Dear Hoang Nhat-Duc,

Please accept my apologies for the slightly slow response to your revised manuscript, this was a delay on my part rather than the reviewers.

I have two reasonably positive reviews of the revised manuscript. Both reviewers identify some grammatical errors. GMD will proof read however it is important that the manuscript is in good shape when accepted for publication. I ask that you consider how to address the reviewers comments in this regard.

Reviewer #1 is concerned with the description of the methodology and notes that this comment was the same in their first review. Please consider how the presentation of the method might be improved for the reader. Perhaps a flow diagram like Figure 5 in section 3.2 that sets out the components of the method prior to the technical discussion would help the reader?

Finally, I agree with reviewer #1 comment regrading hydrological modelling and I wonder if you could also explicitly comment on 1) the record length required by your technique and 2) the potential for data gaps to be a problem, either between Landsat overpasses or due to cloud cover. Perhaps this is intrinsic to the test you have run however a comment on these factors, that are often critical for optical remote sensing based methods, would be a valuable addition to the conclusions I believe.

Best Regards,
Jeff Neal

Dear Dieu Tien Bui,

Thank you for your revised manuscript. You present a very interesting method and some excellent results. However, there are still two issues:

1) The quality of the writing, especially in the introduction and literature review is not sufficient for publication. The grammatical errors make the manuscript very difficult to read.

2) I simply do not understand your argument regarding missing data and data length. For example, if I were to sample flood locations in a particular year, missing many of the floods and assuming some random locations where I had not sampled were non-flood locations I could use the proposed methods to get a flood susceptibility estimate. I do not see how this can be a probability because you have never seen the vast majority of floods that could occur and you would not know how common the flood you have seen is.

Alternatively, I could have gathered data over a 100 year period on every flood in the area and used those data in your model. Under this scenario I would have many more flooded pixels and presumably a higher flood susceptibility, unless the method accounts for data record length and the potential for missing data.

I'm not saying your method lacks merit but it needs to be clear how it differs, and in what respect it might be inferior, to frequency based methods that attempt to estimate the probability of a location flooding in a given year.

Kind regards,
Jeff Neal

Dear Tien Bui Dieu,

Thank you for your response to my previous question. I believe I know understand your method and I would like to recommend your article for publication.
I have consulted with the journals copy editors on the standard of English because it is not yet of publication standard. However they have indicated to me that they will be able to work on the manuscript prior to publication.
I’m sorry to suggest more changes but I my final read through I spotted 11 minor amendments you will want to consider. I also think the justification of the data length you have provided in the response to my question should be written formal into the manuscript.

Specific corrections

1) Line 15: “the posterior probability of flood, which is the output of the BayGmmKda model, is used as flood susceptibility index” should read
“the posterior probability output of the BayGmmKda model, is used as flood susceptibility index” as it is not the probability of flood due to the lack of a frequency component.
2) Line 31: The sentence starting “Particularly, Vietnam is a storm centre” does not make sense, please reword
3) Line 38: “flood forecasting” technically flood forecasting is wrong in this context you should be talking about predicting flood hazard/risk for land use planning
4) Line 50: “enhancing the prediction accuracy” relative to what
5) Line 54: “To yield probabilistic outputs of flood,” should be “To yield probabilistic outputs of flood susceptibility,”
6) Line 72-73: “relatively simple to more sophisticated methodologies” could you give an example of what you mean by relatively simple and sophisticated because its not clear what you mean
7) Line 77: “providing flood predictions” should be “providing discharge predictions”. You will then need a hydrodynamic model in the case of the statistical models and hydrological models to estimate flooding.
8) Line 87: The Neal et al paper doesn’t use a hydrological model. It’s a statistical event generation technique based on river gauge data and probably belongs in the earlier statistical section
9) Line 91: For me this sentence adds nothing and should be removed or used differently “Nevertheless, the complex and nonlinear nature of the flood modeling problem also brings about difficulties for the hydrological methods and techniques (Sahoo et al., 2006).”
10) Line 94: “susceptibility prediction of flood” previously you have used the wording flood susceptibility index. Could you use the same definition for your output throughout the manuscript
11) Line 99: Somewhere here you need to say that the results depend on having sufficient training data