Divergent evapotranspiration partitioning across observations and modeling is dominated by transpiration

Partitioning terrestrial evapotranspiration (ET) into its physical (evaporation, E) and biological (transpiration, T) components is a grand challenge in Earth system science. Wide variation in partitioning methods and their applications, both in space and over time, has previously hindered rigorous cross-comparisons, and uncertainty persists about how E and T components should be determined. This research provides a comprehensive and data-driven synthesis of five principal evapotranspiration partitioning approaches, including stable isotopes, high-frequency turbulence measurements, remote sensing techniques, biogeophysical models, and land surface models, by leveraging standardized, continental-scale datasets from the National Ecological Observatory Network (NEON). Across NEON observations, significant, mechanistically driven discrepancies are evident: isotope methods yield the highest transpiration-to-evapotranspiration ratios (T/ET), while process-based models yield the lowest. High-frequency turbulence and remote sensing methods yield relatively consistent estimates of T/ET. These divergences are amplified systematically in transpiration-dominated ecosystems and during extreme climate conditions. Methodological discrepancies arise from fundamental assumptions and structural limitations of each approach and challenge the interchangeability of T/ET estimates. Broadly, this assessment suggests that Earth system models may be systematically underestimating the role of transpiration in regulating land-atmosphere feedbacks. Our finding provides a foundational benchmark and a mechanistic framework to guide future model development, reconcile methodological divergences, and advance our predictive understanding of the Earth system.