Fire Severity Outperforms Remote Sensing Indices in Exploring Post-Fire Vegetation Recovery Dynamics in Complex Plateau Mountainous Regions
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe title lacks specificity and impact. It does not highlight the innovative aspects. It is relevant to specify the geographical scope and the main contribution.
The abstract lacks detail on the context of the study, including why post-fire recovery is important in this particular region of Yunnan. In addition, the methodological description remains vague and generic. Specifying the comparative basis of the algorithms tested (CART, SVM, NB) and why the random forest was chosen would be useful. It would also be useful to explain how the indices (NDVI, LSWI, NBR) were used to assess vegetation recovery. In addition, the results presented in the summary lack depth and figures. Finally, the summary should explain how local managers can use the results or to guide restoration policies.
The keywords are poorly aligned with the specific content of the study.
The introduction lacks structure and clarity. It should begin with the importance of forests in climate regulation and biodiversity, followed by global threats (deforestation, climate change) and local threats (forest fires), supported by recent references. Existing solutions, such as remote sensing indices (e.g. NDVI, LSWI), should be better discussed, particularly their limitations in long-term analyses and complex mountainous contexts. The shortcomings of previous studies, such as the lack of data on long-term post-fire dynamics and the comparative performance of classification algorithms, should be identified. Finally, the objectives of the study must be aligned with these shortcomings, accompanied by precise hypotheses, and the methodology justified to enhance overall relevance and consistency.
Materials and methods
The description of the study area lacks information on local vegetation and activities, and the map (Figure 1) does not indicate the geographical position of the study area.
The methodology is unusually long and less concise: to gain relevance, the section on remote sensing data could include additional references. The process of merging Landsat 5 TM, 7 ETM+, and 8 OLI data is based on a coefficient conversion method proposed by Roy et al. but the technical details, such as Table 1, seem surplus to requirements and could be summarised. In addition, the authors use the Normalized Burn Ratio (NBR) to detect burnt areas, and the delta NBR (dNBR) to assess fire intensity. Although these indices are commonly used, their application could benefit from justification specific to the local characteristics of the area studied. Finally, the sections on the evaluation of machine learning (ML) methods for fire intensity classification delve into exhaustive technical details regarding the parameters of the RF, CART, and SVM algorithms, making the section unusually long and able to be summarised by focusing on aspects directly related to the objective of the study.
Results
The results provide a comprehensive description of variations in fire intensity and vegetation recovery but lack in-depth interpretation and concise contextualization. The authors need to focus on key trends, anomalies, and practical implications.
Discussion
The section lacks detail on methodological limitations and implications for sustainable forest fire management. The analysis of possible biases and associated errors is not sufficiently thorough. In addition, although the performance of the ML algorithms is discussed, the environmental and social impacts are not considered.
Conclusion
The conclusion presents an overview of the results obtained but lacks perspectives for future studies.
Author Response
Please see the attachment.Thank you.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors
This manuscript evaluates Machine Learning algorithms such as CART, SVM, NB, and RF for Post-fire vegetation recovery dynamics in central Yunnan from 2005 to 2020 based on dNBR and random forest.
Although, I believe the article addresses an important research topic, it has a number of shortcomings. The list of identified shortcomings is as follows:
1. In the abstract, the abbreviation dNBR, like other abbreviations, should be expanded. The abstract of the manuscript should be revised as the new contributions and the explanation are not clear. Also, the abstract part should be shortened and the main results should be presented.
2. It will be more understandable for readers to mention what exactly is meant by visual interpretation, which parameter is visually interpreted. This should be taken into account.
3. In lines 57-65 of the Introduction, it should be provided information about the use of NBR and dNBR, as well as NDVI and LSWI, but it is not explained why the dNBR method was chosen.
4. In lines 92-93, the bracket is not closed.
5. Formulas must be written and numbered as required by MDPI.
6. The literature is not well reviewed in the manuscript, for example, very few researches on long-term Post fire vegetation recovery dynamics are mentioned (https://www.sciencedirect.com/science/article/pii/S0378112724004389, etc.), but they are not listed, although they exist and work similar to this article has been done in them. There are also works comparing ML methods (https://www.mdpi.com/2072-4292/13/4/792 etc.) and they are not mentioned in order to increase the importance of the article. Special attention should be paid to this issue to increase the level of the article.
7. References 6, 29, 31, 32, 36, 57 not found, need to be revised.
8. The process of citing literature in the text should be done as required, for example, 57 cite as superscripted or 58 cite is added to the word. The process of citing should also be reviewed.
9. In Fig.2, 6 annual fusion image are directly selected from the Landsat 5 TM… data and transferred to the Optimal pretrained classifier model. If these images are not pre-processed, dNBR and visual interpretation processes are not performed, the image cannot be directly transferred to the ML model. This should be reconsidered.
10. In Fig.2, the decision is being made based on the combined data from the last two blocks. However, this is not reflected in the article. The decision is made based on data from only one side. What does this actually mean? The author(s) should clarify this misunderstanding.
11. Figure 2 shows that Post fire vegetation recovery dynamics is estimated based on Change of forest fire intensity and Changes in LSWI, NDVI, NBR, but while reading the article it becomes clear that only Changes in LSWI, NDVI, NBR were used to estimate Post fire vegetation recovery dynamics (Section 3.3). How should this be understood? The authors should explain this situation or reconsider this process.
12. Table 2 is not fully visible, as well as the writings are confused.
13. It is desirable that the size of the letters in the drawings correspond to the size of the letters in the text of the article. In some places it is small and in some places it is large. Small letters can be understood, large text needs to be revised, for example, fig.7 (years, areas) or fig.8 (LSWI, MDVI, NBR)
14. Ground truth data should also be provided in Fig.6 and Tab.2 to assess the performance of the methods.
15. Why is dNBR and visual interpretation mentioned in the Abstract and NBR and visual interpretation mentioned in the Conclusion? It should be clearly stated which method was used. Is it NBR or dNBR in line 158?
Author Response
Please see the attachment.Thank you
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe article should be clear and easily understood
Comments for author File: Comments.pdf
Author Response
Please see the attachment.Thank you.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsAll the comments made have been taken into account. I would like to thank the authors and believe that the manuscript is publishable at this stage.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe current version of the manuscript has been revised and meets the requirements for publication.
Reviewer 3 Report
Comments and Suggestions for AuthorsThank you very for the revisions
