General comments

Sauquet et al. presented in their manuscript the results of French national project called Explore2 to provide hydrological climate projections in France for streamflow and groundwater changes until the end of this century. It is a successor project of Explore2070 which took place in in the early 2010ies. The motivation to update projections were to extend projections to the end of the century with newer climate model data and to increase the locations for projections. Many institutions were involved over several years from science, administration and stakeholders in water management. The objective of the project also to provide a better accessibility and understanding of the data for stakeholders compared the predecessor project. The main objectives of the manuscript were to describe the ensemble of the simulations, to evaluate the ensemble in past climate and present main results of the climate projections.

They used a subset from the EURO-CORDEX CMIP5 climate chain ensemble, two bias correcting methods, nine surface hydrology models and four groundwater models. Models were evaluated in the past both using observed forcing and climate model data at large number of observation location, i.e. > 600 gauging stations for surface hydrology models > 200 piezometers for groundwater models over many years. Some of the models were calibrated using similar data as for evaluation. Not all models cover whole France. The authors present evaluation under multiple aspects and mention thresholds for acceptable performance.

The projections cover up to approx. 4000 simulation points for streamflow projections covering whole France and parts of France for groundwater recharge projections. Results for evaluation and projections focused on the surface hydrology models as results of the groundwater models are (or will be) presented elsewhere. However, there is a sufficient overview of the groundwater part presented in this manuscript.

The substantial content of the manuscript is, however, not matched with the quality presenting the data. Most importantly I miss a literature review on the current knowledge of hydroclimatic projections in France (and Europe). The question must be answered, to my opinion, how the presented results embed in and expand existing knowledge.

The second largest issue is the readability of the manuscript, which suffers under many aspects. Most problematic are the large number of the results, the imprecise language which requires the reader to interpret what could be meant and the manuscript structure. To my opinion the manuscript will largely improve with using a commonly used structure to separate data/methods and results/discussion. I will detail the arguments below. With this impression I recommend to major revisions for this manuscript.

Detailed comments

  Literature review

As mentioned above the manuscript is not largely referring to previous work. This should also include regional studies. This aim was mentioned in the overview of the manuscript to be part of section 7, but I could not find a relevant discussion with previous literature. This literature review should also include model decisions (e.g. the impact of bias correcting methods …). Section 7 rather adds new own results not presented before.

  Readability

First is the large number of results. The objective was to show only main results, and I think this can be more condensed. The large number of results in the figures and tables is supplemented with additional numbers in the text, which makes it difficult to read (e.g. on page 28). Generally, I advise that these numbers in the text could be presented in a table, however, given the need to condense results for this manuscript, I would vote for a more qualitative text description to put the chosen results in tables and figures into value.

Second is the imprecise language. As it is a complex data set covering many dimensions it is essential to be precise. The reader may be able to follow what is meant but it requires a very thorough reading. A table and a figure, however, should be self-explanatory to a certain extent. This is often not given. As an example for many other parts is the presentation of Table 6. The unit is not provided, so one cannot say if the numbers are percentage changes or absolute changes. It can only be assessed from the context that the “interquartile range of the median change …” in the caption description refers to an interquartile range between simulation points (space) and the median change to a set of simulations varying in climate models, bias corrections and hydrological models.

Additionally, I could often not follow what the authors interpret from figures, either by wrong references to figures or that they were imprecise with mentioning patterns or regions. For example, in lines 482-483 “The most significant and positive changes with an agreement between projections can be found for QDJF in the Alps and Pyrenees (Fig. 8) for the three RCPs.” I could follow this statement with looking at Figure 7. Similarly, I could not follow the impressions of the elasticity values (lines 338-342) shown in Fig. 3: Are the values for temperature good, or for precipitation in the summer?

  Manuscript structure

The reading flow is often interrupted by switching between methods and results as the manuscript is more structured by content (evaluation, future projections …) than separating data/methods and results/discussion. For example, the results on projected changes start with a description of a multi-model index of agreement (Eq. 2). Ideally the manuscript would keep the content structure but replicate this structure for the data/methods section and the results/discussion section.

  Focus of the manuscript

In the abstract the authors mention prominently that they want to present the data availability and understanding the results through visualisation tools. This part is in the manuscript rather short.  If the authors want to present their data availability, the understanding thereof and their visualisation tools as promised in the abstract I think this needs to be more elaborated in the manuscript.

For example, the fact sheet in the supplemented is quite complex and is not put in value in the manuscript text. I would have liked to see more details on the description of these sheets and how these sheets help to make informed choices pre-select hydrological models (mentioned in section 5)

The data availability is only presented with a link (section Data availability). There is no guidance in the manuscript or on the webpage how I can get netCDFs of a certain figure for example. The webpage is available in an English version but 99% of the text is still in French. Besides the language barrier (which can be solved with translation tools nowadays) the data is hidden in multiple links. There is a link for each model for each RCP for each subregion, which leads to a new webpage with new links with filenames, which relate to different climate chains and bias correction methods.

