The objective of the article titled "CAMELE: Collocation-Analyzed Multi-source Ensembled Land Evapotranspiration Data" is to create a daily merged evaporation product using a collocation-based data ensemble method. This method takes into account 117 non-zero error covariance conditions to merge multiple ET (Evapotranspiration) products, resulting in the Collocation Analyzed Multi-source Ensembled Land Evapotranspiration data. In general, the article is clear and well-written, and it falls within the scope of this journal. Below, I provide some points and comments that, in my opinion, can further enhance the manuscript:

1. It would be beneficial if the Scenario 1 product (at 0.10 degrees) could be extended until 2020. As far as I can see, PMLv2 is available until 2020 (please verify the link to the product). Additionally, it is important for the authors to outline their plans for updating the product and whether it will become operational. This is crucial, as many datasets become obsolete after publication.

2. I recommend expanding the introduction to clarify the implications of non-zero error covariance between different products. This will help readers better understand the importance of considering this aspect in merging strategies, especially when the assumption of error independence is violated.

3. Please consider using the modified Kling-Gupta efficiency proposed by Kling et al. (2012) instead of the KGE of Gupta et al. (2009).

- Kling, H., Fuchs, M., & Paulin, M. (2012). Runoff conditions in the upper Danube basin under an ensemble of climate change scenarios. Journal of hydrology, 424, 264-277.

4. The authors employ slightly different products for different periods in the development of the Scenario 1 and Scenario 2 CAMELE products. It's important to discuss the implications of this choice. Can the authors evaluate if there are changes in performance in the different periods selected to construct the datasets? Does this affect the trends reported in Section 4.3?

5. If CAMELE performs similarly to other products, why should it be used? The goal in merging datasets is to outperform the products used in the merging procedure and thus better represent spatio-temporal evaporation patterns. The authors could focus on the fact that even though CAMELE may not outperform all products in all metrics, it performs better when considering all metrics. This suggests that it is a robust product and that the method can generate a product that leverages the complementary strengths of different datasets to some extent.

6. The multi-year comparison is interesting as it highlights variations in the datasets. The authors might consider excluding the trends comparison, as it may lack significance without a comparison with in-situ-based trends. This change would also help to reduce the manuscript.

7. What are the implications of adding FluxCom for 2001-2015 in Scenario 1? It might be more suitable to use FluxCom as a benchmark and produce the Scenario 1 product solely with ERA5-Land and PMLv2, using only the IVD method. If the authors choose not to follow this suggestion, they should explain, evaluate, and discuss the implications of using two different methods with an additional product for different periods in the Scenario 1 product.

8. Similarly, it would be beneficial to address the transition from GLDASv20 to GLDASv21 in 1999 for Scenario 2. Can the authors discuss the implications of changing the product versions?

9. The discussion regarding the impact of underlying assumptions in collocation analysis could be more closely related to the development of CAMELE. As it currently stands, it seems to be a comparison of evaporation datasets. Readers would benefit from a more direct connection between the performance of CAMELE and the assumptions of the methods used in its development.

10. It's worth exploring why the merging scheme did not significantly improve the performance of CAMELE. Could this be attributed to non-linear relationships between evaporation magnitudes and their respective errors? The authors should consider expanding on this in the discussion section.

Some additional minor comments:

Line 28: I would recommend caution in using qualitative terminology like "excellent performance." Additionally, I find this statement a bit misleading because the merged products performed closely to the products used in their merging. Please revise carefully the manuscript to avoid these overstatements.

Lines 58-59: The authors mention the following: "...previous research has predominantly focused on regional-scale ET estimation, necessitating a more straightforward and reliable global simulation method." It would be helpful for the authors to clarify what they mean by a "straightforward and reliable simulation method."

Line 224: A space before the reference is missing. It should be added for proper formatting.

Line 254: Was the IGBP classification obtained from a dataset? If so, how were the functional types calculated? Do they change during the period of analysis? Please provide details regarding the source and methodology for classifying the functional types.

