Temporal Analysis of Mangrove Forest Extent in Restoration Initiatives: A Remote Sensing Approach Using Sentinel-2 Imagery

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

The authors employed a time series of Sentinel-2 images for mangrove mapping over two restoration initiatives in Madagascar and Abu Dhabi. Several comments from the previous round have been fully considered, and the manuscript has been improved. The manuscript has the potential to be published in Forest, but before that, four below comments should be addressed.

1- Line 96, Please add the full form of GEF in its first presence.

2- Lines 155-156, Why is there a huge difference between Sentinel-2 satellite images over the two study areas?

3- Section 2.3, Please add the reference of the land cover map, GMW, that includes four land cover classes. 

4- Further comparison is required to demonstrate the superiority of the UNet over the SVM. Therefore, add a spatial comparison between the outputs and discuss the differences. Such analysis (as a new Figure) can show the advantages and limitations of each model for mangrove mapping. Currently, no explicit spatial difference is observable in Figure 3. The new analysis should clearly represent where the UNet obtained better results.

Author Response

Please see the attachment

 

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 4)

Comments and Suggestions for Authors

Dear authors and editors. I am grateful for the opportunity to re-review the scientific article "Temporal Analysis of Mangrove Forest Extent in Restoration Initiatives: A Remote Sensing Approach using Sentinel-2 Imagery". I have already noted the merits of a scientific article earlier and I will not dwell on this. As for the comments, their corrections were partially corrected after re-submission.

1. The authors use Google Earth Engine. In the appendix, the authors provide a code snippet. At the same time, the further actions of the authors are unclear. The authors only declare the use of certain methods. With this description of the technique, it is impossible to reproduce and obtain the results. In general, the essence of the methodology is clear. However, its implementation is not clear. In particular, nothing is said about how the indexes were calculated – using GEE or ArcMap? How were the data presented in table 3 obtained? Using which programs?

2. Clearly distinguish in the research methodology what was done in which program. Figure 2 does not give an idea of this. It is completely impossible to review a scientific article without this data. Do the authors declare that they used GEE, and in the methodology indicate that they also used ArcMap?

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The paper introduces an intriguing case study employing machine learning classification approaches to map mangrove forest extents using Sentinel-2 imagery. However, the methodology lacks adequate explanation, and information regarding restoration projects in the study sites is insufficient. Without this information, it is not possible to evaluate the paper's contribution to the journal. Additionally, some questions and recommendations are made to enhance the paper’s overall quality and presentation.

1 - In the abstract, the aims of the paper must be clear.

2 - Between lines 92 and 93, it is highlighted that scientific literature suggests better overall classification accuracy with Planet imagery compared to Sentinel-2 images. Consequently, it is imperative to clarify why the authors opted for Sentinel-2 instead of Planet images.

3 - In the methodology, it is crucial to mention, at least as supplementary material, all Sentinel-2 images used, including their respective dates. Additionally, the paper should incorporate keys of interpretation used to select classified samples, revealing, at the very least, the reference image and the corresponding Sentinel-2 image.

4 - Mangrove environments are generally sensitive to temperature and precipitation variations throughout the year. However, the paper does not explicitly detail the dates of the utilised images, and a "median reducer" is used. How were the season variability of the mangrove forest avoided in the paper? The key of interpretation should encompass insights into how the ecosystem behaves throughout the year, especially if images from multiple months were utilised.

5 – In general, the seasonal change in the forest impacts the vegetation indices. How was this avoided in the paper?

6 - It is advisable to present the algorithms used in the paper either as supplementary material or on a developer platform.

7 - In the results section, it is important to illustrate the weight of each variable for the classifications made. Additionally, testing the classifications without vegetation indices to assess their contribution to the classification is recommended. The inclusion of graphical representations (typically boxplots) of the standard deviation of the bands and index values from the samples used for each class would enhance clarity.

8 - The rationale behind choosing only SVM and U-Net over algorithms like RF and ANN, which often yield better results, is not adequately explained in either the introduction or the methodology.

9 - In the results section, it would be valuable to rank classifications based on overall accuracy and Kappa values.

10 - Section 5.2 is pivotal in this paper. However, it is essential to clearly delineate the stakeholders, financing details, and proposed goals for each study area. This clarity would facilitate a comprehensive debate to determine whether the objectives of mangrove restoration projects are being realised.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

I appreciate the authors' efforts in addressing the comments. The manuscript has improved accordingly. Before finalizing, I recommend authors clarify the black rectangles in the caption of Figure 3.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The authors responded adequately to the raised questions. However, I would like to emphasise that the choice of using Sentinel-2 and the selection of machine learning algorithms need to be explicitly stated in the text, similar to how it is addressed in the responses to points 2, 4, and 8 in the coverletter.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I reviewed the manuscript "Temporal Analysis of Mangrove Forest Extent in Restoration Initiatives: A Remote Sensing Approach using Sentinel-2 Imagery" by Farzanmanesh et al. The topic is interesting and important in the context of climate change and the restoration of blue carbon ecosystems. The paper is well-written and easy to follow, and most parts include sufficient explanations. However, several serious issues exist that should be addressed to demonstrate the significance of the current study.

