General comments

This study evaluates two approaches for simulating over one century-long daily streamflow at different Upper Rhône River catchment locations. The analysis examines the potential of atmospheric reanalyses simulations (ERA-20C), conceptual glacier-hydrological model and two downscaling approaches for providing temporal variations of river discharges, low flow sequences and flood events. The results indicate that bias correction is crucial for both precipitation and air temperature and both downscaling models. While the observed multi-scale variations of discharges (daily, seasonal and inter-annual) are well reproduced, the results for low flows are less satisfactory.

The manuscript is within the scope of the journal and has a good structure. Still, the novel scientific contribution can be better formulated. The Introduction says that the comparison and evaluation of different downscaling approaches have been the subject of multiple previous studies, so it is not clear what are still the research gaps and how this manuscript contributes to some novel findings/knowledge. The manuscript will also benefit from some more detailed justification of why such one-century runs are needed and/or beneficial. A discussion of the value of the proposed approach with some alternative approaches (e.g. stochastic weather generator) will be useful as well. The proposed modelling chains are quite complex, so a rigorous description of the methods is challenging, resulting in difficulty in reproducing the proposed experiments exactly.

My second suggestion is to add more process-related interpretations of results and findings. For example, to discuss and present which processes lead to low flows and floods in the study area and to demonstrate that the suggested chains represent well such runoff generation processes in the (sub) catchments. For example, I’m not sure whether individual correction of the bias for air temperature and precipitation does not result in some artificial combination which can affect, e.g. low flow simulations of the model. If the low flows occur in winter, the impact of air temperature bias correction will be more important than precipitation corrections or vice versa. So it will be interesting to provide more process interpretations in the results.

My third comment is on the validation of the results. It will be interesting to see how the procedures work in some independent (validation) time periods. I wonder what is the relative contribution of each individual member of the proposed chain on the final results.

Specific comments

Abstract: Please make the description of the results consistent with the conclusions. In the abstracts is written: “… the hydrological situations of low frequency (low flow sequences and flood events) are reasonably well reproduced.” However in the conclusions is written: “he results for low flow activity are less satisfactory.”

Figure captions. The abbreviations used in the figures are not always explained in the captions, so it is sometimes difficult to understand them (without a detailed reading of the text)

The manuscript addresses the important topic of downscaling and bias-correction for hydrological modeling by applying different configurations of a “simulation chain” over an alpine river over the last century. I found the paper interesting and including meaningful results. Nevertheless, the paper lacks both important details on the hydrological modeling and discussion on the uncertainty related to the hydrological modeling within the applied simulation chains.

Regarding the hydrological modeling methodology:

• L118: please consider adding a diagram of the GSM-SOCONT model in the paper or at least in the Appendix.
• L122: are these “ice-covered” and “ice-free” parts of the studied catchments dynamics in time? Indeed, the ice-covered part of the upper catchments may have significantly changed over the last century.
• L212: what are these additional criteria considering streamflow availability?
• L124 to 126: the calculation of the potential evapotranspiration (PE) time series needs to be more deeply presented in the paper: which CRU dataset has been used for this calculation, what is the spatial resolution of this CRU dataset, and how relevant is this database over this region and in this hydroclimatic context? What about this regional temperature-PE relationship in a non-stationary context?
• L147 to 148: the routing part of the hydrological model needs to be more deeply presented in the paper. What are the hypotheses? How many parameters are devoted to the routing part of the model?
• L149 to 151: please state in the paper the total number of parameters that needs to be calibrated for each sub-catchment and add a list of these parameter in the paper (in the Appendix?).
• L183 to 184: please give more details on the regionalization procedure applied, and list the ungauged catchment studied here.

