Reconstructing climate dynamics in the Cradle of Humankind, South Africa, using machine learning and mammalian fossil assemblages.

International audience Fossil fauna recovered from palaeoanthropological sites can be used to reconstruct past climatic and environmental conditions when other traditional palaeoclimate proxies such as pollens, diatoms, lacustrine or marine sediments are not available. These reconstructions can help explore if climate-induced dynamics have shaped hominin diversification and influenced the development of new behaviors. Numerous palaeoenvironmental reconstruction methods have been developed over the years, and their application to the fossil faunas of the Cradle of Humankind (COH), South Africa, has produced diverse, sometimes contradictory results according to the methodology and fauna used. To investigate climate evolution over time within a unified methodological framework, we developed a novel palaeoclimatic reconstruction method based on random forest models, using the presence/absence data of rodents and bovids —two clades commonly employed in palaeoenvironmental reconstructions— as climate proxies. The application of artificial intelligence to the fossil record necessitates methodological adaptations to account for its inherent incompleteness and the absence of direct modern analogues for past faunal communities, which we discuss here. With this approach, we reconstructed evolution of temperatures and precipitation over the last 4 million years in the COH. Our reconstructions indicate relatively stable environmental conditions, closely resembling those of the present-day Sterkfontein Valley. We find no support for a shift from closed, humid, woodland-dominated environments to more open, arid, grassland-dominated landscapes over time, as is frequently proposed in the palaeoecological literature. Instead, the variability in reconstructed climatic conditions across different sites reflects fluctuations in faunal community composition, whose temporal resolution may limit direct correlations with long-term palaeoclimatic records from southern African marine archives.