Reduced water loss rather than increased photosynthesis controls CO-enhanced water-use efficiency.

Numerous leaf-level experiments suggest that plant intrinsic water-use efficiency (iWUE) increases under elevated CO because of reduced stomatal conductance and enhanced photosynthesis. However, it remains elusive whether this response can be extrapolated to the ecosystem scale, because confounding factors and compensating feedbacks are often involved in ecosystem iWUE variations. Here we develop a machine learning-based framework to disentangle the ecosystem-scale CO effects on iWUE and its two components, canopy conductance (Gc) and gross primary productivity (GPP), based on global networks of long-term eddy covariance observations. Our results show widespread CO-induced enhancement of iWUE across diverse ecosystems, driven predominantly by Gc reduction rather than GPP stimulation. Moreover, three divergent response types are identified across the studied ecosystems, based on the strength and significance of CO-driven Gc reduction and GPP enhancement, indicating spatially non-uniform responses to rising CO. Nutrient supply, water availability and biome types are found to be critical factors regulating this spatial heterogeneity. Overall, our study provides observational insights into ecosystem-scale CO fertilization effects. Such understandings are essential to inform terrestrial biosphere models for better projections of carbon and water cycles given the intensified changing climate in a CO-rich future.