Review of “Towards a global actual evapotranspiration product for the Copernicus Land Monitoring Service” (Manuscript ID: egusphere-2025-4342)

The manuscript presents a comprehensive overview of the design, development, and initial validation of a global operational ETa product set to be released by the Copernicus Land Monitoring Service (CLMS). This work addresses a gap in the current portfolio of global land surface monitoring products. The development of a 300m resolution, dekadal, open-access ETa dataset based primarily on Copernicus data will be a valuable resource for applications in hydrology, agriculture, and climate science.

The manuscript provides a detailed account of the entire processing chain, from input data pre-processing (cloud masking, biophysical trait retrieval, LST sharpening, weather forcing) through the application of two distinct ET models (TSEB-PT and ETLook) to gap-filling and validation. The decision to use an ensemble of the two models is well-supported by the results, which show improved performance over the individual models. While the manuscript describes a product of considerable importance, several major points require clarification and further analysis to strengthen the paper and ensure its full scientific rigor before it can be accepted for publication.

Major Comments

Justification of the Ensemble Approach: The results clearly show that the ensemble mean of TSEB-PT and ETLook outperforms either model individually. However, the manuscript lacks a clear physical or methodological justification for why only the two models were used? Why not include more ET models to produce a more accurate ensemble product?

EB-ET is also based on energy balance method and global product at 5 km resolution. The model can be implemented at any spatial and temporal resolution if the forcing data can be prepared. It also produced latent heat fluxes and sensible heat fluxes. So this paper should clearly specify why the EB or SEBS model were not implemented in this activity. Actually, the simple EF-gap filling method in Chen et al. 2021 can be also used to improve the temporal resolution of this study.

Chen, X., Su, Z., Ma, Y., Trigo, I. and Gentine, P., 2021. Remote sensing of global daily evapotranspiration based on a surface energy balance method and reanalysis data. J. Geophys. Res. Atmos, 126(16): e2020JD032873.

The enhanced SEBS model (ref. Chen et al, JGR 2019) can be also applied to the CLMS. The model code was also publicly available. Table 2 and 3 show that all the input data for running SEBS model were satisfied. Then please explain why it was not tested at the beginning. I think the ETa community have agreed that there are no one ETa model can satisfy the accuracy or requirement from all kinds of applications. Hereby, an ensemble of ETa products from different models are encouraged to used. CLMS has the resource to merging more datasets produced by models which can be quickly and easily merged in the system.

Line 52-57, I don`t like the analysis of current ETa gaps. There are many other ETa products which has a daily temporal resolution, which is better than dekadal resolution of this study. I don`t agree that “Other ETa datasets have either much lower spatial and temporal resolutions.”. The ETa in this paper also has the same problem. You cannot say “Other” ETa datasets….

There are more ETa products should be reviewed and compared in this study, such as EB-ET, MOD16, etc…A comparative table or schematic contrasting ETC with other methods (e.g., MOD16, GLEAM, SEBS) would strengthen the novelty claim.

I don`t agree that ETa from ETMonitor are not produced operationally. Any models can be put in the Copernicus Land Monitoring Service. This is not a proper way show the advantage of the ETa produced introduced in this study.

The calculation of soil resistance should also take into account the soil texture and soil classification. The equation 13a only consider top-soil moisture. In addition, the value of b and c were not listed in this paper. I suggest to change this equation and make the new equation can consider the soil texture. This is important for bare soil evaporation. Yuan et al. 2024 have investigated how to improve soil evaporation resistance. Please check equation 14-17 and table 1 in Yuan et al.

Yuan, L. et al., 2024. Long-term monthly 0.05° terrestrial evapotranspiration dataset (1982–2018) for the Tibetan Plateau. Earth Syst. Sci. Data, 16(2): 775-801.

Eq. 12a and 12b, how ras and rac were calculated should be also described in this paper, since the model was applied at global scale. I think some important issues about ras and rac should be explained.

