Winter Wheat Canopy Height Estimation Based on the Fusion of LiDAR and Multispectral Data

Round 1

Reviewer 1 Report (New Reviewer)

Comments and Suggestions for Authors

This paper addresses the combination of LiDAR and multispectral data to estimate canopy height of winter wheat. It addresses a very current and important topic in terms of precision agriculture and remote sensing technologies. Increasing the accuracy rate by combining LiDAR and multispectral data can be considered as an innovation that can contribute to the literature in the field. The paper presents a model that integrates multispectral data to overcome this deficiency by demonstrating that existing LiDAR-based approaches are insufficient for height estimation of low height and densely planted crops. In particular, different machine learning methods such as Optuna and the Optimized Random Forest (OP-RF) model are compared and the best performing model is identified. This shows the methodological strength of the study. The authors have handled the paper carefully, but the following corrections are important for the clarity of the paper.

 

Abstract: The Abstract section is carefully written. It presents the purpose of the study, the results and the interpretation of the results.

 

Introduction: In the first paragraph of the Introduction section, growth status information is given but not explained in detail. For this explanation, the article 10.26833/ijeg.1035037 can be examined. Examples of lidar studies are given in the article, but a general evaluation of lidar altimetry is not given. https://doi.org/10.1007/s00468-022-02378-x article may give you an idea about this.

 

Materials and Methods: The Materials section is very well explained. In the Methods section, the explanations are sufficient, but it may be difficult for readers to perceive because it is examined under too many headings. Drawing a workflow diagram will increase the comprehensibility of the methods section.

 

 

Results: Figure 5: The units of RMSE and MAE values are missing.

 

The results are not sufficiently interpreted. You could provide more analysis of what the tables and figures are saying.

Discussions and Conclusion: The study results could discuss more how much improvement it provides compared to other existing studies. In particular, the comparison with other forecasting methods could be deepened.

Author Response

I sincerely thank you for your valuable comments and suggestions. I have carefully studied the relevant literature recommended by you, which provided me with rich insights and greatly assisted me in refining the introductory part of the paper. Based on your feedback, I have meticulously revised the relevant content and completely updated the entire manuscript.

Commentss 1: [Abstract: The Abstract section is carefully written. It presents the purpose of the study, the results and the interpretation of the results.]

Response 1: [Thank you very much for your affirmation and support.]

Comments 2: [Introduction: In the first paragraph of the Introduction section, growth status information is given but not explained in detail. For this explanation, the article 10.26833/ijeg.1035037 can be examined. Examples of lidar studies are given in the article, but a general evaluation of lidar altimetry is not given. https://doi. org/10.1007/s00468-022-02378-x article may give you an idea about this.]

Response 2: [This section has been updated and revised with reference to the literature you provided (lines 43-50, 61-63).]

Comments 3: [Materials and Methods: The Materials section is very well explained. In the Methods section, the explanations are sufficient, but it may be difficult for readers to perceive because it is examined under too many headings.Drawing a workflow diagram will increase the comprehensibility of the methods section.]

Response 3: [An overall workflow diagram has been added as shown in Figure 2 (lines 149-150).]

Comments 4: [Results: Figure 5: The units of RMSE and MAE values are missing.]

Response 4: [The figure has been adjusted according to your suggestions. However, the order of the figure has been changed and has been changed to figure 6 (lines 435-436).] 

Comments 5: [The results are not sufficiently interpreted. You could provide more analysis of what the tables and figures are saying.]

Response 5: [The results section has been revised (lines 339-492).]

Comments 6: [Discussions and Conclusion: The study results could discuss more how much improvement it provides compared to other existing studies. In particular, the comparison with other forecasting methods could be deepened.]  

Response 6: [The discussion and conclusion sections have been revised (lines 494-511).]      

 

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

Hi,

Please find my comments in the attached pdf file.

Good luck

Comments for author File: Comments.pdf

Author Response

Thank you for your valuable comments and suggestions. Revisions have been made in response to the feedback, and the manuscript has been updated accordingly.

Comments 1: [Line 84: Still need more clarification to show the novelty of your work.] 

Response 1: [The section has been revised (lines 86-103).] 

Comments 2: [Line 132: It would be easier to create a flowchart for each processing technique.] 

Response 2: [An overall workflow diagram has been added as shown in Figure 2 (lines 149-150).]  

Comments 3: [Line 209: You should mention those references.]   

Response 3: [Relevant literature has been added (line 230).] 

Comments 4: [Table 4: Although all of these formulas are frequently used in scientific papers but you should cite them in your manuscript.]   

Response 4: [Literature has been cited in the text (line 256).]  

Comments 5: [Table 8: why didn't you use Zadocs growth stage standard to express the growth stage in more detailed method.]  

Response 5: [We thank the reviewers for their valuable suggestions. Zadocs growth stage standard is a widely used method of coding crop fertility with good standardization and comparability. In this study, we adopted the terms Green-up Stage (GUS) , Jointing Stage (JS)) and booting stage (BS), taking into account the field observation habits of researchers and the expressions in the existing literature. These terms are basically consistent with the current common expressions in the field of remote sensing and farmland monitoring, which facilitates comparison and articulation with related studies. For example, the use of these terms is addressed in these articles: doi: 10.3389/fpls.2023.1272049, doi: 10.1111/gcb.16414 and doi: 10.3390/foods9030353, etc.]

Comments 6: [Figure 5: change the color of 1:1 line to make it more recognizable in the graphs.]  