1- Lines 60-62; The authors claimed that limited studies have used CNN-based methods for mangrove mapping. The assertion is not true, considering many recent publications that have used CNN-based methods. Therefore, this part of the introduction section should be revised and recent studies should be included. Moreover, as the authors stated the use of UNet as one of their contributions/objectives of the manuscript, they should revise the contributions/objectives accordingly. Clearly, they should highlight the contribution of their study in comparison with recent publications that used CNN-UNet for mangrove mapping. You can check the below papers that used CNN-UNet methods for mangrove mapping. You can see that modified versions of UNet have also been implemented for mangrove mapping.

1- https://doi.org/10.3390/app13148526;
2- https://doi.org/10.3390/rs13020245 
3- https://doi.org/10.1080/19475683.2018.1564791
4- https://doi.org/10.3390/rs14215554
5- https://doi.org/10.3390/rs12193270

2- Section 2.2; Why is there a notable difference between the number of Sentinel-2 images in the two study areas?

3- Section 2.3; The explanations in this section should be expanded for clarifications. First, please introduce the source of land cover data that contained four classes (Lines 133-139). Second, how did you separate CC and OC classes using the high-resolution images? Please provide further details. Third, please provide further explanations about the manual reference sample collection. Fourth, please revise the sentence in lines 143-144.

4- Section 3.3; Why did you use base UNet? Why didn't you reinforce the UNet architecture with recent backbones?

5- Table 3; Please revise the caption (dates) and add (a) and (b).

6- Section 4.1; I recommend adding more comparisons to make this section more exhaustive. The only statistical comparison is not enough. Please compare the classification results of the UNet and SVM, considering the spatial domain. In this case, you can provide further explanations about their performance and demonstrate their limitations in different parts of the study areas. You can make new figure(s) and show the spatial differences between these two maps. 

7- As the study is focused on restoration initiatives, I recommend enriching the discussion section by including more detailed information about the utility of geospatial and remote sensing data/techniques for the conservation and restoration of mangroves.

Author Response

Please see the attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript titled "Temporal Analysis of Mangrove Forest Extent in Restoration Initiatives: A Remote Sensing Approach using Sentinel-2 Imagery" focuses on using Sentinel-2 imagery to map and assess mangrove extent and distribution in two major restoration projects. It evaluates the effectiveness of these initiatives in mangrove conservation and restoration using advanced remote sensing techniques and machine learning algorithms. The manuscript presents data and analyses related to mangrove cover changes, employing methods such as U-Net convolutional neural network and Support Vector Machine (SVM) for image classification.

1. The abstract provides a clear overview of the study's objectives, methods, and key findings. However, it could be enhanced by briefly stating the significance of the study in the broader context of mangrove conservation and climate change mitigation.

2. The use of Sentinel-2 imagery and the application of U-Net and SVM algorithms for classification are methodologically sound. It would be beneficial to include a more detailed explanation of why these particular methods were chosen over other possible techniques.

3. The manuscript adequately details the data preprocessing steps. However, a more thorough discussion on the selection criteria for imagery, particularly in terms of cloud coverage and other potential distortions, would strengthen this section.

4.  The comparison between U-Net and SVM models is insightful. It would be helpful to include a discussion on the limitations and potential biases of each model to provide a more balanced view.

5. The results are clearly presented. A deeper analysis of the causes behind the observed mangrove cover changes, considering both natural and anthropogenic factors, would add value to the discussion.

6. The description of the methodologies, particularly the U-Net and SVM algorithms, requires more detailed explanation. Specific parameters used in the algorithms should be clearly stated.

7. The interpretation of results could be improved by providing deeper insights into the implications of the findings, especially in the context of mangrove conservation.

8. Some references seem outdated or irrelevant. Ensuring that the most current and relevant literature is cited would strengthen the manuscript's credibility.

Author Response

Please see the attached.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you it was interesting study. If it is possible improve some figure qualities that some places can be viewed with magnification and check figures captions in terms of font similarity.

Best,

 

 

Author Response

Please see the attached.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Dear authors and editors. I am grateful for the opportunity to review the scientific article "Temporal Analysis of Mangrove Forest Extent in RestorationInitiatives: A Remote Sensing Approach using Sentinel-2 Imgery". I would like to note the comparative approach that the authors use for two key research areas. I would like to note that the subject of the scientific article corresponds to the research area of the scientific journal. The article is devoted to an urgent topic. There are several comments, the elimination of which will significantly improve the quality of the scientific article.

1. The authors use Google Earth Engine. At the same time, the code according to which the research is carried out is not given anywhere in the manuscript of the scientific article. The authors only declare the use of certain methods. It is completely impossible to review a scientific article without this data.

2. Specify the limitations of the study in more detail.

3. In section 1, provide a more detailed review of the literature.

4. In section 4, compare the data you have received and their dynamics with other regions of the world.

Author Response

Please see the attached.

Author Response File: Author Response.pdf