Evaluating the Transferability of Spectral Variables and Prediction Models for Mapping Forest Aboveground Biomass Using Transfer Learning Methods
, Zhaohua Liu 1,2,3, Peisong Yang 1,2,3 and Tingchen Zhang 1,2,3Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsBy integrating remote sensing images with a small number of ground measured samples to map forest AGBs can significantly reduce time and labor costs. To evaluate the potential of transfer learning methods in mapping forest AGBs, spatial-temporal transfer of spectral variables (SVs) and prediction models (PMs) was evaluated in the manuscript based on Landsat imageries. The results are informative, however, the manuscript needs to be put more efforts for further improvement.
1. Key words: “prediction models” should be in Keywords list.
2. The transfer learning has been successfully applied in many fields broadly, it is suggested to cited more references in “background” about forest parameters mapping.
3. It is very important to evaluate the transferability of Transfer learning. How has this problem been dealt with in previous studies? and what is new in your study?
4. The tree species should be listed in standard names coming with their Latin names in Study Area and Data”, Please check and improve. In addition, please keep consistency in tree species names in the manuscript. For example, “Fir” or “Chinese Fir”?
5. There is a mistake about the name of tree species in Figure 1, “Quercus” should be “Oak”, please correct it.
6. Please check the abbreviations usage in the manuscript.
7. In “Ground Data”, it is necessary to describe the methods of measuring forest parameters and elaborate the process of above-ground biomass calculation in the sample plots.
8. Equation 4 is incorrectly written, RRCRS should be RCRS according to the context, please check and correct it.
9. In the study, the RCRS was proposed to assess the transportability of spectral variables and predictive models.
10. In previous studies, the incorporation of texture features has been proposed as a means to enhance the accuracy of mapping forest aboveground biomass (AGB), but it is noted that the texture features are not considered in the study, are there any specific differences?
11. Could you propose the possible methods to improve the transferability of variables and models with low transferability?
12. Among plots at high AGB level, it was observed that the AGB value were overestimated in some samples with relative lower AGB distribution. This discrepancy represents a systematic error, and it is suggested to further discuss this phenomenon in “Discussion”.
Comments on the Quality of English Language
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe background and methods for transfer learning are not sufficient for a reader unfamiliar with the approach, such as myself, to provide a good review
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis study presents a method to improve the accuracy of forest AGB mapping. Overall, the authors provide a good overview of the estimation of forest AGB at different temporal and spatial scales using machine learning methods and remotely sensed spectral variables. However, the current discussion and content still has some serious shortcomings. Firstly, the content of this study doesn't align with transfer learning. Typically, the essence of transfer learning is to apply existing data or models to other unknown areas, but the discussions pertaining to both the temporal and spatial dimensions level retraining the model with their respective training samples, which doesn't conform to the basic principles of transfer learning. Studying the application of the same model in different regions is more in line with the theme of discussing the model's generalization capabilities. Therefore, this paper should reconsider the definition of its content and change its focus to explore the model's generalization capabilities, should be on discussing the generalization capabilities of the same model for forest AGB at two different levels. In addition, a reorganization of the method and discussion sections to ensure a smoother logical is necessary.
1, the format of the article is kept uniform
2, the abbreviated form of proper nouns only needs to be stated once at the beginning of the article
3, Describe in more detail the content of chapter 2.5, specifically 2.5.3, which needs to elaborate on each step of the process involved in the methodology used. In addition the elements in figure 2 and figure 3 need to be described in detail.
4, How was the test dataset set up?
5, All conclusions in the discussion need to be rigorously justified, or citations added.
Comments on the Quality of English LanguageThe English expression needs further polishing.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsComment for remotesensing-2662538
This manuscript addresses evaluating the transferability of spectral variables and prediction models for mapping forest aboveground biomass using transfer learning methods. The results could provide information for forest management. I review this manuscript and provide comment as follows.
1. Title and Abstract are suitable that can reflect whole text. But I have a slight suggestion for Abstract. If possible, providing some quantitative results here.
2. Introduction provide adequate background information and significance of this study. I suggest that authors might use point by point in the end of Introduction.
3. Figures 2 and 3 should have more descriptions in text.
4. Result is abundant. However, it is not so easy to read. If follow study purposes, it might be improved.
5. Figures 4, 9 and 12, I suggest deleting all number above bars from these Figures because it let bar chart look too busy. If it is necessary, I suggest that authors might state it in text.
6. From Results I still could fully understand what is the main contribution of this study? I suggest authors to emphasis key contribution in Discussion. Moreover, the limitation of this research also should be mentioned.
7. In Conclusion, authors pointed out “In future work, there is a need to explore interpretable remote sensing variables and develop ecosystem models that incorporate multiple variables to explain the spatial distribution of forest AGB.”. What is incorporate multiple variables? It should be discussed in Discussion chapter!
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsLooks good