Response 6: [The color of the 1:1 line has been changed to blue (lines 435-436).]        

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The article is devoted to the assessment of the canopy height of winter wheat using a combined method based on combining data from a UAV from a Lidar sensor and multispectral data using machine learning.
The article has scientific and practical interest.
To publish in the journal "Agronomy", the authors must make the following adjustments:
1) At the beginning of the article, it would be good to make a research scheme in the form of mind maps or a graphical annotation.
2) Write more clearly what knowledge gap your research covers
3) At the end of the section, clearly formulate the purpose of the study and its objectives in the format: 1,2,3 ...
4) Conclusions should also be clearer and correspond to the objectives.
5) What is the practical significance of your research? You wrote that it is possible to predict the harvest, but the question is not disclosed beyond this phrase. Give an example in quantitative and qualitative terms. What is the significance of your research for agronomy? What is the economic effect? ​​Give examples.
6) Considering that the article is published in the journal "Agronomy", it is necessary to pay more attention to the significance of the study for this industry. This should find a place in the abstract and in the conclusions.
7) It would also be good to provide comparative photographs of winter wheat canopies of different heights from close up, this will be interesting to the reader.
8) I think that the article is overloaded with various abbreviations, names of methods (sections 2 and 3). It is better to show comparisons using practical examples with drawings, graphs and comment in simple language. Perhaps some of the tables should be moved to the appendix (tables 6, 8).

Author Response

Thank you for your valuable comments and suggestions. We have made changes based on the feedback and have made a second round of updates to the manuscript.

Comments 1: [At the beginning of the article, it would be good to make a research scheme in the form of mind maps or a graphical annotation.]   

Response 1: [An overall workflow diagram has been added as shown in Figure 2 (lines 149-150).]  

Comments 2: [Write more clearly what knowledge gap your research covers.]  

Response 2: [This has been revised in the introductory section (lines 86-96).]            

Comments 3: [At the end of the section, clearly formulate the purpose of the study and its objectives in the format: 1,2,3 ...]     

Response 3: [The specific objectives of the study have been articulated in accordance with the recommendations (lines 92-103).]       

Comments 4: [Conclusions should also be clearer and correspond to the objectives.]   

Response 4: [Conclusions have been revised (lines 542-559).]   

Comments 5: [What is the practical significance of your research? You wrote that it is possible to predict the harvest, but the question is not disclosed beyond this phrase. Give an example in quantitative and qualitative terms. What is the significance of your research for agronomy? What is the economic effect? ​​Give examples.]   

Response 5: [Updates on the relevance of this study have been made in the introduction, discussion, etc (lines 43-49, 494-501).]           

Comments 6: [Considering that the article is published in the journal "Agronomy", it is necessary to pay more attention to the significance of the study for this industry. This should find a place in the abstract and in the conclusions.]  

Response 6: [Updates to the industry significance of this study have been made in the summary, conclusion, etc (lines 37-38, 562-565).]           

Comments 7: [It would also be good to provide comparative photographs of winter wheat canopies of different heights from close up, this will be interesting to the reader.]  

Response 7: [A corresponding diagram has been added to the overall workflow diagram Figure 2 (lines 149-150).]      

Comments 8: [I think that the article is overloaded with various abbreviations, names of methods (sections 2 and 3). It is better to show comparisons using practical examples with drawings, graphs and comment in simple language. Perhaps some of the tables should be moved to the appendix (tables 6, 8).]  

Response 8: [We understand your concern that the high number of acronyms and method names in Sections 2 and 3 may affect readability. In the field of remote sensing and crop phenotyping research, especially in related work involving point cloud feature parameters, multispectral vegetation indices, and machine learning models. Relevant terms are usually presented in an abbreviated form to improve the standardization and conciseness of expression. On this basis, we have scrutinized the full text and adjusted some of the abbreviations. For example, in Table 6, the full names of some abbreviations have been added, and the abbreviations of Analysis of variance (ANOVA) and other related parts have been further clarified to improve the overall readability and comprehensibility (lines 343-344, 348-366).

Your suggestion to move Tables 6 and 8 to the Appendix has been carefully considered. However, after careful consideration, we believe that these two tables are important to support the research ideas and results of this paper. It is therefore recommended that they be retained in the main text. The reasons are as follows: Table 6 demonstrates the ANOVA results of the point cloud features and multispectral feature parameters, which are the key basis for screening the features. Table 8, on the other hand, demonstrates the canopy height estimation accuracy of the four models. By comparing the performance of these four models, we can clarify which model performs better in canopy height estimation, thus providing a strong basis for selecting the optimal model.]  

Reviewer 4 Report (New Reviewer)

Comments and Suggestions for Authors

Summary

This paper presents a method for estimating the canopy height of winter wheat by fusing LiDAR point cloud features and multispectral vegetation indices obtained from a UAS platform. Data were collected at three critical growth stages (green-up, jointing, and booting). After preprocessing and aligning both LiDAR and multispectral datasets, feature extraction was performed using ANOVA and Pearson correlation. The authors tested four machine learning models (OP-RF, Elastic Net, XGBoost, and SVR), finding that the Optuna-optimized Random Forest (OP-RF) yielded the best results, with R² values of 0.921-0.936 on test sets across the three stages. The study shows that the fusion of data modalities leads to more accurate estimations compared to using either dataset alone.