Line511 reported that the bias in λE is positive in the majority of the sites and H estimates has dominant negative bias. The lower estimation of sensible heat fluxes and overestimation of latent heat fluxes at global scale has been reported by many other papers. Chen et al. 2019 has solved this issue from energy balance models by considering the canopy-air turbulent diffusion process. This paper did not introduce the method for calculating ras and rac at all. The evaluation results have demonstrated the general problem which reported before. Hereby, I suggest to use the solution of Chen et al. 2019 to update the calculation of ras and rac. Chen et al. 2019 has produced a uniform roughness and aerodynamic resistance calculation method for all kinds of canopy types. The model code was freely shared, Hereby, the aerodynamic resistance scheme should be described and evaluated, otherwise the scheme from Chen et al. 2019 was suggested to be applied in this study.

Chen, X., Massman, W.J. and Su, Z., 2019. A column canopy-air turbulent diffusion method for different canopy structures. Journal of Geophysical Research: Atmospheres, 124: 488–506.

Chen et al. 2021 has adopted an assumption of constant EF during consecutive days. If CLMS can provide daily surface radiation components for both cloudy and clear days. Then the EF gap-filling technique can be also applied to CLMS. This will help CLM to produce a daily ET product not 10 days product. The accuracy of available surface radiation is important for ETa accuracy on cloudy day. This is due to that ET on cloudy days was limited by energy not by moisture availability. That`s why the ET gap-filling method in Chen et al. 2021 can be successfully used for global daily ET calculation.

Chen, X., Su, Z., Ma, Y., Trigo, I. and Gentine, P., 2021. Remote sensing of global daily evapotranspiration based on a surface energy balance method and reanalysis data. J. Geophys. Res. Atmos, 126(16): e2020JD032873.

The significant discrepancy between TSEB-PT and ETLook in forested areas (Figs. 10, 11) is a critical finding. The discussion attributes this largely to differences in transpiration estimates. The manuscript would be strengthened by a more in-depth analysis of the potential causes. For example, how sensitive are the models to the parameterization of aerodynamic resistance (highly dependent on canopy height in TSEB-PT) versus the soil moisture stress factors (derived from an LST-fractional cover trapezoid in ETLook)? A focused discussion on the challenges of modeling forest ET, particularly for these two modeling approaches, is needed.

The poor performance in urban areas is acknowledged, but simply stating that the models were not designed for this land cover is insufficient for a global product description. The authors should briefly discuss the specific challenges (e.g., impervious surfaces, complex energy balance, irrigation of urban vegetation) and outline, even if just as a perspective for future work, how this might be addressed in a potential future product version (e.g., via a dedicated urban land cover class with adjusted parameters or a post-processing step).

Near-Real-Time (NRT) vs. Reanalysis Trade-off: The manuscript identifies the use of forecast meteorological data and one-sided (NRT) gap-filling as a key difference from the reanalysis WaPOR product, contributing to more gaps and potentially higher uncertainty (Fig. 16, Table 11). This is a fundamental design choice with clear implications for users. The authors should more explicitly discuss the trade-offs between timeliness (NRT) and accuracy/completeness (reanalysis). A quantitative estimate of the accuracy gain expected from a future reanalysis version, based on the results in Table 11, would be highly valuable for the user community.

Spatial Representativeness of Validation: While the use of 104 sites is commendable, the geographical bias towards East Asian, and Australia is a limitation (Fig. 4). The authors rightly note that some climates are represented by proxy (e.g., dry regions in Spain), but the validation remains weak in the tropics and parts of Africa and Asia. This should be explicitly stated as a limitation of the current validation. The authors should also comment on the potential impact of this bias on the reported global performance metrics.

Minor Comments

Abstract: Those data were then used to drive two evapotranspiration models: TSEB-PT and ETLook. A prototype implementation of the ETa processing chain was used to produce ETa data across a globally representative range of climatic zones and plant functional types. Two models were used but here did not inform the readers how the two models were combined to produce the ETa product. A transfer or connection needed between these two sentences.

Abstract and Introduction: The abstract could be slightly more specific about the key innovation—namely, the operational, Copernicus-based ensemble approach at 300m resolution—and its validation outcome.

Section 2.2 (Cloud-Masking and Gap-Filling): The decision not to gap-fill LST is well-argued. However, for the operational product, how will persistent cloud cover (e.g., in equatorial regions, as shown in Fig. 2) be handled? A brief comment on the expected data availability in these regions would be useful.

