⁷Sr/⁸⁶Sr associated with higher δ¹⁸O values and lower ⁸⁷Sr/⁸⁶Sr associated with lower δ¹⁸O values, but this intratooth profile likely does not capture a full seasonal cycle (based on the shape of the oxygen curve). Intra-tooth variation in other taxa is displayed in Supplementary Fig. 4. ⁸⁷Sr/⁸⁶Sr variation in Hippotragus sp. (0.0011) and Kobus cf. ellipsiprymnus (0.0001) is very small, consistently with limited seasonal mobility over a homogenous substrate. In contrast, ⁸⁷Sr/⁸⁶Sr variation in warthog Phacochoerus africanus (0.0047), kudu Tragelaphus strepsiceros (0.0056), and white rhino Ceratotherium simum (0.0029) is larger, suggesting either a larger home range or the use of habitats situated at the interface of different substrates.

Fig. 5: Sequential intra-tooth ⁸⁷Sr/⁸⁶Sr and δ¹⁸O values in zebras.

δ¹⁸O values and ages from Bertacchi et al. (in review). Enamel goes from formed earlier to the left to formed later to the right (note axis inversion depending on preservation of the cervix). Calibrated ages: #11736: 3.8–3.6 ka cal BP; #9354: 12.5–12.0 ka cal BP; #41169: 15.7–15.3 ka cal BP; #12711: 16.6–16.3 ka cal BP; #46459: 22.0–21.5 ka cal BP; #82032: 17.4–17.1 cal BP; #67935: 17.0–16.6 cal BP. Silhouettes from phylopic.org.

Fig. 6: Sequential intra-tooth ⁸⁷Sr/⁸⁶Sr and δ¹⁸O values in alcelaphins.

δ¹⁸O values and ages from Bertacchi et al. (in review). Enamel goes from formed earlier to the left to formed later to the right (note axis inversion depending on preservation of the cervix). Calibrated ages: #152874.8: 10.7–10.5 ka cal BP; #75753: 15.3–15.0 ka cal BP; #101126: 16.9–16.5 ka cal BP; #153057.5: 15.4–15.0 ka cal BP. Specimens #67572, 64737, and #67485 are not directly dated. Silhouettes from phylopic.org.

Table 2 Maximum, minimum, and range of ⁸⁷Sr/⁸⁶Sr ratios in sequentially sampled teeth

The EBK-generated isoscape is shown in Fig. 7. The standard error of the predicted ⁸⁷Sr/⁸⁶Sr is relatively high, especially at the undersampled margins of our study area. This highlights the heterogeneity of ⁸⁷Sr/⁸⁶Sr in, and the geological complexity of, the Kasitu Valley. Visual inspection suggests that several low values occur in the higher elevation southeastern portion of the study area. We assessed variations of this isoscape but found this version to perform better overall (Supplementary Text 7).

Fig. 7: ⁸⁷Sr/⁸⁶Sr isoscape of the Kasitu Valley.

Left: predicted ⁸⁷Sr/⁸⁶Sr in the study area based on an Empirical Bayesian Kriging (EBK) model. Right: standard error of the predicted ⁸⁷Sr/⁸⁶Sr. Basemap from ESRI World Imagery.

