Identifying Winter Light Stress in Conifers Using Proximal Hyperspectral Imaging and Machine Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Major revision is required.

The experimental design contains methodological inconsistencies that compromise result interpretation. The temporal sampling scheme shows irregular intervals (7-10 days) without proper justification for this variability, and more critically, the transition from controlled laboratory conditions to field crown measurements introduces uncontrolled variables that are inadequately addressed. The authors acknowledge that crown-level classification failed to distinguish between "cold stress" and "optimal condition" states, essentially reducing a three-state classification problem to a binary one. This represents a fundamental limitation given that the practical application of such technology would require robust performance across different measurement scales and environmental conditions. The lack of standardized protocols for crown pixel marking, as admitted by the authors, further undermines the validity of field validation attempts.

The statistical approach requires substantial strengthening in multiple areas. First, the vegetation index selection appears somewhat arbitrary despite the literature review - no systematic feature selection or correlation analysis is presented to justify why these specific 28 VIs were chosen over others available in the literature. The multicollinearity analysis shown in Figure 10 suggests several VIs provide redundant information, yet no dimensionality reduction techniques were applied. Second, the machine learning validation lacks proper cross-validation strategies. While 10-fold cross-validation is mentioned for LDA, the random forest models use only out-of-bag error estimation, which can be overly optimistic, particularly with the relatively small dataset size. The temporal nature of the data also raises concerns about data leakage - training and testing on temporally adjacent samples may inflate performance metrics. Third, the threshold selection for ROI identification (Carter5>1.4) lacks proper justification or sensitivity analysis, and this preprocessing step could significantly impact downstream classification performance.

Author Response

Dear Reviewer!

Thank you so much for taking the time to review our manuscript! We appreciate your valuable feedback and constructive suggestions on our work. Your comments have made it much better. We retained revision marks on the revised manuscript in the ‘stresses-3899267 — Revised (tracked changes).docx’ file. The authors hope, the current version of the manuscript will meet your expectations regarding clarity and quality of presentation.

 

• The experimental design contains methodological inconsistencies that compromise result interpretation.

Response: Thank you for your comment! We acknowledge that the experimental design is not ideal. Initially, we conducted a multi-year study to study the spectral phenology of conifers in detail, based on detailed series of hyperspectral images of shoots, supported by an assessment of photosynthetic pigment content. Preliminary analysis of the research results (annual dynamics of carotenoid-sensitive vegetation indices, carotenoid content and chl_a/car values, specific response to winter and summer conditions of Platycladus orientalis) showed that if we additionally assess F_v/ F_m and expand the hyperspectral imaging objects (add shaded shoots to sunlit shoots), then an extremely interesting task can be solved in parallel – the identification of winter light stress using hyperspectral phenotyping. This is one of the first studies on the diagnosis of winter light stress in conifers using hyperspectral imaging, so we believe that the preliminary results obtained will be of interest to researchers in the field of spectral phenotyping of plants. In 2025, we started a new experiment with a more ‘strict’ design. Your comments will also be taken into account.

 

• The temporal sampling scheme shows irregular intervals (7-10 days) without proper justification for this variability, and more critically, the transition from controlled laboratory conditions to field crown measurements introduces uncontrolled variables that are inadequately addressed.

Response: Thank you for your comment!  We did not use DOY (day of year) as independent variables in the models. However, with constant time intervals, the study design would have appeared more ‘'strict’. The uneven step in the hyperspectral imaging time series is primarily due to technical reasons. Since the main objective of the study is to establish the fundamental possibility of diagnosing winter light stress using proximal hyperspectral imaging, we assume that the factor of interval variation is not so fundamental. Constant and narrower time intervals will be important in determining the exact timing of the onset and end of winter light stress.

Indeed, the transition from laboratory hyperspectral imaging to field hyperspectral imaging of crowns introduces uncontrolled variables that the laboratory random forest model does not take into account. However, this provides a good opportunity to test the fundamental suitability of the chosen approach for determining winter light stress. We understand that the model for determining winter light stress on crowns must be trained on field hyperspectral imaging data, taking into account laboratory results. This is a task for future research.

 

• The authors acknowledge that crown-level classification failed to distinguish between "cold stress" and "optimal condition" states, essentially reducing a three-state classification problem to a binary one. This represents a fundamental limitation given that the practical application of such technology would require robust performance across different measurement scales and environmental conditions.

