Remote Sensing Applications for Mapping Large Wildfires Based on Machine Learning and Time Series in Northwestern Portugal
, Soltan Galano Duverger 2, António Bento-Gonçalves 1,*, Washington Franca-Rocha 3, António Vieira 1 and Georgia Teixeira 4
Round 1
Reviewer 1 Report
1. Literature research and summary are detailed and comprehensive.
2. It is recommended to clearly state the difference (or improvement) between this method and the existing literature methods in the introduction. It is hoped that the contribution of the method in this paper will be clearly stated in the introduction.
3. Hope to improve the clarity of Figure 1-12 to ensure that the smallest letters can be recognized.
4. Figure 02 in line 130 is suggested to be changed to Figure 2.
5. Hope to supplement the explanation of the content of Figure 2.
6. It is suggested to change Ttable 03 to Table 3, which is consistent with the citation in the paper.
7. Line 251 suggests changing the expression, for example "developed by Breiman et al. [44]".
8. Hope to carefully check the formula number.
9. Figure 8 suggests supplementary scale and coordinate system.
10. It is recommended to add Section Discussion to discuss the experimental results and further explain the advantages and disadvantages of this method.
11. It is recommended to add a description of the transferability of the method to other areas in Conclusion.
Author Response
Dear Editor and reviewers,
Thank you very much for the time you took to review our article. The revisions were helpful to improve the article.
Bellow, we send the answers to your comments. In the text, the changes are highlighted in yellow.
Sincerely
The authors
- Literature research and summary are detailed and comprehensive.
Thank you very much for your positive view.
- It is recommended to clearly state the difference (or improvement) between this method and the existing literature methods in the introduction. It is hoped that the contribution of the method in this paper will be clearly stated in the introduction.
DONE - Added in the last paragraph of the introduction.
- Hope to improve the clarity of Figure 1-12 to ensure that the smallest letters can be recognized.
DONE - Figures 2, 3 and 8 were redone.
- Figure 02 in line 130 is suggested to be changed to Figure 2.
DONE
- Hope to supplement the explanation of the content of Figure 2.
DONE
- It is suggested to change Ttable 03 to Table 3, which is consistent with the citation in the paper.
DONE
- Line 251 suggests changing the expression, for example "developed by Breiman et al. [44]".
DONE
- Hope to carefully check the formula number.
DONE
- Figure 8 suggests supplementary scale and coordinate system.
DONE (We did not add a coordinate system because of the size of the images and can be seen in figure 1)
- It is recommended to add Section Discussion to discuss the experimental results and further explain the advantages and disadvantages of this method.
DONE - see point 4. Discussion
- It is recommended to add a description of the transferability of the method to other areas in Conclusion.
DONE - Added in the last paragraph of the conclusion
Reviewer 2 Report
The paper presents the contents in an efficient and organized manner. The paper begins with a brief introduction that gives a general overview of the topic, and then proceeds to a more detailed description of the mapping of the burnt areas in Portugal.
However, the bibliography concerning the Google Earth Engine platform should be increased, as in the last two years there has been a great increment of studies on the use of classification tools, especially the Random Forest by GEE and time-series analysis for mapping burned areas.
Moreover, it is necessary to specify the algorithms used in GEE for cloud filtering, as the use of the Landsat's optical sensor is affected a significant degree by the coverage of clouds, in fact this is a consequence of the application of the harmonic series. In the graph shown in the figures 5 and 6, there are some jumps/gaps between the different points, this means there is a NO-DATA, most probably due to CFMask discarding some images from the time-series in the code script.
Please review the text format, the hyphenation and the inter-word spaces.
The pictures are not high quality, it is recommended to insert them with higher resolution.
In conclusion, the paper comes out well written, with a good syntax and appropriate style and offers a repeatable and applicable method by GEE, thereby make it a valuable methodological reference in the literature.
Author Response
Dear Editor and reviewers,
Thank you very much for the time you took to review our article. The revisions were helpful to improve the article.
Bellow, we send the answers to your comments. In the text, the changes are highlighted in yellow.
Sincerely
The authors
The paper presents the contents in an efficient and organized manner. The paper begins with a brief introduction that gives a general overview of the topic, and then proceeds to a more detailed description of the mapping of the burnt areas in Portugal.
