Dear authors,

The three referees are all supporting the publication of your manuscript, but they provided a range of open questions and detailed suggestions for improving the article. Thank you for thoroughly considering all the referee comments and also correcting mistakes in the data analysis. The referees asked very valid questions and your thoughtful responses are very valuable. For the benefit of future readers, could you include the information provided in your responses also in the main paper, so that these questions do not arise again?

Please revise the manuscript as proposed and also taking into considerations my comments below. Since the revision includes altering the results, the revised manuscript will be sent for a second round of reviews. To facilitate a speedy review, could you compile all responses in a single document and point to the lines in the revised manuscript (and/or a track-changes document) where modifications were made in response to each comment? Thank you.

Below, I comment on some of your responses, where I think that additional clarification is needed.

- In your response to Comment (1) of Referee #1, you state that in the hourly data, data gaps of at most one hour were linearly interpolated. For a half-hourly dataset, this means that gaps of a maximum of two consecutive data points were linearly gap filled, right? Perhaps this would be a clearer way of describing the process. On the other hand, I am still wondering why hours with no data at all (i.e. one hour data gap) were gap-filled at all. Doesn't this create the synthetic data you wanted to avoid? For the daily averages, would it not be better to calculate them directly from the half hourly data set and describe how you dealt with gaps separately from the generation of hourly data? The criterion of at least 66% valid hourly data could otherwise be reached with in fact 33% valid half-hourly data. In addition, is the criterion of 66% valid diurnal data sufficient to calculate reliable daily averages? For example, if the 66% of the data was measured predominantly during daylight, the average would be quite different to a day where 66% of the data was measured predominantly during the night, even if the total daily evaporation was the same. It might be good to check if the data gaps were random or clustered.

- In your response to Comment (3) of Referee #1, you propose to write: "As a result, the water temperature at the surface is not directly related to the instantaneous energy balance at the surface, which is how Penman’s equation can be interpreted, but rather it is subject to a delay following the large heat capacity of the water body. We think that the Penman model does not present the full story in case of a water body of several meters deep." I think this is misleading, as Penman (1948) did explicitly address the problem of heat storage, stating: "Over a period of several days, and frequently over a single day, the change in the stored heat, S, is negligible". This means that he did not claim that his model should necessarily be useful at the daily scale, and certainly not at the hourly scale. This is omitted on P9L15 in the manuscript where you stated that Penman's derivation assumes negligible G for shallow water surfaces. As the reviewer stated, the time scales of integration are paramount here and it is not clear enough in the article at what time scales each model can be used. Please clarify.

- In your response to comment (6) of Referee #1, you explain the meaning of the adjusted coefficient of determination. When reading this, I realised that it is not mentioned in Section 2.5 how the regression analysis was performed. Could you please add this information, referring to the R-packages used in this analysis and explaining what metrics were used and how they were calculated? On P8L13, you mention that you aim to "find the best simple model that uses maximum two variables, while still able to explain the dynamics of E water well." This might be a good place to add details about the metrics and how they were used to achieve this goal.

- Your response to Reviewer #2's comment on P3.L2-6 prompted me to re-read that paragraph and I believe that the following sentence is inaccurate: "This assumption results in the essential difference between the models of Dalton and Penman, where Dalton uses the vertical gradient of vapour pressure, while Penman uses the vapour pressure deficit at 2 metres height." Firstly, Dalton (1802) did not speak about gradients, only differences. Secondly, the main difference introduced by Penman was that the surface temperature needed to calculate saturation vapour pressure at the surface was replaced by the need to account for the net radiation and heat exchange between the surface and the water body. The assumption that the latter is negligible is NOT integral to the Penman equation, as the equation can equally be applied if that component is measured. I agree with you, though, that the explanation of the Penman equation should not be left out, as suggested by the referee, as the background and underlying assumptions are not immediately present in the readers' minds, even if the equation is well known.

- Your response to Reviewer #3 regarding bias in remotely sensed sub-skin water temperature: Would it not be relatively easy to propagate the reported uncertainty through your analysis to confirm your assertion that the effect is not significant? Could you also mention the strong temperature gradients near the surface mentioned by Reviewer #3 to alert the reader to the fact that the correlation found here may be an exceptional case?

Dear authors,

Referee #3 pointed out one remaining issue about not including air temperature in the analysis and I agree with him that your explanation is not clear enough. It would be good to clarify to the reader that the fundamental driver for evaporation is considered to be the water-air vapour pressure gradient, whereas air temperature affects the water temperature through sensible heat flux, and hence affects the evaporation indirectly through its effect on the vapour pressure gradient. Since VPD is a function of vapour pressure and air temperature, the reader may wonder why it was included as a separate variable in addition to the vapour pressure gradient (P8L19), while air temperature was not.

Dear authors,
Thanks for addressing the remaining concern about not including air temperature (Ta). On P8L28, you explain that you did not want to include both VPD and Ta because of their cross-correlation, which I fully understand. However, if avoiding redundancy is important, I would use vapour pressure, Ta and surface temperature as separate variables, rather than VPD and the vertical vapour pressure gradient. I assume that the choice of the latter two was motivated by the apparent drivers of the Dalton and Penman equations and that the choice of VPD over Ta was due to its explicit mention in the Penman equation, whereas air temperature features only implicitly in the definition of s (and VPD). Perhaps it would be good to clarify this, rather than using the argument of independent drivers of evaporation and assumed lack of response to Ta (P8L25 and P14L10), which seems a bit inconsistent here. If evaporation was not affected by air temperature in this case, why would you use VPD over vapour pressure? If vapour pressure and Ta were used as independent variables, the data could reveal for itself in how far this assumption is justified.

Dear authors,
Thanks for the quick response. Perhaps on P15L1, you would like to briefly discuss the consequences of not explicitly considering Ta in the analysis, as propsed by the referee? I leave it up to you at this stage.