A Clustering–Connection Algorithm for Coarse Root System Architecture Reconstruction Based on Ground-Penetrating Radar

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,
Please find the report below, according to the provided guidelines.

1. Summary of the paper
The manuscript presents a clustering-connection (CC) method for reconstructing plant root system architecture (RSA) using Ground-Penetrating Radar (GPR). The proposed method addresses existing limitations in GPR-based RSA reconstruction by improving adaptability to survey line layouts, reducing reliance on extensive parameter settings, and enhancing automation. The accuracy and validity of the CC method are evaluated through simulated and field-measured data, showing significant improvements in root system reconstruction. The paper contributes to non-invasive root system studies and ecological parameter extraction.
The study presents an innovative and practical method for RSA reconstruction that enhances GPR-based analysis. What I find more relevant is that the CC method is adaptable to different survey line arrangements, making it stronger than existing techniques. The manuscript provides extensive quantitative validation using both simulated and field-measured data. The methodological framework is also well-structured and logical, aiding reproducibility

2. General concept comments
I find some weak parts that should be improved.
Improvement 1 : Methodological justification
The CC method improves RSA reconstruction, but the reasoning behind certain parameter choices (e.g., clustering parameters, depth thresholds) should be further explained. Consider discussing how these parameters were selected and if they have been tested under varying soil conditions to ensure robustness. Sensitivity analyses could be conducted to assess the impact of different parameter values on reconstruction accuracy, and comparisons to established benchmarks would help validate the choices made. For example, the choice of the minimum cluster size and depth threshold could be better justified by providing empirical or theoretical support from earlier studies. Sensitivity analyses on these parameters would also strengthen the validity of the method.
Improvement 2: Comparison with existing methods
Some comparative results are provided, but I would recommend a more detailed discussion comparing the CC method with alternative approaches (e.g., Wu et al. 2014, Fan et al. 2022) to  strengthen the argument for its superiority. Specific aspects such as computational efficiency, adaptability, and accuracy should be explicitly compared through quantitative metrics or case studies.
Improvement 3: Root system complexity consideration
The study primarily focuses on coarse root systems. The limitations regarding root overlap, fine root detection, and potential applications to more complex root architectures should be acknowledged in greater detail. A discussion on whether the method can be extended to finer roots or denser root networks would provide useful insights.
Improvement 4: Figures
Some figures (e.g., Figures 10–12) should improve clarity. I would recommend using higher resolution and better contrast to improve interpretability. Adding a discussion on how GPR signal quality varies with soil conditions and root density would be beneficial, as these factors significantly influence data accuracy and visualization. Annotations or zoom-in views on critical regions of interest could also make the figures more informative and accessible to readers unfamiliar with GPR data.


3. Specific comments
Introduction & Literature Review
Improvement 5: The discussion on existing GPR-based RSA methods lacks a critical evaluation of their shortcomings. Consider elaborating on why specific methods fail in certain scenarios.
Improvement 6: The importance of non-invasive RSA studies is well-stated, but adding a brief comparison of GPR to other techniques (e.g., X-ray tomography, MRI) would enhance the introduction.

Materials and Methods
Improvement 7:  Justification for setting min_cluster_size = 2 in HDBSCAN should be provided based on past literature or empirical testing.
Improvement 8:  Figure 3: The preprocessing steps for GPR data are outlined. However, a brief discussion of potential signal noise and how it impacts root identification would be beneficial.

Results & Discussion
Improvement 9:  Table 1: It would help to include confidence intervals or standard deviations for accuracy metrics to assess variability in performance.
Improvement 10:  The discussion of RSA continuity issues in field data is important. Have alternative interpolation or smoothing techniques been considered to mitigate discontinuities?
Improvement 11:  The manuscript suggests that the CC method is applicable across various survey line layouts. Have any tests been conducted with alternative layouts beyond circular and grid arrangements?

Conclusion
Improvement 12:  The conclusion summarizes the study’s contributions but I recommend to briefly mention future work directions, such as integrating multi-frequency GPR or applying machine learning for enhanced pattern recognition.

4. Overall recommendation
Major revision. The manuscript presents a valuable contribution to remote sensing applications in forestry, but revisions are necessary to strengthen its methodological justification, comparative analysis, and discussion of limitations. With these improvements, the paper will be well-suited for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Manuscript Title: A Clustering - Connection Algorithm for Coarse Root System Architecture Reconstruction based on Ground-Penetrating Radar

1. Try to improve your abstract section with a more result-oriented analysis.
2. The concluding part of the study is missing in the abstract section. 
3. In detail, the significance of the study is missing in the introduction. The introduction section must have a background, significance, and novelty of the study. 
4. What is the novelty of the study? define in the last section. 
5. Limited information on the study area. 
6. The method section is understandable to the reader. 
7. Try to discuss more recent analyses in the discussion section. 
8. Add some future research direction to the manuscript. 
9. What is the social applicability of this study? 
10. The conclusion section needs a more in-depth concluding part. 
11. This study is not important for the forest-related analysis. Also, needs extensive revision otherwise scientific soundness is very low. Therefore, based on the current analysis, I must recommend a major revision of this manuscript. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript presents a clustering-connection method for the reconstruction of coarse  root system architecture (RSA) using  Ground-penetrating radar (GPR). The article is in good shape and could be accepted for possible publication after addressing the comments given below:

Abstract: The abstract is very weak and it seems like introduction section. The authors should revised abstract and clearly mentioned main results of study and its significance.