Regarding the discussion about the hydrological modelling uncertainty within the applied chains:

• L172 to 174: “For gauged catchments for which the hydrological behavior is significantly altered over P1 and for which "natural" flow observations are available prior to 1950, parameters were estimated based on hydrological signatures (Sivapalan et al., 2003; Winsemius et al., 2009). In the present case, parameters are calibrated so that simulated signatures reproduce at best observed ones but observed and simulated signatures come from different periods following Hingray et al. (2010) (e.g. 1961-2015 and 1922-1963 respectively for the Viège basin).We thus assume that the weather regimes and the natural hydrological behavior of the catchment have not significantly changed over the last century, which seems a reasonable assumption to make in first approximation.” This strong hypothesis needs to be more deeply investigated in the paper: what about potential significant interannual / decadal hydro-climatic variability in the region? What about potential impacts of such interannual / decadal hydro-climatic variability on the hydrological model parameters? What about using adjustment on long-term climatic information (e.g. Nicolle et al., 2013, doi: 10.1002/2012WR012940)?
• L180: Table S1 needs to be presented in the paper and not in the Appendix, and more deeply discussed, since the performance obtained on several subcatchments are poor (e.g. Arve@Genève, BDM). What are the potential reasons for these differences in performance between the studied catchments? Please also consider producing the Figure S2 for every studied catchment in the Appendix.
• L279 to 284: “As many sub-basins have altered hydrological regimes, the "hydrological reference" used for the comparison is the discharge time series obtained via hydrological simulation with the "observed weather" as input. For some upstream sub-basins, which hydrological behavior can be considered as roughly natural, the evaluation could also rely on a comparison with discharge observations. We however choose to use the simulated reference. This first makes the evaluation homogeneous for all URR sub-basins. This additionally allows to only focus on the ability of downscaling chains to simulate hydrologically relevant weather scenarios. In other words, this allows to not distort the evaluation by intrinsic errors introduced by the hydrological model.” This point is critical and needs to be more deeply discussed: what about “simulated reference” that are not hydrologically relevant, i.e., highly different from observed streamflows (e.g. Arve@Genève, BDM catchment, with a NSE after calibration equal to 0.44)? What are “intrinsic errors introduced by the hydrological model” in this context?
• L424 to 429: this delayed dynamics between air temperature, snow accumulation / melting and thus simulated streamflows might be compensated (for good or bad reasons) by the hydrological model parameters during calibration. This point has to be discussed in the paper.
• L440 to 450: what are the potential impacts of these changes (on the air temperature lapse rate) on the calibration of the hydrological model parameters? Did you try to do another calibration of the model considering these changes? Please discuss this point in the paper.
• L461 to 466: what are the potential impacts of these changes (on the orographic precipitation enhancement) on the calibration of the hydrological model parameters? Did you try to do another calibration of the model considering these changes? Please discuss this point in the paper.
• L481 to 488: did you check these hypotheses by comparing simulated streamflows with observed ones (and not “simulated references”)?

Other specific comments:

• L90: what is a “Binn-Simplon” situation?
• L92: what is a “retour d’Est” situation?
• L114: please consider highlighting the spatial resolution of the ERA-20C dataset on Figure 1 (in the topleft panel?)
• L213: please consider highlighting the spatial resolution of the MAR model on Figure 1 (in the topleft panel?)
• L240: how many previous days are considered? Please add details on this point.
• L243: please details what is a “non-parametric method of fragment” in this context.
• L336: the comparison of the Figure 8 and 9 is not easy: please consider adding another figure that present flow differences on each catchment and configuration to ease the comparison.
• Figure 8: please add the observed flows when available. What is the plateau observed for the “reference” streamflow series of the Rhône@Geneve in 07/1982 (and for other dates)?
• L398 to 402: please add some references on the use of ERA-20C for hydrology or other references related to this potential ERA-20c drawback.

This is a very well written  paper (with high quality figures ) on discharge simulations based on climate model outputs, with “a typical simulations chain” (abstract first line). The paper was easy to read, which is a great achievement for such a complex modelling paper. The topic is probably undersold: it is not just a case study but one of the few attempts to produce long simulations of the past, with a high number of challenges to overcome. I am not aware of long simulation studies back in the past in mountainous environments with climate model outputs (I might not be on top of that literature but I suspect there are very few.

Accordingly, the added value to the literature should be formulated in a much more straightforward way as early as possible in the introduction and again in the conclusion. 

The method is not well captured in the abstract or in the paper overall and calling this a “typical simulation chain” is perhaps underselling. A critical step in such simulation chains is how to combine reanalysis data with a hydrological model, especially in mountainous areas. What does this study do differently than other works? Why is the approach more interesting?