Major comments:

Lines 244–261: The authors describe the use of ANOVA and Pearson correlation for feature selection, but no thresholds are specified for the correlation coefficient (e.g., |r| > 0.5) or for statistical significance in ANOVA (e.g., p < 0.05). Additionally, the manuscript does not address the potential issue of multicollinearity between input features, which could affect model performance and interpretability. It is recommended to include these selection criteria and a brief analysis of inter-feature correlation (e.g., VIF or correlation heatmap).

Lines 423–426: the study is limited to a single geographic location, which the authors acknowledge. However, additional validatio - either by including external datasets or by a simple transferability analysis - would enhance confidence in the model's robustness.

Table 8: although OP-RF performed best, the training R2 is quite close to 1 in some cases, which raises concerns about potential overfitting. Including learning curves, residual plots, or validation on unseen data (e.g., a different field or year) would help assess generalization.

Minor comments:

Line 14: “Accurately predicting wheat canopy height helps improve field management...”; consider replacing “helps” with “can improve”.

Line 199: “...to guide the plot segmentation and delineation of the multispectral data...” ; “delineation” is slightly awkward; consider “definition” or “mapping”.

Lines 327-330: the sentence is understandable, but consider rephrasing slightly for clarity, e.g.: “The highest Pearson correlation coefficient at the BS stage was observed for MTVI1 (r = 0.40).” This avoids any ambiguity and standardizes the reporting format across growth stages.

Author Response

Thank you for your valuable comments and suggestions. The manuscript has been revised accordingly, and the necessary updates have been made.

Comments 1: [Lines 244-261: The authors describe the use of ANOVA and Pearson correlation for feature selection, but no thresholds are specified for the correlation coefficient (e.g., > 0.5) or for statistical significance in ANOVA (e.g., p <0.05). Additionally, the manuscript does not address the potential issue of multicollinearity between input features, which could affect model performance and interpretability. It is recommended to include these selection criteria and a brief analysis of inter-feature correlation (e.g., VIF or correlation heatmap).]   

Response 1: [Variance inflation factor has been added and the content has been revised (lines 283-295, 340-388).]  

Comments 2: [Lines 423-426: the study is limited to a single geographic location, which the authors acknowledge. However, additional validatio -either by including external datasets or by a simple transferability analysis -would enhance confidence in the model's robustness.]  

Response 2: [Added distribution of Root Mean Square Error (RMSE) estimates and distribution of MAE estimates from Figure 7, and updated and revised content (lines 465-492).]            

Comments 3: [Table 8: although OP-RF performed best, the training R² is quite close to 1in some cases, which raises concerns about potential overfitting. Including learning curves, residual plots, or validation on unseen data (e.g., a different field or year) would help assess generalization.]     

Response 3: [Added distribution of Root Mean Square Error (RMSE) estimates and distribution of MAE estimates from Figure 7, and updated and revised content (lines 465-492).]       

Comments 4: [Line 14: "Accurately predicting wheat canopy height helps improve field management..."; consider replacing "helps" with "can improve".]   

Response 4: [The narrative has been revised (line 15).]    

Comments 5: [Line 199: ".to guide the plot segmentation and delineation of the multispectral data..."; "delineation" is slightly awkward; consider "definition" or "mapping".]    

Response 5: [We agree that the term “delineation” may sound slightly abrupt in some cases. However, in this case we aim to describe the process of identifying field plot boundaries in multispectral data. To make this clearer, the sentence has been modified to read: “to guide the segmentation and boundary delineation of plots in the multispectral data.” (lines 221-222).]           

Comments 6: [Lines 327-330: the sentence is understandable, but consider rephrasing slightly for clarity, e.g.: "The highest Pearson correlation coefficient at the BS stage was observed for MTVI1 (r = 0.40)."This avoids any ambiguity and standardizes the reporting format across growth stages.]  

Response 6: [This section has been revised (lines 382-387).]           

 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Authors are recommended to read this article. 

UAS Quality Control and Crop Three-Dimensional Characterization Framework Using Multi-Temporal LiDAR Data.

In this article, authors showed how the crop height can more effectively measured and what is the influence of external and internal parameters on crop height estimations. 

Line: 62 - 73 I am disagreeing with authors - UAS LiDAR provide very high point cloud density enough to measure the crop heights with ref. to above article however external factors such as phenological stages and LiDAR sensor operational parameters therefore can impact. Authors are advised to include some information from above article to rewrite this paragraph. specially, environmental factors as authors of the above article discussed in detail in the article. 

Line 81-82: I am not certain and does not make sense how vegetation index can be used to estimate height. Generally, vegetation indices are useful to monitor the crop health and greeness. The height can be measured using photogrammetric modeling such as structure from motion (SfM) techniques.

Line 108 - 109: photosynthetic activity can be used to measure height how? I am not certain about it.

Line 94 - 117: Not enough refs. are provided, please include more refs. from past studies. the articles from above mentioned authors might be useful to revise this section.

Fig. 1 the fonts and size of subplots are too small to be understood. 

windless weather is NOT correct statement - there is always wind right? yet authors can mention UAS flight favorable wind conditions not impacting the flight and and data quality.

Line 159 - 166: The field measurements can be cross referenced from above mention article.

line 184: instead of placing symbols authors can provide textural descriptions such a square layout of X layout. Show these markers in Fig.1 and then cross reference them in main body text and fig. captions. 

the sub-headings sub follows the number format consistent.

(1) should be 2.3.1. Image control point extraction and follows the same sequence.

In its current configurations - Fig.3 makes no sense at all - the resolution and labeling is too small hardly visible. 

