Inundation Assessment of the 2019 Typhoon Hagibis in Japan Using Multi-Temporal Sentinel-1 Intensity Images

Round 1

Reviewer 1 Report

The authors present a well-written approach for the delineation of flooded areas as a result of heavy rainfalls from Hurricane Hagibis in October 2019.
The study includes comparison of two SAR-based appraoches (one-date and two-date identification), an extensive field survey, the modelling of backscatter of flooded areas and a systematic accuracy assessment based in independent reference data.
While the first part ist clearly-structured, the presentation of results, the validation and the discussion are a bit mixed between the chapters 5, 6 and 7. I recommend using more sub-sections for a clearer separation of these aspects so the reader is guided through the findings, their evaluation and the open points.

General remarks

• As you correctly outline based on your literature review, many approaches already exist on SAR-based flood mapping. So the reader might ask why your study is relevant. So maybe you can make it more clear after the literature review (line 103), why your study brings something new or better. I see great potential because it is easy to adapt and does not need SLC data, therefore automatable and quickly to reproduce after a flood event. Also the systematic comparison of two approaches for the same case is a plus, as well as the backscatter models of the two buildings from the field survey. I just recommend to highlight this to your own benefit.
• Instead of "current authors", you can simply write "authors" or "authors of this study"
• The various steps to delineate different flooded areas are interesting, but I am missing a flow chart which puts the steps into an order, marks decisions, inputs/outputs ect. to achieve and validate the combined flood mask (Figure 12d). Placed at the beginning of chapter 4 it would help the reader follow along your method description. Especially the different average+/-std cases should be part of the summary chart to visualize at which step you applied which assumption or threshold.
• As both methods are based on finding a threshold (Table 2), the terms "differencing" and "thresholding" might not be ideal, and they are a bit misleading. To have a clearer distinction between both approaches, would it make sense to call them "monotemporal" and "multitemporal" or "single-date" and "dual-date"? My recommendations might not be the most elegant, but I hope you get my point and I would suggest to find names which point on the actual difference of the methods. Maybe a combination, "multi-date differencing (MDD)" and "single-date thresholding (SDT)" or something like that. If you think the current terminology is sufficient, I won't insist on changing it, but I was rather confused when I first read the manuscript.
• Table 4 is used twice, the second should be Table 5

Specific points (adressing the line numbers in the manuscript)

