A Highland Barley Crop Extraction Method Based on Optimized Feature Combination of Multiple Phenological Sentinel-2 Images
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors of this paper aim to map the spatial distribution of highland barley using a Random Forest model with several Sentinel-2 images taken at multiple phenological stages. In general, the manuscript is quite well-written, However, there are major methodological concerns with the methodology followed in this work as currently presented.
Comments:
1) Could you include the locations of the sample points in Figure 1 to visualize the spatial distribution of both the validation and training data?
2) Table 1 has not been used in the text before its appearance.
3) You don't need to put both Sentinel-2A and Sentinel-2B surface relectance details together because their differences are minimal and only significant if extremely important but tis is not the case of your study.
4) you should provide Texture feature equations in Table 3.
5) I know what does GLCM stand for but most of readers no. Please add the description for this abbreviation.
6) You should include an additional column in your table that provides the reference for each vegetation index you use.
7) I miss the information about cloud cover, how did you deal with them.
8) Could you explain why the Gini index is chosen over other methods, particularly the Mean Decrease in Accuracy (MDA) who is common to gives a more detailed and robust evaluation of feature importance compared to gini.
9) You did a good job explaining the random forest model, but I noticed there was no discussion on how to tune the model, specifically in terms of selecting the optimal number of trees. This aspect is crucial because it can greatly influence the model's performance, affecting its accuracy, generalizability, and computational efficiency.
10) I'm surprised that the section on accuracy assessment is entirely missing from the methodology .
11) To randomly divide your data into training and validation sets, you should ensure that the split is done in a way that maintains the representativeness of both subsets. How did you make sure to adress this issue and to limit overfitting.
12) The Salience score has not been described in the text in any place ?
13) You have already specified in Figure 4 that the jointing stage was identified as the most significant feature, but this is absent in the optimal combination of features in Table 4. This is confusing. How did you select the optimal combination of features? If you didn't use the importance index, why did you include it in the previous figure?
14) Fig. 5 needs to be zoomed out.
15) Since your objective is to map highland barley, why didn't you consider including other auxiliary data such as slope, elevation, or aspect?
Comments on the Quality of English LanguageModerate 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 study in question addresses the use of Sentinel-2 satellite images for identifying areas cultivated with barley. Overall, the text is well-structured; however, for publication, I suggest a more thorough revision.
In the introduction, the objective of the study is not clearly defined and needs to be improved to provide a clearer understanding of the study's goals and hypotheses. The literature review should be expanded to include a more comprehensive analysis of relevant previous studies. It is important to incorporate references that discuss the use of satellite images in identifying agricultural crops and to detail the methods used in these studies. Additionally, a discussion of the specific uses of Sentinel-2, including its capabilities and applications in agriculture, should be included to better contextualize the current research.
It is not clear in the text whether the training and validation classification was performed using UAV images with prior visual classification.
A critical aspect of using Random Forest (RF) models is the risk of overfitting and underfitting. How did the authors address these issues in their study? Additionally, using the same data for both training and classification may compromise the validity of the results and the model's ability to generalize. It is important to assess whether the training samples were properly separated from the samples used for classification, or if validation data were inadvertently included in the training process.
Including an image showing the location of the points used for training and validation could provide valuable insights into the data distribution and help evaluate whether the model was appropriately tested in areas not previously known. Furthermore, it is essential to analyze the model's performance in relation to the areas it was able to classify and determine if these results are consistent with the expected outcomes.
A more detailed analysis of the model's accuracy and precision is needed. In particular, an assessment of false positives and false negatives should be included. Analyzing false positives is crucial for understanding the impact of areas incorrectly classified as positive. It is important to investigate how these errors affect the model's effectiveness in identifying and segmenting the desired targets. Additionally, consider how the presence of false positives may influence the interpretation of the results and decision-making based on the classifications generated by the model.
Comments on the Quality of English Languagecan improve
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors explored the “extraction method based on optimized feature combination of multiple phenological Sentinel-2 image”. Though the work is largely well-written, the article still needs significant improvement in the introduction and methodology and conclusion sections. Considering my observations as follows, I suggest major revisions before considering it for publication.
· Please provide more concise and clear information regarding the background, objective methodology, and conclusion with policy recommendations in the abstract.
· In the introduction, the authors reviewed the past literature but failed to analyze the limitations of past research and clarify the innovation of their own research.
· The aim and objectives of the study are missing in the introduction section.
· Why was PCA used in this study? Is PCA used for feature selection? I am not clear about it. Please clear it in the methodology section.
· In the case of the RF model, what amount of data was used for training, validation, and testing of the model? I mean, how was the model built? How was it calibrated? The most important parameters and the choice of values for the model were not explained. Moreover, what were the response and predictor variables?
· Conclusion can be improved by highlighting the innovation content of the paper, future research direction, and recommendation for policy formulation.
Comments on the Quality of English LanguageModerate editing of English language required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsI appreciate the authors' hard work and thoughtful responses to my comments. But in some cases, I am not convinced, therefore I suggest that the paper must undergo major revision before considering it for publication
Comments
· The authors did not correct my previous comments (below), as it remain the same as before.
“Please provide more concise and clear information regarding the background, objective methodology, and conclusion with policy recommendations in the abstract. Rewrite the abstract”.
Abstract started with the first line “Previous studies have primarily focused on the extraction of highland barley crops using 12 single phenological images” not looks sound good. Please rewrite the abstract again.
· Page 18 (line-186): In this study, PCA was used to extract spectral features from the set of feature variables. Please show the PCA result in the manuscript.
· The authors answered my previous comments (below) in the author response sheet but I did not find it in the manuscript. Please correct it in the manuscript.
“In the case of the RF model, what amount of data was used for training, validation, and testing of the model? I mean, how was the model built? How was it calibrated? The most important parameters and the choice of values for the model were not explained. Moreover, what were the response and predictor variables?” 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
