Dear Authors,
thank you for having uploaded your replies, where you have addressed the referees’ comments.

In your reply you have already identified what in the paper should be modified/integrated in order to improve it and I encourage you to proceed with the identified changes.

As I have already written in my Editor Comment, more details are needed on the data (Ref#2) and especially on the method (Ref#1), in order to allow the understanding of the work. Many doubts raised by the referees are due to a lack of clarity in the meaning and functioning of the SOM network.

On the specific comments/replies:

Ref #1 Comments and replies

a) I would indeed suggest to move (or to add) at the beginning of the paper the clarification that you are dealing with fluvial flooding only: that would also help to put in the correct perspective the state-of-the-art and would avoid any possible misunderstanding (see the first set of comments by Ref#1)
b) QE, TE and Davies-Bouldin indexes should be added as equations: the text now includes their meaning only
c) since the SOM methodology is far from familiar to the large majority of the readers, more detailed information on the method, on its meaning and on how to interpret its results are needed (see also doubts raised by Ref#2)
d) spatio-temporal: I do agree with Ref#1 that it would be better using a different term if not dealing with actual space dimensions.
e) the ‘cluster discussion’ issue will probably be solved when more detailed on the use and meaning of the SOM and on how to interpret the SOM results will be added.

Ref #2 Comments and replies

a) again, I believe that all the first comments made by Ref#2 (and probably also the comment on p. 8, lines 5-15) are due to the lack of clarity in the description of meaning and functioning of the SOM
b) as I said, I fully agree that more information on the database is necessary, too
c) I suggest highlighting also in the conclusions section (and not only at p. 4) the shortcoming due to lack of information on the actual protection level.

I look forward to the revised manuscript, that will be submitted again to the referees evaluations.

My best wishes,
Elena Toth
HESS Topical Editor

Dear Authors,
many thanks for having further elaborated on your replies, trying to address both my and the referees' comments in a constructive way, and for the subsequent changes to the manuscript.
As anticipated, I am asking the referees' comments on the revised version of the paper.
Best regards,
Elena Toth

The paper is to be appreciated since it focusses on the application of Self-Organising Maps that is a powerful data exploration tool in the analysis of such complex data set as those that as hydrologists we deal with every day, linking natural and human-induced drivers that are very difficult to disentangle and understand.
It presents an application of some methodological improvements (introduced by the authors in previous works in different applications) to a very difficult problem in water resources researches, that is an analysis of the patterns of evolving flood impacts over a set of cities at global scale. Since HESS readers have a clear-cut hydrological background and are not SOM experts, the novelties of the work are twofold. First, of course (and as highlighted by the Authors), the introduction of a city level to the analysis: in fact in the previous works by the research group (Ward et al., Winsemius et al) who has prepared such an important global data-set and carried out the main analysis on it so far, the data have been analysed at watershed scale and not for single cities. Such change of objective is actually a very delicate point too, due to the need to take into account the uncertainties inherent in using a global database at urban scale and in particular to consider the specific flood defense measures that are crucial al local scale, whereas their overall impact may probably be averaged at watershed scale (at least this is the assumption of the works, by Winsemius et al. and Ward et al. on which the presented application is based). The other, and in my opinion more important (and also less questionable) strong point of the paper is the application of the SOM approach (including the temporal extension to assess the expected changes in time) for analysing the Acqueduct dataset, which is indeed new. And I believe such strength should be stressed more in the final paper, from its very beginning.
The discussion paper was revised, as we asked, in terms of both the explanation of the data set and the methodology applied to analyse it.
I do agree with the referees that some additional improvements are needed to improve the clarity and the ‘acceptance’ of the paper, better highlighting its limitations and its strengths:
1- even more emphasis should be put, and as soon as the city level is introduced as one of the main novelty of the work, on the strong limitations of an application at city scale (and with small catchment areas upstream , in some cases), partly due to the spatial scale of the applied models and partly due to the specific defence measures, that are much more important than at watershed scale, where, due to overall averaging, it may be more acceptable assuming them as proportional to the national income level.
2- as said above, more stress, in abstract and Introduction, on the novelty of the application of the SOM approach (including the temporal extension to assess the expected changes in time) for analysing the Acqueduct dataset, which is indeed new.
3 - as remarked by Ref#2, the introduction of the Acqueduct data set and the similarity and differences of the work with those by Ward et al and Winsemius et al, should be very clear since beginning, too.
Following the above three points, the Introduction section is certainly to be deeply revised (and especially since more weight is to be given to the above points, I do agree with Ref#2 that the first two paragraphs of the current one may be summarised in a few lines). And also the conclusions should stress more the potential and also the strong and weak points related to the application of the SOM networks, in order to better help the readers to understand how they be used also in other similar applications.
I look forward to seeing the final version of the manuscript.

Best regards,
Elena Toth

University of Bologna
HESS Editor

Dear Authors,
thank you indeed for your last revision where you addressed the last points we raised.
I believe that the current version of abstract and Introduction is able to clarify the background, the main novelties and also the main limitations of the work and will help the readers to better understand and appreciate it.
Actually, unlike Ref#2, I believe that the two comments he asked to move in the Discussion section were better placed in the Conclusion section (as in the first revised version), so I would suggest moving them back, but I leave the decision to you.
Many thanks for having published your work on HESS.
Best wishes,
Elena Toth