Sections that will be re-numbered 2.3.1, 2.3.2, and 2.3.3., 2.4., 2.4.1. 2.4.2, 2.4.3. and following headings and sub-headings makes no sense at all - entire methodology should be rewritten where each sub-heading and following headings should make sense. this can be done with providing the data processing steps and corresponding Figures to show how the methodology proceeded.

As far as I know, the processing is straight forward, RGB and MS images were stiches then projected, radiometric or geometric corrections were applied and all datasets projected to same coordinate system. this is just a one liner, and authors need to explain each step purely from remote sensing/GIS perspective. the current textural description is not scientifically correct.

2.5.1.

putting a one liner in brackets () does not explain the whole thing, authors need to explain each step in data processing sections. 

just like Table 1. authors need to create a table for features extracted from LiDAR and provide the mathematical formulae for each step along with its description and citations. 

instead, the notation of GUS, JS, BS use the terms which are more coinvent e.g., early growth stages, intermediate growth stages, and mature growth stages. 

Fig.5 is also has very poor resolution and I am unable to read any information out of it.

I am unable to understand how the height was obtained from LiDAR datasets. it is pretty simple to create a canopy height model (CHM) from LiDAR point clouds which provide crop heights pretty consistent.

what the GRB and multispectral features played role in crop height estimates does not provide any evidence. 

For model training what was the ground truth or reference data and how it was linked to the features extracted from LiDAR and MS datasets.

 

 

Comments on the Quality of English Language

this is general English language writing i.e., plain-text writing which does not carry the writing style and grammar of scientific English writing. 

Author Response

Thank you for your valuable comments and suggestions. I have carefully read the article you recommended, and it has provided me with many useful insights, offering a new perspective and stronger support for the introduction of my paper. Revisions have been made based on your feedback, and the manuscript has been updated accordingly.

Comments 1: [Line: 62 - 73 I am disagreeing with authors - UAS LiDAR provide very high point cloud density enough to measure the crop heights with ref. to above article however external factors such as phenological stages and LiDAR sensor operational parameters therefore can impact. Authors are advised to include some information from above article to rewrite this paragraph. specially, environmental factors as authors of the above article discussed in detail in the article.]

Response 1: [Based on your suggestions, I have revised and expanded the section of the article related to crop height measurements using UAS LiDAR. In the revision process, I have focused on adding the influence of external factors (e.g., phenological stage) on the accuracy of LiDAR measurements. Existing studies have shown that when estimating the height of low crops, LiDAR point cloud data can be affected by crop growth characteristics and environmental disturbances, leading to a decline in data quality (lines 58-69).]

Comments 2: [Line 81-82: I am not certain and does not make sense how vegetation index can be used to estimate height. Generally, vegetation indices are useful to monitor the crop health and greeness. The height can be measured using photogrammetric modeling such as structure from motion (SfM) techniques.]

Response 2: [Based on your feedback, I have revised and added the relevant content to the article. In response to your question regarding the use of vegetation indices to estimate crop height, I have included the relevant literature and analyzed the results of some studies that have used vegetation indices for height estimation. Although vegetation indices are mainly used to monitor crop health and greening, some studies have shown that an estimation model can be constructed to indirectly estimate crop height through correlation analysis between specific vegetation indices and crop height. For example, Li Z. et al. found that a vegetation index based on different spectral bands can be used to estimate plant height using a multispectral platform carried by UAS. Xie Z. et al. found that in the case of uneven terrain, vegetation index-based crop height estimation has some advantages. However, when the canopy is closed, vegetation indices mainly reflect canopy surface information and are not sensitive enough to small changes in crop height, limiting their application in specific situations. Multispectral data, on the other hand, can complement the degradation of data quality caused by point cloud data affected by low crop growth characteristics and environmental disturbances by providing additional spectral information (lines 70-79).]

Comments 3: [Line 108 - 109: photosynthetic activity can be used to measure height how? I am not certain about it.]

Response 3: [The section has been revised (lines 80-92).]

Comments 4: [Line 94 - 117: Not enough refs. are provided, please include more refs. from past studies. the articles from above mentioned authors might be useful to revise this section.]

Response 4: [The section has been modified and supplemented, and the existing literature has been reviewed. Studies on the use of fused LiDAR and spectral data to estimate forest and maize height have been retained. However, through literature screening, no studies were found that explicitly mention the use of LiDAR and spectral data fusion for estimating the height of low and dense crops (e.g., wheat). The research on forest and maize height estimation may provide some reference for estimating wheat height (lines 80-92). ]

Comments 5: [Fig. 1 the fonts and size of subplots are too small to be understood.]

Response 5: [Changes have been made to Figure 1 (line 111).]

Comments 6: [windless weather is NOT correct statement - there is always wind right? yet authors can mention UAS flight favorable wind conditions not impacting the flight and and data quality. ]

Response 6: [The section has been revised to include a description of weather conditions (lines 121-122).]

Comments 7: [Line 159 - 166: The field measurements can be cross referenced from above mention article.]

Response 7: [The article has been revised with reference to your recommended article, "UAS Quality Control and Crop 3D Characterization Framework Using Multi-temporal LiDAR Data" (lines 129-141).]

Comments 8: [line 184: instead of placing symbols authors can provide textural descriptions such a square layout of X layout. Show these markers in Fig.1 and then cross reference them in main body text and fig. captions.]

Response 8: [Revisions have been made to figure 1 and the text (lines 152-156).]

Comments 9: [the sub-headings sub follows the number format consistent.  

should be 2.3.1. Image control point extraction and follows the same sequence. ]

Response 9: [In the text, “(1)” is used to describe one of the three steps of the data alignment process, not a subtitle. Therefore, changing it to "2.3.1" would disrupt the original structure and logic. To maintain clarity, I have kept the existing formatting and adjusted the numbering of other sections to ensure consistency (lines 152-156). ]

Comments 10: [In its current configurations - Fig.3 makes no sense at all - the resolution and labeling is too small hardly visible.]

Response 10: [Figure 3 has been revised (line 233). ]

Comments 11: [Sections that will be re-numbered 2.3.1, 2.3.2, and 2.3.3., 2.4., 2.4.1. 2.4.2, 2.4.3. and following headings and sub-headings makes no sense at all - entire methodology should be rewritten where each sub-heading and following headings should make sense. this can be done with providing the data processing steps and corresponding Figures to show how the methodology proceeded.]

Response 11: [The methods section has been completely revised, and the subheadings have been reorganized. I have also added descriptions of the data processing steps and provided corresponding figure (lines 149-262).]

Comments 12: [As far as I know, the processing is straight forward, RGB and MS images were stiches then projected, radiometric or geometric corrections were applied and all datasets projected to same coordinate system. this is just a one liner, and authors need to explain each step purely from remote sensing/GIS perspective. the current textural description is not scientifically correct.]

Response 12: [Each step of the data alignment process has been completely revised (lines 149-169).]

Comments 13: [2.5.1. putting a one liner in brackets () does not explain the whole thing, authors need to explain each step in data processing sections.

just like Table 1. authors need to create a table for features extracted from LiDAR and provide the mathematical formulae for each step along with its description and citations.]

Response 13: [A new table has been created, based on Table 1 in the original article, listing the feature parameters extracted from the LiDAR data and providing the corresponding mathematical formulas and descriptions for each step (line 276).]

Comments 14: [instead, the notation of GUS, JS, BS use the terms which are more coinvent e.g., early growth stages, intermediate growth stages, and mature growth stages.]

Response 14: [Descriptions of wheat growth stages, such as the green-up stage (GUS), jointing stage (JS), and booting stage (BS), are commonly used in the relevant literature; therefore, these descriptions were adopted in this study. ]

Comments 15: [Fig.5 is also has very poor resolution and I am unable to read any information out of it.]

Response 15: [Figure 5 has been revised (line 363).]

Comments 16: [I am unable to understand how the height was obtained from LiDAR datasets. it is pretty simple to create a canopy height model (CHM) from LiDAR point clouds which provide crop heights pretty consistent.]

Response 16: [CHM, as a common method for generating canopy height from LiDAR point cloud data, can provide more consistent crop height estimates in many cases. However, its accuracy may be affected by the quality of the point cloud, vegetation characteristics, and topography. With this in mind, this study constructed a canopy height model by fusing point cloud feature parameters with multispectral feature parameters.

In this study, we attempted to construct an Optuna-optimized random forest (OP-RF) estimation model using the CHM method. For GUS, JS, and BS, the coefficient of determination (R²) for the canopy height models constructed using only CHM were 0.485, 0.636, and 0.441, respectively. In contrast, the R² for the models fusing point cloud features with multispectral feature parameters was significantly higher: 0.910, 0.857, and 0.822, representing improvements of 0.425, 0.221, and 0.381, respectively. This indicates that using feature parameters to construct wheat canopy height models has significant advantages. ]

Comments 17: [what the GRB and multispectral features played role in crop height estimates does not provide any evidence. ]

Response 17: [In this study, a wheat canopy height estimation model was constructed by fusing point cloud features and multispectral feature parameters. The fusion of multi-source data effectively compensates for the limitations of single data sources, thus improving estimation accuracy. It has been demonstrated that the fusion of LiDAR and multispectral data offers certain advantages in forest and corn height estimation. Relevant literature has been cited and discussed in detail in lines 83-88 of the text.]

Comments 18: [For model training what was the ground truth or reference data and how it was linked to the features extracted from LiDAR and MS datasets. ]

Response 18: [In this study, the Ground Truth refers to the actual wheat canopy height measured during the experiment. The reference data are the model estimates, derived from a model constructed by fusing feature parameters extracted from LiDAR and multispectral (MS) data. ]

 

Reviewer 2 Report

Comments and Suggestions for Authors

Winter Wheat Canopy Height Estimation Based on the Fusion 2 of LiDAR and Multispectral Data

Review

The work presents an interesting approach for canopy height estimation from remote sensing UAS-based analysis. I suggest a major revision due to the different improvements I have highlighted in the specific comments section below.

Specific comments

Line 22: The abstract is well written and clearly presents the study’s goals, methodology, results, and main findings. I recommend using the official term “Unmanned Aerial Systems (UASs)” instead of “Unmanned Aerial Vehicle (UAV)”, since you are referring not merely to the vehicle but to the entire monitoring system, even if your focus is on a low-altitude UAS flight in this instance.

Lines 37-38: I suggest the authors replace the keywords already used in the title to enhance the work’s impact and improve its discoverability.

Line 132: I propose giving paragraph 2.1 a different title that more clearly indicates its content.

Lines 141-143: The current explanation for how the authors divided the experimental field is unclear. Please rephrase and provide a larger, more intuitive figure. When referring to the 7 × 1.4 treatment areas within the experimental field, you might consider using the term “plot” rather than “experimental field.”

