Analysis and Quantification of the Distribution of Marabou (Dichrostachys cinerea (L.) Wight & Arn.) in Valle de los Ingenios, Cuba: A Remote Sensing Approach
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper develops a method to classify populations of an invasive weed across the landscape utilising satellite imagery and freely available Google Earth Engine. This is an important problem for which the authors have developed an effective solution.
Such a tool is very useful to allow a landscape scale understanding of the spread and abundance of invasive species that is often not otherwise available. It also allows analysis of the impact of land management practices, which the authors undertake.
Major issues
The authors need to provide more information relating to how the training data and sampling plots were used to derive the random forest classifications. It is not at all clear to me how this has been done. This is a critical part of the paper as it underpins the classification accuracy presented by the authors, and, as far as I can tell, is the only ground truthing that has been done for the paper. Section “2.2.4 Field data” presents information about the ground truth parcels that were collected in 2014. No information is provided about the process for classifying these parcels and how pure they were, e.g. is Parcel 1 pure sugar cane, or sugar cane spread around roads, farm infrastructure, housing etc? Furthermore, the land cover classes used here (listed lines 206 to 207) are not the same as those listed in Table 1, nor are they the same, or even similar, to the random forest classifications presented in section “3.3 Confusion matrix” and “3.4 Coverage maps and data”.
Another problem is that I am not clear what marabou distribution associated with Fig 6 represents. Is there a cover threshold for the associated pixels, e.g. dense marabou infestations? Or are the pixels those with any amount of marabou present? Surely there is a minimum threshold.
Minor issues:
Citation 10 listed in line 67 and does not have last names listed in the reference list (line 772).
Figure 1 left is a very low-resolution image. It is not helpful as is.
Appendix B not A – Line 332 refers to Appendix A instead of B.
Bold instead of yellow highlight would be more appropriate for Table 2
Comments on the Quality of English LanguageEnglish language quality in fine.
Author Response
Dear Sir,
I sincerely thank you for the time and effort you have invested in reviewing our manuscript entitled "Analysis and Quantification of the distribution of marabou (Dichrostachys cinerea L. Wight & Arn.) in Valle de los Ingenios, Cuba: a remote sensing approach". Your constructive feedback has helped us to refine our work and we greatly appreciate your insightful comments.
I am glad that you recognised the importance of the problem addressed in our study and that you think the proposed solution is a good one. Your positive feedback on the usefulness of the developed tool for understanding invasive species dynamics at the landscape level and their impact on land management practises is truly encouraging.
In response to your thorough review, we have made the following changes:
Key points:
- To address this concern, we will provide a more detailed explanation in Section "2.2.4 Field data”. The original field worksheets from our Cuban partner are now included in Appendix A to provide a better overview of the classes for which extensive fieldwork was conducted. This addition is intended to increase the transparency of the data collection process.
- The classification criterion used for the distribution of marabou in Figure 6 is: if it contains any amount of marabou, it is classified as marabou, as we have proceeded to detail in point 2.2.5.
Minor issues:
- The error in the reference to Appendix A has been corrected, and the correct reference has been inserted.
- The low-resolution image in Figure 1 (left) has technical limitations as it originates from our Cuban partner.
- The citation error in reference 10 has been corrected.
- The highlighting in Table 2 has been changed from yellow to bold to reflect your preference for more appropriate formatting.
We believe these revisions will address your concerns and contribute to the overall improvement of our manuscript. We appreciate your attention to detail and your commitment to improving the quality of our work.
Thank you for your time and consideration. We look forward to any further dialogue and hope that our responses will meet your expectations.
Best regards,
Eduardo Moreno Cuesta
University of Huelva
Reviewer 2 Report
Comments and Suggestions for AuthorsRemote sensing of invasive plant species is an important research topic worldwide, though some regions remain understudied. Moreno et al. use the Google Earth Engine platform to map the invasive Dichrostachys cinerea in Cuba from 2000-2018 with high accuracy applying the random forest algorithm.
Main comments:
Training and validation of the model:
If 29 data points were collected (L205), how does this translate to 136 points in the confusion matrix?
Parameterization and application of the RF algorithm needs to be elaborated. E.g., what were the input variables, and were any parameters optimized?
Results:
Avoid and remove citations in this section.
Avoid and move any interpretation to the discussion (e.g., L333ff).
Discussion
Please discuss also potential limitations of your study, e.g. due to the use of different satellite platforms, the projection of the 2014 model to other time steps without validation, the relatively high accuracy using only the NDVI.
Further comments:
Introduction:
L27-30: Elaborate in more depth about the topic of invasive species and their ecological impacts. There is a rich body of literature on this topic.
L50ff: There are recent reviews on remote sensing and invasive species that need to be considered. See e.g, Müllerova et al. 2023.
