The paper presents an interesting contribution on the use of deep learning to identify flood extents in urban areas from night-time images.  As such it is relevant to NHESS, but it has a strong focus on the deep learning methods.  Many readers of the journal, like myself, will not be familiar with the detailed language and concepts used in deep learning and the paper needs to be re-written with this audience in mind.  Sections 2.1 and 2.2 need particular attention.

Much of the language in the paper is opaque and uses terms that are not common in scientific discourse (some of these are set out below).  There also needs to be greater clarification of what the authors set out to do in the research, which of the methods they developed themselves, and what the conclusions mean for those working on urban flooding.  For example, on Line 82 the aims are listed, but these are in fact a description of what was done without a justification.  Further Line 74 says that NWseg is "proposed" and Line 91 says that the NWseg is "contributed", but was it developed by the authors or taken from other research work?

Significantly more than half of the references that I tried to read online returned an error message or a message in Chinese characters.  Further, many of these are to non-peer reviewed sources.  Whilst the latter is acceptable in a few cases, the former is not at all acceptable in an international journal.

Specific points are:

Line 50: a more scientific term that "remarkable" would be more appropriate.

Line 54: it is stated that such surveillance is "ubiquitous", but whilst this may be true in the authors' experience is not true in all countries.  This should be acknowledged and it limits the usefulness of these methods.

Line 55: of these three references two (Cheng, Yang) have links that do not work and one is to a publication that is not peer-reviewed.  The remaining one does not mention whether this methodology has been tested in more than one country.  Please clarify.

Line 89: the term "ablation" is common in machine learning, but is an example of a term that needs explaining to a different audience.

Line 99: there is no subject in this sentence so we cannot see who proposed the model.

Line 102: many terms in this paragraph need clarifying for a non-expert in machine learning, some examples are: SOD, illumination-independent reflectance, semantically supervising the training of the de-entanglement module, Retinex, Illumination-Aware Parser (IAParser), pyramid pooling module and a convolutional layer to construct an attention mask.

Line 115: a diagram showing how all these methods fit together would help readers understand what you are doing.

Line 189: what is a "Labelme tool"?

Line 191 says that the work was done by three graduate students.  Rather than describing who did the work, it is necessary to explain how they did it and how the quality of the analysis was checked and ensured.

Line 193: "waterlogged" refers to soil saturation.  I think "inundated" would be a better word.

Line 196: this and other figures captions need more details.

Section 4.1: I think there should be less discussion of the three inferior methods as the differences between them are minor compared to their differences to NWseg.

Line 250: I don't see any experimental results in the text i.e. data that was collected through physical measurements on site.

Line 252: what is "social inundation"?

Line 254: "exceptional" is too strong a word here.

Line 285: these images, and later one, are too small.

Line 328: if the conclusions state that there is a high computational demand, this should be investigated and reported on in the results section.  How much greater is it?  How long did it take?  What sort of computer was used?  Does this allow for practical use of NWseg?