Section 3.2.2 (Instantaneous Fluxes): The positive bias in λE and negative bias in H for TSEB-PT (Fig. 12) suggest a potential issue with energy balance closure at the instantaneous time step, even if the daily/daily aggregate performs well. This warrants a brief discussion.

Section 4.4 (Potential Improvements): The suggestion to use a temporal running mean for biophysical parameters is interesting. The authors should clarify if this is planned for the initial operational release (end of 2025) or for a future reprocessing.

Line 112-114, In addition, LST under clouds is different to LST in clear-sky conditions and using gap-filled values can lead to energy imbalance at the land surface. Therefore, LST is usually not gap-filled, especially if it is to be used as input into ETa models. If the gap-filled LST values under clouds were not accurate, then this can cause energy imbalance. But this does not mean the gap-filled LST can not be used for ETa calculation.  These two sentences should be rephrased.

Line 214, canopy structure (LAI and Campbell (1990) leaf inclination distribution parameter).

Line 337, series resistance network (in analogy to electrical systems) which depend on aerodynamic and meteorological conditions….. meteorological conditions influence aerodynamic, one is about mesoscale, the other is about microscale. Both meteorological and aerodynamic were parallel used, they are partly overlap. Rephrase this sentence.

Line 350, can CLMS produce Sdaily? If yes, then EF gap-filling method adopted in Chen·s paper can be also used to produce a daily ETa.

Line 382-383, I can understand how did you get Ksc for cloudy days or target date, then I got lost how did you derive ETa for cloudy days or target date using continuous Ksc? Please add description on how to use daily Ksc to derive daily ETa or accumulated ETa for each 10 days.

Line 388, a simple water-balance approach indicates that the soil is wet, what kind of water balance approach can be used to indicates the soil is wet?

Figure4, there are no flux measurement collected from eastern Asian, I suggested to use the following flux measurement to verify the ETa product

Ma, Y. et al., 2020. A long-term (2005–2016) dataset of hourly integrated land–atmosphere interaction observations on the Tibetan Plateau. Earth Syst. Sci. Data, 12(4): 2937-2957.

Figure10 shows that the ET from model and measurement have high difference. How can the data users diagnosis the error source of the ET product? The paper should give some demonstration on this aspect.

Please find my review in the attached pdf. 

The article describes the procedure to develop the new ET product of the Copernicus Land Monitoring Service (CLMS), which is currently missing. The description of data inputs and the modeling chain are well described, but several points need to be addressed before the paper can be accepted.

Comments

• In the abstract the main innovation could be more explicitly included
• Temporal and spatial resolution: 300 m is not the original resolution of either the vegetation and the land surface temperature products. 10 days of temporal resolution is probably not enough to capture the variations of ET, especially in agricultural areas or for water management applications, as the authors state. If daily meteorological forcings are available, with the gap filling technique described in the paper, an almost daily product cloud be produced. A detailed explanation of these aspects is needed to identify the weakness of the product.
• The choice of the two models (TSEB-PT and ETLook) should be better justified. Both models are based on the energy balance, limiting the strength of the analyses. The research community have developed and compared many models based on different approaches (energy balance, water balance, machine learning …) agreeing that it doesn’t exist a model that clearly outperforms all the others (like the OpenET approach). A clear justification from a methodological, hypotheses and inputs data analyses is missing, when the selection of only these two models has been done.

• It’s interesting to see that the ensemble approach outperforms each of the two models. But I think the authors should justify why not more ET models based on different assumptions could not create a more accurate ensemble ET.
• Lines52-55: actually there are many other models available in literature. A more comprehensive analysis should be done, especially many models have higher temporal resolution.
• Line 59: ETmonitor is operational
• Line 60: who decided and why only two models should be considered?
• I suggest adding a section describing the proper selection of only two models based on the energy balance appraoch
• Line 188: please explain “that Ln is usually computed internally by each ET model”. How the models compute especially the incoming longwave radiation is important due to the high uncertainty. In addition, how emissivity is computed/measured?
• 12 how the aerodynamic resistances are computed?
• Equation 13.a add the reference of the soil resistance equation. A definition of the empirical parameters a,b,c, is missing. Are they defined with soil texture? Setop is a stress factor or soil moisture?
• From figure10 and 11, it seems that the biggest errors are present in the forest areas, especially big differences between the two models are found in transpiration during summer time. Probably the parametrization of aerodynamic resistance is not proper for high vegetation. In addition, it always seems that ET from ETLook is always higher than TSEB-PT, leading to an ensemble which is an average of the two.
• In general there is a long description of the errors between models/climatic conditions/vegetation, but a more detailed explanation of the models hypotheses on the obtained results is needed to understand the model accuracies.
• Line 509: The high errors in the cities should be better discussed, as both models don’t have a specific module for the urban energy balance.
• Figure13: the text in the figure is not readable
• The use of weather forcings in respect to reanalyse data should be clearly defined, and a discussion on the possible loose of accuracy in the model ET estimates discussed.

