Rapeseed Yield Estimation Using UAV-LiDAR and an Improved 3D Reconstruction Method

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear author,

The manuscript presents a solid and innovative methodological framework for volume modeling and yield estimation using UAV-LiDAR and the HybridMC-Poisson algorithm. The originality of the approach and its technical rigor are evident. Below are several recommendations aimed at broadening the scope and enhancing the scientific impact of the work.

The study is currently based on a single sampling site, which limits the demonstration of the model's generalizability. It would be valuable to consider including trials in multiple fields or geographic locations to validate the model under diverse agronomic conditions. Additionally, evaluating yield performance under different planting density regimes, while optional, could provide further evidence of the model’s practical applicability.

Furthermore, the current analysis focuses on structural data obtained via UAV-LiDAR. Recent literature emphasizes that combining these data with spectral information (such as vegetation indices derived from multispectral or hyperspectral sensors) significantly improves the accuracy of biomass and yield estimation. Integrating structural, textural, and spectral data could help capture key physiological traits that volume alone may not fully reflect, especially at later crop growth stages.

Since the HybridMC-Poisson algorithm is one of the core contributions of the manuscript, a more detailed description is recommended. It would be useful to elaborate on how the algorithm handles noise and variable point cloud density, and how its performance (in terms of accuracy and runtime) compares with other current methods such as NeRF or machine learning-based biomass estimation approaches. The discussion on point cloud stitching efficiency could also be expanded, as this remains a key bottleneck in UAV-LiDAR applications.

Regarding data and code availability, although the manuscript states that they are available upon request, leading remote sensing journals increasingly require open data and code as part of standard transparency and reproducibility practices. Publishing raw datasets and source code on platforms such as GitHub or Zenodo would reinforce the credibility of the study and facilitate its adoption by the broader research community.

Lastly, while the introduction appears to reference classical 3D reconstruction methods appropriately, it could benefit from the inclusion of recent studies on deep learning architectures applied to plant phenotyping. Models such as NeRF, specialized CNNs, or point cloud completion networks are driving a shift in how complex crop architectures are modeled. In addition, many yield prediction studies now implement multi-year, multi-location, or multi-phenological stage validation schemes, which could also be cited to strengthen the scientific context.

This work has high potential, and with these refinements, it could reach a broader audience and achieve greater impact within the fields of remote sensing and agricultural phenotyping.