Figure 4: The quality of Figure 4 could be improved. Consider enhancing the clarity and readability of the figure. You might want to simplify the information presented or consider moving some details to a supplementary figure.

Line 557: The authors mention that based on the results of Figure 4, CAMELE performs well at 0.10 and 0.25 degrees, and all products have similar performance. The phrase "performs well" may sound like it performs better compared to other products, which could be misleading. Consider rephrasing this to clarify that CAMELE performs similarly to other products.

Line 585: While it's understandable that the authors want to promote their product, it might seem a bit odd to say that CAMELE performs exceptionally well and closely resembles two of the products used in the merging scheme. The desired outcome in merging datasets is to outperform the products used in the merging procedure. Consider rephrasing this to maintain objectivity.

Figure 6: The quality of Figure 6 could be improved for better clarity and readability. Consider reducing the information presented in the figure or moving some details to a supplementary figure. Another option is to highlight the top three performing products for each IGBP class with color coding.

Line 863: Remove the word "excellent" from this line to maintain a more neutral tone

The manuscript aims to develop a new evapotranspiration (ET) product using collocation techniques. It has the potential to be a useful contribution to the literature and to the broad userbase of ET products. Nonetheless, a few major issues need to be addressed before it can be considered for publication.

First, the construction of the products should be justified by a clearly outlined rationale. The new ET product is built from multiple ET solutions with different temporal coverage. How are those individual products selected for each analyzed period, and when three instead of two products are selected, what is the corresponding gain in terms of performance? Second, the manuscript demonstrates the consistency of the proposed product with its peers, but I believe it is more important to highlight the unique strength and weakness of the new product. When and where does the new product outperform or underperform its peers? Does it improve upon its individual constituents in terms of characterizing the long-term trend, seasonality, inter-annual variability, or the extremes of ET? Third, the organization of the manuscript can be improved, too. I list some of my suggestions in the detailed comments below. For example, the discussion of the non-zero ECC spreads across two subsections in the Discussion, and I think they should be merged and moved to the result section. I think addressing these issues will strengthen the scientific robustness of the manuscript and facilitate the adoption of the new product.

Detailed comments:

L173-175. The rationale of choosing GLDAS-2.0/2.1 is questionable. Given the difference in the underlying modeling/reanalysis schemes, the error structures of GLDAS and ECMWF-based ET estimates are inherently different regardless of what sets of meteorological forcing are used. When selecting the GLDAS products, one could arguably choose the ET products that are driven by the more reliable forcing.

L252. It would be ideal to show the distribution of the selected sites on a map.

L271. The methodological detail for Sections 3.1 can be trimmed as they are widely available. Highlighting aspects that are either implemented or discussed in this study would be sufficient, e.g. the assumptions, especially regarding the cross-correlated errors.

L393-L400. This should go to the intro. Overall the method session needs to focus on clarifying the rationale of the chosen methodology and how they are directly implemented for this study.    

L430. One of the PMLv2 should be GLDAS.

L430-439. I understand this is a prior work that is directly related to this study, but the summary of this prior finding should go to the intro. Only the key assumptions made based on this prior work need to be highlighted in the method section (in this case, the non-zero ECC pairs).   

L454. I don’t fully understand the rationale of grouping different products within each scenario. For Scenario 1, e.g., is the goal here to include as many products as possible within a given period? If that’s the case, it will be helpful to clarify the gain in terms of performance of doing so.

L477-484. This should go to the Method section.

L485. What about the correlated errors?

L621-629. This is more informative than the global statistics. I think it will be useful for the readers to adopt the new product if the authors can highlight when/where and over what spatiotemporal scales that CAMELE outperforms other products substantially. From a practical standpoint, establishing the unique strength of the proposed product is more important than showing its consistency with its peers.

L635. How does the proposed product and its peers compare with the FluxNet in terms of long-term average and trend?

L677. I think only the statistically significant trends should be shown.

L745. This should go to the result section. It is not clear to me how different treatment of the cross-correlation pairs will impact the final product.

