Theoretical Development of Plant Root Diameter Estimation Based on GprMax Data and Neural Network Modelling

Round 1

Reviewer 1 Report

Review on “Research on plant root diameter estimation based on GprMax and neural network”

 

This paper describes the in-silico design and testing of a modelling approach for the estimation of root diameter in ground penetrating radar data.  The paper is much improved from the first detail, and now comprises a logical progression that can be understood in the main by non-mathematical  / GPR users.   I would suggest the authors review the title.  Perhaps something along the lines of “Development of plant root diameter estimation based on GprMax data and neural network modelling”

I also have the following suggestions.

Abstract “Time interval” specify what the “time interval” refers to.

This study verified the feasibility and effectiveness of in-situ non-destructive estimation of plant root diameter parameters using GPR   - I disagree with this statement.  I would say it demonstrates the potential, feasibility to me is something that has to be demonstrated in a practical way rather than in silico.

Line 27 :  “the circulation of terrestrial ecosystems” of what within terrestrial ecosystems

Line 49: 3D laser scanning method [11,12] – These references seem inappropriate here as they deal with tree stems rather than roots.  I suggest instead methods such x -ray ct and  MRI methods used particularly in root research are mentioned (and referenced appropriately)

Line 87 “tent, soil clay content and so on”, add references.

Line 124: Paraview – Reference….

Line 126: “Figure 1 simulates the vertical” Figure 1 shows a simulation of the vertical…..the graphics do not simulate the show a simulation.  This needs to be corrected throughout the manuscript

Figure one needs some sort of axis labelling and generally all the Figure legends need more detail.

Lines 132 : “The roots were scanned by moving the antenna to the right using a pair of small rectangular analog GPR transmit and receive antennas located at the surface” Are these represented by the tiny grey and black dots in Figure 1 ? If so would be worth pointing this out.

Line 137  “by scanning”  - This should really read “by simulating scanning of”

Line 138 : “due to the difference of dielectric properties between root and soil, the electromagnetic wave was partially reflected at the interface between root and soil”   it will be clearer if the part of this sentence before the comma is swapped with the part after the comma.  The change “and the rest continued to” to “and the rest of the wave continued to”

Line 221 I find the title Experimental data acquisition a strange use of language.  Perhaps : - “Assessment of model input parameters“

Line 443 “the diameter of 90 actual roots”, while this is within the context of the manuscript the statement still leads the reader to get the impression that you have swapped to real physical roots, I suggest this be changed to the “root input diameter”.  

A suggestion for an improvement to the Figure 13 would also be a graph of the residual as a percentage of the root input diameter.

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Reviewer 2 Report

The mathematical method should be improved

"GprMax is an open source software that can simulate electromagnetic wave propa-119 gation for GPR."
"A center frequency of radar 252 was set to 1.5 GHz."
Observation: for this frequency, the wave length is 0.2metres.  
 "For each effective combination 250 of the root and soil dielectric constants, seven simulated cylindrical roots were set, with a 251 radius of 5 mm beginning and increasing at 3 mm intervals."  What about the ddiffraction phenomenon?
"Any object that has dimensions comparable to the wavelength will be bypassed from where" (Jean Augustin Fresnel -1819)
The conceptual setup is wrong!

on the other hand, the solution (10) for equation (8) is WRONG. The solution for the second order least square method is in the attach

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Reviewer 3 Report

This is an interesting study evaluating the suitability of ground penetrating radar for root diameter estimation with a least squares regression and a neural network in a simulation-based approach with GprMax. Technological progress has broadened the set of methods in forest ecosystem research, thus researchers more and more focus to develop and optimize non-destructive methods such as ground penetrating radar. To facilitate optimization of such methods, simulation model studies make it possible to test them for a wide range of influencing parameters. Therefore, this study matches the scope of Forests. However, the authors especially need to clarify the difference between model assumptions and results and work on the general structure of the manuscript. In the following, I provide comments on major and minor issues regarding these aspects.

