Global Runoff Partitioning Based on Budyko‐Constrained Machine Learning

Abstract Understanding the partitioning of runoff into baseflow and quickflow is crucial for informed decision‐making in water resource management, guiding the implementation of flood mitigation strategies, and enhancing drought resilience measures. Methods that combine the physically based models with machine learning (ML) have demonstrated potential for global runoff estimation. However, such “hybrid” approaches remain unexplored for baseflow estimation. Here, we develop a ML approach combined by the physically‐based Budyko framework for baseflow estimation by incorporating the baseflow coefficient (BFC) curve as a physical constraint. Parameters of the original Budyko curve and the newly developed BFC curve are estimated based on 13 climatic and physiographic properties using boosted regression trees. BRT models are trained and tested in 1,461 catchments worldwide and subsequently applied to the entire global land surface at a 0.25° grid scale. The models exhibit strong performance during the testing phase, with R 2 values of 0.96 and 0.91 for runoff and baseflow, respectively. Results indicate that, on average, 35.3% of global continental precipitation (819 mm yr −1 ) is partitioned into runoff (292 mm yr −1 ), comprising 19.6% as baseflow (162 mm yr −1 ) and 15.7% as quickflow (130 mm yr −1 ). Among the 13 climatic and physiographic properties analyzed, vegetation properties emerge as primary controls for Budyko parameter α while soil properties dominate parameter Q b , p in BFC curve, although significant spatial variability is observed across global catchments. Overall, the proposed framework provides a global data set and methodology for runoff partitioning while revealing how catchment properties control this process.