The manuscript “CAMELE: Collocation-Analyzed Multi-source Ensembled Land Evapotranspiration Data” presents an ensemble product compiled from collocation analysis/weighting of five global evapotranspiration (ET) products (ERA5-Land/GLEAM/GLDAS/FluxCom/PML). The authors illustrate that by using non-zero ECC collocation weighting, multiple independently-sourced ET products can be merged resulting in enhanced accuracy. Generally, the paper is properly written and structured. It is well suited for this journal. Some remarks:

• The authors should consider [consistently] defining all abbreviations before use (more below). While many of the abbreviations may be obvious to the authors (and for many in the sub-field), they may be misinterpreted by other readers. Some terms, such as EC may be misinterpreted by most interested in the (experimental) observation and modeling of the surface energy budget.
• 5 ET products (ERA5L/GLEAMv3/GLDAS/FluxCom/PMLv2) are applied in this study. What was the criteria used to select these 5? Have the authors considered including other ET products, such as the MERRA, MOD16, WaPOR, SSEBop, …, in their analyses. If not, why?
  
  All the ET products described here (and consequently the ensemble CAMELE product – ~1°, 0.25°) are rather coarse. Most of the local characteristics that influence the local flux interactions are therefore averaged out. For purposes that involve local/field-scale applications, and in terms of accuracy (i.e., evaluation scale mismatch with FluxNet local footprints), a discussion of the scale limitations is necessary.
• One interesting outcome from this study is that the CAMELE product appears to perform comparatively well over most of IGBP-based plant functional types (PFTs). While commendable, the authors only touch on this without really discussing why it performs better. What are the implications of selecting one product over the other over different PFTs, especially with respect to real applications.
• A weighted average of the ensemble members (ET products) is described and discussed. Nowhere, though, do the authors detail the weights quantitatively, even for a few example scenarios. In Park et al. (2023), for instance, the weighting factors calculated in a triple-collocation study (with no consideration of non-zero ECC) were analyzed. It would be interesting to provide the readers with a sense of such weights between the 5 products used within CAMELE, especially given that non-zero ECC is considered here. Are they close to equal weighting? Are different weights assigned depending on the season, e.g. as is/was done for the DOLCE product? How much do the weights vary with the plant functional type?

Specific comments :

L34: Do these metrics imply the CAMELE ensemble ET is fit to be applied as a benchmark reference of choice? maybe state that it may be a suitable ‘reference’ candidate for ET product evaluations.

L42: add to read “soil moisture and air temperature/humidity

L45: “… evapotranspiration, resulting in many datasets” - Maybe some citation is necessary here?

L66: TC and EIVD used before being defined - Full names given further below (lines [71] and [79]). Consider describing the abbreviations here instead.

L96: “(i.e., IVS, IVD, TC, EIVD, and EC)” – note that some of the abbreviations here have not been described earlier (e.g. EC)

L109: “… error covariance (ECC)” - defined in L78 as “error cross-correlation”. Consistency.

L137, 139:  “referred to as ERA5L” – you call it ERA5L instead of the common ERA5-Land. Ok.

“… ERA5-Land …” – Consistency. Continue using ‘ERA5L’ since that is how it is abbreviated in this study

L174-175: “… potential error homogeneity issues between GLDAS-2.2 and ERA5L” - Have these potential ‘homogeneity errors‘ due to use of equivalent meteorological forcings been documented anywhere? There should still be differences between the two ET estimates/products since: 1) GRACE data is assimilated (L171-172), and 2) different LSMs are used (i.e. lines [159-160] and [141-143] for GLDAS and ERA5-Land, respectively)

Looking at Figure1 of Jiménez et al. (2011) where 3 GLDAS models (NOA, Mosaic, CLM) are inter-compared -among others) shows that relatively large variations can be observed between the NOA, MOS, CLM flux estimates; these can generally be attributed to the differences in the models (parameterization, structure, physics, …). As such, the non-homogeneous error condition (as required in TC) will generally still be met between different LSMs - even with equivalent forcings.

