Biodiversity varies predictably with environmental energy around the globe, but the underlying mechanisms remain incompletely understood. The evolutionary speed hypothesis predicts that environmental energy shapes variation in speciation rates through temperature- or life history-dependent rates of evolution. To test whether variation in evolutionary speed can explain the relationship between energy and biodiversity in birds, mammals, amphibians, and reptiles, we simulated diversification over 65 million years of geological and climatic change with a spatially explicit eco-evolutionary simulation model. We modeled four distinct evolutionary scenarios in which speciation-completion rates were dependent on temperature (M1), life history (M2), temperature and life history (M3), or were independent of temperature and life-history (M0). To assess the agreement between simulated and empirical data, we performed model selection by fitting supervised machine learning models to multidimensional biodiversity patterns. We show that a model with temperature-dependent rates of speciation (M1) consistently had the strongest support. In contrast to statistical inferences, which showed no general relationships between temperature and speciation rates in tetrapods, we demonstrate how process-based modeling can disentangle the causes behind empirical biodiversity patterns. Our study highlights how environmental energy has played a fundamental role in the evolution of biodiversity over deep time.

These data were generated to investigate the role of temperature- and body size-dependent rates of speciation on global biodiversity patterns in tetrapods (birds, mammals, amphibians, and reptiles). Using large, published datasets of spatial, phylogenetic, and body size data, we estimated 54 metrics that summarise different kinds of biodiversity patterns. We used a spatially-explicit simulation model to generate spatial, phylogenetic, and body size data under four alternative models of diversification in which speciation-completion rates were dependent on temperature (M1), life history (M2), temperature and life history (M3), or were independent of temperature and life-history (M0). The simulation model used published paleo-reconstructions of temperature and aridity over the past 200 Ma as input. From the simulated spatial, phylogenetic, and body size data, we estimated the same set of 54 biodiversity metrics. To assess the agreement between simulated and empirical data, we performed model selection by fitting supervised machine learning models to multidimensional biodiversity patterns.