A Semi-Automated, Hybrid GIS-AI Approach to Seabed Boulder Detection Using High Resolution Multibeam Echosounder

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Lines 85-86: seems like an editing error. Parenthetical doesn’t make sense.

Several places in intro: inconsistent punctuation and spacing, especially with references.

Lines 92+ seem to be copies of lines 37+.

Suggest returning to authors for a more careful job editing before a review is warranted.

Author Response

Author Reply

We appreciate Reviewer 1’s concerns and we have worked to improve the manuscript in the areas defined by the checkboxes above.  We have also addressed the specific suggestions and have noted the completed revisions below.  We note that Reviewer 1 has supplied fewer substantive comments than other reviewers, providing less direction for the improvement of the actual research content; however, we are confident that the comments from the other three reviewers have helped to satisfy not only this reviewer’s concerns, but also improved the presented study. 

Comments and Suggestions for Authors

• Lines 85-86: seems like an editing error. Parenthetical doesn’t make sense.

Amended see line 86.

• Several places in intro: inconsistent punctuation and spacing, especially with references.

Amended. Comments marked and changed tracked in “TrackChanges” word document.

• Lines 92+ seem to be copies of lines 37+.

Amended, duplicated lines 92+ removed.

• Suggest returning to authors for a more careful job editing before a review is warranted.

We accept that improvements were necessary and, as stated above, we are confident that implementing the edits suggested in the comments from this reviewer and the others have addressed some inconsistencies present in the original submission.  However, we respectfully disagree that the manuscript suffered from an overall poor quality of English.  We also note that the other three reviewers reported opinions aligned with our own on this matter.  Despite this, we have striven to improve the English throughout the manuscript. 

 

Reviewer 2 Report

Comments and Suggestions for Authors

     This study proposes a semi-automated approach combining GIS and a Random Forest (RF) classifier to detect seabed boulders and their clusters from high-resolution multibeam echosounder (MBES) data. The authors validated the method in a 0.24 km² test site in Long Island Sound, achieving a recall rate of 83%, precision of 73%, and an F1-score of 77, slightly outperforming the average of expert manual annotations. Its advantages include efficiency improvement, and it is suitable for fields such as offshore wind power planning and ecological research. Though some details require further refinement.

1. Fig 4(c) displays a heatmap of the areas with the most intense clustering of contacts. Please briefly describe what the color gradients in the figure represent.
2. In Section 3.5.3, the authors state: "This approach was selected for its demonstrated effectiveness in geospatial applications, particularly in classifying remote sensing data involving smallscale features such as boulders." Please provide a concise explanation of why this algorithm is more suitable than others.
3. In line 431, regarding the Gini index, please supplement with an explanation.
4. In lines 572 and 717, the expression of "numerical units" is inconsistent. According to the International System of Units, a space should be left between the number and the unit, e.g., "0.5 m." Please carefully review the rest of the article for similar issues.
5. The citation format in the references is inconsistent. For example, the date formats in references [12] and [48] differ. Please thoroughly check other instances.
6. The references include few recent publications (within the last two years). It is recommended to add at least three more recent references.

Author Response

Author Reply

Reviewer 2 has demonstrated a good understanding of the manuscript and we appreciate their insights.  Their comments have helped us improve the presentation of results and their relationship with the following conclusions.  We have also addressed the specific suggestions and have noted the completed revisions below. 

Comments and Suggestions for Authors

1. Fig 4(c) displays a heatmap of the areas with the most intense clustering of contacts. Please briefly describe what the color gradients in the figure represent.

The colour gradients in Figure 4(c) represent the relative density of detected contacts across the study area, calculated using kernel density estimation. We have added this information into the text, see line 474 – 477. The warm colours (e.g., red) indicate areas of higher concentration of detected contacts, while the cooler colours (e.g., white) indicate lower density regions. This figure has been updated (Page 14), and the heatmap legend title has been changed to “Contact Density” with a new colour ramp title of “High to Low”. Figure 4c caption has been updated to “KDE heatmap of contacts”.

2. In Section 3.5.3, the authors state: "This approach was selected for its demonstrated effectiveness in geospatial applications, particularly in classifying remote sensing data involving smallscale features such as boulders." Please provide a concise explanation of why this algorithm is more suitable than others.

An explanation for this statement has been provided in the text from lines 365 – 372.