Please find my comments in the attached PDF

This paper describes the development of Copernicus Land Monitoring Service (CLMS) ETa data products and validation of the prototype product. I appreciate the authors’ transparency in documenting these choices and opening the discussion on EGUsphere. However, the paper reads more like a combination of Algorithm Theoretical Basis Document and Product Quality Assessment Document, which should be published with CLMS data products anyway. Therefore, I’m questioning the scientific significance of this paper.  

General comments

1. I suggest the authors focus on specific research questions, rather than merely presenting the paper as “describing the design choices”. Some lengthy details could be included in Annex to keep the focus on such research questions. For example, the paper could focus on investigating

• Why the average of two models perform better than individual model?
• Where and when the ensemble average neutralizes biases of two models?
• What are the uncertainties and gaps in input data and processing that could influence model performance?

1. The paper lacks a clear definition of ETa. You referred to Bojinski et al. (2014) for the definition of ECVs. However, in that article, “water use” was mentioned instead of “actual evapotranspiration”. Table 1 shows that CLMS products do not include evaporation from canopy interception. Since interception contribution to total evaporation is substantial (even more than soil evaporation in many cases) (Savenije, 2004), and very an important process to hydrology (Dingman, 2015), I think it should not be neglected by CLMS.
2. Since canopy transpiration [mm/day] and soil evaporation [mm/day] are also CLMS products, effort should be made to evaluate their quality. You mentioned in L407 that these could not be contrasted against in situ However, there are references that could be used as proxy, e.g., partition of Eddy Covariance measurements (Nelson et al., 2020), and recently SAPFLUXNET (Poyatos et al., 2016). The ETLook model overestimates transpiration (compared to TSEB-PT) created large bias in ETa for specific cases, as you showed. The reason could also be that ETa from EC tower is underestimated due to energy balance closure (which was not discussed). In any case, investigating the accuracy of both models in transpiration estimation could add significant values in the discussion.
3. The validation of the prototype is most limited in the tropical climate (Figure 5). The validation results also show lower performance than other regions. Meanwhile, the number of cloud-free images are lowest in this climate (Figure 2), even limiting eastern swath does not change much this fact. This points to a serious gap in a global data product like CLMS, which should be highlighted.