Figure 1: The text inside the images is too small and difficult to read. Please increase the font size. Make sure to check other figures as well.

Lines 150-151: It is not clear whether the AA10 model refers to the UAS or to the LIDAR technology produced by CHN company. Please provide more details about both the UAS and the LIDAR models used in the survey.

Lines 158: The statement “The flight trajectory was in an ‘S’ pattern” needs more detail about why and how this pattern was chosen.

Lines 163-166: The dataset used appears relatively small.

Lines 174-177: How did the authors ensure accurate and consistent overlay between the LIDAR and the multispectral (SfM-derived) digital model? Did you use RTK, Ground Control Points (GCPs), or another method? More explanation would strengthen this section. Figure 2 offers some insight, but these details should also appear in the main text.

Line 180: I would recommend including a brief introduction (just a few words) explaining the pipeline points to make this section clearer.

Line 184: Replace the “□” symbol with something more conventional (e.g., “squared”) if necessary. Also, consider whether specifying the shape is essential, since stating that you placed four GCPs and explaining their purpose may suffice.

Lines 185-187: The meaning is unclear; please give more details or consider adding a new figure.

Line 189: Provide a brief explanation of how spike processing works, or cite a relevant scientific publication that describes the process.

Line 234: How did the authors establish the vegetation index threshold?

Line 253: This step lacks clarity. Please enlarge Figure 3 and improve its quality so that the information is easier to interpret.

Line 260: As suggested earlier, provide a clearer subdivision of the experimental field (e.g., Experimental field, plot, subplot).

Line 271: How did the authors determine the vegetation index thresholds?

Line 280: What is the orientation (length direction) of the experimental field? Please indicate it clearly in the figures.

Lines 288-299: How were the geometric point cloud-derived features measured in practice? Providing references for each variable would help readers understand how the analyses were conducted and would increase the method’s repeatability.

Lines 317-319: Which statistical test was employed? What was the research hypothesis?

Lines 331-333: Please explain the process behind selecting these models and clarify why the proposed models were chosen. Readers often look for the rationale behind choosing specific analytical models or tests. Being transparent about your reasoning can significantly strengthen the paper.

Table 4: There is a typographical error in the last column.

Lines 436-439: It is not clear where the joint estimation, based on both point cloud data and multispectral indices, of canopy height is presented. Please specify which table contains this information.

Lines 440 and 443: Insert a reference to the relevant table.

Conclusions: I recommend reorganizing this section so that all critical conclusions are stated in a coherent, discursive format rather than in bullet points. Additionally, include a concise “take-home” message that highlights the main findings.

Comments for author File: Comments.pdf

Author Response

Thank you for your valuable comments and suggestions. Revisions have been made in response to the feedback, and the manuscript has been updated accordingly.

Comments 1: [Line 22: The abstract is well written and clearly presents the study's goals, methodology, results, and main findings. I recommend using the official term "Unmanned Aerial Systems (UASs)"instead of "Unmanned Aerial Vehicle (UAV)", since you are referring not merely to the vehicle but to the entire monitoring system, even if your focus is on a low-altitude UAS flight in this instance.]
Response 1: [The term 'Unmanned Aerial Vehicle (UAV)' has been revised to 'Unmanned Aerial Systems (UASs)' in the text (line 20).]

Comments 2: [Lines 37-38: I suggest the authors replace the keywords already used in the title to enhance the work’s impact and improve its discoverability. ]

Response 2: [Keyword changes have been made (lines 34-35).]

Comments 3: [Line 132: I propose giving paragraph 2.1 a different title that more clearly indicates its content. ]

Response 3: [The title of 2.1 has been revised (line 102). ]

Comments 4: [Lines 141-143: The current explanation for how the authors divided the experimental field is unclear. Please rephrase and provide a larger, more intuitive figure. When referring to the 7 × 1.4 treatment areas within the experimental field, you might consider using the term “plot” rather than “experimental field.”  ]

Response 4: [The explanation of the experimental field division has been reformulated and described based on the experimental field, the 16 experimental plots, and the 80 subplots, as shown in Figure 1 (lines 108-110). ]

Comments 5: [Figure 1: The text inside the images is too small and difficult to read. Please increase the font size. Make sure to check other figures as well. ]

Response 5: [The text size in Figure 1 has been adjusted, and the text in the other charts has been checked and modified accordingly (Figure 1). ]

Comments 6: [Lines 150-151: It is not clear whether the AA10 model refers to the UAS or to the LIDAR technology produced by CHN company. Please provide more details about both the UAS and the LIDAR models used in the survey. ]

Response 6: [The relevant elements have been revised. A multirotor UAS equipped with a LiDAR surveying system (model: AA10) was used to collect LiDAR point cloud data of the wheat canopy. The AA10 system is manufactured by Shanghai Huace Navigation Technology Co., Ltd., Shanghai, China (lines 116-118). ]

Comments 7: [Lines 158: The statement “The flight trajectory was in an ‘S’ pattern” needs more detail about why and how this pattern was chosen. ]

Response 7: [The reasons for the 'S' shaped pattern of flight trajectories and the specific implementation have been explained in detail in the text (lines 124-126). ]

Comments 8: [Lines 163-166: The dataset used appears relatively small. ]

Response 8: [The size of the current dataset is relatively small, primarily limited by the number of experimental fields. However, the following measures were taken to ensure the reliability of the study results and minimize the impact of dataset size on model performance:

• Multi-model validation: During the modeling process, four different models—Optuna-optimized RF (OP-RF), Elastic Net regression, Extreme Gradient Boosting Tree (XGBoost), and Support Vector Machine Regression (SVR)—were used for analysis. This allowed for a comparison of the applicability of different methods and enhanced the robustness of the study through result comparison.
• Cross-validation optimization: 5-fold cross-validation combined with the grid search method was used to optimize the model's hyperparameters, improving its generalization ability and reducing the risk of overfitting due to the small dataset.
• Feature screening and correlation analysis: Before model construction, ANOVA and Pearson correlation analysis were performed on the extracted point cloud and multispectral feature parameters to identify those significantly correlated with canopy height. This step aimed to improve prediction accuracy by reducing the interference of redundant features, thus optimizing the model's stability and generalization ability.
• Multi-source data fusion: This study used the fusion of point cloud and spectral features to combine information from different data sources, improving the model's ability to estimate crop canopy height. Compared to using a single data source, the fusion of different feature types provides richer vegetation structure and spectral information, which helps mitigate the problem of insufficient data due to smaller dataset sizes.

Subsequently, we plan to further validate the research methodology on larger experimental fields to extend the model's applicability and improve its reliability under different environmental and crop conditions.]

Comments 9: [Lines 174-177: How did the authors ensure accurate and consistent overlay between the LIDAR and the multispectral (SfM-derived) digital model? Did you use RTK, Ground Control Points (GCPs), or another method? More explanation would strengthen this section. Figure 2 offers some insight, but these details should also appear in the main text.]

Response 9: [In this study, the data alignment process involves LiDAR point clouds, visible light imagery, and multispectral imagery, but does not use the Structure-from-Motion (SfM) method. Ground Control Points (GCPs) are used to ensure accurate alignment between data sources. The specific steps include GCP extraction, alignment of GCPs with visible images, and inter-image alignment. A brief overview of the alignment process is shown in Figure 2, and these steps, along with the methodology used, are described in detail in the main text (lines 149-169).]

Comments 10: [Line 180: I would recommend including a brief introduction (just a few words) explaining the pipeline points to make this section clearer.  ]

Response 10: [This section has been revised in the text, and brief statements have been added to explain the key steps in the data processing (lines 149-171). ]

Comments 11: [Line 184: Replace the “□” symbol with something more conventional (e.g., “squared”) if necessary. Also, consider whether specifying the shape is essential, since stating that you placed four GCPs and explaining their purpose may suffice. ]

Response 11: [The "□" symbol has been removed from the text, and the content has been adjusted accordingly. Regarding the description of the shape, it is clearly shown in the figure, so we do not believe it is necessary to further emphasize the specific shape (lines 152-156). ]

Comments 12: [Lines 185-187: The meaning is unclear; please give more details or consider adding a new figure.  ]

Response 12: [Changes have been made to this section (lines 152-169).]

Comments 13: [Lineverability. ]

Response 13: [Changes have been made in the text (lines 152-169). ]

Comments 14: [Line 234: How did the authors establish the vegetation index threshold ? ]

Response 14: [The vegetation index threshold method used here is based on the spectral reflectance properties of the red (Red), green (Green), and blue (Blue) color channels in the raw point cloud data. It leverages the absorption properties of plants, which strongly absorb red light, weakly absorb green light, and absorb blue light to an intermediate degree. Vegetation and non-vegetation are differentiated by evaluating the reflectance intensities of these color channels (lines 196-206).]

Comments 15: [Line 253: This step lacks clarity. Please enlarge Figure 3 and improve its quality so that the information is easier to interpret. ]

Response 15: [Changes have been made to Figure 3 (line 234).]

Comments 16: [Line 260: As suggested earlier, provide a clearer subdivision of the experimental field (e.g., Experimental field, plot, subplot).  ]

Response 16: [Figure 3 has been modified to more clearly delineate the concepts of the experimental field, experimental plots, and subplots. ]

Comments 17: [Line 271: How did the authors determine the vegetation index thresholds?  ]

Response 17: [This section has been revised. The threshold ranges for determining individual vegetation indices in the text were initially set based on the descriptions in the references for vegetation areas. Subsequently, the thresholds for specific vegetation indices such as NDVI, EVI, and NDWI were determined in the experimental plots of this study through direct observation and analysis of different vegetation types, in conjunction with the descriptions in previous studies (lines 246-253). ]

Comments 18: [Line 280: What is the orientation (length direction) of the experimental field? Please indicate it clearly in the figures. ]

Response 18: [Line 106 of the text mentions that the dimensions of each test plot are 7m x 1.4m (length x width). Therefore, the lengthwise direction refers to the 7m dimension and has been clearly labeled in Figure 1.]

