Oil Spill Detection with Dual-Polarimetric Sentinel-1 SAR Using Superpixel-Level Image Stretching and Deep Convolutional Neural Network

Round 1

Reviewer 1 Report

The article is well written and conceived. The topic is very current, and the discussion and results will serve many researchers in improving their work. The figures, tables and graphs contained in the article are of satisfactory quality. Congratulations to the authors on a very interesting and current work.

Author Response

Dear Reviewer 1, thanks a lot for the very positive comments and for recognizing the value of this work.

Reviewer 2 Report

I have reviewed manuscript remotesensing-1802763 by Zhang et al. Whenever I review a manuscript, the first question I ask is whether the manuscript describes an application of the scientific method. The scientific method involves asking questions and/or testing hypotheses. This manuscript does neither. No questions are asked; no hypotheses are posed. I think it should be possible to rationalize this work based on answering a question or testing an hypothesis, but the question or hypothesis needs to be clearly stated in the Introduction. Near the end of the manuscript, the authors need to clearly state the results in terms of the answer to the question or whether the hypothesis was accepted or rejected. The other big concern I see with this manuscript is that “ground truth” is never defined. If the question being asked or hypothesis being tested has something to do with detecting an oil spill, then the authors need to clearly define what ground truth means in the context of an oil spill. Presumably an oil spill means that there is oil floating on the surface of the ocean, but how do we know that there was oil floating on the surface of the ocean? Was there someone in a boat taking samples? What concentration of oil on the surface is defined to be a spill? Without a clear definition of ground truth, I do not see that this study proves anything. For purposes of this study, it is absolutely essential that ground truth be defined. That information is completely lacking. Finally, a study titled “Chronic oiling in global oceans” was recently published (Dong et al., Science 376: 1300-1304 (2022)). The conclusion of that study is “Our findings reveal that the present-day anthropogenic contribution to marine oil pollution may have been substantially underestimated”. Based on the information in this manuscript submitted to Remote Sensing, do the authors feel that the conclusions of Dong et al. (2022) are justified?

Author Response

Thank you for your valuable comments on the article. We have revised the article according to your comments. The attachment details the parts and contents of the changes, and answers your questions. Express our gratitude again!

Author Response File: Author Response.docx

Reviewer 3 Report

Review of “Oil Spill Detection with Dual-polarimetric Sentinel-1 SAR by Superpixel Level Image Stretching and Deep Convolutional Neural Network” by Zhang et al.

 

Proposed is an oil spill detection method using image stretching based on superpixel and convolutional neural network. It was found that the method could effectively improve the classification accuracy. The topic is interesting. Overall, I like the paper. I would like to recommend the manuscript be accepted for publication after some minor revision.

 

Specific comments:

 

Some abbreviations are used in this manuscript, they may save some space, but they make the paper unnecessarily hard to read. For example, I have to go back several times to check what “CS” and “OS” refer to. A paper should be different from a computing code. I would strongly suggest to get rid of these abbreviations: CS, OS, LA, LAND, SH.  Note that CS, OS, and LA are widely used to stand for computer science, operating system, Los Angeles, respectively.

 

Line 18, “this paper proposed” should be changed to “we propose”. Please note that “this paper” can’t propose anything. Also, present tense should be used instead of past tense in this case. These apply to many other places in the manuscript as well.

 

Lines 30-34, the introduction starts with the story of the famous 2010 oil spill in the Gulf of Mexico, and cited a PhD thesis. I think the seminal work – the AGU book (Liu et al., 2011) – should be properly cited instead.  The book was the original work contributed by the first responders of the oil spill incident, including the monitoring and modeling of the oil spill. The monitoring part of the book has several chapters covering the topic of oil spill detection. There are many follow-up studies of that oil spill, including the cited PhD thesis. It is important to cite the original work.

 

Define “MIoU” in the abstract.  It is not a commonly understandable word.

 

Line 121, “this paper proposed” should be changed to “we propose”.

 

Line 124, “this paper designed” should be changed to “we design”.

 

Line 128, “MIoU” should be defined when it first appears in the main text.

 

 

Reference:

 

Liu Y., MacFadyen A., Ji Z.-G., Weisberg R. H. (Eds.) (2011). “Monitoring and Modeling the Deepwater Horizon Oil Spill: A Record-Breaking Enterprise” Geophys. Monogr. Ser. vol. 195 (Washington, D.C., USA: AGU/geopress), 271 PP. doi:10.1029/GM195.

 

 

Author Response

Thanks a lot for your recognition of this paper’s contributions and the positive comments for our manuscript. We really appreciate it. Hopefully the revised version can meet the request for publication. According to your suggestion, we have revised the article and reference. The details of the modification are listed in the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I have reviewed the revision of remotesensing-1802763 by Zhang et al. The last sentence of the Introduction could serve as an hypothesis: The proposed method could significantly improve the accuracy of oil spill detection, and alleviate the inconsistency of accuracy caused by imaging differences. The null hypothesis would therefore be: The proposed method cannot significantly improve the accuracy of oil spill detection and cannot alleviate the inconsistency of accuracy caused by imaging differences.

 

The first few sentences of section 3.1 Dataset are very poorly written. The authors begin by saying that they are going to use SAR data from the Persian Gulf and then abruptly launch into a discussion of oil spills in the Gulf of Mexico without offering any explanation of what the relevance of the Gulf of Mexico is to the research. If they have SAR data from the Gulf of Mexico, why does the first sentence mention only the Persian Gulf?

 

The biggest concern I have with this manuscript is the ground truth. The authors say that (lines 261–263) “Oil spill areas in these data were verified based on in-situ investigations and recorded by Scanex. According to this information, we manually interpreted and labeled different areas on the image as ground truth.” I infer that the authors have ground truth when there was spilled oil on the water. However, the following four possible outcomes all need to be investigated to determine whether the proposed method can significantly improve the accuracy of oil spill detection and can alleviate the inconsistency of accuracy caused by imaging differences.

 

1.     It correctly identifies an oil spill when there is an oil spill

2.     It fails to identify an oil spill when there is an oil spill

3.     It correctly identifies an area where there is no oil spill when there is no oil spill

4.     It incorrectly identifies an oil spill when there is no oil spill

 

The ground truth available to the authors enabled them to investigate only the first two of these possible outcomes. They have no ground truth for the absence of an oil spill. Therefore the images they characterize as “look-alike areas” are not areas where in-situ investigations confirmed the absence of an oil spill. The only ground truth they have are instances where in-situ investigations confirmed that there was an oil spill.  This would have allowed them to test for false negatives (the second possibility above) as well as true positives (the first possibility above) but would have given them no way to test for true negatives (the third possibility above) or false positives (the fourth possibility above). If the only ground truth the authors have is verification of oil spill areas (and that is what the authors say in lines 261–263), I cannot see that they had any rigorous way to determine whether what they are calling “look-alike areas” were really “look-alike areas” or instances of oil spills that did not happen to be confirmed by in-situ sampling. It is hard to believe that every oil spill that occurred in the Persian Gulf and Gulf of Mexico was verified by in-situ sampling.

Author Response

Thank you for your suggestions, especially the part about ground truth. We have carefully read your comments and revised the article. The specific modifications are written in the attachment.

Author Response File: Author Response.pdf