Specific comments

1. L14 “overall best” => “average” (of all site-dekads)
2. L16 which are the similar global ETa dataset mentioned?
3. L21 What is the definition of “operational” in this context? Does it mean “near-real time monitoring” or something else?
4. L45-47 Could you provide some references?
5. L48 What are the other global CLMS products? What are their spatial and temporal resolution?
6. L48-L59 This paragraph seems to justify the spatial and temporal resolution of CLMS (300m, 10-d). If continuing WaPORv3L1 is the main factor, it should be expressed more explicitly. Other products can be presented as reference for state-of-the-art.
7. L60 What are these preparatory activities? Who recommended two ET modelling frameworks? Why were these two selected?
8. L66 Related to my first general comment, if the paper only “aims to give an overview of the design choices”, it could be just a technical documentation. To make a clearer scientific significance, this paper should focus on investigating further the two model frameworks (L61).
9. Section 2. I would start with 2.8 first to provide an introduction about the two modelling frameworks, clearly indicate which input variables are required for both/each framework. Then continue with 2.1 and link required variables with input data sources in Table 2, indicating which one is used in which model.
10. Table 2. How were 100m wind speed, air temperature, and water vapour pressure converted to near surface level (e.g., 2m is commonly used).
11. L81 Is this in the mandate of CLMS?
12. If Copernicus data product should be based on Copernicus data, why the canopy height map was derived from a static map by ETH Zurich and not derived from other Copernicus data dynamically (e.g., Sentinel-2)?
13. Table 3. The native resolution of these input data sources should be included.
14. L104 missing reference for semi-Bayesian approach
15. L108 What if LAI changes in case of crop harvesting, forest fire?
16. L109 Can the gaps in LAI be suitable for filling in areas where it is cloudy most of the year?
17. L112 needs reference
18. L185 needs references. Not clear to what degree of accuracy, and from which EO data.
19. L187 How do these models differ in Ln calculation?
20. L2555 needs reference for “most approaches”
21. L269 needs reference
22. L276 Is this statement based on sensitivity analysis? I would expect aspect and slope influence shaded relief (related to shadow effect, intensity of sunlight exposure), which influence LST. (e.g., Peng et al., 2020)
23. L291 Only topographic correction?
24. L313 resampled by averaging?
25. L398 How close is the closest non-gap-filled dates? Do you apply a threshold for when the date is too far?
26. L416 Although it is mentioned here that you searched for more diversity in the reference datasets, only Eddy Covariance methods are considered. Meanwhile, many other references have been used to (qualitatively) validate remotely sensed ET, especially in regions lacking Eddy Covariance measurements (Tran et al., 2023)
27. For example, a large area in Figure 4 has no stations at all. These areas would require alternative effort and methodology to evaluate CLMS data products. For examples, see de Andrade et al. (2024) for South America, see Weerasinghe et al. (2020) and Blatchford  et al., (2020) for Africa, see Athira  et al. (2025) for India, see He et al. (2025) Southeast Asia, see Cogill et al. (2025) for South Africa. I think at least, the authors should refer to these studies for future validation of CLMS data products.
28. L440 Which quality flag was excluded? And what was considered realistic? Based on what?
29. L446 What I missed here is whether ETa values were extracted at the 300-m pixel containing the station or flux footprint was considered. The mismatch in spatial support although not affect overall results but would affect some cases where the flux site is located in heterogenous area (e.g., IT-BCi).
30. Table 8. rBias should also be included for site level
31. L462 The overall statistics in Appendix C would not change much, but in my experience, bias and rbias would differ remarkably for some sites that energy balance disclosure is high.
32. L481 “better than” what?
33. L483 what do you mean by “the variability of the fluxes”? temporal or spatial variability?
34. L484 reference to Figure 7?
35. L486 Interesting. I think this should be investigated further in future development.
36. Figure 10. Could you also plot the EB-corrected data for tower? What does the error band around “Tower” line represent?
37. L491, ETLook is not visible in the Taylor plot (Fig9-EBF).
38. L496 Here it says the ensemble formula is a suitable option in the case of DBF (Figure 11). But in the case of EBF (Figure 10- IT-Cp2), the ensemble is worse than TSEB-PT, so would you say the ensemble formula is not applicable everywhere?
39. L525 The description of WaPOR global ETa dataset should be mentioned before the validation results (Section 3)
40. L530 a low bias presents high accuracy, so I would say “performance” instead of “accuracy” here.
41. L566 The selection of tiles seems arbitrary to me. I would suggest more strategic selection, focusing on tiles without flux sites, tiles with high heterogeneity (e.g., indicated by coefficient of variation of vegetation index), and tiles represent climatic zones (especially Tropical and Arid, since these are lacking validation sites).
42. Figure 15. Although it might make the histogram look crowded, I’d suggest including the ensemble.
43. Figure 17. What does ETA stand for?
44. L618-619 This explanation is not clear to me. The uncertainty in LAI at higher LAI should affect ETLook model more since LAI is used to partition available energy for transpiration and soil evaporation in ETLook. I think saying that it has “minimal effect on ETa modelling” requires some local sensitivity analysis or uncertainty analysis.
45. L628-630 I’m connecting this larger bias in LAI in dense vegetation (e.g., broadleaf forest) with the lower performance of ETLook in EBF and DBF discussed in L486-489. Maybe a simple local sensitivity analysis could help address this.
46. L633-656 Similarly, the inconsistency of CLMS LAI at SAV site could contribute to why the r2 of ETLook is lower than 0.5 for SAV (Figure 7)
47. L660 How were these AOIs selected? Based on which criteria? Why not selecting AOI based on level of heterogeneity/homogeneity?
48. L672-L680 The site-level results were discussed but not presented or referenced.
49. L690 Interesting. Could you show some examples? Where were these artifacts commonly observed? I think this would be important for users to know since the use of products like CLMS are often for spatial analysis.
50. L696-L700 Do these 45 sites mainly in Europe also represent Tropical and coastal regions?
51. Figure 22. Could you say something about the cluster of points with EC below 100 Wm-2 but spread from 100 to 350 Wm-3 for ETA?
52. Overall Section 4.3 presents a lot of methodological details and could be more focused on the impact on the prototype products. I would suggest methods or trials of methods to be described in Methodology section or Annex.
53. L733 ECOSTRESS could also be considered, then.
54. L747 Also canopy interception
55. L761 Could you say that the pre-processing methods are “applicable globally” when the evaluation in Tropical climate (areas with least validation sites and most data gaps) was not included?