Based on our isoscape, we generated posterior probability density maps representing spatial assignments for our faunal enamel samples. Most maps usefully show areas with high probability of origin, while also including very small pockets of probability around sampling points that are likely methodological artefacts. E. quagga from MAZ-1, dated to ~12.3 ka cal BP, likely originated from the area today occupied by Afromontane grasslands in the southeastern quadrant of our study area (Fig. 8). Conversely, 25–15 ka cal BP E. quagga from both MAZ-1 and HOR-1 appears to have had a narrower habitat preference along riparian corridors (Fig. 8). E. quagga from MAZ-1, dated to ~3.5 ka cal BP, possibly occupied both the southeastern quadrant and the upper Kasitu River (Fig. 8).The only E. quagga individual yielding large intratooth variation (#67935 from HOR-1, dated to ~16.7 ka cal BP) appears to have ranged along the upper Kasitu during a wetter part of the year and along a different riparian corridor during a drier part of the year (Supplementary Fig. 9). Alcelaphins from HOR-1 have an even more restricted likely provenience, with most posterior probability concentrated along the Luwewya River to the west of the site (Fig. 9). Among the other large bovids, all from MAZ-1, Hippotragus sp. dated to ~18.8 ka cal BP likely originates from the riparian habitats along the upper course of the Kasitu River or the area today occupied by Afromontane grassland around KAD-1; T. strepsiceros dated to ~10 ka cal BP originates from the southeastern quadrant; T. oryx dated to ~6.4 ka cal BP likely originates from a riparian habitat other than the Kasitu (Fig. 10). The likely provenience of reduncin bovids is shown in Supplementary Fig. 10. Some of the equifinality in the density maps can be resolved by considering that these antelopes are unlikely to have ranged far from dependable sources of water²⁵. Thus, both Pleistocene Kobus vardonii from MAZ-1 and Kobus ellipsiprymnus from HOR-1 were likely procured from the upper Kasitu River, while 10–8 ka BP K. vardonii from HOR-1 derived from a secondary riparian corridor or dambo. Among the small bovids, duikers at KAD-1 and HOR-1 appear to have been procured from different localities (Supplementary Fig. 11). Precise proveniences are difficult to determine, but local procurement near the sites cannot be rejected. Klipspringers Oreotragus oreotragus from KAD-1 were likely procured locally near the site, while the provenience of the individuals from HOR-1 is more ambiguous (Supplementary Fig. 13). Equifinality in spatial assignments also affects other small bovids, but in general, local procurement remains the most parsimonious interpretation (Supplementary Fig. 16). 14–10 ka BP Ph. africanus from HOR-1 likely originated from the area that today is Afromontane grassland, while 6.5–0 ka BP individuals from HOR-1, HOR-5 and MAZ-1 all have their most likely origin in riparian corridors other than the Kasitu (Supplementary Fig. 18). Pleistocene Potamochoerus larvatus from HOR-1 also originated from a riparian habitat, but this is more likely to be the upper Kasitu (Supplementary Fig. 18). C. simum dated to 25–15 ka BP from HOR-1 and to 6.5–0 ka BP from MAZ-1 could have been procured from a relatively large area in the southeastern quarter of the study area, with clusters of probability in the upper Kasitu and in the habitats around KAD-1 (Supplementary Fig. 21) Conversely, 10–8 ka BP C. simum from HOR-1 likely inhabited lower elevation areas in the west or northeast of the study area. Using mean faunal ⁸⁷Sr/⁸⁶Sr ratios from HOR-1 and MAZ-1 as a proxy for the hunting catchment of each site, it appears that the foragers occupying both sites focused their foraging along the riparian corridors of the Luwewya and Kasitu, as well as the southeast of the study area (Fig. 11).

Fig. 8: Posterior probabilities for equids.

Posterior probability density maps overlain on satellite imagery of the study area showing the likely provenience of plains zebra E. quagga from MAZ-1 and HOR-1. 25–15 ka maps combine multiple individuals; for individual maps, see Supplementary Figs. 6 and 7. Basemap from ESRI World Imagery.

Fig. 9: Posterior probabilities for alcelaphins.

Posterior probability density maps overlain on satellite imagery of the study area showing the likely provenience of alcelaphins from MAZ-1 and HOR-1. The 25–15 ka map combines multiple individuals, for individual maps see Supplementary Fig. 9. Basemap from ESRI World Imagery.

Fig. 10: Posterior probabilities for tragelaphins and hippotragins.

Posterior probability density maps overlain on satellite imagery of the study area showing the likely provenience of tragelaphines and hippotragins from MAZ-1 and HOR-1. Basemap from ESRI World Imagery.

Fig. 11: Posterior probabilities for pooled fauna by site.

