| 04 Mar 2026
Reviews and syntheses: Eddy covariance-based evapotranspiration partitioning
Abstract. The task of reliably partitioning evapotranspiration (ET) is imperative so that we can better understand how individual components of the terrestrial water flux are contributing to the global hydrological cycle and changing under a warming climate. By constraining how evaporation (E) and transpiration (T) separately adapt to increased global temperatures, we can make more accurate predictions in land surface models, further our understanding of plant water use, and better manage our limited water resources. Eddy covariance (EC) is a globally used technique that measures net biosphere-atmosphere fluxes, including ET, and if reliably partitioned, presents a promising way to constrain E and T values and trends across ecosystems. Several EC-based ET partitioning methods exist, and there is a need for an updated comprehensive guide to the available approaches. This systematic literature review was conducted with the objectives of 1) identifying EC-based ET partitioning methods and categorizing them based on underlying ecosystem assumptions, 2) determining the main advantages and disadvantages of each method dependent on their assumptions and data requirements, and 3) evaluating how broadly these methods have been tested based on geographic location and ecosystem type. The review identified 10 independent partitioning methods tested across 123 studies. Methods using assumptions of underlying water use efficiency (uWUE) and ecosystem conductance all use the relationship between ET and gross primary production with vapor pressure deficit (VPD) to estimate the transpiration ratio (T/ET). Additionally, two machine learning based methods, one method assuming a linear relationship between ET and gross ecosystem photosynthesis, and three methods using high frequency EC data to estimate T/ET were identified. The uWUE methods, while the most frequently used partitioning approach, consistently predicted the lowest T/ET estimates when compared to both other EC and non-EC based partitioning methods. The machine learning methods predicted the highest T/ET values compared to other EC-based methods which agreed well with values estimated with independent methods. Savannas and evergreen broadleaf forests had the highest T/ET of all ecosystem types while deserts and wetlands had the lowest. Leaf area index and soil water content were found to be the most important drivers of T/ET values and trends with VPD and air temperature also displaying significant effects. Of the global studies identified in this review, an average annual T/ET value of 0.58 was found, a value that falls within the range of other studies using isotopic partitioning, remote sensing methods, and global estimates from various ecosystem models. More testing, specifically increased paired analyses of two or more EC-based ET partitioning methods on the same dataset, is needed in order to fully understand the applicability of each method, their differences, and to better constrain global T/ET dynamics.
Referee Comment— Cochran et al.
Summary and Overall Assessment
Cochran et al. present a timely and useful review of eddy‑covariance (EC)–based methods for partitioning ecosystem evapotranspiration (ET) into evaporation (E) and transpiration (T). Because EC is widely regarded as the most direct approach for quantifying ET at the ecosystem scale, a synthesis focused on EC‑based partitioning methods is both important and valuable. The manuscript summarizes roughly ten methods developed over the past two decades and classifies them by theoretical assumptions and data requirements. This organizational framework is a strong element and will be very helpful to readers.
The discussion is less strong, particularly where it aims to generalize across method classes and draw broader conclusions. In its current form, the manuscript would benefit from clearer articulation of conceptual links among method families, earlier and more consistent treatment of ecological drivers (e.g., SWC, LAI), and a more definitive set of take‑home messages and recommended directions for future work.
Overall, this review addresses an important need and has strong potential. The specific comments below are intended to strengthen clarity, rigor, and practical usefulness.
Major Comments
Please provide a more explicit introduction to EC and why it is unique relative to other ET estimation approaches. Many methods that separately estimate E and T rely heavily on modeling assumptions, whereas EC directly constrains total ET (the sum of E and T). Framing the motivation this way will help readers understand why an EC‑based synthesis is needed as opposed to getting the values “cleanly” from LSM’s or remote-sensing based algorithms.
Rising temperatures directly influence atmospheric demand and ET, but also indirectly through changes in LAI, plant functional type composition, and phenology.
LAI is a primary explanatory variable for T/ET variability across many studies, yet it appears late in the manuscript. Consider integrating LAI into the background/motivation and method summaries, so ecological context and methodological choices are connected from the outset.
The manuscript frequently uses “interception evaporation,” which specifically refers to water evaporated from the wet canopy. In many contexts you likely mean evaporation more broadly (including wet leaves, rocks, and moist soil layers). Adopting the broader term where appropriate will reduce confusion.
Please define “high‑frequency” early and clarify why such data are not frequently available. Noting that 10 Hz data represent ~18,000 samples per half hour will help readers understand storage/processing constraints and why FluxNet typically provides aggregated half‑hourly records. Explain how one would obtain high-frequency data.
In many instances, methods are applied to estimate T/ET rather than tested against independent observations or via cross‑method comparisons. Replacing “tested” with “applied” where appropriate would improve accuracy and avoid implying validation that may not have occurred.
There are clear relationships among the regression‑type, uWUE, and physically‑based methods. For example, uWUE methods can be viewed as special cases of linear regression; TEA18 is physically based but follows the same core T = GPP/WUE structure; many ML approaches embed VPD information and use similar screening/upper‑envelope logic (e.g., 75th vs. 95th percentile). Making these connections explicit could help unify the review a bit.
A practical addition would be qualitative guidance on where each method family tends to perform best (e.g., forests vs. grasslands, humid vs. arid sites, seasonal snow, low‑stature canopies). Even high‑level recommendations would be valuable for practitioners.
Consider reframing the Conclusions as “Conclusions and Recommended Future Directions.” This review is well positioned to identify priorities (e.g., validation datasets, better access to high‑frequency data, harmonized screening protocols, sensitivity to WUE assumptions), and to lay out what the community should do next.
Minor Comments
Moving Figure 2 earlier may make the manuscript more compelling. In Figure 1 (which is overall quite nice), please clarify “records excluded” (row 2) and “reports not retrieved” (row 3).
The point that systematic GPP bias may not strongly affect T/ET in some methods is excellent.
Define “domain knowledge” where used (especially in ML contexts), and ensure consistency across the manuscript. Is this ML jargon or an ecological category?
Line‑Specific Comments
AI Disclosure: Microsoft CoPilot was asked to review the text above to reduce redundancies and provide standardized formatting. All content was written by the reviewer.