28-33: If all the statistics on flooding are from your first resource, it might be good to name it in the beginning of the list, for example, "According to numbers of the Centre for Research on the Epidemiology of Disasters (CRED), the number of...". If the numbrs come from different sources, please list them accordingly.
34: Please give a reference for the estimated economic loss of 26 billion dollars.
48: "and a wide area was flooded" - can you maybe specify what kinds of areas (agriculture, natural, urban, or equally distributed over all classes) were predominantly flooded?
53-55: Maybe some introducing words on the Charter would help readers which are not familiar with the term.
56: Please shortly give examples what these maps were used for (evacuation, damage assessment) and why the quick provision of such maps is important to underline the relevance of your paper. After that, you can start a new paragraph on remote sensing in general ("Remote sensing is an effective tool..."
58: Can you please shortly summarize the core findings highlighted by the three review papers (Klemas, Lin and Koshimura)? Maybe there is a consensous between all of them which could be presented at this point.
68: One or two more sentences on the nature of thresholding would be nice (flooded areas are darker because land has more roughness and is therefore brighter in the image, histogram can be split into dark and bright parts, representing flooded and non-flooded land)
77-80: I agree with the downsides of coherence-based flood detection, but the paragraph lacks a short explanation what is different from intensity based methods and also a reference for the listed disadvantages.
82: If the study on Kanto and Tohoku [is based on intensity-based thresholding, I recommend to present it before you mention coherence-based approaches.
83: If you mention the integration of elevation data, you should also shortly mention why or how it can be used to improve flood mapping.
120-142: Very nice description and illustration of the flood event - nothing to add here from my side.
144: Although it is displayed in Figure 1, you should mention in the beginning, that both pre and post flood dates were covered by an ascending and descending track at the same date, because this is not very common to have in all situations. Maybe you intentionally selected these dates based on this fact, so it's good to communicate to the reader.
156: Is there a reason why you applied radiometric calibration after the projection of the data (which already involves resampling of the pixels)?
157: sigma nought (not with an a)
158: Please add a reference for the enhanced Lee filter.
175: Thank you for creating red-blue composites (instead of green-yellow) to include deuteranopic readers.
212-215: Feel free to make the red dashed line more visible in the field photographs - very impressive.
225-237: The iterative approximation of an ideal threshold (for section 4.1) based on two comparably simple overall image statistics is interesting. Yet, I do not understand why it works. Maybe a short hypothesis or technical background (exemplified based on a histogram) would be helpful, also references to other studies based on this assumption are welcome. For example, would it also work if the sea areas were not masked?
272: "The threshold value was set as the average value plus twice the STD of these pixels." This was already explained in line 266 where you refer to the previous study. Maybe you can rephrase it in a way that you follow the good experiences from this study.
276: More importantly on this paragraph: What do you recommend users which have not conducted previous studies. How can they narrow down the factor to multiply based on the STD to find an ideal threshold?
294: Why did you not use any of the EMS products for validation? There might be good reasons to use the maximum inundation boundary by GSI, but it would be good to shortly explain your choice among the available reference datasets.
304: What areas were taken as "non-flooded" reference? All pixels outisde the mask?
334: The title of chapter 5 is a bit misleading: You aim at the identification of flooding in built-up areas, not the extraction of the buildings themselves, right?
354: "This increase in backscattering was caused by the disappearance of a radar shadow." I don't think this is the case. Maybe you can rephrase it a bit to make it clearer.
400: The moving window index is interesting - was it applied to the flood image or the difference image?
403-417: As stated above, a introducing chart with the different avg/std settings would help to get through the paper.
432: The flood boundary of GSI is barely visible in Figure 12b and c
450: Would it make sense to present a bar graph comparing Precision and Recall for the two classes of Table 3, 4 and 5 so the readers see how it has improved (also with and without the high water channel)?
451-544: Chapter 6 sounds logic and comprehensive, but maybe it could be divided into two sections - one about validation and one about general shortcomings? I'm not sure either, but I got lost a bit in between. For example, it does not make so much sense to me to present accuracy assessments in both chapter 5 and 6. If there is a reason why you distribute them over two chapters, please make it more clear earlier so the reader knows what to expect.
539: As stated above, Table 4 should be Table 5.
545: The conclusions are rather a summary. This is ok, but at this point readers might also be interested in if and how they can apply your approach for their flood events. The following questions come to my mind
- How likely is the chance to get a Sentinel-1 image from the day of the flood event at this latitude (including both ascending and descending tracks)?
- How quick after image acquisition the data is provided by ESA?
- How should users find the ideal thresholds? Is there a way for automating your workflow? Where is the main need for adjustment? Again, the flow chart could assist readers in replicating your ideas.
Some of these things are partly addressed in chapter 6, again raising the need for a separation of results, and open points/challenges.

Author Response

Thank you for your kind review.
We responded to your comments in the attachment file.
We also revised the manuscript according to your comments.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,  I have reviewed the article "Inundation Assessment of the 2019 Typhon Hagibis in Japan Using Multi-Temporal Sentinel-1 Intensity Images".

The article is well written and the results are clearly presenting showing the potential of Sentinel-1 intensity images to extract flood inundation extent.

There is a minor issue that needs to be fixed.

Line 175. Figure 3 was not available in the manuscript, therefore, I can provide any feedback.

Good luck

Author Response

Thank you for your kind review. Figure 3 is below Table 1 on page 6.

Reviewer 3 Report

Dear Authors,

The research presented aimed at developing a simple procedure to create accurate inundation maps for an emergency response using Sentinel-1 SAR intensity images. The case study was a severe inundation due to Hagibis Typhoon that struck Japan in 2019. The extraction of completely inundated areas was performed using two methods: the difference of backscattering intensity and thresholding. Moreover, you propose a procedure of extraction of the partly inundated buildings based on backscatter model.

The presented research topic is an interesting issue, in particular, due to climate change as well as more and more violent and more frequent sudden weather phenomena such as typhoons. At the same time, the research methodology is very universal and can also be applied in other regions of the world that are hit by floods. The use of InSAR inundation mapping data - due to their global coverage and free availability - is another very innovative and promising research issue.

The article was written very correctly, using correct English. Its structure is linear and coherent, the following chapters are written very well. The authors present an interesting introduction and give many examples of global methods of flood mapping, then very broadly present the research area and the research methods used. On several pages, they describe the results obtained, analyze their accuracy and present consistent conclusions.

I have no comments on the merits. I enclose to the Authors only a few minor editorial notes, the introduction of which will improve the already good manuscript. After a minor revision, I believe that the article is worth publishing in Remote Sensing and will be valuable research into inundation studies worldwide.

