Global Climate and Vegetation Controls on Evapotranspiration Change Revealed by Budyko-informed Machine Learning

Evapotranspiration (ET) is a central component of the water and energy cycle, jointly affected by climate and vegetation. Under ongoing climate change and large-scale greening, a key challenge is to quantify how much of observed ET change is driven by vegetation, rather than by co-varying precipitation and atmospheric demand. Strong multicollinearity among climate, ET, and leaf area index (LAI) in observations hinders robust estimation of ET’s sensitivity to vegetation. Even though the classical Budyko framework can separate climatic and vegetation controls, it relies on fixed functional forms, limiting its ability to represent the strongly state-dependent and region-specific ET responses across climate and vegetation regimes. Therefore, we incorporate the Budyko framework within a data-driven model. This enables us to disentangle the influence of climate and LAI on ET, by combining global observations and established water-energy balance constraints. Using local derivatives and counterfactual experiments, we estimate the sensitivities of ET to climatic factors and LAI, and decompose ET changes into contributions from climate and vegetation. Our results show that ET sensitivity to LAI increases and then decreases with rising LAI, peaking in transitional regimes where neither water nor energy fully dominate. At the global scale, climatic contributions to ET change are spatially diverse, whereas LAI increases almost everywhere enhance ET. Even though climatic effects are typically stronger locally, their opposing signs across regions cancel out when aggregated globally. Therefore, during 2001-2020, the global ET increases correspond mainly to LAI trends. By leveraging this Budyko-informed model, which reduces the influence of multicollinearity, we can obtain a more robust separation of climatic and vegetation drivers of ET. These findings highlight the dynamic role of vegetation in regulating terrestrial water loss, which directly affects how reforestation and greening impacts on water resources are interpreted.