Comments 19: [Lines 288-299: How were the geometric point cloud-derived features measured in practice? Providing references for each variable would help readers understand how the analyses were conducted and would increase the method’s repeatability. ]

Response 19: [Geometric features extracted from the point cloud are used, rather than those obtained from physical measurements. Therefore, the calculation of feature variables is based on the geometric analysis of the point cloud data, not on physical measurements. We have further clarified the relationship between point cloud geometric features and canopy height in the paper and cited relevant literature to support the analytical approach (Table 2) (line 276). ]

Comments 20: [Lines 317-319: Which statistical test was employed? What was the research hypothesis? ]

Response 20: [In this study, one-way ANOVA was used for statistical testing. The null hypothesis (?0) stated that there was no significant difference in the mean canopy heights across the levels of the different characteristic parameters. The alternative hypothesis (?1) posited that there was a significant difference in the mean canopy height at least at one level of the characteristic parameter. Statistical analysis was conducted using a significance level of ? = 0.05, with significance assessed by p-value: p ≤ 0.05 indicated significance (*), p ≤ 0.01 indicated high significance (**), and p ≤ 0.001 indicated extreme significance (***). The null hypothesis was rejected if the p-value was less than the significance level, indicating that the characteristic parameter had a significant influence on canopy height. For Pearson correlation analysis, the hypothesis testing was as follows: ?0 stated that there was no significant linear correlation between two characteristic parameters (i.e., ρ = 0), and ?1 stated that there was a significant linear correlation between the two parameters (i.e., ρ ≠ 0). Significance was assessed based on the p-value, where p ≤ 0.05 indicated a significant linear correlation between the two variables (lines 294-297). ]

Comments 21: [Lines 331-333: Please explain the process behind selecting these models and clarify why the proposed models were chosen. Readers often look for the rationale behind choosing specific analytical models or tests. Being transparent about your reasoning can significantly strengthen the paper.  ]

Response 21: [In processing the data, this study referred to several methods mentioned in the existing literature, including multiple linear regression, ridge regression, partial least squares regression, multilayer perceptron regression, decision tree regression, and over ten other models, all of which were trained and evaluated. Due to space limitations, we ultimately selected four representative models for detailed analysis, based on model performance, computational efficiency, and interpretability.]

Comments 22: [Table 4: There is a typographical error in the last column.  ]

Response 22: [This issue has been revised. ]

Comments 23: [Lines 436-439: It is not clear where the joint estimation, based on both point cloud data and multispectral indices, of canopy height is presented. Please specify which table contains this information.  ]

Response 23: [The results of fitting the canopy height model, constructed by fusing point cloud features and multispectral feature parameters, are shown in lines 368-393 and in Table 7. ]

Comments 24: [Lines 440 and 443: Insert a reference to the relevant table. ]

Response 24: [The relevant table has been inserted (line 404).]

Comments 25: [Conclusions: I recommend reorganizing this section so that all critical conclusions are stated in a coherent, discursive format rather than in bullet points. Additionally, include a concise “take-home” message that highlights the main findings. ]

Response 25: [The conclusions have been revised (lines 468-489). ]

 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Winter Wheat Canopy Height Estimation Based on the Fusion 2 of LiDAR and Multispectral Data

Round 2

Review

Authors improved the general quality of the manuscript and addressed to all issues, but there are still small improvements to be made.

Specific comments

Line 21: Do not forget to add the acronym after defining the “unmanned aerial system”.

Lines 107-108: you changed the measurement unit form “plants/ha” to “plants/hm²”, I suggest keeping the first one: “plants/ha”.

Lines 116-118: it is still not clear if the AA10 is the UAS or LIDAR model’s name. Please define them.

Line 149: DJI terra is a structure form motion software used to align 2D photos in order to obtain a 3D object/orthomosaic.

Lines 202-206: it is clear what a VI is, what is not reported is how the authors determined the threshold.

Response 21 referring to lines 331-333 of Round 1 manuscript: I agree with the authors that it is better to keep a concise writing, but it is essential to report this information you gave me in your response and to clarify how you decided which models were the more representative.

Conclusions: I suggest the authors to also quickly highlight the critical aspects of their work and future challenges.

 

Comments for author File: Comments.pdf

Author Response

Thank you for your valuable comments and suggestions. We have made changes based on the feedback and have made a second round of updates to the manuscript.

Comments 1: [Line 21: Do not forget to add the acronym after defining the "unmanned aerial system".]   

Response 1: [The relevant elements have been revised (line 21).] 

Comments 2: [Lines 107-108: you changed the measurement unit form "plants/ha" to "plants/hm2", I suggest keeping the first one: "plants/ha".]   

Response 2: [Changes have been made to this section (lines 108-109).]

Comments 3: [Lines 116-118: it is still not clear if the AA10 is the UAS or LIDAR model's name. Please define them.]     

Response 3: [Additions and revisions have been made to the relevant elements of UAS and LiDAR (lines 117-118).]

Comments 4: [Line 149: DJI terra is a structure form motion software used to align 2D photos in order to obtain a 3D object/orthomosaic.]    

Response 4: [The narrative of the relevant elements has been optimized and revised. In this study, DJI Terra was used for image alignment of the UAS multispectral data. After alignment, radiometric correction was performed by comparing the reflectance values of 25%, 50%, and 75% standard reflectance calibration plates with the corresponding regions in the multispectral images. The corrected image data were then used to complete the stitching of the multispectral images (lines 147-152).] 

Comments 5: [Lines 202-206: it is clear what a VI is, what is not reported is how the authors determined the threshold.]   

Response 5: [This section has been revised (lines 208-210).]  

Comments 6: [Response 21 referring to lines 331-333 of Round 1 manuscript: I agree with the authors that it is better to keep a concise writing, but it is essential to report this information you gave me in your response and to clarify how you decided which models were the more representative.   

Response 6: [This has been revised in the text (lines 309-318).]

Comments 7: [Conclusions: I suggest the authors to also quickly highlight the critical aspects of their work and future challenges.

Response 7: [This section has been revised as suggested (lines 499-511).]

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

I accept the manuscript in present form.