Discrimination of Larch Needle Pest Severity Based on Sentinel-2 Super-Resolution and Spectral Derivatives—A Case Study of Erannis jacobsoni Djak

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The imagery picture resolution needs to be significantly improved. It is difficult to see the different level of Severity. 

Did the authors try cross testing this Spectral derivative to some new ground data to check for the validity and significance of the method? How can you justify its significance give the limitations you mentioned? 

Figure 4. Graph resolution need to be improved

Line 240. Rewrite the sentence and revise manuscript for sentence structure errors

Author Response

Please check the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Manuscript "Discrimination of Larch Needle Pest Severity Based on Sentinel-2 Super-resolution and Spectral Derivatives-An Example of Erannis jacobsoni Djak" presents a promising approach to pest monitoring, integrating the latest remote sensing technologies with machine learning and statistical methods. The research is highly relevant due to the growing demand for effective methods of forest pest management. However, there are a few aspects that need refinement and further clarification, especially regarding the methodology, statistical analysis, and results validation.

Here are some line-by-line comments and suggestions to improve the scientific quality and clarity of this manuscript:

1.      Line 40–60: The introduction provides a good overall view of the problem but then lacks substantial discussion on previous studies' application of both remote sensing and machine learning towards forest pest monitoring in the case of Erannis jacobsoni Djak. Discussion of prior recent studies using similar methods or similar pests should be extended towards a comprehensive background.

2.      Line 138–140: "We believe that Sentinel-2 data have the potential to be used in refined monitoring."  This is a very strong statement, for which solid evidence is required. Provide references to recent works dealing with advantages of Sentinel-2 data against other remote sensing platforms in support of pest monitoring..

3.      Line 198–201: The methodology for defining the severity classes is well described, though the thresholds used for the classification are not statistically validated. It is recommended that statistical validation of the severity classes be done, such as inter-observer reliability or error rates from the field survey. A justification for choosing those thresholds would strengthen the methodology.

4.      Line 229–236: "The Sentinel-2B L1C data is a top-of-atmosphere reflectance product..." Use of L1C data is appropriate, but it does not mention the application of atmospheric corrections, apart from basic calibration steps. Please state whether additional pre-processing other than that described has been carried out prior to the analysis, to homogenize the data set..

5.      Equations 2–4 (Lines 276–279): The spectral derivatives are calculated, but it is not clear whether the band spacing is uniform. This is important because non-uniform band spacing can affect derivative calculations. Clarify if the band intervals Δλ are uniform across Sentinel-2 bands or indicate if adjustment was made.

6.      Line 283–299: "ANOVA and Tukey's HSD test... followed by RF-RFECV algorithm..."  The statistical tests used for feature selection are appropriate but the assumptions underlying ANOVA and Tukey's HSD-that is, normality and homoscedasticity-are not discussed. Address the assumptions of these tests. If assumptions are violated, consider applying appropriate data transformations or non-parametric alternatives.

7.      Line 316–317: The data is split into a training set and a test set in a 70:30 ratio, but it is not mentioned whether this is done randomly or by stratification. Stratification would ensure that all severity classes are well-represented in both the training and test sets. Specify whether the split is stratified or provide information on how the split was performed.

8.      Equations 5–8 (Lines 327–334): The applied accuracy metrics are standard; however, no confidence intervals were given regarding these metrics, which is a must for assessing the reliability of the results. Provide confidence intervals of accuracy and Kappa values, preferably by bootstrapping or cross-validation.

9.      Lines 338–359 (Super-resolution Metrics): In this section, the super-resolution processing is evaluated by Q, ERGAS, and SAM. However, the paper does not discuss what thresholds for these metrics are considered acceptable. Include thresholds for Q, ERGAS, and SAM to describe the quality of the super-resolution enhancement. Also, compare these values to those reported in similar studies.

10.  Figure 4 (Lines 401–403): The classification accuracy increases with the number of features. However, it is not immediately obvious why exactly seven features were chosen. It would be nice to see how performance varies for less or more features. Motivate why seven features were chosen, and consider doing a sensitivity analysis showing how performance degrades when using fewer features.

11.  Figures 5–7 (Lines 417–449): While the confusion matrices are eloquent, it would be more informative to get more complete evaluations with other metrics such as F1-scores, Precision, Recall and MCC, especially in the case of imbalanced datasets. Add metrics in order to assess the models by applying additional metrics such as F1-score and MCC.

12.  Line 491–516 (Super-resolution Evaluation): This gives insights in a super-resolution enhancement, while not taking into account such a processing possibly introduced artefacts. Discuss how super-resolution can introduce noise or other artifacts in the results, while providing ways of quantifying distortions this may introduce.

13.  Line 531–575 (Feature Sensitivity Analysis): The role of D8a and NDVIswir is discussed but maybe more in relation to their ecological relevance. Elaborate how these features are related to the EJD infestation, for example, by referring to studies that linked these spectral bands with pest stress or defoliation.

14.  Line 603–607: Interesting suggestion to use different amounts of Sentinel-2 data for different conditions; however, it needs more specific examples or scenarios. Give real examples of how multi-temporal or multi-source data could enhance the monitoring of pests.

15.  Line 645–646: This is a good conclusion-that the super-resolution processing increases the monitoring of pest severity-but this is not supported through specific data or metrics of comparison. Strengthen this conclusion by referring to specific improvements in accuracy or in the discrimination of pest severity after super-resolution.

16. Line 649–650: Here, the combination of spectral derivatives and vegetation indices is mentioned to improve model accuracy, but quantitative results are not shown. Quantify the gain in model accuracy when these features are combined to give clear evidence of added value.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have significantly improved the manuscript by addressing most of the comments. Below, some line-by-line comments that further improve the manuscript are provided.

Line-by-Line Comments:

1. Lines 55–62: Extended introduction is good but will be further improved if some discussions on Sentinel-2 applications other than NDVI and pest monitoring are given.

2. Line 206: Justification of severity classes may be further improved referring to recent studies or validation of thresholds with more field data.

3. Line 242: While mentioning atmospheric corrections here, a short comparison of different correction techniques is warranted, e.g., Sen2Cor vs. others.

4. Lines 293–306: It was good to substitute ANOVA with non-parametric tests. Shortly discuss here how this impacts the interpretability of the results.

5. Lines 490–493: Good, confidence intervals are included. To further enhance the visual robustness, include bootstrapped distributions of these intervals.

6. Lines 531–575: Features are well explained in terms of ecological relevance. The role of environmental factors, however, could be elaborated more, as in soil type or microclimate.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf