Peatlands and their fossilized counterpart, coal, are key indicators of past and present climate. However, tools for predicting their potential global distribution in the geological past remain limited. Here we use machine learning to build a climate-driven peatland distribution model, PeatClim v1.0, and to identify key climatic controls on peatland formation. The model is trained on bioclimatic variables in regions of modern peatland occurrence, aiming to estimate potential peatland distributions, rather than to reproduce observed maps. Results show that partitioning the global peatland dataset into low- and high-temperature subsets and training them separately improves model predictive performance and aligns better with observations. Diagnostic analysis reveals distinct dominant climatic controls for the two subsets: low-temperature peatlands (northern peatlands) are mainly controlled by annual temperature range, whereas high-temperature peatlands (tropical peatlands) are primarily controlled by annual precipitation. PeatClim v1.0 is designed for use with palaeoclimate model outputs, facilitating the prediction of potential coal deposits in Earth’s history and palaeoclimate model-performance evaluation.