Materials and Methods: The authors should explain about study area in section 2.1.2 Field measurement data first, then explain simulated Root system data.

The authors should explain YOLOv4 in details in separate section.

In section 2.2, he authors should cite some related studies about clustering-connection method.

Figures 3 and 4 could be merged to make an interactive flowchart.

There are too many figures, authors should reduce number of figures.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,
You have adequately addressed all reviewer concerns. Your responses improve the manuscript’s clarity, justification, and discussion of limitations. While minor enhancements (i.e., mentioning resolution differences in Q6, referencing interpolation techniques in Q10) could further strengthen the responses, these are not critical.
Please find below a response point by point to your comments:

Q1: Methodological justification
The authors have provided a well-structured sensitivity analysis for the key parameters (min_cluster_size and depth threshold). They explain the inverse relationship between parameter magnitude and classification outcomes and justify the choice of min_cluster_size = 2 with empirical data and references. 
The response is enough. The justification is clear, and sensitivity analysis adds robustness. However, a brief discussion of how different soil conditions (e.g., sandy vs. clayey soils) may affect optimal parameter selection would further improve the section.

Q2: Comparison with existing methods
The authors have expanded their comparative analysis, incorporating explicit accuracy metrics (R values) for the CC method versus Wu et al. (2014) and Fan et al. (2022). 
The response sufficiently addresses the reviewer's concerns. While the exclusion of computational efficiency comparison is reasonable due to manual steps in the other methods, a qualitative statement on the computational workload of the CC method would further strengthen this section.

Q3: Root system complexity consideration
The revised manuscript now acknowledges the challenges posed by highly complex root architectures, including overlapping root signals in GPR data. The authors explain that while the CC method improves accuracy, connectivity issues may persist in dense networks.
The response is appropriate. Adding this limitation improves transparency. If possible, the authors could briefly suggest potential ways to adapt their method for finer root networks (e.g., hybrid clustering techniques, higher-frequency GPR).

Q4: Figures
The authors improved the resolution of Figures 10 and 11 and restructured Figure 12 to highlight critical differences using boxed annotations. 
No further revisions are necessary.

Q5: Critical evaluation of existing GPR-based methods
The expanded introduction now discusses the limitations of earlier methods, highlighting issues such as survey line dependency, reliance on root attributes, and parameter tuning challenges.
No further changes are required.

Q6: Comparison of GPR with other techniques
The authors have added a discussion on how GPR compares to techniques like X-ray tomography and MRI, emphasizing GPR’s advantages for field-based, non-invasive detection of coarse roots. The references provided support this distinction.
The response is enough. The added section differentiates GPR from other methods. However, briefly mentioning the spatial resolution differences among these techniques would provide more clarity.

Q7: Justification for min_cluster_size = 2
The authors have now justified setting min_cluster_size = 2 based on earlier research (Cui et al., 2021) and empirical testing.
The response is appropriate. No further changes are needed.

Q8: Discussion of GPR signal noise and root identification challenges
The response addresses issues like irregular and overlapping hyperbolas, as well as noise interference in GPR data, which can lead to missed root points. The discussion now acknowledges errors in B-scan processing and the need for further improvements.
The response adequately expands on the challenges posed by signal noise. No further revisions are necessary. 

Q9: Confidence intervals or standard deviations for accuracy metrics
The authors explain that due to the limited dataset (single simulated and measured dataset), confidence intervals and standard deviations were not included. They acknowledge the importance of these metrics for future studies.
This is a reasonable justification. 

Q10: Addressing RSA discontinuity through interpolation or smoothing
The authors acknowledge that discontinuities arise due to limitations in GPR detection and measurement errors. They propose a post-processing method to identify "short roots" and use interpolation to improve continuity.

The response effectively addresses the issue and proposes a clear future direction. Referencing interpolation techniques from related research would further support this proposal.

Q11: Adaptability of CC method to other survey line arrangements
The authors clarify that only circular and grid survey lines are used in GPR-based root surveys. However, they argue that the CC method’s clustering and connection steps are independent of survey line arrangements and should be adaptable to other layouts.
The response is reasonable. 

Q12: Future work directions in the conclusion
The authors have expanded the conclusion to discuss future directions, including multi-frequency GPR image fusion, deep learning for noise reduction, GPS-integrated survey automation, and cross-species validation.
The response is excellent. It outlines clear and relevant future research directions. No further changes are necessary.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The authors are trying to improve the manuscript, but still, some minor comments are not addressed in the revised version. 

1. Write one line background of the study in the abstract section.
2. The limited result section needs more detailed analysis. 
3. The conclusion section needs more attention with the concluding part of the study. 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have improved manuscript by incorporated my comments. I think manuscript is ready for publication.

Author Response

We sincerely appreciate your constructive feedback, which has enhanced the manuscript's readability and underscored the importance of concise expression in refining our scientific narrative.