Evaluating refugia in recent human evolution in Africa: Supplementary information

Homo sapiens have adapted to an incredible diversity of habitats around the globe. This capacity to adapt to different landscapes is clearly expressed within Africa, with Late Pleistocene Homo sapiens populations occupying savannahs, woodlands, coastlines and mountainous terrain. As the only area of the world where Homo sapiens have clearly persisted through multiple glacial-interglacial cycles, Africa is the only continent where classic refugia models can be formulated and tested to examine and describe changing patterns of past distributions and human phylogeographies. The potential role of refugia has frequently been acknowledged in the Late Pleistocene palaeoanthropological literature, yet explicit identification of potential refugia has been limited by the patchy nature of palaeoenvironmental and archaeological records, and the low temporal resolution of climate or ecological models. Here, we apply potential climatic thresholds on human habitation, rooted in ethnographic studies, in combination with high resolution model datasets for precipitation and biome distributions to identify persistent refugia spanning the Late Pleistocene (130-10 thousand years ago). We present two alternate models suggesting that between 27-66% of Africa may have provided refugia to Late Pleistocene human populations, and examine variability in precipitation, biome, and ecotone distributions within these refugial zones.

Methods
Analytical Code: All data used are available from open access online portals (e.g. OSF) or directly from R packages listed in the code.

Table 1: All data are summarised from the referenced literature. The upper and lower age estimates derive from the confidence interval of chronometric dating, with the mid-age used for analysis calculated as a simple average of the two. For individual dates, the mid-age matches the central age estimate. Where an archaeological assemblage is constrained by two or more dates, the upper-most and lower-most confidence interval are used, integrating assessment from original reports.