This paper assesses the use of convection-permitting regional climate models (CPMs) to improve the outputs of regional climate models (RCM), as climate projections could incorporate simulation of convection system storms in CPMs. CPMs could be applied to improve climate projections in the Mediterranean area, where floods are usually generated by convective storms. Simulations generated by a CPM are compared with the climate projections supplied by a RCM. Climate projections are bias corrected. Two hydrological models are used to transform climate projections into flood time series.

The paper is well written and organised. It can be considered for publication after addressing the comments included below.

General comments

• Convection-permitting regional climate models are a potential tool to improve the characterisation of convective storms in climate projections, as such events cannot be represented by the current climate models, given their spatial and temporal resolution. However, the analysis is limited to one climate model and one climate change scenario. Therefore, the conclusions could be highly limited to such a climate model and scenario. Why only one climate model is used? In addition, the selection of such a climate model and no other should be discussed. Why only the RCP 8.5 scenario is used?
• Furthermore, the paper should clarify if the climate model used in the study belongs to either AR5 or AR6 of IPCC.
• The CREST distributed hydrological model has been calibrated fixing most of its parameters. A discussion should be included about why some parameters are fixed. A sensitivity analysis of model parameters could be useful to improve the understandability of the paper.
• Section 4.1. The results are assessed qualitatively by using figures to compare outputs from different simulations. However, quantitative assessment of the results would be very useful for the reader. For example, comparison between model outputs shown in Figure 3 in terms of cumulative probability distribution could be improved by using similar metric to those used with the hydrological models.
• Table 3 shows that the CREST performance is lower for all the metrics. It could be assumed that a distributed model could represent better rainfall-runoff processes in the catchment that a lumped model. However, if a distributed model has a lower performance than a lumped model, the lumped model should be preferred as it is simpler. Therefore, a comment should be included about why the CREST model is used in this study despite having a lower performance than a simpler lump model. In addition, maybe the lower performance of the CREST model could be increased by improving its calibration.

Specific comments:

• 74. Do French Mediterranean region face intense floods with important damages and casualties every year? I think that such great floods happen from time to time, not every year.
• 129. Kendon (2010) cite is missing in the References Section.
• 147. What are CCLM and WRF? These acronyms should be either introduced in the paper or explained to readers.
• 224-225. Some previous studies have used CNRM-AROME for assessing climate change, such as Monteiro et al. (2022). Such an application of CRRM-AROME should be investigated reviewing published works. In addition, the statement should be changed.
• 297. Vrugt et al. (2009) cite is missing in the References Section.
• 303-306. No information is supplied about how the soil parameters in both CREST and GR5H are estimated from available soil information.
• 312. Only a few parameters of the CREST model have been considered in the calibration process. A description about why some parameters are fixed and others not should be included in the paper. In addition, a sensitivity analysis of the model parameters could be included in the paper to identify the parameters that can be fixed and the parameters that should be calibrated. Finally, a discussion should be included in the paper about what hydrological processes are fixed and what processes are considered in the calibration process.
• 361-362. Simulations are corrected in each cell separately. A sensitivity analysis could be included comparing the results correcting bias in climate projections at the regional scale and cell by cell. In addition, has spatial correlation between cells been considered? Correcting bias cell by cell could change spatial correlation patterns in precipitation.
• Table 1. Why nQ equals n plus 1? A comment could be included in the paper.
• 483-489. After bias correction, both models underestimate precipitations associated with the highest quantiles. In addition, in Figure 3 for the period 2000-2018 both distributions are similar for quantiles close to 99.99. The main improvement of CPMs are supposed to be a better characterisation of the most extreme events. However, such an improvement cannot be seen in Figure 3. A comment should be included in the paper.
• 614-616 and Figure 6. How simulated discharge with observed data represented by the green line in Figure 6 is obtained? The paper should include a description about how these simulations were obtained.
• 619-623. This discussion should be clarified to improve the understandability of the paper.
• 623-624. CDFs of both climate models are similar after bias correction. In addition, it seems that ALADIN performs somewhat better than AROME, as ALADIN CDF is closer to observations than AROME CDF. A deeper discussion about why CPM improves flood characterisation in this catchment should be required.
• 744-745. There are two brackets after a and b. One bracket should be deleted.
• Section 4.5. Comments about np and POT boxplots in Figure 8 are missing.
• Figure 8. In the boxplot of FI, the y-axis scale should be changed, as boxplots are excessively small to assess the results.
• 813-815. The added value of CPM can be seen before bias correction. After bias correction, outputs of CPM are similar to those obtained with RCM. This should be clarified in the conclusions.
• 821-824. It is stated that similar results are obtained regardless the hydrological model used. A comment about the limitations of the distributed model to simulate floods in this catchment should be added, given that a simpler lumped model supplies better results.