Posterior probability density maps overlain on satellite imagery of the study area using the mean ⁸⁷Sr/⁸⁶Sr ratios from archaeological fauna at HOR-1 and MAZ-1 as a proxy for the main catchment territory of each site. Basemap from ESRI World Imagery.

Discussion

The broad ⁸⁷Sr/⁸⁶Sr ranges and the high intra-locality variation recorded in the Kasitu Valley are in accordance with the global observation that regions with higher mean ⁸⁷Sr/⁸⁶Sr also have higher variability¹⁴. This is likely due to the differential weathering of different components of old rocks, which also weather overall slowly making the contribution of exogenous sources proportionally larger¹⁴. This is evident in our samples, where dust has lower ⁸⁷Sr/⁸⁶Sr than the plants it was deposited on, suggesting it originated from the erosion of rocks that were not as old and K-rich as local rocks. Since 23 of 25 leaf-dust pairs come from non-riparian or dambo localities and the upslope soils in our study area are acidic, it is also possible that the relatively large contribution of atmospheric Sr to the bioavailable pool is a product of low Sr concentration in those soils²⁶. Intra-locality variability is also comparatively large^{cf 27}, although not as high as the variability of 0.0052 noted by Snoeck et al. in plants from Ireland²⁸. Non-riparian plants yielded a higher mean ⁸⁷Sr/⁸⁶Sr than riparian and dambo plants, likely because the latter average across different sources in their watershed. While our plant sample records a very large variability, the true range of bioavailable ⁸⁷Sr/⁸⁶Sr in the Kasitu Valley is likely to be even larger than the one captured by our sampling strategy. For instance, two coffee beans from Ludwing–Mzuzu (~39 km northeast from our study area) yielded ⁸⁷Sr/⁸⁶Sr lower than our lowest plant sample (0.7131 and 0.7148)²⁹, although this might be influenced by fertilizers. The water of lake Malawi also has ⁸⁷Sr/⁸⁶Sr lower than that recorded in our plant sample (0.7166)³⁰.

Our local-scale isoscape is broadly in agreement with a recent continental-scale isoscape computed through random forest regression³¹. The differences between the two isoscapes are mostly within ±0.01, especially along rivers and in the southeastern quadrant of the study area, whereas larger differences exist in undersampled areas (Supplementary Fig. 22). Although our range of ⁸⁷Sr/⁸⁶Sr in plants is greater than those directly measured at most other localities worldwide, an even larger range (0.7054–0.8780; Δ = 0.1726) is recorded in southern Zimbabwe, where plants growing on substrates with Archaean gneisses and massive granites yielded ⁸⁷Sr/⁸⁶Sr > 0.7552 (ref. ³²). Much lower values are recorded in most regions of southern Africa—for instance the southern Cape (⁸⁷Sr/⁸⁶Sr < 0.7242)^16,33 and the Maloti-Drakensberg mountains (⁸⁷Sr/⁸⁶Sr < 0.7173)³⁴—and the Ca-rich volcanics of eastern Africa—for instance Kenya and northern Tanzania (⁸⁷Sr/⁸⁶Sr < 0.7280)¹⁵, and Olduvai Gorge (⁸⁷Sr/⁸⁶Sr < 0.7048)³⁵. There are exceptions, such as the quartzites, basalts, and shales in the Sterkfontein valley (⁸⁷Sr/⁸⁶Sr up to 0.774)¹³ and the Archaean rocks of the Angola Block (⁸⁷Sr/⁸⁶Sr up to 0.7681)³⁶. Trying to make inferences about the past circulation of people, animals, and goods between eastern and southern Africa, e.g.³⁷, it is tempting to use the highly radiogenic ⁸⁷Sr/⁸⁶Sr observed in Malawi and Zimbabwe as a relatively strong proxy for provenience. For instance, a single Late Pleistocene ostrich eggshell bead from Sehonghong rock shelter in Lesotho yielded a ⁸⁷Sr/⁸⁶Sr of 0.7555 (ref. ³⁴) and may have been imported from southeastern-central Africa. However, it is likely that the recorded ⁸⁷Sr/⁸⁶Sr variation across sub-Saharan Africa is biased by research being concentrated around famous paleoanthropological localities and future work will demonstrate more widespread equifinality in radiogenic Sr ratios, as predicted by the continental isoscape³¹. For instance, modern land snails from the Ntinini Valley in KwaZulu-Natal yielded unexpectedly high ⁸⁷Sr/⁸⁶Sr ranging from 0.7245 to 0.7843 (ref. ³⁸). Further assessment of bioavailable Sr ratios will likely expose new areas with elevated ⁸⁷Sr/⁸⁶Sr.