References

Athira, K.V., Rajasekaran, E., Boulet, G., Nigam, R. and Bhattacharya, B.K., 2025. Multiscale evaluation of three remote sensing-based evapotranspiration models across the humid to arid tropics: a study over India. Hydrological Sciences Journal, pp.1-22.

Blatchford, M.L., Mannaerts, C.M., Njuki, S.M., Nouri, H., Zeng, Y., Pelgrum, H., Wonink, S. and Karimi, P., 2020. Evaluation of WaPOR V2 evapotranspiration products across Africa. Hydrological processes, 34(15), pp.3200-3221.

Cogill, L.S., Toucher, M., Wolski, P., Esler, K.J. and Rebelo, A.J., 2025. Evaluating the performance of satellite-derived evapotranspiration products across varying bioclimates in South Africa. Remote Sensing Applications: Society and Environment, p.101612.

de Andrade, Bruno Comini, Leonardo Laipelt, Ayan Fleischmann, Justin Huntington, Charles Morton, Forrest Melton, Tyler Erickson et al. "geeSEBAL-MODIS: Continental-scale evapotranspiration based on the surface energy balance for South America." ISPRS Journal of Photogrammetry and Remote Sensing 207 (2024): 141-163.

Dingman, S.L., 2015. Physical hydrology. Waveland press.

He, S., Yang, Q., Zhang, L., Shi, Z., Wang, X. and Lv, B., 2025. Consistency Assessment and Uncertainty Analysis of Spatial-temporal Characteristics of Evaporation Data in the Greater Mekong Subregion. Journal of Hydrometeorology.

Nelson, J.A., Pérez‐Priego, O., Zhou, S., Poyatos, R., Zhang, Y., Blanken, P.D., Gimeno, T.E., Wohlfahrt, G., Desai, A.R., Gioli, B. and Limousin, J.M., 2020. Ecosystem transpiration and evaporation: Insights from three water flux partitioning methods across FLUXNET sites. Global change biology, 26(12), pp.6916-6930.

Peng, X., Wu, W., Zheng, Y., Sun, J., Hu, T. and Wang, P., 2020. Correlation analysis of land surface temperature and topographic elements in Hangzhou, China. Scientific Reports, 10(1), p.10451.

Poyatos, R., Granda, V., Molowny-Horas, R., Mencuccini, M., Steppe, K. and Martínez-Vilalta, J., 2016. SAPFLUXNET: towards a global database of sap flow measurements. Tree physiology, 36(12), pp.1449-1455.

Savenije, H.H., 2004. The importance of interception and why we should delete the term evapotranspiration from our vocabulary.

Tran, B.N., Van Der Kwast, J., Seyoum, S., Uijlenhoet, R., Jewitt, G. and Mul, M., 2023. Uncertainty assessment of satellite remote-sensing-based evapotranspiration estimates: a systematic review of methods and gaps. Hydrology and Earth System Sciences, 27(24), pp.4505-4528.

Weerasinghe, I., Bastiaanssen, W., Mul, M., Jia, L. and Van Griensven, A., 2020. Can we trust remote sensing evapotranspiration products over Africa?. Hydrology and Earth System Sciences, 24(3), pp.1565-1586.