3. In line 431, regarding the Gini index, please supplement with an explanation.

An explanation has been added into the text (Line 384 – 387).

4. In lines 572 and 717, the expression of "numerical units" is inconsistent. According to the International System of Units, a space should be left between the number and the unit, e.g., "0.5 m." Please carefully review the rest of the article for similar issues.

Amended, see lines 340, 434 and 532. Document has been thoroughly checked for this.

5. The citation format in the references is inconsistent. For example, the date formats in references [12] and [48] differ. Please thoroughly check other instances.

Amended see references [15,64,67].

6. The references include few recent publications (within the last two years). It is recommended to add at least three more recent references.

Section 5.3 of the paper has been updated to include three more recent publications. See lines 673 – 680 and 693 – 708.

Reviewer 3 Report

Comments and Suggestions for Authors

1 The paragraph from line 66 to line 74 is duplicated with another paragraph from line 122 to line 130. Please revise it.

2 In line 336, there is an expression of c.5 km. I'm sorry I don't understand it, please provide an explanation to make it easy to understand.

3 The training process and the test do not exist at the same time in the detection process, and generally after the training is completed, the training is no longer in use. So The workflow in Figure 2 should be revised.

4 In line 638, there is an expression of c.11 hours. I'm sorry I don't understand it.

5 In line 758, there is an expression of km2. I think it should be revised.

Author Response

Author Response

Reviewer 3 has provided good feedback; except they have indicated that the research design and description of methods could be improved.  This is important feedback.  We note that it seems only their third comment, addressing the training process, pertains to these points.  We have revised the manuscript to address this comment and acknowledge that it does make an improvement to how our methods are presented. 

1 The paragraph from line 66 to line 74 is duplicated with another paragraph from line 122 to line 130. Please revise it.

Amended, duplicate lines 122 – 130 removed.

2 In line 336, there is an expression of c.5 km. I'm sorry I don't understand it, please provide an explanation to make it easy to understand

The “c.” in c.5 km is a standard abbreviation of “circa” 5 kms. We note that there should be a space between “c.” and “5” to address a previous comment. This has been amended (line 280).

3 The training process and the test do not exist at the same time in the detection process, and generally after the training is completed, the training is no longer in use. So, the workflow in Figure 2 should be revised.

The workflow has been revised to clearly show that the training process takes place separately to the test site dataset. Once the model is trained the test site SFPs are passed through the model and predictions are made. Figure 2 “RF Classifier” section has been updated to reflect this. See page 10.

4 In line 638, there is an expression of c.11 hours. I'm sorry I don't understand it.

See comment 2 response. In this context we have updated the phrasing to “approximately 11 hours”. See line 598.

5 In line 758, there is an expression of km2. I think it should be revised.

Amended, see line 741.

 

Reviewer 4 Report

Comments and Suggestions for Authors

Review of “A semi-automated, hybrid GIS-AI approach to boulder detection using high resolution multibeam” by Downing et al.

 

This study developed the new data processing procedure to detect potential of clusters of seabed boulders from the 0.5-m resolution MBES dataset by using GIS-based filtering functions and machine learning techniques. The authors results were verified by comparing with the manual detections by the experts. Although the ground-truth that the boulder clusters are truly exhibited at in-situ points was not be shown, their procedure can contribute to construction of offshore infrastructures such as wind power plants. Overall, the presentation is well organized and the contents are interesting. So, this manuscript can be published in the journal of Remote Sensing; however, some line comments would be raised to be addressed before publication.

 

Lines 101 – 142: The same phrase is repeated with lines 45 – 85. Probably the former paragraphs should be removed. Anyway, the authors should reconsider this contextual issue.

 

Line 361: I guess that Fig1 may be Fig2.

 

Line 391: There is no Table 2 in the manuscript.

 

Line 502: Table 2 here should be referred to the appropriate table.

 

Appendix: Overall, the Appendix section seems not completed, e.g., the caption of Table A1 and no captions for all figures.

Author Response

Author Reply

Reviewer 4 has provided a summary of the manuscript that demonstrates a good understanding of our study.  We appreciate their effort, and the insights provided. 

 

Comments and Suggestions for Authors

Lines 101 – 142: The same phrase is repeated with lines 45 – 85. Probably the former paragraphs should be removed. Anyway, the authors should reconsider this contextual issue.

Amended, duplicate lines 101 – 142 removed.