Response: Thank you for your comment!  We used the classification of crowns using a model obtained from laboratory data to test it. Of the three states, we were able to classify two. In this classification configuration, this is a good result. We understand that in order to bring the research to the level of technology, it is necessary to accumulate a large amount of data obtained remotely over different years and under different conditions.

 

• The lack of standardized protocols for crown pixel marking, as admitted by the authors, further undermines the validity of field validation attempts.

Response: Thank you for your comment!  Indeed, with high spatial resolution, this task seems very difficult to solve. We have also not found any solutions to this problem in the literature.

 

• The statistical approach requires substantial strengthening in multiple areas. First, the vegetation index selection appears somewhat arbitrary despite the literature review - no systematic feature selection or correlation analysis is presented to justify why these specific 28 VIs were chosen over others available in the literature.

Response: Thank you for your comment!  When choosing VIs for the study, we relied on previous research results on plant stress. Most researchers agree that carotenoid-sensitive VIs are most effective for recording stress. However, we agree with you that the use of chlorophyll-sensitive VIs can improve the effectiveness of the model. We note the absence of such VIs in the group of independent variables as a limitation of the study. Currently, there are more than 100 VIs that are sensitive to chlorophyll or describe other plant properties indirectly through chlorophyll content. Analysing such a volume of data is essentially a separate study. At the same time, we agree that such work needs to be done.

 

• The multicollinearity analysis shown in Figure 10 suggests several VIs provide redundant information, yet no dimensionality reduction techniques were applied.

Response: Thank you for your comment! When analysing the correlation matrix, we found that two indices, CCRI and CTRI/CI_red-edge, have a high correlation strength. We excluded CTRI/CI_red-edge from the independent variables, as noted in the first version of the manuscript. Following your comment, we calculated the variance inflation factor (VIF) for the remaining 27 VIs. The results of the multicollinearity assessment of VIs are presented in Table 2. For all VIs, the VIF value is below the critical level. The changes can be tracked in a track-changed version.

 

• Second, the machine learning validation lacks proper cross-validation strategies. While 10-fold cross-validation is mentioned for LDA, the random forest models use only out-of-bag error estimation, which can be overly optimistic, particularly with the relatively small dataset size.

Response: Thank you for your comment! Changes have been made to section ‘4.9. Data analytics’. A 5-fold cross-validation method was used to adjust the hyperparameters and to assess efficacy. RF hyperparameters: number of trees = 100 and number of variables tried at each split = 5. The changes can be tracked in a track-changed version.

 

• The temporal nature of the data also raises concerns about data leakage - training and testing on temporally adjacent samples may inflate performance metrics.

Response: Thank you for this valuable information! We have not encountered a description of this effect in other researchers' publications. At the same time, we would like to note that on adjacent dates, the VIs values (both for shoots and crowns) are not statistically related in terms of pixels.

 

• Third, the threshold selection for ROI identification (Carter5>1.4) lacks proper justification or sensitivity analysis, and this preprocessing step could significantly impact downstream classification performance.

Response: Thank you for your comment! The rationale for two-stage ROI selection using a Carter5>1.4 threshold on hyperspectral images of conifer shoots is provided in articles DOI: 10.18522/2308-9709-2023-46-3. and 10.1007/s12145-023-01118-0 (https://www.researchgate.net/publication/374872520_Classification_of_invasive_tree_species_based_on_the_seasonal_dynamics_of_the_spectral_characteristics_of_their_leaves. We are attaching an English translation of the relevant section of the Russian-language article.

 

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Reviewer's response to the authors

This manuscript describes a study on Identifying Winter Light Stress in Conifers using Proximal Hyperspectral Imaging and Machine Learning. The study is well structured but requires significant clarification and improvement in Introduction, methodology, results, discussions and conclusions.

Major Revisions

Keywords:

• Use a maximum of five or six keywords and ensure they are not identical to the words used in the manuscript title.
• Introduction:
  Avoid using bibliographic references older than the last five years. The paragraphs contain an excessive number of citations; please reduce them.
• The introduction is very extensive; try to merge paragraphs and synthesize ideas. It should clearly state why WLS is difficult to detect with traditional methods and focus on the relevance of WLS for remote sensing.
• In the title of Figure 1, avoid acronyms such as P or HSI; use full terms instead.

