Sentinel-1 Change Detection Analysis for Cyclone Damage Assessment in Urban Environments

Round 1

Reviewer 1 Report

Dear authors,

the manuscript is interesting but needs a deep revision. First of all, the English have to be revised.

The figures have to be improved because some are difficult to read and others miss some elements (norht arrow, scalebar, legend).

The introduction have to be improved with a state of the art.

Discussion and conclusions are very short and have to be strongly enhanced.

Please carefully revised the manuscript before the resubmission.

Some detailed comments are in the attached pdf.

Comments for author File: Comments.pdf

Author Response

Individual responses:

the manuscript is interesting but needs a deep revision. First of all, the English have to be revised.

We have gotten the paper proof read.

The figures have to be improved because some are difficult to read and others miss some elements (norht arrow, scalebar, legend).

Additional description has been added to the captions and the compass, scalebar, etc. have been added.

The introduction have to be improved with a state of the art.

Som extra references have been added and the text has been revised.

Discussion and conclusions are very short and have to be strongly enhanced.

A more thorough discussion and conclusion is now added.

Some detailed comments are in the attached pdf.

thank you for the detailed comments attached.

Thank you very much for your constructive comments. We have strived to meet all your requests to improve the paper as much as possible.

Best Regards,
The authors

Reviewer 2 Report

The paper focuses on the use of SAR data for disaster emergency response. An already existing change detection method, developed by Nielsen et al. is applied to this research, focusing on a specific case study of the cyclone Idai that hit Mozambique in March 2019.

The question is well defined, although it is hard to understand at which level the results provide an advance in current knowledge. The results are interpreted appropriately, even if there are some flaws in the discussions and conclusions, primarily related to the fact that the authors aim at damage detection at the building scale, although the resolution of the employed SAR data is 20x22 meters (and a UA of 115x115 m has been employed). The article requires an English spell check, some of the typos have been highlighted in the attached pdf file. The study is technically sounding, and it will be interesting for the readership of the journal. The conclusions should be expanded, as now it appears that SAR change detections can provide an assessment of damages to the house level, which is not entirely true without considering the limits in the employed methodology. I am confident that there could be an overall benefit to publishing this work if the revision comments will be satisfactorily addressed by the author (minor revision). Here follow some remarks. Please also inspect the attached pdf file as it contains some more comments mainly related to the grammar check.

• To place the study in the broader context, it would be beneficial for the readers if the authors discuss better also other damage assessment methodologies; there are also remarkable experiences on the use of not only optical satellite data, but UAV deployment after a disaster. I am aware that, for example, some application has been tested in Mozambique by the World Food Programme. The authors are encouraged to extend the introduction part, providing more citation of related/similar studies.
• Please opportunely cite the Copernicus service at line 27.
• Please revise a missing citation [?] at line 82 and provide a reference for the statement “but 6-day intervals are occasionally available” in ln 83.
• The need for comparing SAR data with optical data as ground truth makes (at the current state of the art) optical data an essential piece of knowledge for building damage assessment. Even if the adverse weather conditions make this impossible for the specific case study presented in this research, the authors are encouraged to cite similar studies in damage assessments where SAR and optical data have been compared.
• Line 138 it would be more clear to specify if the “6am-6pm” is an interval of time for obtaining the daily average, or there are two average for each day (at 6 am and 6 pm); this should be specified particularly for non-specialized readers who are not confident with the World weather online database.
• Line 200, the tests for defining the range 0.01%-10% have been performed on the same dataset? Are they related to the same case study? Even if this value has been obtained empirically, it would be beneficial to understand if it is expected to work for other similar datasets or not.
• Concerning the previous point, have the authors been thinking of setting this range comparing two different SAR data both before the disaster? What I mean is that to set a threshold value to turn data into change/no change related to the cyclone, the easiest way could be to compare two different moments both before the cyclone, as in that comparison there should be no significative changes expected (I see you have compared 02-14 mar in figure 5, but it is not completely clear the empirical approach for setting this threshold).
• It is not clear to me how the authors ensure that the changes detected are independent of natural vegetation and moisture in the ground between houses, as stated in lines 270-273.
• In lines 288-291 it is discussed a change in the scale of the analysis; detecting a crack on the roof is also tricky in case of 30 cm resolution optical images, I believe here it would be more beneficial if the authors discuss the limits of their methodology considering that Beira does not have big buildings. Moreover, the majority of buildings are tiny compared to the SAR resolution and for sure smaller than the 115x115 m adopted UA. Maybe this method is more efficient for building blocks instead of single buildings?
• Please provide a reference for the “some suggested” at line 314.
• As already anticipated, conclusions should be more comprehensive of the pros and cons of the adopted methodology and the obtained results.

