Dear Authors,

Your manuscript has received three reviewers that raise important issues to improve your paper. You have addressed all reviews and I recommend taking them into account in a revised version of the manuscript, which I will review before final acceptance for publication.

We thank you again for submitting your work to the special issue on seasonal forecasting, and look forward to receiving the revised version of your paper.

Best Regards,
Maria-Helena Ramos.

Dear authors,

I read carefully your answers to three reviewers and your revised version of the paper. I agree with the reviewers that the paper is worth publishing and I consider that you have addressed well the remarks and suggestions from the reviewers, which greatly improved your manuscript.

I have just a few remarks that I think would be interesting to be considered for the final paper (I consider them just technical remarks with no need for a new round of revision, but I would appreciate if you could consider addressing them). They are listed below.

Best regards,
Maria-Helena Ramos

Remarks:

1) I agree with the first remark of reviewer #3 about using a single-value forecast to post-process streamflow predictions. Technically, ensemble dressing with model errors can consider all members of an ensemble without the need for reducing the ensemble to a mean or median value. I can understand your choice not to use the full ensemble members, but I have more difficult to find your arguments convincing. I think the sentences on lines 4-8, Page 10, need some rethinking. I do not think that the applications you cited for aggregating ensemble predictions are used in the correct context. The study of Matte et al. (2017), for instance, focused on a utility model, not a post-processing development approach. Besides, the ESP approach is common in operational seasonal forecasting for water management and it is based on ensembles of historic precipitation used as input to a hydrological model. I would consider writing this sentence differently: “Although the use of aggregation approaches for single-valued streamflow forecast from ensemble predictions has been seen in operational applications (e.g., Lerat et al, 2015…), we note that this approach may result in some information loss.”
Then, the sentence on lines 7-8 also raises a question: do you mean further work is needed within your study or in general? I think you could have considered the full ensemble with further work within your study, but I also think that we can find ensemble post-processors in the literature, so it is not as if there is nothing done on the topic. Post-processors that use the full information from ensembles do exist. I suggest that you either delete or you explain better what you mean with this sentence.
These are suggestions to clarify your opinion on the topic as expressed in your paper, so the reader can better apprehend your viewpoints.

2. I also have an issue with the calculation of what you call “sharpness” and I think it comes back to a remark of reviewer #1 on using “sharpness versus precision”. As indicated in Eq. 10, it is not the same as calculated in forecast verification studies, as presented in the reference books: Jolliffe, I.T., and D.B. Stephenson, 2012: Forecast Verification: A Practitioner's Guide in Atmospheric Science. 2nd Edition. Wiley and Sons Ltd, 274 pp. , and Wilks, D.S., 2011: Statistical Methods in the Atmospheric Sciences. 3rd Edition. Elsevier, 676 pp.
Sharpness is supposed to be an attribute that measures the degree of variability of the forecasts or concentration of the predictive distributions. In forecast verification, sharpness is a property of the forecast only. In Eq. 10, if I understand well, you calculate a type of coefficient of variation with the standard deviation of the time series of (median?) predictions and the mean of the observations. It could be understood, I guess, as a normalized "standard deviation", but I think it is misleading to call it “sharpness”, and I suggest you to call lit otherwise to avoid misunderstanding from the forecast verification community.

3) Minot issues are:
- Page 1, line 13 (abstract): consider changing to “… compared to a longer calibration period…”
- Page 2, line 12: you explain in your reply to the reviewers that you prefer to use “forecast” instead of “forecasted”. Maybe you should do the same here for the sake of consistency.
- Page 2, line 15: I would consider changing to “… user need for monthly to seasonal streamflow forecasts…”. In fact, the use of “seasonal” alone in the paper is appealing, but the study is dealing with “monthly” forecasts and not seasonal forecasts. I think you should not mix both and should keep to using “monthly” when referring to your study and application. For instance, I suggest to consider this remark in these other parts of the paper: Page 4, line 12; Page 4, line 22; Page 17, line 8.
- Page 2, line 20: consider changing to “… calibration; Mount et al., 2016; …)”
- Page 2, line 24: consider changing to “…(Wu et al; 2013); however this is not always the case (for example, Brigode et al., 2013)”
- Page 3, line 4: consider changing to “…etc.); see, for example, Westra et al., 2014).”
- Page 3, lines 10-11: consider changing to “…calibration period, which exposes….of a shorter calibration period, which exposes the model…”
- Page 3, line 17: consider changing to “…and Maranzo, 2004) and disaggregation approaches…”
- Page 4, line 4: “… forecasting applications in France.”
- Page 4, line 17: consider changing to “... of this study are to:”
- Page 5, line 3: consider changing to “…(Figure 1), which conveys…”
- Page 8, line 5: consider replacing “big” to “important”
- Page 8, line 28: I would write “10-year calibration period” or “one-year warm-up period”, etc.. Consider checking if that would be correct in English and change over all the uses in the paper.
- Page 11, line 13: with only one reference, I suggest deleting this part of the sentence “and are common in forecasting applications” and only mention “… monthly time scale (see another example in Lerat et al., 2015)”.
- Page 11, line 19: consider adding a comma: “For all scenarios, observed…”
- Equations on Pages 12-13: in general, I think the terms of the equations are not clearly presented. For instance, “Qt in bold” for predictions and “Qt~” for observations is only clearly stated in line 16 for Eq. 12, while it was already used in Eq. 10. Then in Eq. 14, we have a “Qt Teta”: is it the same as “Qt in bold”?
- Metrics: some metrics seem to sum/average over the time series before computation (Eq. 10 and 11), while other compute at each time step (or forecast monthly lead time) and then take the average (Eq. 12 and Eq. 14). Maybe it is worth untangling this issue when introducing the equations.
- Page 13, line 16: “These changes…”: what changes exactly are you referring to? Please, clarify.
- Fig. 5: how do you explain the simulated peak flow at the end of 2002 that is not observed? Isn’t there a problem with the observations? Could you say something about it in the text?
- Fig. 8: for C1, at least, the parameter “split” seems to evolve around 0.9, therefore, finally, it is worth considering it as a free parameter to calibrate? Could you say something about it in the text?
- Discussion: please note that state updating was used for seasonal forecasting in “Crochemore, L., Ramos, M.-H., and Pappenberger, F., 2016: Bias correcting precipitation forecasts to improve the skill of seasonal streamflow forecasts, Hydrol. Earth Syst. Sci., 20, 3601-3618, doi:10.5194/hess-20-3601-2016”. The authors also discuss about trade-offs between forecast attributes and it could be worth having a look at it.
- Page 16, line 26: I am not sure about using the word “appreciably” here. I suggest deleting it or making more explicit what you mean here.
- Page 17, line 8: “change to “These catchments were…”