Author Response

Response to Reviewer 1 Comments
1. Summary
We sincerely appreciate your time and effort in reviewing our manuscript. We are grateful for the valuable and constructive comments provided, which have greatly helped us to improve the overall quality of our work. We acknowledge that some issues raised in the review were due to our oversight during the initial submission. In this revised version, we have carefully addressed all the comments point by point and made the corresponding revisions throughout the manuscript. We believe that these changes have significantly enhanced the clarity, rigor, and overall presentation of the paper. We truly appreciate the reviewers’ insightful suggestions and the editor’s consideration, and we hope that the revised manuscript now meets the publication standards of the journal through our joint efforts. The revised manuscript has been uploaded as an attachment for your kind review.
2. Questions for General Evaluation	Reviewer’s Evaluation	Response and Revisions
Does the introduction provide sufficient background and include all relevant references?	Can be improved	Improved as suggested.
Are all figures and tables clear and well-presented?	Must be improved	Improved as suggested.
Is the research design appropriate?	Can be improved	Improved as suggested.
Are the methods adequately described?	Can be improved	Improved as suggested.
Are the results clearly presented?	Yes
Are the conclusions supported by the results?	Yes
3. Point-by-point response to Comments and Suggestions for Authors
Comments 1: The study is currently based on a single sampling site, which limits the demonstration of the model's generalizability. It would be valuable to consider including trials in multiple fields or geographic locations to validate the model under diverse agronomic conditions. Additionally, evaluating yield performance under different planting density regimes, while optional, could provide further evidence of the model’s practical applicability.
Response 1:Thank you for this valuable comment. The experimental site in Dongloushan, Yi County, was selected as it represents a typical winter rapeseed cultivation area in northern China, characterized by representative topography, soil fertility, and management practices. This ensures that the methodological validation of the proposed framework was carried out under highly representative agronomic conditions. To address this comment, we have added a paragraph in the Discussion (Future Work section, p. 17,lines 501-506) emphasizing that, although the current site is representative, future studies will include multi-site and multi-density trials to verify the model’s generalizability under diverse agro-ecological environments.
Comments 2:The current analysis focuses on structural data obtained via UAV-LiDAR. Recent literature emphasizes that combining these data with spectral information (such as vegetation indices derived from multispectral or hyperspectral sensors) significantly improves the accuracy of biomass and yield estimation. Integrating structural, textural, and spectral data could help capture key physiological traits that volume alone may not fully reflect, especially at later crop growth stages.
Response 2: We sincerely thank the reviewer for this insightful suggestion. Our team’s current platform and expertise are centered on UAV-LiDAR–based 3D reconstruction, which provided the foundation for this study. We fully agree that integrating spectral information can complement structural features and enhance yield-estimation accuracy, especially at later growth stages. Accordingly, we have added a statement in the Future Work section of the Discussion (p. 16-17,lines 495-500) outlining the planned integration of multispectral and hyperspectral data to achieve structure–spectrum fusion for more comprehensive crop characterization.
Comments 3: Since the HybridMC-Poisson algorithm is one of the core contributions of the manuscript, a more detailed description is recommended. It would be useful to elaborate on how the algorithm handles noise and variable point cloud density, and how its performance (in terms of accuracy and runtime) compares with other current methods such as NeRF or machine learning-based biomass estimation approaches. The discussion on point cloud stitching efficiency could also be expanded, as this remains a key bottleneck in UAV-LiDAR applications.
Response 3: Thank you for this valuable comment. The description of the HybridMC–Poisson algorithm has been expanded in Section 2.3 Single-Plant 3D Reconstruction Algorithm(p. 6-8,lines 212), with added details on the adaptive density-trimming mechanism and rigid-registration fusion used to handle noise and point-density variation. Regarding comparisons with deep-learning approaches such as NeRF, these models require large, multi-site datasets beyond the scope of this study; however, we plan to explore such comparisons in future work as data accumulation increases.
Comments 4: Regarding data and code availability, although the manuscript states that they are available upon request, leading remote sensing journals increasingly require open data and code as part of standard transparency and reproducibility practices. Publishing raw datasets and source code on platforms such as GitHub or Zenodo would reinforce the credibility of the study and facilitate its adoption by the broader research community.
Response 4:We sincerely thank the reviewer for stressing this essential aspect. We fully support open-science principles. Because the manuscript is still under review, we have created a private repository that contains the core processing code, a minimal example dataset, and relevant documentation. Upon acceptance, we will make this repository public and archive a release on Zenodo to ensure long-term accessibility and assign a DOI.
Comments 5: while the introduction appears to reference classical 3D reconstruction methods appropriately, it could benefit from the inclusion of recent studies on deep learning architectures applied to plant phenotyping. Models such as NeRF, specialized CNNs, or point cloud completion networks are driving a shift in how complex crop architectures are modeled. In addition, many yield prediction studies now implement multi-year, multi-location, or multi-phenological stage validation schemes, which could also be cited to strengthen the scientific context.
Response 5: We sincerely thank the reviewer for this constructive recommendation. We have updated the Introduction to include a concise overview of recent deep-learning-based plant phenotyping and canopy reconstruction methods, such as NeRF, CNN-based trait extraction, and point-cloud completion networks. In addition, we cited representative multi-year and multi-location yield-prediction studies to strengthen the scientific context. Revisions made: Added new references and text summarizing recent advances in deep-learning approaches for 3D crop modeling (Introduction, p. 3,lines 113-128). Added corresponding discussion linking these works to our future large-scale validation plans (Discussion, p. 17,lines 501-506).
4. Response to Comments on the Quality of English Language
Point 1:The English is fine and does not require any improvement.
Response 1:The revised sections have been carefully polished to ensure precise and professional academic language throughout.
5. Additional clarifications
We have also improved the figure resolution and consistency of units throughout the manuscript. All changes have been marked in Track Changes mode for clarity.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript presents a well-validated UAV-LiDAR reconstruction framework using the HybridMC–Poisson algorithm for rapeseed canopy modeling. The authors have significantly strengthened the study by providing comprehensive quantitative benchmarking (volume error 2.96 %, Hausdorff = 1.88 cm, Chamfer = 1.05 cm, runtime ≈ 3 s) across multiple algorithms. These results confirm a clear engineering improvement and strong applicability to precision agriculture.

