Annular Neighboring Points Distribution Analysis: A Novel PLS Stem Point Cloud Preprocessing Algorithm for DBH Estimation

Round 1

Reviewer 1 Report

The novelty of PLS and the new algorithm tested deserves its publication but right now the paper is under a format far from being at that level. I am encouraging the authors to undertake the listed improvements below. The description of methods is rather redundant and the present order of the methods restrict the readability of the paper. I encourage the authors to re-design the method section and be consistent with the removal of the repetitive parts. Sometimes it is very hard to follow the text due to the huge amount of vague and meaningless terms used. In the results section, the lack of assessment is especially worrying but the weird thing is that authors are presenting results in the discussion section. There is a high need to better balance the contents between these two sections. The quality of the figures must be improved and the presented workflow as well. An additional work-flow including the steps involved in the data processing scheme using Computree and Matlab would be gladly welcome in the next revised version of the research. The novelty of the paper is clear and the importance of the new PLS method is of the greatest interest. Now it is time for the authors to upgrade the scientific soundness of the paper and allow the reader to better understand the interesting findings of the paper.

 

Here I list some general comments on the work

Detailed comments:

Abstract

Compositions such as DBH estimation accuracy advantage are tough to follow. Please, ease your grammar and let do not overcomplicate the readability. What is drift error? Use common and simple words in the abstract and place the more challenging terms and concepts for the Intro and Methods. L19-L22. Is the term “fragments” really needed? L23-L24: See the high complexity of the compositions: i) iterative outermost point removal or ii) annular neighboring points distribution analysis. Please, make it easier. L25-L27: …the ANPDA. L25-L29: I would summarize the main results in just one sentence as there is no need to provide such a level of details in the abstract. On the other hand, I am missing some more general conclusions in the abstract and some discussion about the methods and results. Please, consider a substantial edition of the last part of the section.

 

Introduction

Use “feed” rather than inform Use a better term than “put forward”. Has been instead of was. Check the tense. Operational rather than practical L46-L48: The authors state that DBH measurements are mainly based on TLS and MLS. Actually, that is very irrealistic. It would be fair to mention that TLS and especially MLS are not still fully implemented in operational forestry worldwide. The case of MLS is especially notorious as there technology has just become  available so it is very hard to belive MLS in mainstream in forest inventory. Maybe in some areas, the MLS approach is being now implemented but still far from being the business as usual method. Please, soften the expression to make it realistic. L48-L49: Are TLS and MLS traditional methods? Please edit in line with the previous point. L56: change unacceptable for unfeasible. L60: penetration rate of the laser echo sounds better than pass ability. Edit the paragraph accordingly. L64-L65: Split it in two sentences. See your intro:” Due to its high data collecting efficiency, flexible scanning path selection strategy, omnidirectional data point distribution, and good pass ability [4],...”. Ease the readability of the paper. L74-L76: ”Compared with the TLS system, the hand-held PLS had similar DBH estimation accuracy (with an RMSE of 1.11 cm compared to 1.3 cm) but higher tree detection rate (98% for PLS compared to 96% for TLS)”. The results seem very very similar. L74-L87: The reported studies and their finginds are worth of further analyses. It seems PLS and TLS perform nearly equal but the time devoted to data collection in PLS is always clearly shorter than when using TLS. My question is: how about the time to merge nd combine the resulting point clouds from the walking? Is that process as straightforward as commercial grade TLS methods do? The introduction would benefit from a more enriched discussion on the pros (already here) but also the pitfalls and shortcomings of PLS right now. L90: ...and this is one of... L91: While scanning,.. L91-L92: “While operating, the PLS laser scanner constantly collects data and generates point cloud fragments”. Isn´t this statement rather obvious? L96: The term drift error is not so mainstream. Please, explain it or replace the term by another more understandable for the greater audience. L100-L101: Complex compositions such as “Appropriate coordinate transform matrixes for point cloud fragment co-registration” challenges so heavility the flow when reading. L105-L106: ” Although previous researchers have tried various approaches to reduce the drift error, the inaccurate co-registration problem has not yet been properly resolved.” It seems relevant to provide more detials on how transform matrix is being estimated in those studies as the authors suggest that step is the cornerstone process to filter the quality of point clouds. L107-L108: ”According to prior research, the PLS-based DBH estimation accuracy is significantly influenced by co-registration quality [4,6].” Repetitive. Authors commented on this in the previous paragraph. Figure 1 requieres from an upgrade in resolution. The problem is constant throughout the manuscript. I suggest the authors to edit the figures in ArcMap or R to improve the resolution of the images (Figure 1b is a good example). Otherwise, their interpretation is not fully guarantee from the reader. Overall the introduction is ok. However, the text would benefit from simpliciation to reduce the preence of complex expressions. See the beginning of the parapraph in line 119 ” As PLS-based DBH estimation faces far different issues from the previous methods,..”. Certianly there is room for improvement when it comes to English composition.