Comments for author File: Comments.pdf

Author Response

Individual response:

The results are interpreted appropriately, even if there are some flaws in the discussions and conclusions, primarily related to the fact that the authors aim at damage detection at the building scale, although the resolution of the employed SAR data is 20x22 meters (and a UA of 115x115 m has been employed).

It was not our intention to promise house level detection of damages and we have tried to clarify this. We simply look into how the average radar intensity in a resolution cell is affected by the underlying destruction from the cyclone. When we accumulate to 115x115m cells it is simply to do statistics over areas, correlations, and highlighting "worst hit" cells.

The article requires an English spell check, some of the typos have been highlighted in the attached pdf file.

We have gotten the paper proofread.

The conclusions should be expanded, as now it appears that SAR change detections can provide an assessment of damages to the house level, which is not entirely true without considering the limits in the employed methodology.

We have expanded the conclusion.

Here follow some remarks. Please also inspect the attached pdf file as it contains some more comments mainly related to the grammar check.

• To place the study in the broader context, it would be beneficial for the readers if the authors discuss better also other damage assessment methodologies; there are also remarkable experiences on the use of not only optical satellite data, but UAV deployment after a disaster. I am aware that, for example, some application has been tested in Mozambique by the World Food Programme. The authors are encouraged to extend the introduction part, providing more citation of related/similar studies.
  This has been added
• Please opportunely cite the Copernicus service at line 27.
  We are not sure which reference to add.
  
  
• Please revise a missing citation [?] at line 82 and provide a reference for the statement “but 6-day intervals are occasionally available” in ln 83.
  We do not as such have a reference for this, but when request the data via the Scihub API, 6 days intervals were occasionally available.
  
  
• The need for comparing SAR data with optical data as ground truth makes (at the current state of the art) optical data an essential piece of knowledge for building damage assessment. Even if the adverse weather conditions make this impossible for the specific case study presented in this research, the authors are encouraged to cite similar studies in damage assessments where SAR and optical data have been compared.
  We have added a reference to a study that simalarly to us compares SAR data with manual optical assessed data. (Washaya, Prosper et al., 2018)
  
  
• Line 138 it would be more clear to specify if the “6am-6pm” is an interval of time for obtaining the daily average, or there are two average for each day (at 6 am and 6 pm); this should be specified particularly for non-specialized readers who are not confident with the World weather online database.
  This has been clarified in the text now.
  
  
• Line 200, the tests for defining the range 0.01%-10% have been performed on the same dataset? Are they related to the same case study? Even if this value has been obtained empirically, it would be beneficial to understand if it is expected to work for other similar datasets or not.
  This range just comes from general statistical practices where significance levels usually are chosen between 0.1%-10%. We simply tried different values in this range to make sure our method was not overly sensitive to this parameter.
  
  
• Concerning the previous point, have the authors been thinking of setting this range comparing two different SAR data both before the disaster? What I mean is that to set a threshold value to turn data into change/no change related to the cyclone, the easiest way could be to compare two different moments both before the cyclone, as in that comparison there should be no significative changes expected (I see you have compared 02-14 mar in figure 5, but it is not completely clear the empirical approach for setting this threshold).
  We did consider this but reached the conclusion that it is difficult due to the fact that there always are some levels of changes in urban environments. Of course the level of changes are much lower outside cyclone impact periods, but generally it is hard to say what a normal level of changes in a city should be.
  
  
• It is not clear to me how the authors ensure that the changes detected are independent of natural vegetation and moisture in the ground between houses, as stated in lines 270-273.
  We did not mean to state that we are independent of these things, but simply that we reduce the effect by filtering changes with building footprints.
• In lines 288-291 it is discussed a change in the scale of the analysis; detecting a crack on the roof is also tricky in case of 30 cm resolution optical images,
  We did observe cracks in the roof where marked in the manual tags from UNITAR. Clearly, this was not small cracks.
  
  I believe here it would be more beneficial if the authors discuss the limits of their methodology considering that Beira does not have big buildings. Moreover, the majority of buildings are tiny compared to the SAR resolution and for sure smaller than the 115x115 m adopted UA. Maybe this method is more efficient for building blocks instead of single buildings?
  See response to first comment.
  
  
• Please provide a reference for the “some suggested” at line 314.
  We have added a reference here.
  
  
• As already anticipated, conclusions should be more comprehensive of the pros and cons of the adopted methodology and the obtained results.
  Conclusion has been expanded.
  

Thank you very much for your constructive comments. We have strived to meet all your requests to improve the paper as much as possible.

Best Regards,
The authors

Reviewer 3 Report

I read the work several times and started making comments and appointments but I have the view that the main problem with the paper is on that needs to be reorganized.

I have some comments for the authors as follow:

Small changes in the Abstract (for example in line 3 I propose to use For some natural disasters or For some extreme weather disasters or in line 6 "first (?) information" or in line 11 I propose to use another term for "combined").