A concise summary of the method would help (take a hydro model, calibrate on observed meteo and streamflow data, run with meteo scenarios generated based on reanalysis data; contrary to many other works, no simple spatial downscaling of reanalysis data but use of a weather generator to produce mean areal precipitation scenarios).

Details:

• Method: it should be made very clear at what scale MAP (mean areal precip) and MAT (temperature) are defined. I see from the sentence “bucket-type model that uses time series of Mean Areal Precipitation (MAP) and Temperature (MAT) as inputs for each hydrological unit.” that MAP and MAT is defined at the elevation band scale (?);

4.1.1:

• “a daily potential evapotranspiration time series is derived from temperature”, how? Linear?
• “For each RHHU, daily MAP and MAT are estimated from neighboring weather stations using the Thiessen’s weighting”, why Thiessen? Probably very inappropriate for mountain environments? And how is the Thiessen obtained between points (weather station locations) and elevation bands?

4.1.3:

• what is the storage equation in this model? Is it the storage-volume relationship? Where do you get if from?

• how do you construct the linear outflow relationship? As far as I see from one year of observed data (extract from the pdfs https://www.hydrodaten.admin.ch/en/seen-und-fluesse/stations/2028 and https://www.hydrodaten.admin.ch/en/seen-und-fluesse/stations/2606) any outflow is possible for low water levels (see figures below) and a linear relationship only holds for high water levels and flow beyond 500 m³/s
• 
• 
• Can you say something about the evaporation equation, what does it include? I have not seen the Rohwer’s equation (Rohwer, 1931) before

4.2.2.:

• “In the present work, SCAMP was used to generate 30 time series of daily spatial weather scenarios for the 1902-2009 period from ERA-20C reanalysis outputs.” At what spatial scale? At the scale of the 7 km x 7km of the MAR model? This would be in contradiction with Fig. 3 that says that you generate MAP (mean areal precip) estimates and not first spatial estimates? So, do you first produce spatial estimates, i.e. daily time series per pixel? If yes, how do you combine the pixels into MAP estimates? By taking the pixels within the elevation band? Does this make any sense?
• Method overall: how do you have confidence that the conceptual model calibrated on observed streamflow data with observed station meteo does a good job if used with weather generator scenarios produced at a different scale? What part of your method ensures this?
• Model calibration on signatures: did you check for the catchments with observed concomitant streamflow if the signature calibration gives good results?

4.3, bias correction

• Do I understand correctly that bias correction is done independently for temperature and precipitation, which are also downscaled independently. How do you ensure a good link between the two for the simulation of snow accumulation and melt?

Results

• In the MAT series, we seem to see nicely the 1980 global regime shift (Reid et al., 2015), , with a sharp increase of temperatures, perhaps worth commenting?
• Figure 11c would certainly contain some very interesting information but I cannot see anything. I would split into two figures, and rescale to 200 and 500 m3/s; furthermore, you could perhaps some coefficient of variation and autocorrelation at lag 1 instead of simply the figure?
• Figure 11b: it is written “The results for low flow activity are less satisfactory, especially for the 1920-1949 sub-period (Fig. 11b). The number of low flow sequences below the threshold in both simulations is indeed twice that of the reference. This suggests a limitation of both downscaling models to simulate long persistent dry sequences.” This interpretation is perhaps a bit too limited; most readers might not know that there was a e.g. a heat wave in 1946 and a period of strong melting in the 1940 due to enhanced solar radiation (Huss et al., 2009); there were also some very cold winters in this period and some very warm years (see here https://www.meteoswiss.admin.ch/climate/climate-change.html). I would give some more details here and perhaps use other subperiods to better understand what is going on with the modelling chain? This low flow underestimation is really too striking to not discuss it in far more depth. This is really important because many modelling chains do a rather poor job for snow-influenced low flow but this is rarely discussed. And: it is important to be precise here: is the reference for period 1920-1960 simulated or observed? To understand if the lake management can or cannot explain the low flow differences.
• In fact, do I understand correctly that the reference is not always the same for all result plots (which would not be ideal and should be mentioned in all plots, in the caption)? At line 382 following it is written “Simulated year-to-year variations of mean annual discharges are next compared to the "reference" ones over the 1920-2009 period (Fig. 11c). The "references" are observed discharges for the 1920-1960 period and simulated discharges from observed weather for the 1961-2009 period.”

