Automatic Semi-Global Artificial Shoreline Subpixel Localization Algorithm for Landsat Imagery
Round 1
Reviewer 1 Report
This paper presents an appropriate methodology for shoreline detection using a satellite data and the results presented are within an acceptable accuracy for physical use cases. However, the case studies are carried out in only on an artificial shorelines, but the title and conclusion gives reader an impression that methodology presented in the paper can be applied globally for all shorelines. To make this assertion, the algorithm should be tested for different shorelines (including natural shorelines near vegetative growths). If this paper is to be published as it is, the tile and conclusion needs amendment to reflect the actual content of the paper. Otherwise, further tests over other coastal regions needs to be tested for a scientific rigor.
Author Response
Response 1 ：We are very thankful for your careful revision and constructive comments. With consideration carefully, we have changed our title to ‘Automatic Semi-Global Artificial Shoreline Subpixel Localization Algorithm for Landsat Imagery’. And we also have modified our research aim on how to deal with the various geometric morphology artificial shorelines subpixel positioning.
Reviewer 2 Report
The proposed manuscript describes an automaticatic approach to extract subpixel shorelines from Landsat imagery.
The manuscript is written in a fluent English and is easy to read in terms of language but some confusion in remote sensing terminology and in the morphodynamics of natural systems, specifically on how to use shoreline position for erosion studies are detected. The number and quality of the provided references seems appropriate and recent, but along the manuscript they do not always consubstantiate the state-of-the art as much as it should.
My main concerns with this manuscript is the introduction section, as I don´t find it in sync with the rest of the manuscript. The methodology is mostly an image processing problem applied to satellite images – remote sensing data – which is the most accurate detection of the water/land boundary in coastal artificial structures. In this sense it is appropriate to the present journal. But the introduction is more about the specific applicability of automatically extracted shorelines in the study of shoreline erosion. Artificial structures don´t erode, they get damage or destroyed.
As such I identify these main issues:
1) Coastline/shoreline terminology should be substitute by water/land boundary.
2) The title, objectives and introduction should be reformulated as to account the first change and centered on accuracy of detecting water/land boundary in coastal artificial structures.
3) Satellite images are referred as remote sensing images, but remote sensing is a field quite extensive and not limited to satellite sensors.
4) The description of how ground-truth was acquired is essential on the methodology section and not just in the discussion section.
5) Section 5.5 should be removed as there is no relation (background, data or knowledge) to support the results and discussion. To introduce this application an entire new manuscript would be needed, including a thorough shoreline indicators review and its relations with coastal processes.
More particular and smaller issues are highlighted in the main manuscript file.
From all the above it is my belief that the present manuscript would benefit from an in-depth restructure regarding objectives and introduction, including an updated state-of-the art in accordance to the new objectives. With all these in mind my recommendation is to Reconsider after major revision (control missing in some experiments).
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment. And all the modifications have been marked in red in the new manuscript.
Author Response File: Author Response.pdf
Reviewer 3 Report
Thanks for the opportunity to review this paper. The subject matter potentially is of high interest to the research community, especially coastal remote sensing community.
English – needs some corrections for grammar and style. For example, an algorithm cannot have thoughts, can have parts, procedures, functions, etc. Also some sentences are difficult to understand.
Nowhere is commented on the influence of tides in the imagery, that might be more important than anything else when shorelines are concerned, especially if the shoreline is natural, e.g. not armored, or is protected (e.g. have a sea wall or riprap). At least this problem needs to be mentioned, if not address. Do you use only imagery with the same tide level, or do you look at tide gages for the time the imagery was collected and correct for that, or what? In there any tide difference between the Landsat image and the GF-2 image? Is the error introduced by different water levels taken into consideration?
Registration error – 3 m – Is this average error, RMSE, min/max error? How this influence all the error measurements between the shoreline derived from Landsat and the shoreline derived from GF-2?
Ln 130 – it is OTSU an acronym – if so spell it out – or the name of a person? I found Otsu (1979) in the Reference list.
Ln 131 – 132 – The shoreline is divided in what? Smaller segments of homogenous morphology (and what do you understand as morphology, it might be different for a geologist, coastal geomorphologist or a geographer), quasi-straight lines, something else? This need specifying even if it is obvious to you, or explained later in the paper, it might not be obvious to the reader.
Figure 2: Misspelled “Ftiing”
Ln 149 – 150: Why necessarily intensity A smaller than intensity B? Cannot be reverse? Shouldn’t it be intensity A different (dissimilar, not equal) to intensity B? (since it may depend to assigning A or B to water / land?)
Equation 6: It is correct but wouldn’t be more elegant to say that SUMi* = AxSAi*+BxSBi* = (A-B)SAi*+B(m1+m2+1) if you take into consideration equation 3? Afterwards whoever feels so inclined can use equation 5 to calculate (SAi*).
Ln 185 – 190: Please rephrase the paragraph. It has grammar errors and makes it difficult to understand.
Figure 4b. The rectangles of different line colors are hard to see. The scale bar – should the values below 0 be negative?
Ln 201 – MNDWI acronym needs to be spelled out first time of use.
Ln 219 – LoG acronym needs to be spelled out first time of use
Ln 229 – PAE acronym needs to be spelled out first time of use.
Ln. 243 – Incomplete sentence.
Ln. 250 – 251 – The sentence seems incomplete or needs rewording.
Ln 259 and equation 18: The units of slope k need to be defined (usually angular degrees) – and if that is true, what is the significance of a negative k? Has it something to do with direction, and if so, how?
Ln 289 – 291 – Re-segmentation: This paragraph needs re-wording – it is difficult to understand.
Ln 311. It seems there is confusion about what “ shoreline morphology” means for a coastal geologist, geomorphologist or geographer. Usually when we speak about shoreline morphology we refer to natural shores such as natural beaches, natural unarmored shoreline, natural wetland, upland shoreline, etc, or armored shoreline with sea walls, bulkhead, riprap. All these types of shoreline morphologies will react differently to tide regimes, sea level rise, and storms. In Fig 6a and Fig 6e it seems you have 2 shorelines with sea walls, one with a concave point of inflection and one with a convex point of inflexion. How are those different shoreline morphologies besides the convex / concave point of inflexion, especially when you segment them in straight lines? Each line in this case has the same morphology. Also, we may understand morphology as referring to how complex the shoreline is, it is a straight line, it is gently curving being convex or concave towards water / land, does it have deep pocket beaches or promontory, or it is very irregular, changing direction with high frequency? Not to mention that this is very dependent on the scale we are looking at the shoreline.
Ln 313 – Is it not Figure 6b and 6f instead of Figure 6b and 7f?
General comments for methodology: It seems that majority of segments for shorelines in 6a, 6e and 6i (if not all) are straight lines or extremely close to it, so is it not an overkill to model these lines with a cubic polynomial instead of a straight line (1st order polynomial)? It is relatively very easy to compare your segments to a straight line and judge if a cubic polynomial is really needed. Also, do you have any example of a natural unarmored shore? Maybe in Figure 6q?
Ln 330. Line matching – how was it calculated? The readers are very familiar with MAE, RMSE or SD, but line matching can mean different things including matching line features in stereo imagery for example with SLAM (Simultaneous Localization and Mapping) for structure from motion applications.
Ln 335 – 336. Your error stats are between 2.5 and 4.77 m and already in the text is mentioned that the registration errors are 3 m, although you don’t mention what 3 m represent (RMSE, MAE, min/max error ….). In this case a misalignment error of +/- 3 m for example is big enough to have more than a subtle effect. Also, do you have any bias in your errors?
Ln 350: Acronyms NDWI, MNDWI and AWEI need to be spelled out first time of use.
Ln 380: Intensity slope – does that mean the slope of intensity gradient?
Figure 9. Is any of the sample windows over a natural unarmored shore? All examples from 9a left are over sea wall shores (it seems) and the same can be said for 9a right image (maybe with a possible exception for the window on top right, but the scale of the image make the identification uncertain).