L181: Note that, while not yet documented, they now have v3.8a available

L185: “…from 1980 to 2022” - Note that v3.7b (based on satellite data) only runs from 2003

L194, 196: “into actual transpiration or bare soil evaporation” – maybe replace ‘or’ with ‘and’? for total actual ET. “by (Martens et al., 2017)” >> “by Martens et al. (2017)”

L198: Add this abbreviation in L194 above or define Actual Evapotranspiration (AET) here

L210: “…, white sky albedo, …” - Do they really only use the white sky albedo in their computations of available energy ? Normally the broadband albedo is applied, which is a combination of white- (diffuse) and black-sky (direct) albedos (see MODIS albedo data for reference - https://lpdaac.usgs.gov/documents/97/MCD43_ATBD.pdf, i.e. pg.11, EQ 32).

In Figure1 of Zhang et al. (2019), they indeed indicate “White Sky Shortwave Albedo”, but the same is not mentioned anywhere else in their article. Since it might have been a misplaced error in that figure, you should drop ‘white sky’ here unless you can confirm from them that only WS albedos are used in PMLv2 calculations - which would then mean an additional source of uncertainty in PMLv2 ET products.

L213: “(𝑃𝑠𝑢𝑟𝑓, 𝑃𝑎, 𝑈, 𝑞), and” - These [meteo] variables have not been defined elsewhere.

L238-241: “latent heat values to evaporation using ERA5-Land” - latent heat [*energy] values to *evapotranspiration using *ERA5L.

“using ERA5-Land aggregated daily air temperature” – how? were the air temperature data used to rescale the latent heat of vaporization used for the energy-to-mass flux conversions?

replace “evaporation” in L241 with evapotranspiration.

L251: “ET data were corrected.” - Maybe clarify how? residual method? bowen? …?

L252-261: That makes 12 PFTs and 206 sites. What of the other 212-206 sites (13-12=1 PFT)?

L270: Again, EC here has yet to be defined. It is defined further below [L342]. Note that EC in ET circles may be interpreted to mean Eddy Covariance, so consider defining EC further up to avoid confusion.

L315, Equation 7: NSR is Noise to signal ratio? why write it here if it will not be used elsewhere?

L316: “In comparison to the conventional coefficient of determination 𝑅𝑖𝑗” - Is it common to write the standard/conventional coefficient of determination as R instead of R^2?. R is generally reserved for correlation.

L395: “… CCI” is not defined

L410: “… superior …” – your ensemble ET product performs somewhat similarly to the others, so the authors should be a bit modest here. Use another word; otherwise detail the aspects that make it superior.

L418: “… PMLv2 and FluxCom have an original resolution of 0.083° and an 8-day average - note that for FluxCom, energy balance fluxes are also available at the daily scale, i.e. denoted ‘RS_METEO’.

L420: “… and the values for each data period of 8 days are kept consistent. For example, the values for March 5 to March 12, 2000, are the same”

It is not clear what ‘8 days’ means here. Going by L420 it is appears like one value is replicated for the 8 days. [Actual] ET is influenced by radiation, atmospheric vapour demands as well as surface water availability. These do not usually remain constant throughout an 8-day period. So using an 8-day average to represent the temporal dynamics should ideally introduce further uncertainties.

Also, why do the authors only use the FluxCom 8-day dataset (which employs only remote sensing data)? there is also the ‘RS_METEO’, which is available at daily timesteps.

L430: “ERA5L/GLEAMv3/PMLv2/FluxCom/PMLv2” - PMLv2 appears twice

L468, 469: “shortcomings in Nash-Sutcliffe” such as ? Where>>where

L491: “variation curves of average with latitude” - Variation curves of which state variable ? add *ET after ‘average’. Are these metrics curves really fair since there is quite a bit of missing data, especially over North Africa (and other sub-Saharan regions below the Equator & Australia - e.g. for Fluxcom)

L509,510: “(0.59±0.58 mm/d), GLDAS2.0 (0.37±0.44 mm/d), and GLEAMv3.7a (0.38±0.36 mm/d)…” - Are these reported global (mean±standard deviation) values more or less equivalent to the latitude-averaged values ? looking at the variation curves in Figure 2, I am not really sure, as the GLEAM and GLDAS products qualitatively appear to have higher variation than ERA5 especially from latitude -45_to_-35 and -15_to_10