Thank you very much for your positive view.
However, the bibliography concerning the Google Earth Engine platform should be increased, as in the last two years there has been a great increment of studies on the use of classification tools, especially the Random Forest by GEE and time-series analysis for mapping burned areas.
DONE - 15 papers were added on Random Forest and GEE
Moreover, it is necessary to specify the algorithms used in GEE for cloud filtering, as the use of the Landsat's optical sensor is affected a significant degree by the coverage of clouds, in fact this is a consequence of the application of the harmonic series. In the graph shown in the figures 5 and 6, there are some jumps/gaps between the different points, this means there is a NO-DATA, most probably due to CFMask discarding some images from the time-series in the code script.
DONE – see lines 156-169
Please review the text format, the hyphenation and the inter-word spaces.
DONE
The pictures are not high quality, it is recommended to insert them with higher resolution.
DONE - Figures 2, 3 and 8 have been redone
In conclusion, the paper comes out well written, with a good syntax and appropriate style and offers a repeatable and applicable method by GEE, thereby make it a valuable methodological reference in the literature.
Thank you very much for your positive view.
Reviewer 3 Report
This paper examined a burned area segmentation using Landsat images and the Portugal ICNF dataset, which is a reasonable approach but needs modifications regarding the method explanation and the performance indices. Major concerns are as follows.
*Table 6. The authors have 687 polygons as ground truth. Please show some example maps comparing the polygon label image and the model prediction result.
*Table 7. For image segmentation, precision, recall, and mIOU are the mandatory indices for performance evaluation. Please present these indices additionally. The authors can calculate them using the contingency table in Table 7.
*Section 2.2.3.2. RF is a supervised classification. Was the ICNF dataset used as a true label? If so, how did you divide the training, validation, and test sets? It needs to be clearly explained. Also, were the performance statistics calculated from the test sets?
Author Response
Dear Editor and reviewers,
Thank you very much for the time you took to review our article. The revisions were helpful to improve the article.
Bellow, we send the answers to your comments. In the text, the changes are highlighted in yellow.
Sincerely
The authors
*Table 6. The authors have 687 polygons as ground truth. Please show some example maps comparing the polygon label image and the model prediction result.
We include figure 11
*Table 7. For image segmentation, precision, recall, and mIOU are the mandatory indices for performance evaluation. Please present these indices additionally. The authors can calculate them using the contingency table in Table 7.
We include figure 14
*Section 2.2.3.2. RF is a supervised classification. Was the ICNF dataset used as a true label? If so, how did you divide the training, validation, and test sets? It needs to be clearly explained. Also, were the performance statistics calculated from the test sets?
DONE - Added in the last paragraph of the Section 2.2.3.2.
Reviewer 4 Report
The study aim of this article is to develop a methodology for mapping LW in northwestern Portugal using a Machine Learning algorithm and time series from Landsat images. For the burnt area classification, the authors initially used the Fourier harmonic model to define outliers in the time series that represented pixels of possible burnt areas and then, applied the Random Forest classifier for the LW classification. The results indicate that the harmonic analysis provided estimates with the actual observed values of the NBR index, thus the pixels classified by Random Forest were only those that were masked, collaborating in the processing, and reducing possible spectral confusion between targets with similar behaviour.
some suggetstions:
1. rewite the methods and data. There is some confusion in this part.
2. All figures not clearly, please redraw them.
3. part 2.2 Burned Area Classification Approach, this part should say clearly and give detile information.
Author Response
Dear Editor and reviewers,
Thank you very much for the time you took to review our article. The revisions were helpful to improve the article.
Bellow, we send the answers to your comments. In the text, the changes are highlighted in yellow.
Sincerely
The authors
- rewite the methods and data. There is some confusion in this part.
DONE - As the methodology is described throughout the article, we have introduced corrections throughout the different sections.
- All figures not clearly, please redraw them.
DONE - Figures 2, 3 and 8 have been redone
- part 2.2 Burned Area Classification Approach, this part should say clearly and give detile information.
DONE - We improved the flowchart and added explanations in the subtopics
Round 2
Reviewer 1 Report
Congratulations to the author, now that the article has been revised and improved, I have no other comments.
