General assessment

This manuscript presents a scientifically robust and timely contribution. The ensemble-based EVASPA framework, combining multiple LST, radiation, EF and G datasets, provides a valuable foundation for analysing uncertainties in contextual evapotranspiration (ET) modelling. The methodological implementation is technically sound, and the topic is highly relevant for the remote-sensing and ET communities, especially in the context of upcoming missions such as TRISHNA.

My comments concern almost exclusively the readability and organisation of the manuscript, not the scientific validity, which appears strong. The manuscript is dense, and the large amount of information sometimes makes it challenging to identify the central messages. A clearer narrative and more guidance for the reader would significantly increase the accessibility and impact of the work.

Major comments

1. Readability and narrative flow

The manuscript contains extensive information, often presented in long paragraphs with multiple embedded ideas. This makes it difficult for readers to extract the main points and to follow the progression of the results. A clearer hierarchy of information, distinguishing essential findings from detailed descriptions, would be highly beneficial.

Example: In parts of Section 4.3, very long multi-clause sentences make it difficult to isolate the key conclusions.

2. Clarity of objectives and role of each analysis

The study combines input-data variability analysis, performance evaluation against flux towers, ensemble-based uncertainty quantification, and similarity clustering. All of these components are relevant, but the manuscript does not always clearly articulate how each one contributes to the overarching objective. Briefly restating the purpose of each major section would help maintain coherence.

3. Interpretation and introduction of figures

The figures are well designed, but several require more explanation to be fully interpretable. In some cases, it is not clear how values were aggregated or what specific elements represent.

Examples: 

- For Figure 3, the method used to compute the distributions (means and standard deviations) is not entirely clear, particularly the number of pixels and temporal samples considered.

- Figure 5 would benefit from a clearer explanation of what each column corresponds to in terms of ensemble subsets.

Providing more explicit introductions to figures would greatly help readers understand what to focus on.

4. Spatial representativeness

Most analyses are based on the eight 1-km pixels corresponding to the flux-tower sites. This setup is fully appropriate for a site-based uncertainty assessment, but it does limit spatial generalisation. A brief acknowledgement of this limitation would help clarify the scope of the conclusions.

5. Emphasis on key insights

Some of the most important conclusions - such as the dominant role of LST (particularly overpass time), the notable influence of EF methods, and the different impact of radiation in gap-filling versus instantaneous estimates - are present but sometimes buried within dense text. Highlighting these insights more explicitly would strengthen the communication of the study’s main contributions.

Minor comments

- Some sentences are particularly long and could be split for clarity.

- Acronyms may be reintroduced when they reappear after long intervals.

- The introduction could converge more directly toward the specific objectives of the study.

- The conclusion is informative, but a more concise synthesis of the central messages would improve its effectiveness.

Final recommendation

The science is solid and the manuscript has clear potential. Once clarified and streamlined, it will make a strong contribution. I recommend major revisions, focusing mainly on the readability, structure, and presentation rather than on the scientific methodology.

This manuscript presents a comprehensive analysis of uncertainties in long-term ensemble estimates of contextual evapotranspiration (ET) over Southern France using the EVASPA modelling framework and MODIS remote sensing products. The study addresses an important and timely topic, as uncertainty quantification in satellite-based ET estimates remains a major challenge for hydrological and climate applications.

The ensemble-based approach adopted by the authors is methodologically sound and well aligned with current best practices in remote sensing of land–atmosphere fluxes. The manuscript demonstrates a strong technical foundation, and the comparison with flux tower observations adds credibility to the analysis.

Major comments

• A key issue is the comparison against the daily ET estimates measured by flux towers. Is their footprint being compared with the pixel dimension? A scale issue should always be addressed when discussing such comparisons.

• Furthermore, even though the mail goal here would be the spread of the “error”, more than the error itself, it would be helpful to sort some results in terms of land cover, as you did with the four seasons, in order to assess potential for uncertainty in different environments.

• L185-186. This assumption is quite common throughout the literature. However, it would be helpful to recall somewhere what uncertainties lie behind it, and, if possible, try to include them in the uncertainty quantification you obtain later from your ensembles
• Since the focus here is uncertainty, I would have liked more quantitative references to the standard deviations of the different combinations (e.g., in Fig.5 it seems a missing information), both in absolute and relative (to the main variable values) terms. Furthermore, as seen in Fig.6, a lot of the ensemble results seems to distribute quasi-normally, although a non-negligible quantity (e.g., Aurade, Caoussoul, LeBray, Toulouse) do not. Given these distributions, a skewness indicator or other, more complete parameters would seem to be better fit to describe the results of the different indexes you computed, as opposed to the simple SD, which can be misleading when the sample is not distributed normally.

Minor comments

• No need to replicate the LST and NDVI acronyms, you already mentioned them at L54
• This “relatively low” maybe would be better supported by some mention of relative uncertainty (standard deviation / average), to help the reader in contextualizing the potential error
• On a more general note, the manuscript is at times a bit too dense. Although this is understandable, given the scope of the analysis, this can hamper readability. Some sections, especially Section 4.2, feature long paragraphs with many interlinked concepts. Also some figures (e.g., 8, 10 or 11) feature either a complex caption, too many information or are hard to interpret.