References

Monteiro, D., Caillaud, C., Samacoïts, R., Lafaysse, M., & Morin, S. (2022). Potential and limitations of convection-permitting CNRM-AROME climate modelling in the French Alps. International Journal of Climatology, 42( 14), 7162– 7185. https://doi.org/10.1002/joc.7637

This paper evaluates the skill of the CNRM-AROME Convection-Permitting regional climate Model (CPM) in projecting floods, using the Gardon at Anduze catchment in southern France as a case study. The CPM demonstrates superior accuracy in reproducing extreme hourly rainfall events compared to traditional models. The study underscores the potential of CPMs in future flood predictions in a warming climate.

The manuscript is interesting and generally well written, although some important aspects must be addressed before being suitable for publication in NHESS:

MAJOR COMMENT

1. The focus of the paper is not clear to me. I understand there has been a great effort to run two different hydrological model approaches, with two climate projection standards CPM vs high resolution but non-CPM models. The results and conclusions are too much focused on describing the experiments output with barely no interpretation. Indeed the results sections consist of what should be a caption inserted in the text and a description of the figure. What is the research question attempted to be responded by this study? The outlook for future flood scenarios is clearly not an objective as no attention to uncertainties is put.

MINOR COMMENTS

• The title is too ambitious provided the type and robustness of the conclusions reached in the manuscript. The conclusions do not offer “new perspectives”.

1. , 56 “all floods are projected to increase… the moderate floods are expected to decrease.” More precision in the language is required. Are these sentences referring to the frequency of floods? the magnitude? the flashiness?

2. “Thober et al. (2018) showed a decrease of high flows and flood magnitudes for different levels of future global warming.” Where?

122-127 The argument presented in these lines seems vague and unfounded. Can you identify studies that directly attributes the contradictory results to the underrepresentation of sub-daily extremes?

1. The resolution required to explicitly simulate convective processes is not necessarily determined at 4km. This statement is too absolute.

2. “get rid of” seems too colloquial.

167/176 and 168. floods→ flood and emission→emissions

1. “Evaluate the added value of the CPM on extreme rainfall”, you mean “the simulation of”? “the prediction of”

2. The COMEPHORE is a high quality dataset but is not an “observation”. As the text already states, it is a high resolution analysis gridded field of precipitation. This should be corrected throughout the manuscript.

242 and 243. Can you find alternative expressions to refer to the radars used that avoid the relative reference as “foreign” of the Swiss and Jersey Islands radars. They are not “foreign” to some readers.

243-245. Are there any references that support this statement. “is still considered” should be backed by a referenced source.

1. I think the change of subject towards temperature use for PET must imply a new line/paragraph.

2. Can you find a more rigorous description of the differences between the two hydrological models than “physical concepts”?

3. Describing the use of a hydrological model as a system that “transforms” precipitation into discharge sounds naïve.

296-297. What is the calibration set of flood cases used? Are these extreme floods? How does this affect the application of the model for extreme events in the experiments?

344-345. The presence of biases in climate simulations hampers their use in virtually all impact studies, not only in hydrological applications.

359-362. How do calibrated precipitation maps look like? Is the spatial correlation of the original pcp fields lost?

1. Add “…periods of the same…”

405-406. “The rainfall thresholds are related to our knowledge of the river basin dynamics and hydrological expertise.” Can you provide any hint or supporting evidence?

459-461. Is this spring-autumn connection possible provided the intense hydric deficit that characterizes Mediterranean climate, that acts as a drying reset to the hydric cycle in most surface basins in the region?

Figure 3. Being COMEPHORE an analysis, it very likely underestimates actual precipitation peaks. Despite that, none of the two models reach its extremes. Can you comment on that and the implications for the projected scenarios and the derived hydrological conclusions?

555-558 After asking the reader to keep in mind an aspect, the authors are expected to make an important statement that requires to keep that in mind. What is it here?

Figure 6. The bias corrected results show that AROME underestimates Flood Peak Discharges more than ALADIN. Doesn’t that tell opposite messages than the main point of the paper?

1. Can you provide a better section title?

2. “determine how the flood distribution will evolve in the future” This statement is way too pretentious. I don’t think we currently have tools that can do that. At most, current tools generate projection, but not outputs that “will” occur.

643-645. This is a caption for Figure 7. Avoid using this type of sentence in the argumentation text throughout the results section.

744-745. Idem.

795 if→ whether

797-801 These two sentences seem to state contradictory messages. “Until now” “regional models … cannot” and “In the last 10 years …CPM”. Are regional models that now allow (…) or CPMs?

• Despite the last paragraph of the conclusions section mentions the lack of attention to uncertainties, previous parts of the text attributes predictive capacity to the set of experiments done. For instance, the 3rd paragraph of the conclusions present results of the experiments with 2080-2099 data as “future” predictions.
• The use of the label “future” to describe results is excessive, provided the lack of robustness of the single projection used. I recommend using “projection” and avoid presenting the scenarios calculated as an interesting outcome of the work, provided that not uncertainty analysis is done.