Similarly, the visualisation tool is presented just with a link but not put in value in the manuscript

In a similar note: In the conclusion in line 595 the authors advertise their data set with the ability to step back from the results in case of disagreement and promoting an understanding of the differences between hydrological models. This is indeed possible with such a data set, but it is not presented in the manuscript. The authors show mainly median results between hydrological models (except of a minor subtopic with Figure 5). In Table 3 there are thresholds presented for model performance, but nothing mentioned about the consequences. It would be interesting if the authors would include a detailed description of their fact sheets and how one could potentially use them.

  Minor points

• In the abstract there are no results summarized for the groundwater models
• Section overview is inconsistent with section headlines and content (lines 91-95). Similarly, reference is probably wrong in line 446 (Table 7 is not existent), and section references (line 426).
• Line 103: What are the criteria selecting the subset of climate models? The cited study, which could answer the question is only available in French.
• Line 111: Why is the reference period (sometimes called with different names, try to use consistent naming for better readability) chosen to be 1976 to 2005 opposite to the latest AR5 and AR6 reports? This could be better motivated.
• Line 111: The meteorological forcing is on a 8 km grid, but small catchments up to 64 km2 (line 206) are included. Given that the effective resolution of the climate model is > 12 km and that of SAFRAN also > 8 km, which is more reflected by the underlying station distribution, I have the impression that too small catchments are included in the study. Can you comment on this?
• 1 shows differences in bias correcting methods. This should be mentioned. As they are quite substantial, I would also like to see (despite condensing results) a graph showing not only a median change over all simulations (Fig 6), but also split into bias correction methods. Figure 5 would also be interesting to see and discussed for the bias corrections separately.
• Table 2: Can you include the model resolution in Table 2? Right now, I need to look in the supplement. How did you bridge the gap from 8 km resolution after bias correction to the much finer hydrological model resolution?
• Table 3 (I refer to the second Table 3): Can you include formulas for the metrics (maybe in the assets or provide code)?
• Figure 3: some metrics are not mentioned in the text (aCDC, alphaQA). Consider to put them in value or delete the figure.
• Line 259 & 374: What do the authors mean with “historical runs”, I assume hydrological runs forced with climate model data from the period xxx to xxx?
• Line 389: term “expected” is misleading as this metric can take on any value.
• Line 392: Which are the 17 runs? The owns with RCP8.5 only?
• Fig 5: how do the results differ spatially?
• Line 438: title of section is misleading as also results mid of century are presented
• Lines 439-440: Starting a results paragraph with saying that parts of the results are presented elsewhere reads strange.
• Line 481: What do the authors mean with changes are limited when referring to Fig. 7? I can see substantial changes from ranging from minus to plus 50% covering most and parts of France, respectively?
• In whole section 6: I would recommend not to use the wording “significant change” as no statistical test was applied
• Line 504: The authors say that the most significant changes are in the summer. Looking a Fig. 7 RCP8.5 I do not see a large difference to fall.
• Line 514: North-south gradient. Please be a bit more specific to which season and which variable as I do not agree on the sentence “most maps show a north-south gradient”.
• Line 514: “This gradient is already present in the ensemble of climate projections”. Please add that this was not shown in the manuscript or put a figure in the Supplement.
• Line 546: Please mention that with the storylines only climate model uncertainty is presented (and maybe only a part), while the other sources are presented elsewhere.
• 9: Do you want to discuss why the yellow story line (which should be a moderate one) is more extreme than the median changes presented in Figure 7? This is pretty obvious by design, but I recommend the authors to do not leave these kind of thoughts all readers.
• Fig. 10 is not mentioned in the text.
• Fig. 10: I would suggest plotting also all individual projections with moving averages. The annual variability is not the point here to my opinion.
• Line 610 points to an outdated data set, there is a new doi provided

Review of “A large transient multi-scenario multi-model ensemble of future streamflow and groundwater projections in France” - submitted to HESS

Sauquet et al. have submitted a manuscript detailing the development of a hydrological climate projection dataset for France, named Explore2. This dataset builds on a previous dataset, developed in the early 2000s. The work contains the collection of 36 EURO-CORDEX simulations, bias-correction and subsequent hydrological modelling.

When reading my review, it is relevant to know that I have a climate scientific background, rather than a hydrological background. As such, I have mostly focussed on the climate projections part, the logic of design decisions and and clarity of text/concepts.

In general, I think the project and hence the manuscript are of high relevance. High quality climate and hydrological projections are of high value to any society that aims to adapt to future climate change. As such, I see potential for the submitted manuscript. However, in my opinion the writing of the manuscript has not been done very well at all. There is bad flow, important details are only mentioned as a by-the-way, and I kept having to go up and down to figure out how many and which simulations/datasets were being used at what point. In the discussion of the results I frequently was confused as to what the aim of the manuscript was, given many references to other papers showing the results.

I therefore recommend the editor to request major revisions before considering publishing in HESS.

Major comments

CLIMATE SERVICES

Explore2 is a classic ‘climate service’, I suggest the authors add some literature on good practices of climate services in their paper/introduction. Part of the discussion from line 416 onwards might suit in this part of the introduction.

Also, please clarify the relationship between Explore2 hydrological projections and existing meteorological projections from MeteoFrance. Are the two related and how, if not, how are users meant to combine or choose between the two? Hydrology is a classic cross-border problem, with for example shared river basins and shared aquifers.

CLIMATE SCIENCE BASIS

Section 2 - This section is vital in your Explore2 project/data stream and hence in consequent use of the data. Without understanding of the source of climate data, how they have been bias-corrected, and how they sit in the range of uncertainty projections from CMIP/CORDEX/etc sources, I would argue the resulting data lose a large part of there value for users. The current description of data, data treatment and uncertainties are incomplete and lacking clarity. In my opinion, this needs to be addressed before publication in an international scientific journal like HESS can be considered.

• The most important detail, for me, is which climate data underpin the Explore2 dataset and is not mentioned. The report that supposedly details it is written in French, Section 2.1 does not offer any information beyond a three-sentence standard introduction of EURO.CORDEX.
• Please include more details on your bias-correction methods. Why were these two methods chosen, are they uni- or multi-variate, how are they combined? The information about SAFRAN being the baseline reference should probably appear earlier, where you introduce the climate model data?
• Show me, rather than tell me, that you have assessed the compatibility with CMIP6. If you feel such figures are not the main message of this paper, you can add them to the supplement. For example, I know that CMIP6 models are much warmer than CMIP5 (higher climate sensitivity), how have you dealt with that? 
• I very much like the idea of offering a subset of ‘representative’ storylines for users. Your statements of ignorance and misuse are a bit harsh though, and without examples/references don’t sound very scientific. Furthermore, please add details on how the four storylines were chosen. In Table 1 if the median is shown in brackets, do I assume right that the other values are means (add to caption please), in Figure 1 what is the difference between grey and black markers?

MANUSCRIPT

Writing style is vague. Many sentences are unnecessarily long or round-about. Many sections are relevant but empty (multiple 2 sentence sections!) Please critically re-read and assess the different paragraphs and sentences to improve content and clarity. E.g

• “The availability of an updated and state-of-the-art ensemble of climate projections since the publication of Explore2070 conclusions, and the increased needs of users underpinned the idea of a new project called Explore2.” —> “A new project, Explore2, was initiated to take advantage of the availability of updated climate projections (CMIP6) and to address the increased needs of users since Explore2070.”
• “This section describes the climate projection dataset used for deriving the Explore2 dataset through the use of hydrological models (HMs).” —> ““This section describes the climate projection dataset used for deriving the Explore2.”
• “Projections from the 5th Assessment Report instead from the 6th Assessment Report were selected for Explore2 as they were, at the moment they were retrieved, the only large set of high-resolution projections over France.” —> very unclear sentence?
• Very very short sections: 2.1, 2.2, 3.2, 3.3

Minor comments

Line 60 - “large noise added by internal variability”: whilst I agree that internal variability adds noise when one is trying to quantify the forced climate response, for adaptation design I wouldn’t call it a source of noise. Instead, good adaptation practise includes preparing for the range of possible climate outcomes, which are determined by the combination of internal variability and forced climate change.

Line 64 - can you include references (scientific or other, policy documents etc) that show how Explore 2070 has contributed to developing water management strategies?

Line 68 - “The availability of an updated and state-of-the-art ensemble of climate projections” add a reference to which you mean? CMIP6, EURO-CORDEX, other?

Line 71 - what does ‘two years of maturing’ mean?

Section 1.2 - I’m not sure it is relevant for the paper to understand your work package structure.  Instead I suggest you add more information on the specific requests/boundary conditions that were set that Explore2 should adhere to. Interesting questions: how did you learn the user needs, was there any co-production process set up, did you evaluate HOW the Explore2070 data was used, and what users missed from there (apart from time horizons and spatial detail)?

Line 159 - states that there are 12 models. Why are there only 9 models in Table 2? Or should I add the 4 models in Table 3? That makes 13 models? How have these models been selected?

Line 374 - these are important insights, maybe more suitable for the introduction or discussion than halfway your results. In the results I expect the proper/best way of assessing quality, not a methodological discussion.

Line 439 - I’m confused, 3 other papers describe the results of the projections? I fully expected these to be the main result of this paper.

Line 546 - Another paper that supposedly describes what I expected in this paper?

Line 590 - More diversity is not always better, especially if one completely loses track of everything used (as in this manuscript). This sentence is a bit weird in that way, did the eFLaG and Australian projects have the same aims and scope? Where the assumptions/budgets/relevance similar? E.g. flood protection (studies) in the Netherlands gets far more attention than flood protection (studies) in, lets say, Austria.

Line 597 - This paragraph I expected in the introduction, what is the policy reality where your dataset adds to. Now it is the closing statement, I’m left wondering how it relates and is useful for that reality