Discussion

• What could explain a warm bias in the lower amtospheric layers of the model? Besides: would be interesting to see winter and summer lapse rates separately to gain more insight
• Line 493 following: “Note that in our simulations, the dry bias in the precipitation input is probably corrected via an adjusted parameterization of the hydrological model, (..)”. I fully agree with this explanation but I would make it clearer for non experts; now it reads like if someone adjusted it, perhaps say something like: “the dry bias is probably compensated during hydrological parameter optimization; parameter optimization generally results in forcing the model to close the water balance; conceptual reservoir-based models as the one used here can thereby compensate missing rainfall input by lowering the evapotranspiration losses. This problem is well known but very rarely discussed. An example is the work of Minville  et al. (2014).

References

Huss, M., Funk, M., and Ohmura, A.: Strong Alpine glacier melt in the 1940s due to enhanced solar radiation, Geophysical Research Letters, 36, L23501, 10.1029/2009GL040789, 2009.

Minville, M., Cartier, D., Guay, C., Leclaire, L.-A., Audet, C., Le Digabel, S., and Merleau, J.: Improving process representation in conceptual hydrological model calibration using climate simulations, Water Resources Research, 50, 5044-5073, https://doi.org/10.1002/2013WR013857, 2014.

Reid, P. C., Hari, R. E., Beaugrand, G., Livingstone, D. M., Marty, C., Straile, D., Barichivich, J., Goberville, E., Adrian, R., Aono, Y., Brown, R., Foster, J., Groisman, P., Hélaouët, P., Hsu, H.-H., Kirby, R., Knight, J., Kraberg, A., Li, J., Lo, T.-T., Myneni, R. B., North, R. P., Pounds, J. A., Sparks, T., Stübi, R., Tian, Y., Wiltshire, K. H., Xiao, D., and Zhu, Z.: Global impacts of the 1980s regime shift, Glob. Change Biol., n/a-n/a, 10.1111/gcb.13106, 2015.

This paper by Legrand et al. presents and discusses the results of a modelling chain that uses large-scale atmospheric information (ERA-20C reanalysis) to produce continuous daily river discharge at several stations of the Upper Rhône river catchment (~ 11000 km²). This modelling chain includes a semi-distributed hydrological model (GSM-SOCONT) and a downscaling model. Two downscaling models (statistical or dynamical) are actually used and tested. The results are analysed over several periods (1902-2009, 1961-2009, 1981-1983, 1920-1949, 1950-1979, 1980-2009) and according to various indicators (daily discharges, mean monthly and annual discharges, low and high flows).

This paper clearly corresponds to a huge amount of work. A lot of material and results are presented, the Figures are very rich. The drawback of this very rich content is that it is not always easy for the reader to navigate and appreciate the results in the present form of the manuscript. My main remarks and recommendations to improve the paper are listed below :

1/ Title and focus.

The paper is entitled « Simulating one century (1902-2009) of river discharges, low flow sequences and flood events of an alpine river from large-scale atmospheric information » which is a very general title and does not really correspond to the real focus of the paper. In my opinion, as stated by the authors themselves, the objective of the work presented here is to assess the ability of downscaling chains to simulate hydrologically relevant weather scenarios (p12, l. 282-284). This is consistent with the detailed description of the different steps of the downscaling models, and the presentation of the results as comparison to a reference that is not the observed discharge (but rather the discharge simulated by the hydrological model forced by reference observations). I therefore suggest to make this objective more clear in the Introduction, to change the title accordingly and to trim the parts of the text that do not contribute directly to this objective (a lot of details about the hydrological model description and set up could be placed in supplementary material for example).

2/ Figures.

The Figures contain a lot of information, maybe too much compared to what the reader is able to see and to what is necessary to illustrate the author’s point. They would gain a lot from a bit of trimming. As examples :

- Fig 1 : the names of the gauging stations on the map are not necessary (they are also in Fig 2) and prevent the reading.

- Fig 7 : is it really useful to present both MAT and DMAT ?

