Earthquake Damage Region Detection by Multitemporal Coherence Map Analysis of Radar and Multispectral Imagery

Round 1

Reviewer 1 Report

This work introduces a novel damage detection method using an unsupervised process from Sentinel imagery. 

In my view, the methodology is well described, and the results are sound.

However, before its publication, some issues should be addressed:

General comments:

Regarding the English language:

"proposed method" appears 29 times over the manuscript.

"Damage" 84 times.

Please, revise the comma and article usage.

I prefer using human-made rather man-made, to use inclusive language.

In figures, please fit the coordinates frame to the study area, as it appears as no information (blank).

 

Specific comment:

Line130: I do not understand this expression: "high coherence, low coherence, or high coherence."

L149-150: Please, rewrite using such as instead of etc.

L169: Please, specify the SVM acronym the first time you use it.

In Figure 3: Write the name of the countries or regions. (Iran, or so forth) in the upper images.

In Figure 4 a) and b), the text is not readable.

L235-237: A reference is missing in the sentence: "The optical Sentinel-2 data have good quality and only need atmospheric correction, done to transform TOA reflectance to BOA reflectance via algorithms developed by DLR/Telespazio."

L246: Please, add some references to this sentence: "For our study, we used the radial basis function (RBF) kernel, which is widely utilized in the remote sensing community." Some ideas:

https://doi.org/10.1016/j.cageo.2020.104548

https://doi.org/10.15625/0866-7187/42/3/15355

L272: Please, rewrite "Sarpolzahab" as "Sarpol-Zahab."

 

Author Response

Please see the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Authors,

I have thoroughly read your article. The study aimed to develop a new damage detection method at the unsupervised level, using multi-temporal optical and radar images from the Sentinel mission.

Rapid assessment of the damage caused by natural disasters, including earthquakes, is essential to assist those affected by its occurrence. The methods used to assess this type of phenomena have mainly been based on the expert classification of damages carried out directly in the area of investigation. This type of work was time-consuming and involved, at least in part, a subjective assessment of the damage by the expert. For this reason, the search for algorithms and methods to overcome the problems associated with the traditional method of assessing post-earthquake damage is an important and interesting research issue faced by the authors of the submitted manuscript.

Unfortunately, the submitted manuscript is of very low quality. Although the methodology presented by the authors has innovative aspects, it has been described in a very inconsistent manner. For this reason, there are many doubts as to how to interpret the presented data and algorithms. It is also difficult to understand the content of the article in a relatively poor English language. The manuscript as a whole should be edited and revised by the English native speaker. The structure of the manuscript is not linear, authors in many places jump from one subject to another in their descriptions, use many repetitions and obvious, non-scientific statements.

Although the research issue is very interesting, there are many drawbacks to this article. For this reason, I believe that it should not be published in Remote Sensing, and I, therefore, request that it be rejected as it stands.

Below I am enclosing detailed comments which I hope will help improve the article.

Best regards,

Reviewer

 

Introduction:

Lines 60-62: What do you exactly mean by the term „low temporal resolution”? Do you support this statement considering for example Sentinel-2 imagery?

Figure 1: I see no need for providing such a figure presenting only one aspect of conjugated use of radar and optical imagery considering all limitations that you e to in the manuscript. I would suggest to the authors that they redraw the figure so that it depicts all issues mentioned in the text.

Line 73: What do you mean by „height data” I see what you mean (altitude), however, please be more consistent.

Lines 80-82: I would not say the focus is only on the index of two temporal datasets whereas other aspects are completely ignored. One could say the focus is primarily or mainly… You never know for 100% what the reality is. Therefore, please rephrase the sentence.

Lines 83-95: Please, merge both paragraphs as you repeat the meaning of the first paragraph in lines 87-90.

 

Methodology:

Figure 2 is not readable. Please, redraw it, specifically enlarge the font size. Consider dividing the figure into parts (left and right sides of the original figure could be arranged up and down).

Lines 112-113: What if I would like to apply the methodology presented to optical images acquired using different sensors (not Sentinel)? Would it be possible?

Lines 111-120: I am not sure how exactly man-made objects are extracted from images. It is not known from the text provided. Please, rephrase/modify this section.

Lines 122-126: How exactly do you compute coherence? It is necessary to create an interferogram to estimate a coherence. Therefore, I would like to know what algorithm do you use for interferogram computation. It is crucial what the baseline of radar images is, what orbits you apply, what filtering methods are implemented. These issues have to be addressed in the text.

Lines 130-131: How do you remove or try to mitigate noise that occurs in the coherence map? It is of key importance as one can assume low coherence pixel as damage related to the earthquake, whereas the true reason for low coherence in a specific pixel could be a noise-related for example to atmospheric constraints.