 

Materials and Methods:

• In Section 2.5, the calibration performed with the hyperspectral equipment is not mentioned.
• In Section 2.6 (line 230), “size 15” is indicated: specify the units or clarify what this refers to.
• Consider merging Sections 2.6 and 2.7, as it does not make sense to treat them separately if the indices have already been mentioned earlier in the manuscript.
• In Section 2.8, a spectrophotometer is mentioned; please specify the model, brand, and country of manufacture.
• In Section 2.9 (line 253), clarify whether “+18…+20 ºC” refers to a fluctuation range or fixed values.
• In Section 2.10, the Random Forest procedure and methodology are not described. Please include the details, provide a bibliographic reference, and report the evaluation parameters used.
• Add the number of plant samples used in the Random Forest classification.
• The methodology does not mention whether cross-validation was used, nor does it specify the train/test split percentage.

 

Results:

• In Section 3.3, why was the coefficient of determination used instead of the correlation coefficient? And why was the variance inflation factor (VIF) not applied to assess multicollinearity?
• In Figure 10, clearly indicate in the title what the X variables represent.
• In Figure 11, explicitly define the concepts a, b, c, and d in the caption.
• In Table 2, adjust the format to comply with the journal’s requirements.
• A specific section on the limitations of the study should be added.
• Explain why water stress indices such as NDWI or MCARI were not considered. It is also suggested to extend the study over at least two years and include more tree species, as using a single species and one season is not representative enough to extrapolate to an entire region.
• The validation relies only on the OOB error. It is recommended to include a k-fold cross-validation scheme or independent validation by dates to provide greater robustness to the results.

 

Discussion:

It is suggested to strengthen the discussion with UAV- and Sentinel-2-based studies that demonstrate the detection of WLS at larger scales, in order to link the laboratory findings with operational remote sensing applications.

 

Author Response

Dear Reviewer!

Thank you so much for taking the time to review our manuscript! We appreciate your valuable feedback and constructive suggestions on our work. Your comments have made it much better. We retained revision marks on the revised manuscript in the ‘stresses-3899267 — Revised (tracked changes).docx’ file. The authors hope, the current version of the manuscript will meet your expectations regarding clarity and quality of presentation.

 

• Keywords:
  Use a maximum of five or six keywords and ensure they are not identical to the words used in the manuscript title.

Response: Thank you for your comment! We have reduced the number of keywords in accordance with your recommendations. The changes can be tracked in a track-changed version.

 

• Introduction:
  Avoid using bibliographic references older than the last five years. The paragraphs contain an excessive number of citations; please reduce them.

Response: Thank you for your comment! We have reduced the excessive number of bibliographic references. We have retained references older than five years only for priority studies on the physiology of winter light stress in conifers. In addition, we have retained references older than five years to authors of vegetation indices. The changes can be tracked in a track-changed version.

 

• The introduction is very extensive; try to merge paragraphs and synthesize ideas. It should clearly state why WLS is difficult to detect with traditional methods and focus on the relevance of WLS for remote sensing.

Response: Thank you for your comment! We have tried to shorten the introduction. We have clarified that the main advantage of the HSI method over traditional methods of detecting plant stress is its ability to be used remotely. The changes can be tracked in a track-changed version.

 

• In the title of Figure 1, avoid acronyms such as P or HSI; use full terms instead.

Response: Thank you for your comment! The title of Figure 1 has been corrected. The changes can be tracked in a track-changed version.

 

• Materials and Methods: In Section 2.5, the calibration performed with the hyperspectral equipment is not mentioned.

Response: Thank you for your comment! We have added information about calibrating the hyperspectral camera. A white reference panel was used to calibrate the reflectance. The changes can be tracked in a track-changed version.

 

• In Section 2.6 (line 230), “size 15” is indicated: specify the units or clarify what this refers to.

Response: Thank you for your comment! The Savitsky-Golei filter is usually implemented as a convolution with constant coefficients, requiring a fixed data step on the x-axis (uniform sampling on the x-axis). In our case, the units of measurement are nm. We have added this information to the manuscript. The changes can be tracked in a track-changed version.

 

• Consider merging Sections 2.6 and 2.7, as it does not make sense to treat them separately if the indices have already been mentioned earlier in the manuscript.