To further enhance the clarity and practical value of the paper, I recommend minor revisions in the following areas:

1. Clarify that the HybridMC–Poisson is an applied integration of existing reconstruction algorithms, not a new theoretical method.
2. Provide details on the hardware setup and specify whether the reported runtime refers to a single plant or batch process.
3. Clearly describe the statistical test used for the reported significance.
4. Add a short paragraph discussing limitations and future extensions (e.g., crop types, LiDAR density sensitivity).
5. Contextualize the results agriculturally—explain how improved reconstruction accuracy supports field management (e.g., yield estimation precision, fertilizer planning).
6. Improve figure resolution and labeling, and perform light English polishing.

Overall, this is a solid applied study with practical relevance to precision farming and remote sensing. After addressing these minor issues, the paper will be suitable for publication in Agriculture (MDPI).

Comments on the Quality of English Language

The paper does not require major rewriting; it is linguistically acceptable after minor editing for grammar, conciseness, and consistency in variable notation and units. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript is devoted to the current scientific topic of "Rapeseed Yield Assessment Using UAV-LiDAR and Improved 3D Mapping," as remote yield assessment using non-contact methods is currently very important for forecasting rapeseed yields.
The article thoroughly examines the challenges of existing optical remote sensing methods and the potential for improving the accuracy of biomass and yield assessment under dense canopies during the flowering and maturity phases of rapeseed. To address this issue, the authors propose a hybrid algorithm (HybridMC-Poisson), which combines the advantages of two methods: Marching Cubes (MC) for preserving sharp boundaries and details, and Poisson reconstruction for creating globally smooth and impermeable surfaces.
However, the article has a number of caveats:
1. The description of the HybridMC-Poisson algorithm is insufficiently detailed. Although the general workflow is described, the mathematical calculations are presented fragmentarily.
2. Statistical analysis for comparing the algorithms is lacking. Table 3 and the text provide mean error values, but there are no statistical test results (e.g., t-test, ANOVA) that would confirm that the superiority of HybridMC-Poisson over other methods is statistically significant and not due to chance.
3. Small sample size for plant-level validation. Only 20 plants were used for validation and comparison of 3D reconstruction algorithms. A larger sample size is desirable for reliable statistical inferences, especially when comparing multiple methods.
4. The authors compare their method only with three classical reconstruction algorithms (Ball-Pivoting, Poisson, and Alpha-Shape). There is no comparison with more modern approaches mentioned in the introduction (e.g., neural network-based approaches such as 3D Gaussian Splatting), which prevents a full assessment of its competitiveness.
5. The model linking plant volume to yield is a simple linear regression. Although it yielded good results, this approach does not account for other potentially important factors, such as planting density, phenotypic traits (pod number, height), or environmental variables, which could improve accuracy.
The comments presented primarily concern the presentation and methodological depth. The research itself appears valuable and has clearly defined objectives, but requires revision before publication in a peer-reviewed journal.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Reviewer's Notes

Substantive comments

1. In the introduction, please check the correctness of the literature citations (e.g. lines 96, 113).
2. Figures 1 and 2 do not add any value to the article (they are illegible), they can be removed.
3. The methodology should include formula numbers and a description of all formula components. Are these the authors' own formulas? If they are from the literature, the source must be cited.
4. The methodology needs to be supplemented with the number of plants collected for volume analysis and the number of dried plants used to develop regressions. Is this the information in line 359?
5. The results include fragments of the discussion (e.g., lines 364-368; 372; 382-383). Move to the discussion and expand .
6. In the discussion, the material concerning Figures 7 and 8 in Chapters 4.2 and 4.3 is a further part of the results and analyses described in the purpose and scope of the research, and not a discussion.
7. The discussion chapter requires significant improvement (only three literature items were cited) consisting in an objective confrontation of the obtained results with widely published research on this topic..

 

Technical Notes

This article requires extensive proofreading:

1. Citing literature in the text should be prepared in accordance with the journal's requirements.
2. Correct numbering of literature items (e.g. see lines 96 and further), the whole requires changing the numbers.
3. Unclear text e.g. lines 161, 182 etc. to be deleted.
4. Completely incorrect numbering of tables and figures, which makes the article difficult to read, and some tables and figures are not referenced in the text.
5. Descriptions of tables and figures (text, font, etc.); illegibility of drawings and descriptions on drawings.
6. The description of the literature item in the references (abbreviations, font, bold) is inconsistent with the publisher's guidelines; check compliance with the information provided in the text.
7. Latin names of plants should be written in italics.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf