the research is very good.  the combing of the hydrological model and machine learning is very  Innovative research. 

I have reviewed Flow intermittence prediction using a hybrid hydrological modelling approach: influence of observed intermittence data on the training of a random forest model”, which seeks to combine JAMS -J2000 model with statistical learning to predict the daily state of flow at reach scale. I think the approach is clever and the final results are interesting. I have only small comments:

Abstract: add a summary of model performance (quantitative values)

Line 17 replace home with habitat

Line 20 “These ecological corridors can be disrupted when river beds dry up”  I think this point should be expanded, it is true that some species cannot live in the presence of water, but non-perennial rivers due to the succession of different flow phases represent biodiversity hotspots.

Line  68 I think it is better recall this section “materials and methods”

Line 128 In cases of simultaneous acquisitions from multiple resources. Was the information coincident? For example, did information from traditional measurement stations coincide with google images? Was the 'flow condition information the same?

Line 200  I suggest you use accuracy, sensitivity, specifcity, and true skill statistic to validated RF

This paper aims to simulate flow intermittence using a mix of hydrological modelling and random forest algorithms. The study is performed in three different catchments located across Europe with different climate regimes. The study assesses the impact of the used observed dataset on the performance of the model. 

Overall, I think that the idea and approach of the authors is interesting and that it could provide relevant outcomes on the importance of different observation datasets for developing data-driven algorithms. I think that the paper has the quality of being published but there are some point that the authors should address before publishing it. 

I think that the paper could be strengthen if the following points are considered:

1. Line 23: A word is missing in the sentence, perhaps “limit”? For example: … flow can endanger ecosystems and limit the access to water resources …”.

2. Line 46: I think that the “a” is extra in the beginning of the sentence: “Another a challenge in …”.

3. Line 51: Change “do” into “to”.

4. Lines 61 to 64: This paragraph is a repetition from the previous paragraph. Please remove one.

5. Line 74: Change the sentence to “…, located in the South of Spain...” or “,… located in southern Spain…”.

6. It would be relevant to include the annual mean temperature and total precipitation in table 1 or in the study areas description section. These would help better understand the climate contrast among the different study sites. 

7. Line 83: you could add “input to” and change “modelling” to “model” in the sentence. “.., hydrogeology information needed for input to the spatially distributed hydrological model…”.

8. Please use superscript for units throughout the text (e.g. cubic meters, square meters, etc.).

9. Consider changing the symbology for the Google Earth observations because they are not easily observed as they can be confused with the river network. 

10. In Figure 3. It should be “Daily flow state”, a “t” is missing.

11. Line 147. A period is missing after (HRU).

12. Line 157. Not clear if the period that you mention is for all study sites or whether it is for the calibration or validation step. Please clarify.

13. Figure 4: I recommend that you add the definitions of MPS, DPS, LPS and all the abbreviations that you used in the figure. These could go into the caption of the figure. 

14. Table 2. The Albarine and Lepsämänjoki sites have a continuous simulation period but the Genal has a gap between the calibration and validation period. Would it be relevant to explain why is that?

15. The metrics in table 3 could also include a metric similar to your POD and FAR but for the hydrological model. For instance, set a low-flow threshold and indicate how many times the model succeeds simulating when the river is below the threshold (POD) or when the model is not simulating it accurately (FAR). This would also give information on how good the hydrological model is to simulate low-flow periods, which I think would add relevant information for your study. Is the hydrological model skillful simulating low-flow periods or is the RF algorithm making most of the work?

16. Figure 5. Along the text you refer to Fig. 5a or Fig. 5b, but in figure 5 there is no division between sections a or b. 

17. Lines 167 and 278: Is it possible to infer why the uncertainty for the Genal catchment is higher? Is it that the catchment is very complex to simulate or some other issue?

18. Line 279: Would it be relevant to also include some results for configuration 1 (perhaps in the supplement)? Just for comparison and to include something related to the uncertainty of the results. 

19. Figure 11. The name of the Finnish river needs to be corrected.

20. Lines 356 to 260. It is not clear if this paragraph only refers to the Albarine catchment. Please clarify.

21. Line 382: Should it be “are not-known”?

22. Line 439. You could support this claim using previous expert elicitation studies that looked into expert perception uncertainty. For example: https://doi.org/10.5194/hess-26-5605-2022 or https://doi.org/10.1002/2015WR018461

23. Lines 479 to 483. As you state that you will use this approach to assess impacts of climate change, it might good to add a line about the implications of using it for such kind of studies. For instance, mentioning a bit on what you wrote in line 396.

24. Line 265. You repeated “which shows the” in the sentence. Please remove one.