 

Major issues:

• Considering what the really new findings from this study are, I find the first aim of the study quite problematic (Lines 108-110). This part is more about the design and assumptions of GprMax in terms of the application for root diameter estimation. The results presented in Chapter 3.1 are not really “new results”, it is rather a consequence of how GprMax works and what parameter combinations you assume to be realistic for root diameter estimation. I strongly recommend to clearly distinguish between the assumptions and the results of your model. Therefore, I suggest that you focus on the second and third aim of your study (Lines 110-113) and shorten Section 3.1. Accordingly, in the Conclusion section, Lines 537-540 should be omitted.
• I find it hard to understand what datasets were used for what purpose. In the Methods section, you present model training data (Section 2.4.1) and model test data (Section 2.4.2). However, in Section 3.2.2., you mention that even model training data consisted of training samples, validation samples and test samples (Line 417). Please give a clear overview in the Methods section which datasets you used as training data and which as validation or test data and indicate whether this was different between the methods (and if so why).
• In Section 3.1, you present some Results where I miss the related parts in the Methods section. In general, I recommend to shorten Section 3.1 (see first Comment), but in case you decide to present results such as in Lines 375-390 (including Fig. 9 and Equation (17)), you need to give details on this in the Methods section (perhaps between Sections 2.3 and 2.4). Furthermore, please check that all information related to methods is given in the Methods section, whereas the Results section only presents results. For example, Lines 416-422 belong to the description of methods.
• Please revise your text (especially Abstract and Conclusion sections) so that it is always clear and easily understandable for the reader that your study is based on model simulations, not based on field experiments.
• Please add a discussion to your manuscript. In the current state of the manuscript, what you call “Discussion” (Section 4) consists only of description of new results. This needs to be integrated into the Results section. The Discussion section needs to cover an integration of your results with other studies from the latest literature. Some interesting aspects to discuss should be the novelty of your research (i.e. what do your findings add to the relevant literature), the application of GPR and your model for field research, and critical discussion of your methods choice. I am sure you can think of much more interesting aspects that you find relevant to discuss based on your results.
• You need to work on the general structure of the manuscript. Some parts of the manuscript need to be shortened. For example, Lines 69-103 should be substantially shortened and only the information relevant for your study should be mentioned here. Maybe, you can utilize some sentences for the discussion. In the Methods section, you should only keep information from Sections 2.1 and 2.2 that is most relevant for your research aim, as most of this theoretical basis is more about the software design. I suggest to shift large parts of Sections 2.1 and 2.2 as additional information to a supplementary document (Appendix), including Figures 1-4. Moreover, I suggest to omit Lines 276-290 and 321-332 by providing references where those basic statistical methods are explained. If you like, you could also keep those text parts as a further supplementary document in the Appendix. In the Results section, Section 3.1 should be substantially shortened (or partly shifted to the Methods section or Appendix), as this presentation does not provide new insights, but rather the manifestation of model assumptions. The current Section 4 does not provide a discussion, but results that need to be shifted to Section 3.
• All figures an tables must be self-standing, i.e. they need to be understandable only from the captions. This means that all information on colors, icons, abbreviations etc. need to be explained in a legend within the figure or in the figure/table caption.

 

Minor issues:

• Title: Delete “Research on”. Additionally, I suggest that you integrate that your study is a simulation-based study. “Plant root diameter estimation” could rather point to a study with practical application in field experiments to estimate plant root diameter.
• Line 17: The phrase “the root system with increasing diameter” is hardly understandable, please reformulate.
• Lines 29-30: Please think about whether this is really the main finding from your study. You do not present field experiments, so it seems inappropriate that the main point of your study is the verification of “the feasibility and effectiveness of in-situ” estimation.
• Line 37: Be more precise: “circulation of … in terrestrial ecosystems”
• Lines 41-42: Change “protects” into “protect”, because the reference is “helps… protect”.
• Line 45: Change “methods” into “approach”.
• Lines 49 and 50: Change “Ground-penetrating” into “ground-penetrating” (twice)
• Line 58: Use correct citation, i.e. “Liang et al., Zhang et al. and Wu et al.”.
• Line 59: Change “ground penetrating radar” into “GPR”.
• Line 67: Change into “requires high resolution GPR”.
• Line 68: Change into “GPR can distinguish”.
• Line 69: Missing blank space after “[28]”.
• Line 71: Delete “and” before “introduced”.
• Line 73: Missing blank space after “[29]”.
• Lines 77 and 78: Missing blank spaces: “900 MHz” and “2 GHz”.
• Line 79: Change into “detect objects in 1 m depth and obtain”.
• Line 80: Missing blank space after “[31]”.
• Lines 80-81: Change “which obtained” into “obtaining”.
• Line 86: Delete “effect”.
• Line 87: Delete “and so on” (“include” does not mean that the list is complete).
• Line 88: Change into “is a decisive factor” (obviously, performance also depends on other factors).
• Line 89: Change “comparison” into “contrast” or “difference”.
• Lines 94-95: Change into “roots with a diameter of 2 cm buried 30 cm deep in the ground”.
• Line 97: Missing blank space before “[37]”.
• Line 99: Do you mean “3-5 cm” instead of “3 ~ 5 cm”?
• Line 101: Missing blank space after “[39]”.
• Line 110: Be more precise: What are the “two estimation models”?
• Lines 113-114: Delete “and get a more suitable model for estimating root diameter parameters” (or explain what “more suitable” means and to what reference “more” is related to).
• Line 114: Change “model” into “models” (see “two estimation models”).
• Line 132: In Figure 1, the root is ca. 7-8 cm buried in the ground.
• Line 133: The phrase “moving the antenna to the right” is not understandable here without more information on the two-dimensional design with the antenna starting on the left.
• Figure 1: Explain colors in the caption. Additionally, in the caption, delete “Examine the”.
• Figure 2: Add a label for the x-axis and explain the different lines in the caption.
• Line 151: Delete blank space after “root”.
• Lines 169-173: These information should be shifted to the caption of Figure 3.
• Lines 175-176: Delete “the color changed from … in Figure 3, this corresponds to” and add a reference to Figure 3 at the end of the sentence.
• Line 179: Start a new sentence after “in Figure 4”.
• Line 183: Delete blank space before “Figure 3”.
• Line 183: Change into “time points” (refers to T₁ and T₂).
• Figures 3 and 4: Explain T₁ and T₂ in the captions.
• Line 189: Delete “to investigate”.
• Line 193: Change into “making the root detection difficult”.
• Lines 193: Change into “Liu et al. [46] found”.
• Line 195: Change “his” into “their”.
• Line 199: Change into “Liu et al.”.
• Line 202: Delete “be” (twice).
• Line 203: What do you mean by “were significantly compared”? Please reformulate.
• Lines 211 and 214: Information on colors and abbreviations (R1, …, R15) in Figure 5 need to be shifted to the figure caption.
• Line 213: Change “rest roots” into “further roots” or “other roots”.
• Figure 5: Add labels and units to the axes. In the figure caption, explain the colors and abbreviations.
• Line 222: Delete “acquisition”.
• Line 224: Start a new sentence after “unknown”.
• Line 229: Change into “Wu et al. [48]”.
• Line 229: Delete “fitting”.
• Line 234: Change into “reference for the root”.
• Line 234: Change “constant” into “constants”.
• Lines 235-236: Change into “5 different soil dielectric constants and 10 different root dielectric constants”.
• Line 236: Change into “selected and combined”.
• Figure 6: Add a label to the color legend and explain black color.
• Line 240: Missing blank space after “soil”.
• Line 241: Missing blank space before “the y axis”.
• Line 242: Missing blank space after “root”.
• Lines 244-245: Change into “50 dielectric constant combinations of root and soil”.
• Lines 248-249: Change into “the combinations of the dielectric constants of root and soil with ?_???? < ?_???? and Δε < 3 were discarded”.
• Line 250: Missing blank space after the comma.
• Line 257: Delete “acquisition”.
• Line 261: Change into “from 7 to 24 at interval 1”.
• Line 261: Delete “sets of”.
• Line 267: Add comma before “so that”.
• Lines 269-272: Delete this part, as it is not needed.
• Line 274: Change “section” into “Section”.
• Line 276: What do you mean by “better results”? Please explain what results are better, why are they better and what is the reference for comparison?
• Line 300: Explain abbreviation “BP”.
• Line 302: Start new sentence after “neurons”.
• Line 302: Delete “its own”.
• Line 302: Change “will calculate” into “calculates”.
• Line 303: Missing blank space before “[50]”.
• Line 304: Change “was shown” into “is shown”.
• Lines 304-305: Delete the last sentence of the paragraph, as it can be seen from Figure 7.
• Figure 7: Explain letters / abbreviations and what network layers can be seen.
• Line 308: Change into “in Figure 7”.
• Line 323: Missing blank space after “error”.
• Lines 334-335: Delete these two sentences, that does not need to be explained.
• Equation 15: Explain what D is in this equation.
• Lines 351-352: Add trace numbers (x-axis values) to help the reader find the right positions in Figure 8.
• Line 360: From here on, you always use nanoseconds / ns as time unit. Before, and also in Figure 8, it is always seconds / s. Which one is correct?
• Lines 364-365: This rather seems to be a model assumption / part of the model design of root diameters in GprMax than a result.
• Table 1: Explain abbreviations in the table caption.
• Figure 9: Why do you use a quadratic function for fitting (Eq. 17) when you have a linear correlation coefficient R = 0.9996. With such a high correlation coefficient, it does not seem to be justified that you prefer a quadratic function over the more simple linear one.
• Figure 9: Explain ΔT in the figure caption.
• Line 388: In Equation 17, you have D as dependent variable and ΔT as response. So why do you use RMSE(D_L) here?
• Line 392: Delete “Estimation of”.
• Lines 393-396: These sentences belong to Section 3.1.
• Line 408: Delete “law”.
• Line 414: Change into “the overall relationship between ΔT and D is similar”.
• Line 415: Delete “Estimation of”.
• Line 416: Change “section” into “Section”.
• Line 417: Do you mean “validation samples” instead of “verification samples”?
• Line 417: What is the difference / the different purpose of the validation data and test data? From the presentation of the results, it seems that you treated them equally? Please explain in the Methods section.
• Line 417: Missing blank spaces (four times).
• Line 419: Missing blank spaces after “LM”.
• Figure 11: In the figure caption, change “curve” into “curves”. Moreover, explain abbreviations “E” and “A”.
• Lines 425-428: This information belongs to the caption of Figure 11.
• Line 432: “verification” or “validation”?
• Figure 12: x-axis labels should rather be “Number of root scans” as the total number of samples is equal to 18, i.e. the models work on n = 18 samples?
• Lines 443-446: This information belongs to the caption of Figure 12.
• Line 442: Missing blank space: “Figure 12”.
• Line 443: What do you mean by “actual roots”?
• Table 3: Delete this table, as all information except for RMSE is already in Figure 13. For RMSE, you can add one sentence in Line 461.
• Line 463: Delete “own”.
• Lines 467 and 472: Explain what you mean by “more stable”.
• Lines 473-477 can be deleted, only keep “The percentage of average prediction error is 3.62%.”
• Lines 483-486: What do you mean with this sentence? I suggest that the sentence can be deleted.
• Line 490: Maybe you want to add a reference to Figure 10 after “prediction result”.
• Figure 14: Explain colors in the figure caption and check x-axis labels (see Figure 12).
• Line 499: Delete “law”.
• Lines 500-502: This information belongs to the caption of Figure 14.
• Lines 513-515: I do not agree with this formulation, because you have only 1 mm difference between both models (meaning both models perform similarly well based on the residual span). Please reformulate.
• Line 517: Why “mainly”?
• Line 524: Change “Therefore” into “Altogether” (repetition).
• Line 532: I agree with stressing that your study is simulation-based. However, what do you mean here by “simulation detection”?
• Line 548: Change “under the condition of” into “with”.