 Methods and Materials

The structure of the section is rather ilogicial. I would follow the order Area - Data – Methods and Evaluation of the research hypothesis. In the present format, the manuscript is not properly organized. What is the point of subsection 2.1? Actually, the last part of the Intro is very very similar to this subsection 2.1 of just 10 lines? Remove it or frame it under another subsection. Figure 2 needs to be edited. Point-based maps in ArcMap or R (better). Line 166: The donut expression has no room here. What you want to say is already clear so remove the expression please. L168: No donuts please…. Double buffer? Two concentric circles?... L164-L172: A lot of sentences to express the concept. I am very sure the authors could find a way to shorten the section while keeping the good level of description in this version of the manuscript. L178-L179: From “To reduce the DBH estimation error, ANPDA treats the points of these bulgy parts as the outliers, attempts to identify and remove them” to ”The developed ANDPA method identified the bulgy points as outliers before removig them for a better DBH estimation.”It is important to implement this kind of improvements to better follow the paper. Same comment for Figure 3. Very poor resolution. Unacceptable. L188-L190: I have already read this statement around three times before line 188. “... ANPDA applies an idea of iterative outermost point analysis and removal...”. Authors need to be precise when explaining their methods while avoiding redundance across the paper as much as possible, please. Line 194: Tough start of the sentence “Defining the distance between the outermost point and the fitting circle center as R and a pre-determined thickness value as t which describes the distance between the inner and outer circle,....”. Do not be scared of using points to convert long quotes into short and precise ones. Figure 4. Much better. Figure 5 is fantastic as it is very illustrative of the iterative process. I do like the Figure altohugh I would remove one step to allow the Figure to be one-page. Again in this case, make sure the images are properly edited with the correct mean to improve the resolution. The flowchart presented in Figure 6 must be substantially edited. Right now, it is like introducing consecutive sentences within text boxes. Ideally, the flowchart should be concise using as less text as possible. Please, consider the use of alternative designs and make the flowchart look like a scientific flowchart. L289: Section 2.1.4 should be integrated with previous sections. Figure 5 shows 2.1.4.Polar angle probability distribution analysis so please explain what it is before Fifgure 5. The Methods section must be i) re-structure accordingly and ii) edited to avoid repetition of concepts and expressions. Check formatting in section 2.1.6 Is section 2.1.6 really needed? I am finding it very repetitive. In Equations 2 and 3, I would prefer the terms to be defined after the Equation is shown. Figure 7 must be improved as suggested in previous comments. Section 2.2 must be improved. I already suggested its inclusion as 2.1 before describing the methods. The selected study area is poorly described same as the measured ground data. No information is available of ground records to validate PLS-based estimation (range of DBH, tree height…). I guess those trees were registered using a positioning method too but there is no mentioning in the manuscript. Authors need to better describe the input data. In Reference data they authors say “each stem was indexed and the DBH was manually measured using a diameter tape. The position of each stem in each plot was then measured with a total station and matched with its DBH.” Please provide more info on the total station and how that matching was done. Line 441-442: “To facilitate data acquisition, the experiment was conducted during a less rainy season. All data in this study were collected in January 2018.” Also ground measurement? Indeed, the term experiment could lead to confusion as by experiment you could mean that the sampling plots are firstly established in 2018. Please, clarify. There are two sections 2.2.3. Please, revise the text before submitting. Line 448. Table 2. The reported values of the scanning systems can be found online (https://gexcel.it/en/solutions/heron-mobile-mapping/heron-ac-color) I am more interested in the final statistics of the generated 3D point clouds. Point density , especially. Please, include more information in the next version. And reference the Heron system. I would place those two equally-labelled sections as one single section entitled as something like “Assessment of the PLS-based method”. And I would use less words to express the same points. Figure 9. The selected view might not be the best on to represent the route in the ground. Better use a top view. Or convert the figure into a two-side figure showing the 3D point clouds on the left and the top view on the right. In any case, please inspect better color scales to represent the point clouds. (Coloring by height?). Line 482-485: “Computree performs batch processing for forest point cloud data in a pre-determined order determined by When processing point cloud data, it works automatically without manual intervention. As a supplement, Matlab was used to complete any processes not performed in Computree’s plug-in algorithms.” It is interesting to identofy which tasks were performed in Computree and which ones were done in Matlab. I recommend the authors to present a flowcahrt with the steps and functions used on each data processing step. There is a lot of work to be done in the Methods section.

