Species distribution models fail to predict paleozoological occurrences during the Holocene Green Sahara phase

Paleoclimatic simulations are powerful tools to investigate past faunal biogeographical patterns, but they can fail to capture complex climatic conditions at specific regional or temporal scales. Here we show that species distribution models (SDMs) do not predict the expansion of suitable habitats for mammals that were present in the Sahara during the African Humid Period (AHP) according to radiocarbon-dated paleozoological records. We illustrate this issue by modeling the current and past distribution of the hartebeest (Alcelaphus buselaphus), a typical African savanna antelope with a wide Sub-Saharan distribution. Its Holocene paleozoological record shows that its distribution during the AHP included large areas of the Sahara and the northern African Mediterranean coast, from Morocco to Egypt and the Levant. We use Bayesian additive regression trees (BARTs) with an MCMC algorithm in combination with current climate and occurrence data to generate posterior distributions of habitat suitability, evaluate variable importance, and generate variable partial-dependence plots. From these, we learn that annual precipitation is the most important climatic variable determining the hartebeest’s current distribution. We then projected habitat suitability onto various paleoclimatic scenarios during the AHP and found that the estimated precipitation did not reach the minimum required for the viability of hartebeest populations. These results highlight the potential of the fossil record to test the regional precision of paleoclimatic simulations, ultimately helping to generate more realistic past environmental scenarios.