Lin 23-24 missing reference

Lin 27 missing references

The section "Introduction" should be reorganized. The same the section "Data". The last one contains text that does not describe the data used.

"Results" and "Duscussion" sections are ok but the section Conclusions is very poor.

Finally I would like to warm up a comment. The operational capability of Earth's space observation data is currently in demand. We need to achieve the goals set by the Sendai decision text. Of course, radar data, especially in cases of natural hazards such as extreme weather, are very useful and capable due to their characteristics. But another important parameter for the operational capability is the temporal resolution  of the satellite system. In this case Copernicus Sentinel 1 data offer a systematic collection images with an improved temporal resolution respect to the previous satellites and of course the data are free to download.

Author Response

Individual reponses:

Small changes in the Abstract (for example in line 3 I propose to use For some natural disasters or For some extreme weather disasters or in line 6 "first (?) information" or in line 11 I propose to use another term for "combined").
Corrected.

Lin 23-24 missing reference
It is reformulated now.

Lin 27 missing references
This is a well-known fact, but as household-level damage maps are not generally publicly available (only for affected government and research by direct contact with UNOTAR/UNOSAT), we can unfortunately not easily cite the source. The same is the case in this paper "RAPID DAMAGE ASSESSMENT USING HIGH-RESOLUTION REMOTE SENSING
IMAGERY: TOOLS AND TECHNIQUES", where they can also not cite the source.

The section "Introduction" should be reorganized. The same the section "Data". The last one contains text that does not describe the data used.
We have edited both and hope it satisfy the reviewer. We are not sure what text in the Data section that is reffered to?

"Results" and "Duscussion" sections are ok but the section Conclusions is very poor.
A lot has been added to the conclusion.

Finally I would like to warm up a comment. The operational capability of Earth's space observation data is currently in demand. We need to achieve the goals set by the Sendai decision text. Of course, radar data, especially in cases of natural hazards such as extreme weather, are very useful and capable due to their characteristics. But another important parameter for the operational capability is the temporal resolution of the satellite system. In this case Copernicus Sentinel 1 data offer a systematic collection images with an improved temporal resolution respect to the previous satellites and of course the data are free to download.
We agree with your nice perspective.

Thank you very much for your constructive comments. We have strived to meet all your requests to improve the paper as much as possible.

Best Regards,
The authors

Round 2

Reviewer 1 Report

Dear Authors,

thank you for reviewing the manuscript some of my requests/suggestions were satisfied.

Please check the writes in the figures (also in the legends) because are too small to be readable yet.

The discussion and conclusions need to be further improved.

In addition, the state-of-the-art of this relevant issue (damage assessment by Sentinel-1, or satellite in general, images) in the introduction should be enhanced. It is relevant also for highlight the relevance and the novelty of your work.

Author Response

Thank you for your comments. Below, they are addressed individually.

Please check the writes in the figures (also in the legends) because are too small to be readable yet.
We have increased the size of legends on Figure 5, 6. Figure 7 and 9 has been slightly zoomed, which we hope helps the readability as well?

The discussion and conclusions need to be further improved.

In addition, the state-of-the-art of this relevant issue (damage assessment by Sentinel-1, or satellite in general, images) in the introduction should be enhanced. It is relevant also for highlight the relevance and the novelty of your work.
Regarding state-of-the-art, we have added more relevant references to support the strengthen the introduction.
Novelty of our work has also been enhanced by extra explanation in the second last paragraph of the introduction.

Best Regards,
The authors

Reviewer 3 Report

I believe that the paper has been improved and now only minor comments I have as follow:

In the introduction section the sentence in lines 19-20 must be rewrite for example instead of "destruction" you can use "disaster damages"

In line 25 weather conditions but also day and night acquisitions.

Line 31 ?

line 35 Why ? The detection is based on backscattering properties.

Figure 2 The caption is not clear 

Author Response

Thanks for the comments. Below they are addressed individually.

In the introduction section the sentence in lines 19-20 must be rewrite for example instead of "destruction" you can use "disaster damages"
This has been changed to follow you suggestion.

In line 25 weather conditions but also day and night acquisitions.
That night recording is another advantage of SAR is added to the text now.

Line 31 ?
What is meant is that maps of damages based on SAR can be prepared off-site and sent to the emergency response personnel or authorities.

line 35 Why ? The detection is based on backscattering properties.
According to the classification scheme in (Ge et al.) our method is the intensity approach. One could argue that we use all polarimetric information as well, but the methods in their Polarimetri-based change detection schemes are model-based and involves analysis of scattering properties where ours is a statistical measure on each SAR pixel. We have included extra explanation in the article about this. 

Figure 2 The caption is not clear
This has been improved now.


Best Regards,
The authors