As the world warms, the Earth system moves towards a climate state without societal precedent. This challenges future predictions, as climate models need to be tested and calibrated with real-world data.  One strategy for assessing the effects of greenhouse forcing on climate is to turn to Earth’s past, where changes in climate are stored in the form of sedimentological, paleontological, isotopic, and geochemical proxies. We compile global proxy data across multiple global warming events of the Early Paleogene (66-47.8 Ma) – a period of extreme warmth suggested as a possible analogue for worst-case scenarios of future global warming. We take a novel, multi-proxy approach where we include information on precipitation intermittency and intensity, and integrate the proxies into climate types. The data show surprising hydrological shifts that started well before and persisted well beyond the Paleocene-Eocene Thermal Maximum – the warmest period of the Cenozoic Era and the focus of much of the prior research. We provide a glimpse of an extremely warm world with ever-wet or monsoonal conditions in the polar regions and aridity interrupted by extreme rainfall at mid-latitude continental interiors. We show that extremely warm climates may induce (1) non-linearities in the hydrological cycle’s sensitivity to temperature increase, (2) departures from the wet-gets-wetter and dry-gets-drier response, and (3) a decoupling of mean annual precipitation from precipitation intensity and intermittency. Our results are inconsistent with vegetation boundary conditions and precipitation outcomes from state-of-the-art paleoclimate modelling. Our focus on precipitation intermittency and intensity provides a new perspective on future trends in climate under extremely warm conditions.

This dataset was collected by compiling published literature on terrestrial temperature and precipitation during the Early Paleogene (66-47.8 Ma). Mean annual precipitation (MAP) and temperature (MAT) values were used to assign a climate type with a distinct range of MAP and MAT values. Where quantitative information was not available, qualitative proxies were assigned a range of potential MAT and MAP values based on comparison to modern environments. For a more detailed explanation of the methods, please see the published article's supplementary information.