Congratulations!

Reviewer

Introduction:

Lines 29-30: Please, add a reference.

Lines 34-36: The same comment as the previous one.

Lines 80-90: How about studies carried out by other authors? Just for scientific clarification, it would be advisable to refer to research carried out not only by the Authors of the manuscript presented.

Figure 1: Add a mesh grid and information about the projection of maps presented. Moreover, please erase the following numbers „50” and „2.5” from the scale bar in Figure 1a and 1b, respectively.

Study area and Sentinel-1 imagery:

I would advise dividing this section into two parts. The first part should be a description of the study area – this is well written. However, you should not describe in details how the processing of InSAR data was carried out as this is more like a methodology scheme. Therefore, please move the second part of this section to the next chapter, where you describe the methodology. In my opinion, this could clarify the structure of the entire manuscript.

The field survey:

No comments, this paragraph is well written and structured.

Extraction of completely inundated areas:

Lines 259-261: This is a methodology section with a brief overview of the results obtained, therefore you should not mention about your previous research in this chapter. This sentence should be moved to the introduction chapter.

Extraction of the partly inundated buildings:

No comments, this paragraph is well written and structured.

Final inundation maps and discussion:

Figure 13: I would advise resizing the figure to provide a better quality of the base map. Moreover, one could add a mesh-grid to the figure.

Conclusion:

Line 574: will -> could be overcome. You do not know yet if the procedure based on improving built-up masking for the index will work correctly.

Author Response

Thank you for your kind review.
Our response is shown in the attachment file.
We have revised the manuscript according to your comments.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Thank you for your revisions, I think that the manuscript improved substantially regarding the discussed points. 

Some minor things which are still open

• I really like the workflow in Figure 7, especially its reference to the different chapters, however the quality of the figure is insufficient (probably because of JPEG compression) which makes it hardly readable. Please include a version of higher resolution (PNG, 300 dpi)
• "multi-temporal dual-date" sounds a bit redundant to me, so I would remove one of both terms, as long as you use it consistently throghout the study
• Thank you for implementing the suggestion on the terminology of the two approaches (mono-temporal determination and multi-temporal comparison). Yet, they are not used consistently in the paper
  • Line 120: "thresholding method and change detection method"
  • Line 241: "difference and thresholding"
  • Line 310: "thresholding method...backscatter difference"
  • Line 351: Caption of Figure 10
  • Line 361: "difference or thresholding"
  • Line 479: Caption of Figure 14
  • Abstract "difference and thresholding" <- here I think it is okay because it is an abstract description of what is to come
• You are right about the use of Sigma naught/nought, obviously both notations are accepted, sorry.
• Chapter 6 is now better to read, thank you.
• Line 617: As I understand it, you can remove the sentence "For research articles... must be provided".

Author Response

Response to Reviewers Comments

Wen Liu, Kiho Fujii, Yoshihisa Maruyama and Fumio Yamazaki

Comments and Suggestions for Authors

Thank you for your revisions, I think that the manuscript improved substantially regarding the discussed points.

Some minor things which are still open

I really like the workflow in Figure 7, especially its reference to the different chapters, however the quality of the figure is insufficient (probably because of JPEG compression) which makes it hardly readable. Please include a version of higher resolution (PNG, 300 dpi)

Ans: We replaced Figure 7 with a higher resolution figure (JPG, 300 dpi).

"multi-temporal dual-date" sounds a bit redundant to me, so I would remove one of both terms, as long as you use it consistently throghout the study

Ans: We revised the words as “multi-temporal comparison”.

Thank you for implementing the suggestion on the terminology of the two approaches (mono-temporal determination and multi-temporal comparison). Yet, they are not used consistently in the paper

Line 120: "thresholding method and change detection method"

Line 241: "difference and thresholding"

Line 310: "thresholding method...backscatter difference"

Line 351: Caption of Figure 10

Line 361: "difference or thresholding"

Line 479: Caption of Figure 14

Ans: We revised these words accordingly. We also checked the manuscript again to revise all these words.

Abstract "difference and thresholding" <- here I think it is okay because it is an abstract description of what is to come

Ans: We revised these words as “the change detection” and “thresholding method” to match with the introduction.

You are right about the use of Sigma naught/nought, obviously both notations are accepted, sorry.

Ans: Thank you for your comment.

Chapter 6 is now better to read, thank you.

Ans: Thank you for your comment.

Line 617: As I understand it, you can remove the sentence "For research articles... must be provided".

Ans: We removed this sentence accordingly.