Our spatial assignments were successful in revealing some patterns in the distribution of hunting territories of the foragers who inhabited the Kasitu Valley. First, Sr isotopes support the common assumption, e.g.^39,40, that small game (here, small bovids) in archaeological assemblages represents short-distance hunting in the immediate vicinity of a site. Larger game (equids, suids, large antelopes) was likely hunted over a broader area, but we do not have evidence of long-distance hunting (i.e., extending beyond the confines of our study area). However, it is possible that some individuals were hunted further away, but their teeth were not transported back to the site. The LGM zebras in our sample were restricted to riparian habitats, while the postglacial and Holocene individuals ranged in the Afromontane grasslands further from rivers. This is consistent with oxygen isotopic evidence from the Kasitu Valley showing that the LGM was arid (Bertacchi et al., in review), thus forcing grazers closer to permanent water sources, while the Holocene was more humid, thus enabling herbivores to range more broadly. Alcelaphins are also bound to riparian environments, as these fresh grass grazers need high-quality forage and would not have survived on sparse, nutrient-poor grasses in the understory of the open woodland (see also Supplementary Text 9). Other large bovids, bushpig, white rhinoceros, and Holocene warthog also point at the importance of riparian habitats as focal points for ungulate hunting. A single 14–10 ka BP warthog may be an outlier in that it seems to originate from the Afromontane grassland at a time when aridity would have favored staying closer to the rivers. However, this individual is not directly dated and could have an age closer to 10 ka, or it could have occupied a small riparian pocket close to MAZ-1.

Today, plains zebra and blue wildebeest migrate long distances in the Serengeti-Mara⁴¹, the Kalahari⁴², and other areas in eastern and southern Africa^43,44, but we do not find evidence of migration at any scale in our sample. At most, some of the individuals we sampled show some non-random variation in ⁸⁷Sr/⁸⁶Sr that is consistent with seasonal movement within the study area. For instance, a single zebra with relatively large intratooth variation in ⁸⁷Sr/⁸⁶Sr could have moved seasonally between the upper Kasitu and another undetermined riparian corridor. We cannot determine with the available data why migratory behavior was absent. This could be due to a lack of necessity, with not enough competition to favor long-distance movement. Alternatively, grassland habitats could have been too discontinuous and fragmented to allow obligate grazers to migrate. A recent ⁸⁷Sr/⁸⁶Sr study of Kenyan ungulates 115–12 ka similarly found little evidence of migratory behavior among plains zebra and extant blue wildebeest¹⁷. Multi-isotopic studies of faunas 160–40 ka from the southern Cape similarly could not prove migratory behavior^16,45. While ours and these two studies are separated by thousands of km and cover different time spans, the evidence as it stands today suggests that Pleistocene African foragers were potentially less mobile than previously thought⁴⁶, exploited a broad resource base including small game^47,48,49,50 and were not specifically focused on migrating herds⁵¹. A zooarchaeological study of HOR-1 found broad-spectrum prey selection focusing on small taxa and limited evidence of selective transport of large bovids, both consistent with short-distance hunting⁴.