Ln 461. Finally, the natural shorelines are discussed, and the influence of water levels acknowledged. I think this should be mentioned way up front, because three quarters of the paper really discusses the method only in the armored shoreline condition when the water levels are not a problem. Also, using the same image GF-2 at high resolution and down sampled at lower resolution to assess your algorithm accuracy is not appropriate. You will just get the effect of changing image resolution on your algorithm accuracy. With the satellite imagery you get metadata that will indicate the time of day, not only date at which the image was acquired. Using information supplied by tide gages you can figure out what the approximate difference in water levels might be and use that in your calculations.
Figure 13: The results are interesting, but the amount of noise is relatively small, from 1 to 5%, what happens at higher % or noise? At what % noise level the SGSSL algorithm will perform similar to PAE algorithm?
Table 6. Again, this just demonstrates the influence of down-sampling on your algorithm plus the errors between hand digitizing and an automatic method of extracting the shoreline, since you used the same image. Besides hand digitizing is very subjective and depends on the skill of the analyst that does the digitizing.
Conclusions: Only in this section it is made clear that the method was tested on artificial / armored shorelines mainly, specifically sea wall shorelines. And generally, sea wall shorelines are the easiest case scenario. For the sandy beach, the method was tested on a down-sampled image that was used also at high resolution for digitizing the “real” shoreline. This demonstrates only the image resolution influence on the proposed algorithm and digitizing errors, but not the accuracy of the method itself on sandy shores. For a real test the authors should take into consideration water levels at the time of the image acquisition (both GF-2 and Landsat 8) in conjecture with readings from the local tide gages, and use different images for testing and comparison / validation.
Reference [37] seems to be wrong in the reference list: authors are Mikolajczyk, K, and Schmid C. Please check all the other references to make sure they are recorded correctly.
Author Response
Please see the attachment. All modifications are marked in red in new manuscript.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
These comments refer to the re-evaluation of the revised version o the manuscript Automatic Semi-Global Shoreline Subpixel Localization Algorithm for Landsat Imagery.
Generally, authors have complied to the reviewer’s suggestions and the manuscript improved substantially. A better explanation of how this methodology is going to be used to monitor and manage artificial shorelines would have benefit the overall manuscript but does not hinder my final conclusion:
The present manuscript is ready to be Accept after minor revision. Minor corrections are highlighted in the second revised version of the manuscript.
Comments for author File: Comments.pdf
Author Response
Dear Reviewer，
Thanks for your constructive and helpful comments and suggestions. We have modified the manuscript according to your 'peer-review-4648489.v1.pdf' . We are very grateful for your careful revision. And we are very sorry for our some mistakes.
All the modified sentences and words have been marked the background color cyan. To express the manuscript's correctly, we have conducted a English Edit again by the Letpub professional company.
Author Response File: Author Response.pdf
Reviewer 3 Report
The reviewer is satisfied with the author responses, and the manuscript is markedly improved.
The reviewer appreciates the authors effort to use a specialized service to correct English style and grammar. Unfortunately, prepositions and articles are still sometimes out of place, but that can be attributed to the reviewer personal preferences. Even so, for example, a quantity “approaches zero” but does not “approaches to zero”.
There is one typo in Ln 258: it is “subsection” and not “subsecction”.
Ln. 206 – 208: This sentence is still not clear, and / or grammatically correct: maybe it needs to be change in something similar to: Then, the error expressed by Equation 10 “is different than zero”, or “is not equal to zero”.
Author Response
Dear Reviewer,
We are very grateful for your helpful and constructive comments and suggestions.
In the modified manuscript :
“approaches zero” have been modified to “approaches to zero”.
“subsecction” has been replaced to “subsection”.
Ln 206-Ln 208, the sentence has been modified as: "which leads to the intensity integral error expressed in Equation (10) is not equal to zero."
All the modified sentences and words have been marked the background color cyan.
To express the manuscript's correctly, we have conducted a English Edit by the Letpub professional company again.
Author Response File: Author Response.pdf