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Round 2

Reviewer 3 Report

The authors have made good efforts in revising the manuscript. The quality of the manuscript has improved as compared to the original submission, so that it may be accepted for publication after addressing a few remaining issues:

• Abstract: In the current manuscript Abstract, you use a mixture of present and past tenses. Please check and use consistently present or past tense as much as possible.
• Methods: Please add the methods for Section 3.3.2. These simulations are missing in the current version of the Methods section, i.e. Lines 389-395 (except for the last sentence) belong to the Methods section!
• Line 162: Change into “red circles represent roots (15 roots in total, where R1, …”.
• Line 195: Change “and” into “or” because combinations are discarded, when one or the other or both conditions are met.
• Sections 2.5 and 2.6: I like that this new version of methods presentation in the manuscript is much clearer and further information can be found in the Appendix! However, you should add some references, especially for BP neural networks, but maybe also for Least Square Regressions and Performance Assessment. It would be important for the reader to find the literature, where you took the methods from and/or GPR-specific literature where such methods have already been applied (for comparison).
• Line 235: Change “weight” into “weights”.
• Line 300: Reduce blank spaces around “(Table 1)”.
• Line 379: Change into “considering that the overall residual distribution range”.
• Line 382: Missing blank space before “The root”.
• Lines 432-437: These lines belong to the discussion, as you compare and assess different results. (Probably you can delete these lines, because you have it already in the discussion.)
• Line 482: Change “potential” into “suitable”.

Author Response

Please see the attachment.

Author Response File: Author Response.doc