Response: Thank you for your comment! We have combined subsections 2.6 and 2.7. The changes can be tracked in a track-changed version.

 

• In Section 2.8, a spectrophotometer is mentioned; please specify the model, brand, and country of manufacture.

Response: Thank you for your comment! The DU 730 model, manufactured by BECKMAN COULTER (USA). We have added this information to the manuscript. The changes can be tracked in a track-changed version.

 

• In Section 2.9 (line 253), clarify whether “+18…+20 ºC” refers to a fluctuation range or fixed values.

Response: Thank you for your comment! This is the temperature range. A clarification has been added to the text of the manuscript. The changes can be tracked in a track-changed version.

 

• In Section 2.10, the Random Forest procedure and methodology are not described. Please include the details, provide a bibliographic reference, and report the evaluation parameters used.

Response: Thank you for your comment! We have added a detailed description of RF procedures and methodology to section 4.9. Data analytics. The changes can be tracked in a track-changed version.

 

• Add the number of plant samples used in the Random Forest classification.

Response: Thank you for your comment! In section 2.3 (4.3 in the revised version), we indicated that HSI of crowns was performed for six P. orientalis plants. Section 2.4 (4.4 in the revised version) states that three P. orientalis trees were selected for shoot sampling. Seven shoots were selected from each tree for laboratory HSI. Pixel-based classification was used in the study. The number of pixels per class was aligned to the minimum value. For one class, it was 18,000. The changes can be tracked in a track-changed version.

 

• The methodology does not mention whether cross-validation was used, nor does it specify the train/test split percentage.

Response: Thank you for your comment! In the new version of the manuscript, we have detailed the research methodology. The changes can be tracked in a track-changed version.

 

• Results: In Section 3.3, why was the coefficient of determination used instead of the correlation coefficient?

Response: Thank you for your comment! In this case, the coefficient of determination, like the correlation coefficient, is an indicator of the strength of the relationship. Its use is related to the fact that the relationship between Vis in the form of a matrix can be presented more clearly, since the coefficient of determination does not take negative values. In addition, the coefficient of determination is used to calculate VIF.

 

• And why was the variance inflation factor (VIF) not applied to assess multicollinearity?

Response: Thank you for your comment! Following your recommendation, we calculated the VIF value to increase the reliability of the conclusion about low multicollinearity. The results are presented in Table 2. For all VIs, the VIF value was below the critical value. The changes can be tracked in a track-changed version.

 

• In Figure 10, clearly indicate in the title what the X variables represent.

Response: Thank you for your comment! The symbol ‘χ’ in the matrix indicates the absence of a reliable correlation. The corresponding changes have been made to the manuscript. The changes can be tracked in a track-changed version.

 

• In Figure 11, explicitly define the concepts a, b, c, and d in the caption.

Response: Thank you for your comment! We have corrected the caption for Figure 11. The changes can be tracked in a track-changed version.

 

• In Table 2, adjust the format to comply with the journal’s requirements.

Response: Thank you for your comment! We have adjusted the table format. The changes can be tracked in a track-changed version.

 

• A specific section on the limitations of the study should be added.

Response: Thank you for your comment! We have added a section on limitations to the manuscript. In it, we specify, in particular, that these are preliminary results and that the study was aimed at proving the fundamental possibility of recording light stress using HSI. The changes can be tracked in a track-changed version.

 

• Explain why water stress indices such as NDWI or MCARI were not considered. It is also suggested to extend the study over at least two years and include more tree species, as using a single species and one season is not representative enough to extrapolate to an entire region.

Response: Thank you for your comment! Thank you for your suggestions. We assume that water stress indices are not suitable for identifying winter light stress. They will primarily respond to a sharp loss of water during acclimatisation (when conifers transition from vegetation to dormancy). At the same time, we are recording the dynamics of water content in shoots. These are interesting VIs for determining the state of conifers – vegetation or dormancy.

At the beginning of 2025, we made the experimental design more ‘'strict’. The study is currently ongoing. Following your suggestion, we plan to increase the number of experimental species by adding Thuja occidentalis, Pinus nigra, Juníperus sabína and Juniperus davurica.

 

• The validation relies only on the OOB error. It is recommended to include a k-fold cross-validation scheme or independent validation by dates to provide greater robustness to the results.

