Dear Authors,
thank you for having submitted your work to HESS.

I agree with the two Referees that the manuscript requires major revision. The major concern of both Referees on your work is the lack of novelty in the proposed approach and, as I wrote in my initial comments to your submission, this is my main doubt too: please take particular care in addressing this issue in the revised manuscript, explaining why your work may provide an advance in the real-time streamflow forecasting field.

My best regards,
Elena Toth
HESS Editor

Dear Authors,
thank you indeed for resubmitting the paper and for the detailed reply letter, where you addressed the points raised by the Referees and by myself.
The presentation is certainly clearer and the referees are very satisfied with your explanation of the novelty of the work, but I agree with them that the manuscript may be further improved.
In particular, please consider carrying out the following two additions, that the Referees asked in the first comments but you preferred not to perform in the revision:
1) Improve the analysis/interpretation of the evaluation of the forecast skill, considering other indexes or plots to support the results given by the containing ratio CR
2) Try to estimate the transformation parameters also by means of a likelihood approach, comparing the results with the present ones and attempting to interpret the differences and relative limitations.
(please address also the other points raised by the referees, clarifying the text in order to help all the readers to better understand and interpret your work)
I look forward to receiving the last version.
Best regards,
Elena Toth

Dear Authors,
thank you indeed for having submitted a new version where you adequately address all the comments raised in the second revision phase, including adding the QQ-plots and applying also the likelihood parameter estimation procedure.
I have sent the manuscript to one of the previous referees, who has some doubts on the new Figure 10 plots (see below).
Actually, I believe that the correct number of points in Figure 10 should be the number of hourly data in each season for the entire validation period (5 hydrological years), so around 450*24 (10800 hours) points, and indeed in Figure 10 there seem to be much less dots: can you please explain it?
Best regards,
Elena Toth

Comments by Referee

First of all, I really appreciate that the authors worked very hard to address all questions raised by
reviewers. That is why I did my best to ensure the published version of this paper is of high quality. I
hope my comments are constructive rather than critical.
I suspect that the PIT plots in Figure 10 are incorrect.
(1) The number of points in the PIT plots is not consistent with the number of hourly-data in the
validation period. The case study uses four-year data in the validation period. That makes the
number of the total hourly data is ~35040 or ~8760 per season. If the forecast system issues one
forecast per day, there should be around 365 points in each PIT plot. Based on my visual inspection,
there are much fewer points in the plot. In addition, based on the Kolmogorov critical value is ~0.25
for each plot and the approximation of critical value (critical level is 0.95) for the Kolmogorov test is
1.358/sqrt(n), I estimate that only ~30 points are available in each plot. I suggest you use all hourly
data (~8760 per season) to construct PIT plots.
(2) I am not sure how the authors generated ensemble forecasts to make PIT plots, especially for
the 24h LT forecasts. I only can see the deterministic forecasts defined by Equation (3) and the
forecast interval estimated by LRVE (Page 10 Line 7). The forecast interval only provides information
on forecast spread but the PIT plots provide more. Can you elaborate on how to generate the PIT
plots from the forecast intervals?
Some of Figure numbers are not updated. For example, (Page 13 Line 13) Fig. 4a should be Fig 5.

Dear Authors,
thank you indeed for having corrected the PIT-plots. Actually, I believe that the new plots, providing very different results from the previous ones, would deserve more discussion, especially to explain the causes, in particular trying to explain the reasons why the uncertainty bands are higher at short lead-times (1 hour) than at longer ones (adding some additional explanation of the method used for the probabilistic forecasts would probably be useful).
(and I would also suggest explaining, for the readers’ benefit, how to interpret your PIT-plots, explaining why the shape you obtained means ‘large forecast’)
In addition, the new findings given by the PIT-plots on the limitations of the uncertainty estimation method should also be reported in the Conclusions section, which now is still the same as the previous version.
Minor comment: In the new paragraph (l. 1-3 p. 20), specify to which ‘lower than expected percentage of coverage’ you are referring to.
Best regards,
Elena Toth

Dear Authors,
I am glad you revised and are now presenting the correct results of the uncertainty analysis, which seems more plausible than those of the previous versions and are well interpreted by your remarks.
Provided you do not foresee to do other checks, this version may be published, as far as I am concerned.
Best regards,
Elena Toth