Using high-resolution modelled climatic data, we have conducted more sophisticated analyses of the environments inhabited by MSA-making populations than was previously possible, bridging former disconnects between archaeological and climatic records within a spatiotemporally explicit framework. Figure 1 highlights the challenge of generalising from proxy core records which we have overcome through the parameterisation of temperature and precipitation with simulated models, revealing interpretable patterns in terms of MSA assemblage environments. Our analyses have extended previous work applying models of the LGM and LIG to represent climatic extremes [10,11] and importantly have revealed the climatic bracket of eastern African MSA occupations to around 9–25°C mean annual temperature and 396-1593mm total annual precipitation. MSA assemblages are more prevalent during interglacial phases (MIS 3, 5, and 7) and we found that populations during these periods inhabited more diverse and widely distributed environments. Most assemblages are located within or with access to tropical xerophytic shrubland, and usually are found within proximity to at least one forest or woodland biome. Using spatiotemporally explicit models we also established that precipitation levels experienced by MSA populations had an impact on both behavioural variability and the use of raw materials, which we explore further here using penalised logistic regression.
Our results highlight spatial and diachronic trends in the environments inhabited by eastern African MSA populations, which are demonstrated by Fig. 1, 3 and Supplementary Table S1. MIS 9 and 8 saw occupations that were typical for the eastern African MSA in terms of temperature, precipitation, altitude, and biomes inhabited. During MIS 7, there was a dramatic increase in the environments occupied, spanning much of the climatic range of the later interglacial phases with emphasis on forest/woodland ecotones and tropical xerophytic shrubland. MIS 6 saw the occupation of a cold, high-altitude environment; however, the phase is only represented by a single assemblage, Marmonet Drift_H2, with a large date range so we do not place too much emphasis on its mid-age climatic placement. MIS 5 was marked by another phase of expansion, though with limited presence in woodland habitats and typically warmer, drier and low altitude occupations. MIS 5 has been proposed to represent a transitional point within the MSA stone tool record, due to the augmentation of MSA toolkits with new combinations of stone tools and the colonisation of different landscapes into the Late Pleistocene [9,10,13]. Engagement with these new environments likely occurred at the edge of the logistical landscape, with many eastern African MSA assemblages falling on ecotonal boundaries between tropical xerophytic shrubland and previously unoccupied biome types. An example of this is found with the youngest MSA occupation at Panga Ya Saidi, which is placed at a unique tropical ecotone . MSA occupation of Panga Ya Saidi at the end of MIS 5 is rapidly followed by cultural change, with the appearance of the Later Stone Age here occurring earlier than anywhere else in the region . New constellations of tools within MIS 5 toolkits suggest corresponding changes in behaviour previously unseen in the MSA , which perhaps may have been key for these expansions beyond tropical xerophytic shrubland. Together, this suggests that certain MSA toolkits may have led to adaptation, as opposed to adaptations being a pre-requisite for expansion, with transient ecotonal habitats requiring the adoption of a range of survival strategies and potentially mediating interaction between different groups across different environmental contexts [2,28,29]. Assemblages dated to MIS 4 show a climatic range close to that of MIS 3, where we see a push into higher precipitation woodland habitats with a limited change in temperature and altitude. Habitability modelling for MIS 3 suggests that temperature, rather than precipitation as seen during earlier interglacial periods, constrained expansions into new landscapes. Assemblages across the Middle to Late Pleistocene exhibited comparable levels of temperature variability (MIS 3 = 9–19°C, MIS 5 = 13–25°C, MIS 7 = 12–23°C), however MIS 3 saw a notable extension of the habitable precipitation regime and the occupation of both more arid and humid environments, though this was limited to within the MSA temperature range.
Eastern African MSA assemblages are primarily located within tropical xerophytic shrubland, characterised by arid-adapted species such as that from the Acacia and Commiphora genera. Whilst it is the most abundant biome across modern eastern Africa, and likely also so during the Pleistocene, the intense occupation of tropical xerophytic shrubland and its associated ecotones suggests that eastern African MSA adaptive strategies were centred around engaging with subsistence resources associated with this biome. Our results also confirm the importance of wooded ecotones for sustaining MSA populations , with the most common ecotone occurring between temperate conifer forest and tropical xerophytic shrubland, distributed widely across the region though most concentrated towards the eastern edge of Lake Victoria and the associated region, where most assemblages occupy ecotones between tropical xerophytic shrubland and various forested/woodland biomes. Today, Lake Victoria sits on the junction between central African rainforests and savanna habitats to the east, forming an important boundary for large mammal  and human  populations. Our phased models (Fig. 5) propose that the region to the east of Lake Victoria would have seen sustained and persistent occupation throughout the Middle to Late Pleistocene in the face of climatic fluctuation, implicating it as a potential refugium for hominins in eastern Africa. The availability of freshwater provided by surrounding rivers and springs  and the complex topography of the Rift Valley [25, 33] may also have helped buffer against the strongest effects of climate change in the area . Occupying refugia likely required minimal cultural adaptation to environmental change, which could explain why assemblages from within the Lake Victoria basin show distinct differences from the general eastern African MSA . Moreover, Lake Victoria, which is a relatively young lake but estimated to exist by 500 − 400 kya , is the source of the White Nile, and therefore its refugial position could also have interesting implications for dispersals from the region out of the continent . Together, this highlights the potential role of refugia in structuring MSA human cultural and biological diversity [3,4], laying the critical foundations for later human evolution.
Beyond simply characterising MSA assemblage environments and ecologies, our use of matrix correlations presents an important means to examine the extent to which climatic features influenced behaviour, here characterised by stone tool assemblage composition. Our results firstly demonstrate that precipitation has a significant, independent effect on toolkit composition; further examination via binary logistic regression shows statistically significant decreases in the probability of backed microliths, borers, centripetal technology, platform cores, and scrapers occurring in assemblages as precipitation levels increase, after controlling for all other variables in the analysis (see Supplementary Methods S2 and Supplementary Table S2). Some of these tool types have been found to be significant indicators of either Later Stone Age (backed pieces) or MSA (scrapers) toolkits , , perhaps linking an increase in precipitation tolerance to the development of the LSA, as documented at Panga Ya Saidi. Roughness also has a significant impact on toolkit compsition, with further analyses demonstrating that the probability of backed microliths and Levallois blades, points, and flakes occurring in assemblages decreases as the energetic demands of the environment increase. Further analysis of the site type variable demonstrates that it has significant effects on the probability of backed microliths, bipolar technology, and Levallois flakes appearing in assemblages, with all these technologies occurring more frequently in cave or rock shelter sites. Finally, there are numerous independent effects of the use of particular raw materials on the presence of individual technologies in the assemblages, as summarised in Supplementary Table S2. We also found precipitation has an independent effect on raw material use, alongside cost path, roughness, simple age and toolkit composition. This constellation of geographic-based variables suggests a complex pattern of spatial autocorrelation, likely grounded in the local availability of raw materials across the landscape. Overall, these results demonstrate a complex interaction between the environment and MSA behaviour, with the energetic requirements of the physical landscape and local precipitation regimes likely requiring different constellations of tools from the MSA toolkit.