The manuscript “Temporal downscaling of precipitation time-series projections to forecast green roofs future detention performance” presented by Pons et al. describes the procedure to simulate rainfall through multiplicative random cascades. Six downscaling models, with time-scale, depth, and temperature dependency, have been developed to simulate time-series based on RCP8.5 to be applied on two green roofs with different properties and in different locations in France and Norway. Two green roofs have been modelled. Results show that the green roof performance shift due to climate change highly depends on the location.

Although the manuscript presents an innovative methodology and interesting results, it needs few minor adjustments and modifications before publication in an international journal such as HESS.

Green infrastructure, and in particular green roof, description could be more detailed in the introduction, explaining why green roofs have been chosen for this study and better highlighting the potential benefits of this work. Meaning and implications of RCP8.5 projections should also be included in the introduction.

The use of abbreviations should be reviewed: sometimes abbreviations are not introduced, or not used. An abbreviation and symbol list could help the reader.

The paper is overall well structured, although it could benefit from a review to catch typos (double or missing spaces etc..)

Specific comments [Page, line]

[Abstract] Please avoid unnecessary abbreviations in the abstract

[1, 9-10] Please add an explanation for this result

[1, 14] What’s a?

[1,15] Evaporation processes time-scale could be higher than 24 hours, please add a reference to this sentence.

[1, 20-24] What is the temporal resolution of these data? daily scale?

[2, 40] Please use the previously introduced abbreviation MRC          

[2,45] Reference written twice

[2, 54] What is MC? This abbreviation has not been introduced before

[2, Introduction] Please consider adding some background for green infrastructures and in particular for green roofs.

[3, 66] and in Marsille? what data are available?

[3, 69] Climate projections in France have the same temporal resolution and length of the Norwegian projections? please specify

[5, 100] The acronyms used here have not been introduced before

[8, 170-174] This paragraph fits better in the model description

[9, Figure 2] Please include in the figure the green roof type

[9, 175-185] Consider adding the results related to all locations in the appendix, to help the reader understanding the results

[17, Paragraph 3.5] It would be interesting to have some considerations also for the French case studies

This paper investigates the use of downscaling model to forecast green roofs performance in the context of climate change. It uses a downscaling approach based on multiplicative cascades. The topic is interesting and relevant for the community. However I would not recommend to publish this paper in its current state and suggest major revisions. Indeed it requires significant clarifications on the downscaling model. Indeed, its presentation is hard to follow and should be more detailed. It notably seems that different distributions of weights are used according to the cascade step suggesting they are not scale invariant. The calibration process of the numerous parameters (up to 19 !) needs to be explained. Please also clarify that what is called “observed data” for the various figures is actually simulations with observed rainfall. Am I correct ?

Detailed comments :

- l. 40-44 : It should clearly be stated that canonical cascades ensure conservation on average only while micro-canonical ones ensure exact conservation of intensity at each step.

- l. 54 : should MC be MRC ? In general, the use of numerous abbreviations does not really help the reader. I would suggest to limit their use to words really often used in the paper.

- Section 2.2.1 : I think there is a need to be more specific, notably for the reader not specialist. Index i and j should be consistent between equations 1 – 2 and Fig. 1. Please also clarify the range of possible values (if “i” refers to a time step then it belongs to 1… 2^n where n is the cascade step and j * 2^n = total duration ?). Eq. 2 : S is said to measure a proportion while it has only 3 possible values. Please clarify.

- Section 2.2.2 : Please clarify how the fitting of the models was done. Are the probability distribution used the same at all cascade steps (only for P(W=0) if I understand well table 2) ? Was a scaling break identified in the data ? What would be the consequences of such break. l. 101 : “all included 5”, may be say all included the use of eq. 5 to help the reader. Please explain how the depth or temperature dependency was included. It would also be needed to clarify in Table 2 to what refer the parameters mentioned.

- l. 144-145 : why limiting to lag-1 ?

- l. 152-153 : how do you define “small”, “major” and “extreme” events ?

- Section 3.2 : How to you interpret physically the differences of behaviour in Fig. 3.a ? Is the shape of Fig. 3.b the same for other time scales ? How the fitting of the model was done from this analysis ?

- l. 212-213 and comments on Fig. 4. c and d. Why is the discharge considered to be only slightly underestimated while the observations do not fall in the 5-95% percentile ?

- l. 216-218 and fig 4.e : the discrepancies between models and observations for the lag-1 autocorrelation should be discussed more.

- l. 335-338 : I have trouble to find the figures mentioned in Fig. 5.

- Fig. 5 : last row (extreme events). May be the vertical scale could be split to enable a zoom on the lower part which concentrate most of the information which is not visible now.

- Fig. 6 : “observed data” is not visible in the graphs

- Section 3.5 : without explaining everything, I believe that some details the variational approach are needed for the non-specialist reader. How do authors interpret the fact that the differences between the two approaches are much more pronounced extensive roofs than the detention based ones ?