Response: Thank you for your comment! Changes have been made to section ‘4.9. Data analytics’. A 5-fold cross-validation method was used to adjust the hyperparameters and to assess efficacy. The changes can be tracked in a track-changed version.

 

• Discussion: It is suggested to strengthen the discussion with UAV- and Sentinel-2-based studies that demonstrate the detection of WLS at larger scales, in order to link the laboratory findings with operational remote sensing applications.

Response: Thank you for your comment!  Unfortunately, we were unable to find any papers where winter light stress in conifers was recorded remotely. There are a number of experimental studies based on proximal HSI, which we discuss. Perhaps the use of UAVs for these purposes is limited by the operating temperature range of hyperspectral sensors. However, we will try to overcome this technical obstacle and conduct field HSI this winter season.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

What is the main question addressed by the research?

The main objective of this manuscript is to obtain an accurate qualitative assessment of light stress in evergreen conifers using C_car and C_chl/C_car-sensitive vegetation indices by constructing multivariate models based on machine learning algorithms.

What parts do you consider original or relevant for the field? What specific gap in the field does the paper address?

To study plant light stress, the authors of the manuscript selected the conifer species Platycladus orientalis, which is suitable for hyperspectral imaging. This allowed them to identify light stress in evergreen conifers and address the objective of this study.

What does it add to the subject area compared with other published material?

A peer-reviewed manuscript shows that a 'Random Forest' model trained on laboratory proximal hyperspectral imaging of canopy shoots is good at classifying 'winter light stress' conditions and poor at classifying 'cold stress' and 'optimal conditions'.

What specific improvements should the authors consider regarding the methodology? What further controls should be considered?

Although the authors achieved the primary objective of the study (identifying the state of winter light stress), they were unable to classify the "cold stress" and "optimal condition" states of P. orientalis crowns. This problem could be addressed by using a broader set of VIs. Also, to develop universal classification models for conifers in the MVP and dormant states, it is necessary to move to long-term crown studies. For field work at low temperatures, different equipment for hyperspectral imaging of the conifer crowns should be used.

Please describe how the conclusions are or are not consistent with the evidence and arguments presented. Please also indicate if all main questions posed were addressed and by which specific experiments.

The data presented in Figures 6–15, as well as in Tables 2 and 3 and in the text of the manuscript, substantiate the possibility of obtaining an accurate qualitative assessment of light stress in evergreen conifers using C_car and C_chl/C_car-sensitive vegetation indices by constructing multivariate models based on machine learning algorithms. These results and conclusions address the objective of the paper stated in the introduction.

Are the references appropriate?

The bibliography of 124 titles provides comprehensive information on all aspects of this manuscript's topics. Most references have Dois, allowing easy access to the cited works. All references are relevant to the topic of this manuscript.

Please include any additional comments on the tables and figures and quality of the data.

In many figures there is an excessive number of labels along the axes (Fig. 6 and Fig. 14 along the X-axis, Fig. 7-9, 12 along the Y-axis).

Author Response

Dear Reviewer!

Thank you so much for taking the time to review our manuscript! We appreciate your valuable feedback and constructive suggestions on our work. Your comments have made it much better. We retained revision marks on the revised manuscript in the ‘stresses-3899267 — Revised (tracked changes).docx’ file. The authors hope, the current version of the manuscript will meet your expectations regarding clarity and quality of presentation.

Thank you for your high assessment of our manuscript!

 

• Although the authors achieved the primary objective of the study (identifying the state of winter light stress), they were unable to classify the "cold stress" and "optimal condition" states of P. orientalis crowns. This problem could be addressed by using a broader set of VIs. Also, to develop universal classification models for conifers in the MVP and dormant states, it is necessary to move to long-term crown studies. For field work at low temperatures, different equipment for hyperspectral imaging of the conifer crowns should be used.

Response: Thank you for your valuable feedback! In early 2025, we refined the experimental design. The study is currently ongoing. We have expanded the scope of fieldwork, including hyperspectral surveys of conifer crowns from various angles. We agree with the reviewer that using chlorophyll-sensitive VIs as independent variables in the model could improve their accuracy. We plan to expand the number of experimental species by including Thuja occidentalis, Pinus nigra, Juniperus sabina, and Juniperus davurica. The experiment is planned to run for many years. We are currently searching for equipment capable of operating at subzero temperatures and seeking a different solution to this problem.