L517-519: “average distribution with latitude” – average distribution of variation with latitude

“ERA5L demonstrates a more even distribution, whereas GLDAS and GLEAM exhibit relatively higher uncertainties in tropical regions” – the authors could consider discussing in terms of the model theoretical basing/assumptions/inputs – i.e., surface characteristics/physics

L529: “during this timeframe” - Figure 2 reports on period 1980-1999 while Figure 3 reports for 2000-2022 [both ERA5, GLEAM, GLDAS2.0/2.1]. Is selection of ~20 year periods deliberate? What about 1980-2022?

L543: “In this subsection, …” - you move directly into CAMELE without mentioning how the weighting between the ensemble members is done. At least, refer the reader to Section 3.4. Also note that Section 3.4 does not mention CAMELE even once.

L552 - not all performance metrics in the Figure 4 are unit-less

L568 “not align with the actual situation” - Needs to be discussed a bit more than this. What do the authors mean by ‘actual situation’?

L585, 588: “…exceptionally” – it performs well, not sure if “exceptionally”. “suggests more minor errors” – what do you mean “more minor”?

L643: “multi-year…” - as you have shown in Figures 2 and 3, different estimates can exhibit varied performance when different periods are considered. Why compare estimates from mixed periods here?

L661 – “while GLDAS and GLEAM have weights of approximately close to 1/3 each” - 1/3 each ? not clear. Also could the authors consider providing the readers with a quantitative illustration of the weights between the 5 products used within CAMELE?

L672: “…of average with latitude” – do the authors mean the “average trend with latitude” ? how is this trend over the different periods calculated  also why is there no consistency in the periods considered? the trend in ERA5L which is from 1980-2022 appears to have a negative trend at the tropics (especially in Africa close to the Equator). Additionally, unlike all the other products, CAMELE appears to have pronounced negative trends in the southern hemisphere, why? How can the weighted output (CAMELE) have higher negative trends than the input/ensemble members? can the authors provide the trends of the other 2 ensemble products to aid with interpretation?

L685: “CAMELE exhibits a more pronounced decreasing trend” – why?

L691: “introduce specific impacts” - what specific impacts? using consistent comparison periods would help readers and the broader scientific community better interpret the results.

L693: “characteristics of the data itself influence this” - maybe explain how the grid and data characteristics influence the temporal trend, or point to the section where it is discussed.

L695-744: This section (5.1) is too general …and the authors are really not discussing the results as presented in the previous chapter. Shorten the section OR consider moving to introduction or methodology section as justification of the algorithms selected in this study.

L808: “This could be attributed to the variations in the input products” – what variations?

L817-818: “GLEAMv3.7b and GLDAS2.2 employed the satellite data from MODIS, introducing rand” - Citation needed. Also, in L174 you talk of error homogeneity arising from ERA5L and GLDAS (due to meteorological inputs) but the same is not discussed here.

Reference

Jiménez, C., Prigent, C., Mueller, B., Seneviratne, S. I., McCabe, M. F., Wood, E. F., … Wang, K. (2011). Global intercomparison of 12 land surface heat flux estimates. Journal of Geophysical Research Atmospheres, 116(2), 1–27. https://doi.org/10.1029/2010JD014545

Park, J., Baik, J., & Choi, M. (2023). Triple collocation-based multi-source evaporation and transpiration merging. Agricultural and Forest Meteorology, 331(February), 109353. https://doi.org/10.1016/j.agrformet.2023.109353

Zhang, Y., Kong, D., Gan, R., Chiew, F. H. S., McVicar, T. R., Zhang, Q., & Yang, Y. (2019). Coupled estimation of 500 m and 8-day resolution global evapotranspiration and gross primary production in 2002–2017. Remote Sensing of Environment, 222(December 2018), 165–182. https://doi.org/10.1016/j.rse.2018.12.031