In conclusion, we cannot falsify the hypothesis that large game was available year-round “locally”, i.e., within 25–30 km of the sites. Late LGM and postglacial foragers in Malawi were less mobile than previously thought, practicing most of their subsistence activities at the scale of tens, not hundreds, of km. This spatial scale is consistent with the restricted size of the social and mating networks inferred from aDNA evidence⁷. Here we hint at a potential explanatory mechanism that links the local availability of abundant and predictable resources—large ungulates congregating near water courses—to the restriction of foragers subsistence, social, and mating landscapes, as predicted by behavioral ecological models^8,46,52. Future work will expand the current sample of archaeological faunal remains to include periods older than the LGM and clarify the timing of the transition from long- to short-distance interactions. Our isoscape can also be used as a test for the local origin and small life-time mobility of post-LGM foragers suggested by aDNA studies⁷. Finally, our method can be applied to personal ornaments, particularly ostrich eggshell beads, to test whether the reduced size of the mating and subsistence network extended to other aspects of the social lives of the foragers who inhabited the Kasitu Valley.

Methods

We collected a total of 196 plant samples in the study area (Fig. 1 and Supplementary Text 4), including non-riparian localities (n = 135), non-dambo riparian localities (n = 33) and dambos (n = 28) (see Supplementary Data for plant taxa and details of the localities). Twelve unbroken shells of giant land snails of the family Achatinidae were opportunistically collected, all from non-riparian localities. Plant samples were air-dried in the field and shipped with shells to the Environmental Isotope Paleobiogeochemistry Laboratory of the University of Illinois in Urbana-Champaign (UIUC), where they were prepared as described in Supplementary Text 4. Elemental concentrations of Ca, Sr, Rb, and Ba were measured on dilute aliquots using an iCAP Q Inductively Coupled Plasma-Mass Spectrometer (ICP-MS) (Thermo Fisher) in the Department of Geology at UIUC. Sr was eluted from aliquots of stock solution by cation exchange chromatography using Sr-spec resin (Eichrom) and ⁸⁷Sr/⁸⁶Sr was measured using a Nu Instruments High Resolution Multi-Collector Inductively-Coupled-Plasma Mass-Spectrometer (MC-ICP-MS) as described in Supplementary Text 4. Duplicate measurements (n = 6) indicate a measurement precision of ±0.00003.

Fossil teeth were identified by visual comparison with reference specimens from the Yale Peabody Museum of Natural History (CT, USA), as well as published and unpublished atlases. We selected enamel samples as a subset of the samples analyzed for stable carbon and oxygen isotopes and reported in detail in Bertacchi et al. (in review). After the study, fossil specimens were deposited in the National Repository in Nguludi (near Blantyre, Malawi). In total, we analyzed 166 samples from 45 teeth (19 teeth sequentially sampled, 26 teeth bulk sampled from the total length of available enamel). Time-bin assignment (Table 1) is based on direct-dates (Bertacchi et al., in review) and stratigraphic association with dated charcoal, ostrich eggshell and land snail shell^7,53,54. Further details are in Supplementary Text 5 and Supplementary Data. Enamel aliquots (~5 mg) were processed either at the Yale Metal Geochemistry Center (YMGC) or in the Department of Earth Science and Environmental Change at UIUC, as detailed in Supplementary Text 5. ⁸⁷Sr/⁸⁶Sr was measured either on a Thermo Neptune MC-ICP-MS at YMGC or on a Nu Instruments High Resolution MC-ICP-MS at UIUC. Measurement precision was estimated at ±0.00006 and ±0.000025, respectively (details in Supplementary Text 5).

EBK was computed in ArcGIS Pro v. 3.2.1 using a power variogram, splitting of the entire study area in 250 cells and allowing 10–15 neighbors per cell. Known values having the same coordinates (e.g., 1 m leaves and 2 m leaves from the same plant) were averaged, since the differences are mostly small (Fig. 2). The process was iterated 10,000 times, each time with a subset of 100 measurements. The standard error of the prediction was estimated through a leave-one-out procedure. We generated probabilistic spatial assignments of archaeofaunal remains using the package “assignR” v. 2.4.0 (ref. ⁵⁵) in R v. 4.3.1 (ref. ⁵⁶). For details, see Supplementary Text 6.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.