 

• In many figures there is an excessive number of labels along the axes (Fig. 6 and Fig. 14 along the X-axis, Fig. 7-9, 12 along the Y-axis).

Response: Thank you for your comment! We have made changes to Figures 6-9, 12 and 14. The changes can be tracked in a track-changed version.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for letting me review it again. After a careful read, I believe the author has addressed the issues very well and fully resolved the comments from the previous round. So I suggest to accept this paper.

Author Response

Dear Reviewer!

We appreciate the time and effort you have spent evaluating our manuscript. Thank you! 

Reviewer 2 Report

Comments and Suggestions for Authors

Reviewer's response to the authors

This manuscript describes a study on Identifying Winter Light Stress in Conifers using Proximal Hyperspectral Imaging and Machine Learning.

If you cannot clearly see the changes made or modifications to the manuscript, please send the changes highlighted and in Word if possible. It's incomprehensible.

 

 

 

Comments for author File: Comments.pdf

Author Response

Dear Reviewer!

We appreciate the time and effort you have spent evaluating our manuscript. Thank you!

We have highlighted the changes in the manuscript:

deleted

added

moved

The authors hope, the current version of the manuscript will meet your expectations regarding clarity and quality of presentation.

 

• Keywords:
  Use a maximum of five or six keywords and ensure they are not identical to the words used in the manuscript title.

Response: Thank you for your comment! We have reduced the number of keywords in accordance with your recommendations. The changes can be tracked in a track-changed version.

 

• Introduction:
  Avoid using bibliographic references older than the last five years. The paragraphs contain an excessive number of citations; please reduce them.

Response: Thank you for your comment! We have reduced the excessive number of bibliographic references. We have retained references older than five years only for priority studies on the physiology of winter light stress in conifers. In addition, we have retained references older than five years to authors of vegetation indices. The changes can be tracked in a track-changed version.

 

• The introduction is very extensive; try to merge paragraphs and synthesize ideas. It should clearly state why WLS is difficult to detect with traditional methods and focus on the relevance of WLS for remote sensing.

Response: Thank you for your comment! We have tried to shorten the introduction. We have clarified that the main advantage of the HSI method over traditional methods of detecting plant stress is its ability to be used remotely. The changes can be tracked in a track-changed version.

 

• In the title of Figure 1, avoid acronyms such as P or HSI; use full terms instead.

Response: Thank you for your comment! The title of Figure 1 has been corrected. The changes can be tracked in a track-changed version.

 

• Materials and Methods: In Section 2.5, the calibration performed with the hyperspectral equipment is not mentioned.

Response: Thank you for your comment! We have added information about calibrating the hyperspectral camera. A white reference panel was used to calibrate the reflectance. The changes can be tracked in a track-changed version.

 

• In Section 2.6 (line 230), “size 15” is indicated: specify the units or clarify what this refers to.

Response: Thank you for your comment! The Savitsky-Golei filter is usually implemented as a convolution with constant coefficients, requiring a fixed data step on the x-axis (uniform sampling on the x-axis). In our case, the units of measurement are nm. We have added this information to the manuscript. The changes can be tracked in a track-changed version.

 

• Consider merging Sections 2.6 and 2.7, as it does not make sense to treat them separately if the indices have already been mentioned earlier in the manuscript.

Response: Thank you for your comment! We have combined subsections 2.6 and 2.7. The changes can be tracked in a track-changed version.

 

• In Section 2.8, a spectrophotometer is mentioned; please specify the model, brand, and country of manufacture.

Response: Thank you for your comment! The DU 730 model, manufactured by BECKMAN COULTER (USA). We have added this information to the manuscript. The changes can be tracked in a track-changed version.

 

• In Section 2.9 (line 253), clarify whether “+18…+20 ºC” refers to a fluctuation range or fixed values.

Response: Thank you for your comment! This is the temperature range. A clarification has been added to the text of the manuscript. The changes can be tracked in a track-changed version.

 

• In Section 2.10, the Random Forest procedure and methodology are not described. Please include the details, provide a bibliographic reference, and report the evaluation parameters used.

Response: Thank you for your comment! We have added a detailed description of RF procedures and methodology to section 4.9. Data analytics. The changes can be tracked in a track-changed version.

 

• Add the number of plant samples used in the Random Forest classification.