Table 1: Are all indexes implemented in your methodology? Or this is simply an example of indexes based on the electromagnetic response of various objects? Please, precise it.

Chapter 2.3 presents a well-known and commonly used knowledge. Therefore, there is no need to refer to such an issue, I recommend erasing this paragraph.

Overall, chapter 2 is completely non-linear and should be thoroughly re-written to clarify its structure. You jump from one definition to another. As a result, it is very difficult to follow the methodology and fully understand its content. You should have thought about presenting the methodology scheme step by step as if it was being explained to the potential reader for the first time.

 

Study area and datasets:

Firstly, the third paragraph should be moved backwards in the manuscript (1- Introduction, 2 – Study area, 3 – Data, 4 – Methodology). You should describe the research area in detail, then the data used in the research, and finally, present the methodology of the research. These drawbacks require improvement throughout the manuscript.

Line 195: What do you mean by the term „ground truth”?

Lines 194-199: Expert knowledge will always differ considering various expert that will carry out the analysis and different study areas. Therefore, your algorithm could not allow for obtaining repeatable modelling results, in particular in different study areas.

Figure 3: I am sorry, however, I do not see the top of the figure. Nonetheless, please show the map of Iran on the map of Asia. Moreover, scale bards in Figure 3a and 3b need improvement, please, do not use a division like 0.7, then 1.05, 1.4. Use a repeatable, full division like 1-2-3-4 km, etc.

Sup-chapters 3.1 and 3.2 should be merged.

Do you mention about Figure 4 earlier in the manuscript, that is to say – before inserting the figure into the text? I do not see that. Besides, in my opinion, this figure needs some changes as the text within the figure is illegible.

 

Experiment and Discussion:

In my opinion, it is pointless to describe in detail the pre-processing of your data, because it is obvious that it has to be done. Therefore, it is also pointless to write it down in a sub-chapter. I would recommend that the authors remove or revise this part of the fourth chapter.

Lines 239-252: This is a methodology, not your results. Include the methodology in the methodology section, describe modelling results in the results section.

Line 253: Figure, not figures.

Figure 5: Figures 5a, 5b and 5c should constitute one, coherent whole. They cannot be on separate pages. The division of the scale bar should also be improved. What is the projection of the map?

Figures 6, 7 and 8: The same comment as for Figure 5.

Chapter 4.3 seems more like a conclusion, not the discussion of the modelling results. Are there any disadvantages of your method? Try to be more critical in evaluating the method presented or let others judge what you did.

In the chapter "Experiment and Discussion", at the very end, I miss a broader statistical evaluation of the developed method and a detailed spatial comparison with the results of damage assessment carried out by other institutions. I would recommend that you evaluate the accuracy of the presented method using statistical tests.

 

Conclusion and Future Work:

These chapters are typically merged in research articles. However, Chapter 5 is a very weak summary of the work done and Chapter 6 makes no sense. You repeatably state that the algorithm could be applied in multiple building damage extraction. This is too less considering future applications oy your method and possible research domains. Please, consider enhancing these paragraphs according to my comments.

Author Response

Please see the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript describes a methodology to use remote sensing to detect and map damaged structures after an earthquake. The methodology proposed involved the use of optical remote sensing to generate an impervious surface mask based on a two-step process: first, by generating preliminary mask using a set of spectral indices and hard-coded threshold values. Then, the preliminary mask is used as training data into a SVM algorithm to refine this mask and create the final impervious surface mask. The mask is compared with digitized results to quantify their accuracy. The mask is then used with time series of RADAR images to detect change based on interferometric coherence.

To improve the confidence in the method, it would be helpful to readers, to include information on the spectral and spatial resolution of the optical sensor and spatial resolution for the RADAR sensor. Additionally, results are presented as overall accuracy and kappa coefficient of agreement. It would be helpful to include a confusion matrix. This would inform readers on the total number of pixels used and commission and omission errors.

 

General Comments

• Objectives need to be revisited. The properties 1 and 3 are the same. Please consider removing/revising one of them.
• The manuscript states this is an unsupervised algorithm. However, in the description of step 1 (section 2.1.1 Step 1), it is described that “training data is gathered in two groups”. This is not clear in section 2.1.1. Later in section 4.2.1, this is explained. The training data is generated using hard coded threshold values of spectral indices. Maybe this explanation should come earlier.
• It is assumed the build-up step uses pre-event images. This is not explained in the methodologies and it needs to be clarified.
• It would be helpful to readers to have information on spatial and spectral resolution of each sensor.
• Figure 3. Either switch “a” and “b” in the figure or switch section 3.2 with 3.1. Figure 3.a should come first than figure 3.b.
• It would be helpful to include a confusion matrix for the “build-up” and “damage” results.
• It is not described in the text how the analysis of interferometric coherence is used to generate four levels of damage.
• The discussion section needs to be revisited. Where does the method worked well and where it did not and why? This is a weak spot in the study.