Line 361: I guess that Fig1 may be Fig2.

The reference to Figure 1 is correct. As stated in lines 303-304: “The workflow takes 2 DEM inputs, corresponding to Test and Training Sites (Figure 1) and consists of four main steps (Figure 2). A summarised graphical representation is shown in Figure 3.” In this context the “corresponding to Test and Training Sites (Figure 1)” refers to the geographical location of the DEMs shown in Figure 1.

Line 391: There is no Table 2 in the manuscript.

 Amended see line 337.

Line 502: Table 2 here should be referred to the appropriate table.

 Amended see line 457.

Appendix: Overall, the Appendix section seems not completed, e.g., the caption of Table A1 and no captions for all figures.

Amended see Appendix.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

This study proposes a semi-automated hybrid GIS-AI approach that integrates bathymetric position index filtering and a Random Forest classifier to detect boulders and delineate boulder fields from multibeam echosounder (MBES) data. The method has been rigorously validated against extensive experimental datasets and demonstrates superior efficacy compared to manual expert selection, achieving efficient detection of seabed megaclasts. It holds significant potential for applications in geological hazard assessment, benthic habitat mapping, and offshore infrastructure planning. The specific comments are as follows:

1. Formatting requires revision (e.g., missing indentation in some paragraphs).
2. The reference numbers in the main text are not sequentially ordered.
3. The citation format for Reference [38] is incorrect.
4. Table 1 appears incomplete.
5. The "Author Contributions" section and other such content are only placeholders, and no information has been filled in.
6. The reference list (70 entries) is overly extensive; prioritize relevance to the study’s scope.
7. Academic language should be further refined to enhance rigor.

Author Response

1. Formatting requires revision (e.g., missing indentation in some paragraphs).

We thank the reviewer for their feedback. Indents included at the beginning of all relevant paragraphs, see lines 30, 393, 487. 

2. The reference numbers in the main text are not sequentially ordered.

Reference numbers have been updated and are now in sequential order in line with Remote Sensing Guidelines. We thank the reviewer for their feedback.

3. The citation format for Reference [38] is incorrect.

We thank the reviewer for catching this. The reference has been updated see reference [70]. See line 982-983.

4. Table 1 appears incomplete.

Table 1 has been revised to improve clarity and completeness. Specifically:

1. We clarified the meaning of all detection methods listed (e.g., “Algo” and “Algo with QC” now refer explicitly to semi-automated methods, as noted in the revised table caption).
2. We enhanced the table caption to include definitions for key terms and metrics (TP, FP, recall, precision, F1 score), and we now state that these metrics are expressed as percentages.
3. We added a brief description of how the “Master” dataset was generated, to provide context for its role as a ground truth reference.
4. We ensured consistent formatting and structure across all rows and columns.

We believe these updates address the concern and make the table fully self-explanatory.

5. The "Author Contributions" section and other such content are only placeholders, and no information has been filled in.

The Author Contributions section was in fact fully completed in the original submission. However, we acknowledge that the formatting may have made it appear as a placeholder. In response, we have reformatted the section to improve clarity, removed quotation marks, and ensured consistent punctuation as per MDPI guidelines. The updated section now clearly reflects the specific contributions of each author. See Line 785-790

6. The reference list (70 entries) is overly extensive; prioritize relevance to the study’s scope.

Thank you for this comment. We appreciate the importance of maintaining focus in the reference list and have removed several citations that were only marginally relevant. The Reference was 78 entries, it is now 72 entries. However, we respectfully note that our study addresses a range of interrelated topics—including boulder and boulder field classification, remote sensing-based geomorphology, GIS-based spatial analysis, and machine learning model development. Given the interdisciplinary nature of the workflow, we believe that citing foundational and field-specific work across these domains is important to both contextualize our contribution and properly credit prior studies that underpin our methodology. The current reference list reflects this scope, and we have taken care to ensure all included citations are directly relevant to one or more core aspects of the study.

7. Academic language should be further refined to enhance rigor.

We stand by the scientific rigor of the study and consider this a separate issue to minor linguistic improvements.  However, we appreciate this feedback and have reviewed the manuscript in light of this insight.  As a result we have made grammatical improvements throughout.  Instances of passive language have been removed and some wording has been made more concise.  These edits have mostly been made in the Introduction section. Paragraphs with updates have been highlighted in the track changes document.