Response: Thank you for your comment! In section 2.3 (4.3 in the revised version), we indicated that HSI of crowns was performed for six P. orientalis plants. Section 2.4 (4.4 in the revised version) states that three P. orientalis trees were selected for shoot sampling. Seven shoots were selected from each tree for laboratory HSI. Pixel-based classification was used in the study. The number of pixels per class was aligned to the minimum value. For one class, it was 18,000. The changes can be tracked in a track-changed version.

 

• The methodology does not mention whether cross-validation was used, nor does it specify the train/test split percentage.

Response: Thank you for your comment! In the new version of the manuscript, we have detailed the research methodology. The changes can be tracked in a track-changed version.

 

• Results: In Section 3.3, why was the coefficient of determination used instead of the correlation coefficient?

Response: Thank you for your comment! In this case, the coefficient of determination, like the correlation coefficient, is an indicator of the strength of the relationship. Its use is related to the fact that the relationship between Vis in the form of a matrix can be presented more clearly, since the coefficient of determination does not take negative values. In addition, the coefficient of determination is used to calculate VIF.

 

• And why was the variance inflation factor (VIF) not applied to assess multicollinearity?

Response: Thank you for your comment! Following your recommendation, we calculated the VIF value to increase the reliability of the conclusion about low multicollinearity. The results are presented in Table 2. For all VIs, the VIF value was below the critical value. The changes can be tracked in a track-changed version.

 

• In Figure 10, clearly indicate in the title what the X variables represent.

Response: Thank you for your comment! The symbol ‘χ’ in the matrix indicates the absence of a reliable correlation. The corresponding changes have been made to the manuscript. The changes can be tracked in a track-changed version.

 

• In Figure 11, explicitly define the concepts a, b, c, and d in the caption.

Response: Thank you for your comment! We have corrected the caption for Figure 11. The changes can be tracked in a track-changed version.

 

• In Table 2, adjust the format to comply with the journal’s requirements.

Response: Thank you for your comment! We have adjusted the table format. The changes can be tracked in a track-changed version.

 

• A specific section on the limitations of the study should be added.

Response: Thank you for your comment! We have added a section on limitations to the manuscript. In it, we specify, in particular, that these are preliminary results and that the study was aimed at proving the fundamental possibility of recording light stress using HSI. The changes can be tracked in a track-changed version.

 

• Explain why water stress indices such as NDWI or MCARI were not considered. It is also suggested to extend the study over at least two years and include more tree species, as using a single species and one season is not representative enough to extrapolate to an entire region.

Response: Thank you for your comment! Thank you for your suggestions. We assume that water stress indices are not suitable for identifying winter light stress. They will primarily respond to a sharp loss of water during acclimatisation (when conifers transition from vegetation to dormancy). At the same time, we are recording the dynamics of water content in shoots. These are interesting VIs for determining the state of conifers – vegetation or dormancy.

At the beginning of 2025, we made the experimental design more ‘'strict’. The study is currently ongoing. Following your suggestion, we plan to increase the number of experimental species by adding Thuja occidentalis, Pinus nigra, Juníperus sabína and Juniperus davurica.

 

• The validation relies only on the OOB error. It is recommended to include a k-fold cross-validation scheme or independent validation by dates to provide greater robustness to the results.

Response: Thank you for your comment! Changes have been made to section ‘4.9. Data analytics’. A 5-fold cross-validation method was used to adjust the hyperparameters and to assess efficacy. The changes can be tracked in a track-changed version.

 

• Discussion: It is suggested to strengthen the discussion with UAV- and Sentinel-2-based studies that demonstrate the detection of WLS at larger scales, in order to link the laboratory findings with operational remote sensing applications.

Response: Thank you for your comment!  Unfortunately, we were unable to find any papers where winter light stress in conifers was recorded remotely. There are a number of experimental studies based on proximal HSI, which we discuss. Perhaps the use of UAVs for these purposes is limited by the operating temperature range of hyperspectral sensors. However, we will try to overcome this technical obstacle and conduct field HSI this winter season.

 

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

Reviewer's response to the authors

This manuscript describes a study on Identifying Winter Light Stress in Conifers using Proximal Hyperspectral Imaging and Machine Learning.

 

I enjoyed reading it, although I should review the structure of the manuscript according to the journal, for example: abstract, introduction, materials and methods, results, discussions and conclusion.