Itemized Suggestions

L19. Replace “paper” with “study”

L21-22. Replace “of some of the indices” with “using spectral indices”

L36. “earth” with “Earth”

L77. Remove “and disagreement”

L87. Replace “paper” with “study”

L96-99. This paragraph could be removed.

L101-104. This paragraph could be removed/revised.

Lines 106-107. The sentence “According to the flowchart in Figure 2, the proposed method applies in two main steps after pre-processing.” Needs to be revised. How about: “The proposed method consists of two main steps after pre-processing (Figure 2).”

Line 107. Add “spectral” between using and indices

Line 111. Change “2.1.1 Step 1” to “2.1.1 Step 1 – Build-up detection”

L129-130. The sentence “According to the flowchart in Figure 2, the proposed method applies in two main 106 steps after pre-processing.” needs to be revised.

L139-144. This paragraph can be removed. This is common knowledge in the field of remote sensing.

L150-151. Consider changing the sentence to: “This research used ten spectral indices (Table 1).

L178. The main reference source for the SAM algorithm is “Kruse, F. A., A. B. Lefkoff, J. B. Boardman, K. B. Heidebrecht, A. T. Shapiro, P. J. Barloon, and A. F. H. Goetz. "The Spectral Image Processing System (SIPS) - Interactive Visualization and Analysis of Imaging spectrometer Data." Remote Sensing of Environment 44 (1993): 145-163.”

L191-193. This information was presented before. This is repeated.

L197-198. “imagery available in Google Earth platform”.

L198. Replace “presented” with “generated”

L203. It would be helpful to readers to also have the area of the study site.

L217-218. Consider removing “The evaluation of results is an important part and inevitable for any remote sensing 217 analysis. To this end, having a reliable ground truth is necessary.”

L216. Consider changing “Reference data for damaged area”

L219. Replace “organs” with “agencies”

L219. Replace “is” with “was” (past tense)

L225-229. These sentences do not read well, please consider revising them.

L231. Consider changing to “Results and Discussion”

L237-238. Consider changing to “All pre-processing operations were accomplished using the Sentinel toolbox, a free and open-source software.” Also, please include a reference to this toolbox.

L251-253.Consider changing to  “The proposed approach was compared with a similar method based on normalized differences (ND) between the pre- and co-event interferometric coherence and impervious surfaces extraction using optical imagery [33]”.

L253-255. Please consider revising these sentences.

L263-265. Consider changing to “Results from the proposed method indicate improved performance on both datasets when compared to the ND method.”

L270. Why is the evaluation based on visual analysis only?

L272-273. This sentence does not read well, please consider revising it.

Figure 9 is very difficult to see. Please consider improving its quality.

L318-320. Consider removing it.

L322-329. Consider revising as “Findings from this study have showed: (1) improved accuracy results while not requiring prior knowledge of damages; (2) that this method can provide both binary and ranges of damage level; (3) the potential of incorporating a new temporal series of remote-sensing imagery; (4) the advantages of using a SAR dataset with all-weather capability and day-and-night operation; and (5) that the implementation of the proposed method is simple and has high efficiency with a low computational cost.”

Author Response

Please see the attached file.

Author Response File: Author Response.pdf

Reviewer 4 Report

1. References of some algorithms used in the paper should be given where they firstly appear, such as SAM, Otsu. 2. Since the supervised classifier SVM is used for Sentinel 2 data, why the proposed method is unsupervised? 3. How many images are required when using the proposed method? 4. Details of each part in Figure 2 should be presented in Section 2, including pre-processing. 5. Data and codes of the paper should be released to the public. A link should be given in the paper. 6. Since the proposed method contains several steps and many parameters, experimental results should be given to show how these steps and parameters affect final results. 7. Comparison with only [33] cannot verify the advantages of the proposed method. More state-of-the-art methods should be added for comparison. 8. The authors claim that the proposed method is a rapid one. However, no running time is given in the paper.

Author Response

Please see the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Dear Authors,

Thank you for your in-depth answers to my remarks. As I have just read again your manuscript, it has been thoroughly improved according to the comments provided by all reviewers. Therefore, the current version of the manuscript is much better and could be published in Remote Sensing. 

I would only advise improving the quality of a few images - please make sure the font size makes it possible to follow the information shown in the figures. 

Best regards,

Reviewer 

Author Response

Please see the attached file.

Author Response File: Author Response.docx

Reviewer 3 Report

Please see attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attached file.

Author Response File: Author Response.docx