Müllerová, J., Brundu, G., Große-Stoltenberg, A. et al. Pattern to process, research to practice: remote sensing of plant invasions. Biol Invasions 25, 3651–3676 (2023). https://doi.org/10.1007/s10530-023-03150-z
L50ff: This section includes very basic technical descriptions about remote sensing. Please revise.
Methods:
L238: Rephrase “elevation 100”
L310: Please make it very clear that results were validated only for 2014. There might be several issue, e.g. due to the use of different satellites.
Comments on the Quality of English LanguageQuality of English language is ok.
Author Response
Dear Sir,
We appreciate your thorough review of our manuscript, "Remote Monitoring of Marabou Invasion in the Valle de los Ingenios, Cuba: A Google Earth Engine Approach." Your positive recognition of the global significance of remote sensing in invasive plant species research and the commendation of our use of the Google Earth Engine platform with the random forest algorithm for mapping Dichrostachys cinerea from 2000-2018 is truly motivating.
Now, let us address the specific points and suggestions you raised:
Training and Validation of the Model: We apologize for any confusion caused by the discrepancy in the number of data points. The 136 points mentioned in the confusion matrix include multiple pixels within each of the 29 plots used for ground truth data. This clarification has been incorporated into the revised manuscript (Section 2.2.4).
Parameterization and Application of the RF Algorithm: We have provided additional details on the input variables for the random forest algorithm, including the bands used from Landsat 8 images. While the number of trees was set to 100, we acknowledge that the optimal number may vary depending on the dataset characteristics. The evaluation was performed on a separate dataset, and the overall accuracy of the classifier was 90%. The complete programming code for the random forest algorithm is now included in Appendix A for transparency and reproducibility (Section 2.2.5).
Results: We have revised the results section as per your suggestion, and quotes have been removed except where essential for contextualizing the employed methodology
Discussion: We have incorporated a thorough discussion of potential limitations, addressing concerns about the use of different satellite platforms, the projection of the 2014 model, and the relatively high accuracy using only NDVI. These considerations now enhance the robustness and transparency of our study.
Introduction: We have expanded the discussion on the ecological impacts of invasive species and included recent reviews on remote sensing and invasive species, particularly referencing Müllerová et al. (2023) for a more comprehensive introduction.
Methods/Results: The mention of "elevation 100" has been rephrased for clarity. We have emphasized that the results were validated only for 2014 and addressed potential issues related to different satellites.
We trust that these revisions have addressed your concerns and improved the overall quality of our manuscript. Thank you once again for your valuable feedback.
Sincerely,
Eduardo Moreno
Principio del formulario
Reviewer 3 Report
Comments and Suggestions for AuthorsContributions to improve our knowledge of invasive species are of great importance. The authors of this paper focused on the development of one method to monitor the changes of distribution of marabou in Valle de los Ingenios, Cuba a World Heritage Site of Cultural value as a testimony of sugar cane plantations and colonial slavery. Marabou basically grew over areas dedicated in the past for this crop. In this regard, the aim of the paper is quite valid and significant. However, the paper itself has some aspects that require to be addressed before publication: 1) Introduction is weak, starting with a short paragraph on the global landscape and no references. In this section it would be a good opportunity to provide a view of the global impacts of marabou. 2) Then the next paragraph talks about marabou in Cuba; this long paragraph is based only in one citation. 3) then in moves to the World Heritage Site without a proper explanation of why this region was nominated and a clear description of its location. 4) Then a description of the aims and proposed method. I consider that it would be better to leave a detailed the description of the species, the World Heritage site, and the method presented for the Materials and Methods Section. It would be a good idea to present how the marabou is ranked in current classifications of invasive plants.
In the methods section, a good explanation of why only 29 points were used is necessary. Especially if they are referred to as data points but an area (of different size) is assigned to them. These issues should be clarified since they are the basis for the method in development. Then the classification algorithm and code are presented.
In the results section, I am not convinced that the ii) mapping phase and iii) testing phase were accomplished. Basically, in the paper, the authors do not present color coded maps showing the type of vegetation assigned nor do they present a set of points, distinct to the sampling points, where the classification was verified. If I did not understand, then a better explanation is required so that the reader can see with clarity why this method should be trusted.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsNo comments
Comments on the Quality of English LanguageNo comments
Author Response
Dear Sir,
I would like to thank you very much for your review of our manuscript. Your valuable feedback has significantly contributed to improving the quality of our work.
We thank you again for your time and effort in reviewing our work.
Best regards,
Eduardo Moreno
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have improved the manuscript. However, some points still need to be clarified.
General comments:
I would have preferred a point-by-point reply to my review to have a better idea which points were addressed and how.
Why some text is printed in bold? Are these the changed parts?