- Figs 8 and 9 : these Figs are more illustrative, maybe present more synthetic results before (Fig 10) + it could be worth presenting a Table with a few synthetic values (bias, KGE) so that we can have a general idea of how the simulation chains perform compared to reference. Why presenting simulations with BC before without BC ? + is it really interesting to present the time series of lake level ? + add the names of the stations directly on the Figures

- Fig 11 : too many lines on Fig c) + Figs a) and b) don’t work. It is not clear that there is a « column » for each period + the only thing we see is the comparison between SCAMP and SCAMP-BC for each period

- Fig 12 : Fig a) not legible. What am I supposed to see on Fig b) ?Why not removing MAR simulation (since lapse correction is only done to the MAR-BC simulation) ?

- Fig 13 : Fig b) : why lines and crosses on this Fig for the different simulations ?

3/ Acronyms and notations.

The paper uses many acronyms that are sometimes confusing and could be simplfied. I particularly struggled with MAP / MAR / MAT (the name of the model can’t be changed, but the names of the variables probably can), with basin-scale, RHHU, grid points, station (Fig 3 and accompanying text), see next remark. BC (Bias correction) is also not clear as it is used for different corrections (not the same for MAR and SCAMP, which I can understand, but also for temperature lapse-rate for MAR. About this last correction, if it is included in the results presented in section 5 (which is the case from what I can understand), why is it not presented along with the other corrections in section 4 ?

4/ Spatial resolution / Fig 3.

I really struggled (and did not succeed) at understanding exactly what was done in terms of spatial aggregation / disaggregation for the weather scenarios in SCAMP. In step (1) how do to change from station (= point) observations to basin-scale resampled values (basin-scale = whole catchment or subcatchments?). Are the stations at the end of the process the same as the observation stations ? (I understand rather RHHU = subcatchment but later in the text it is still referred as neighbouring stations ad inputs to the hydrological model, see p 22 l 410-411). Similarly Fig 4 presents the numerical experimentation plan with the hydrological model forced by P and T at 7*7 km resolution. Please, make all this clearer !!

5/ Time periods and indicators for the results.

As said before, there are a lot of results, presented for various times periods and various indicators that change from Figure to Figure. This does not make the reader’s task easy. I would be nice, along with the experimental setup, to define in a Table the various indicators and time periods used and explain why they were selected.

If I am not mistaken, the results of the calibration of the hydrological model for the reference simulation are not presented. It would be nice to add a few elements about this, just to make sure that the reference model is not completely off the track.

6/ Reference MAP and MAP.

I am not a specialist of estimation of weather variables in montainous areas but I am surprised by the methodology used here, which consists of Thiessen polygons with a density of stations that is not so high (62 raingauges for ~ 11000 km², even less temperature stations). There are several other methods for the estimation of areal P and T, from very simple such as inverse-distance weighting to more complicated (kriging). What is the reason of choosing such a simplistic approach ? How confident can we be with these « reference » values ? This should at least be discussed thoroughly.

+ p 24 the authors write « no precipitation-elevation relationship was considered » although p 6 : « and a regional and time-varying elevation-temperature relation ship ».

7/ Floods.

I have a few comments on the « flood » part of the paper. First, I do not really agree with the use of the term « flood » given what is presented here. In 5.3 (Fig 10) what is done is picking the maximum daily flow for each year, which does not necessarily correspond to a flood. Similarly, flood events can be expected to last several days given the size of the catchment so I don’t think that the max daily discharge can be called a flood event. In 5.4 (Fig 11), I suppose that the « flood activity » is defined as the number of days over the threshold, which again does not necessarily correspond to single flood events. The way of calculating the thresholds should also be precised (p 21, l 373-374 : are they defined through a flood frequency analysis as the value of the flood return period 1 year, or from the flow duration curve ? Therefore I would be more cautious with the term « flood » and would preferrably use terms like high flow indicators or maybe flood proxy indicators.

+ p 26, l 483-488 : the unrealistic simulated discharges obtained with altitude-elevation correction can also be due to the calibration of the hydrological model, considering in particular that the model was « high flow calibrated » (4.1.2). A rigorous way to test that would be to re-calibrate the hydrological model with the new corrections added to the observations before testing them on the downscaling models.