The sampling of the reference data and its use for validation needs clarification: Where were the 29 points/plots sampled? How was subsampling in these plots performed to reach 136 points? Are these 136 points single pixel or do they comprise several pixel? See Zhen et al. (2013) for a discussion on sampling schemes, particularly when using a limited amount of polygons, which definitely affects the thematic accuracy of the map even if the calculated overall accuracy is high.
I would also avoid using the terms “points” and “plots” synonymously for the 29 and 236 samples (e.g., L253, L270).
Could you compute variable importance for your RF model, or at least boxplots of the predictors for each class? As reference data is limited, maybe additional diagnostic plots could support the validity of the map, especially regarding the time series parameters. As the time series parameters seem to be important, discuss them in the Discussion section.
Detailed comments
L17: I would relax the statement about the 18-year analysis. It is correct that you analysed remote sensing for that time span. However, the model was trained on data from 2014, and model transfer in time is technically possible, but needs to be carefully evaluated as it could be erroneous.
L22-23: It is not clear why the results highlight the “urgent need”. Further, what is meant by “fragile balance”?
L27ff: Section needs citation. Can be joind with following section.
L31: From a global perspective, I would start the IPBES assessment and then maybe refer to the European commission. However, there is other literature that could be cited for definitions of invasive species.
L33: For costs in Central and South America, see Heringer et al. 2021.
L41: This section seems to be out of context as no invasion hypothesis are tested. However, L44-48 provide interesting information and could be joined with the section in L65. The importance of island ecosystems should be emphasized.
L97-122: This is basic knowledge about remote sensing and can be removed. Better focus on the applications of RS in the context of biological invasions.
L141-151 sounds like a section from the Methods. Rephrase clearly or move.
L152-159 sounds like an outlook at the end of the manuscript. Rephrase clearly or move.
L274: Are the coordinates really needed in Table 1?
L276-278: Please clarify. So every pixel/point/plot that contains marabou is counted as marabou? It is difficult to evaluate of that makes sense without further insights about the data extraction.
L287: What do you mean by phase and magnitude band?
L287: There is no NDVI band. Rephrase.
L293ff: Reporting overall accuracy is a result. You even refer to it as “results”. Please move.
LL321 -340: It does not become clear from the Introduction or Methods section why this delineation is required or what the overall purpose is. Further, the source of the DEM is missing.
L364 (Figure 5): Is this the curve for marabou? Then this needs to be stated in the caption. If so, could you add the curves of other vegetation classes?
L374, L379: Please avoid interpretations in this part of the manuscript. This was already mentioned in the first round.
L384 (and in the following): Limit the number of decimal places. Four decimal places do not sound realistic.
L411: Again, this is interpretation of the results and belongs to the discussion. Please check the whole results section for these kind of phrases.
L449: This section deserves a critical view of the methods. First, a random forest trained on data of 2014 and simply projected to any other year could be erroneous. Second, the limited amount of reference data will have an effect on the accuracy. This needs to be considered. Merge this section with L488ff.
L476: Where and how do you assess the efficacy of different control measures? Is there any spatial data or any other spatial information where which measure has been conducted and for how long?
L498: Can you show the specto-temporal profiles of the other vegetation classes in comparison to marabou?
Literature:
Heringer, G., Angulo, E., Ballesteros-Mejia, L., Capinha, C., Courchamp, F., Diagne, C., ... & Zenni, R. D. (2021). The economic costs of biological invasions in Central and South America: a first regional assessment. NeoBiota, 67, 401-426.
Zhen, Z., Quackenbush, L. J., Stehman, S. V., & Zhang, L. (2013). Impact of training and validation sample selection on classification accuracy and accuracy assessment when using reference polygons in object-based classification. International Journal of Remote Sensing, 34(19), 6914-6930.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe paper shows improvement. I still consider that a better explanation of the success classification rates must be presented. If NDVI predicts Marabou distribution positively in most of the cases; then a statement of such high classification rate must be expressly said because it will increase the punch of the article. In all figures there should be a clear text indicating which color represents Marabou.
Author Response
Thank you for your positive feedback.
Regarding your suggestion about providing a clearer explanation of the classification success rates and including explicit labels in our figures, we have made changes to the figure captions to indicate clearly the color representing marabou.
As for your comment on the need for a clearer explanation of the classification success rates, as well as including explicit labels in our figures, we believe that the current presentation effectively conveyed the main results of our study. Our focus on detecting Marabou was evident throughout the manuscript, and while we acknowledge the importance of highlighting the success rates, we also aimed to maintain the flow and structure of the text as it was.
We have carefully considered your suggestions and believe that the manuscript adequately communicated the significance of our results without the need for additional changes to the text. However, we ensured that the figures were visually clear and included any necessary labeling to facilitate interpretation