 Results

Line 552: I know what is the double trunk problem. Please, introduce the term before and comment on it in the discussion. Line 559: When authors say significant: was it significantly supported by t-tests ANOVAs? If not, just avoid misleading expressions related to significance. Figure 10. Improve the resolution. You could use Excel-based graphs for these simple charts. Please, activate the black-and-white style here. No need to use colors at all. Line 575-577: Combine the paragraph with the previous one. Three lines are rather limited for a single paragraph. Same for line 607-609 There are barely comments on the results in the section. Authors are merely presenting tables and figures. I am missing more enriching comparisons and descriptions of the observed patterns in the figures and in the tables too. Figure 12. Already commented

 Discussion

 Lines 642-647: “These results indicated that ANPDA was generally effective for improving DBH estimation accuracy. The bias of the estimation was from 0.24 cm~2.84 cm. The reduction rate of the bias was 53.80%~87.13% after applying ANPDA, which means ANPDA was effective for improving DBH estimation accuracy in stem level. The reduction rates of MAE and RMSE were 38.82%~57.30% and 27.17%~56.02%, respectively”. Why not to present these results in the Results? Line 648: “It is suggested that..” Remove it. The Discussion is being used as the Results section. Authors can comment on the results in the Discussion section but it does not make sense to find new results in the Discussion. Please jointly edited Results and Discussions Lines 671-674: “It appeared that ANPDA performed better for the point clouds collected on the ground surface more difficult to walk steadily. In prior studies, DBH estimation for PLS-based data was usually from 1 cm~ 4 cm for RMSE [3-9,11-13]. However, it can be seen from the results of Plots 3~6 that this accuracy level could not always be achieved in this study.” Why is that? Comment on bad performance of PLS in the study for those conditions. If conditons refer to soil conditions, just say soil conditions. Otherwise, make it clear to the reader what good states for. ”it can be seen” collapses the section. Upgrade the English composition when reporting. Line 679-684: “The performance of ANPDA for point cloud slices of different quality was also studied. As can be seen in Figure 11, the DBH estimation error of the point cloud in each class decreased after applying ANPDA. Tables 4 and 5 show that expect for the class with a p value of -9~-8, which only had six samples, the effective rate of ANPDA was generally higher than 80%. This indicated that ANPDA was generally effective and reliable for different quality point cloud slices even at the single tree scale.”... Sounds more like Results. Lines 723-736: The first paragraph of section 4.2 is rather vague. Lot of words for a slightly minor adding to the previously described. The findings of the paper are rather innovative and they deserve to be published but the authors need to favor that by a fair, straightforward and easy reporting of the results and the methods used. Same as in the Methods section, the authors used a lot of meaningless words and sentences to just bring nothing new at all. I am missing information regarding on the processing time of the PLS approach. Hopefully, the authors will include the required flowchart and some information of the computational resources and time used to complete the analyses.

Discussions

Line 763: Remove “The evaluation was carried out by comparing the DBH estimation accuracy before and after applying ANPDA. Error reductions of 53.80%~87.13% for bias, 38.82%~57.30% for MAE, and 27.17%~56.02% for RMSE were achieved under different forest conditions after applying ANPDA”. Line 773: “To take better advantage of ANPDA, future researches could apply ANPDA in a hierarchical semi-automatic DBH estimation framework or promote the idea of ANPDA for cylinder fitting-based DBH estimation. Iterative points removal from inside to outside can also be attempted as a supplementary.” Move it to the Discussion. Conclusions should be short and emphasized on the general findings of the work. This is not the room to hypothesize about further research and further novelties to be tested. Authors can do that in the Discussion.

 

 

 

Author Response

Dear reviewer

We are extremely grateful for all the great comments and suggestions that enabled us to further improve the quality of our manuscript. We have done our best to accommodate all these helpful comments and suggestions in our revision.

According to the suggestions of from the reviewers, the manuscript has been totally restructured. It might be a little difficult to identify our revision with Track changes.

We would like to thank you for the great effort you have made for improving this manuscript.

For point-by-point response please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear editor and authors,

The work introduces a new algorithm called ANPDA to estimate tree dbh from LiDAR data collected using personal laser scanning (PLS). The novelty of the work is the approach applied to filter bad co-registered points by removing the outermost ones from the point cloud slices at 1.30 m height. The authors made a fair assessment of the algorithm, showing its quality and the situations where it cannot be used. As presented in the manuscript, it is an ingenious algorithm able to drastically reduce the errors of dbh estimation in different site conditions, although it cannot be applied to trees with forked stems or with great branches at 1.30 m height.

It is an interesting work, but the text must be improved so a major correction is required. All the recommendations are indicated in their respective lines and some of them are suggestions (not mandatory); I’m not a native English speaker, but I believe those suggestions can help to improve the text.

There are other aspects that I would like to highlight to orient the corrections. The introduction is quite lengthy and some parts of it should be adapted to the discussion section to improve readability. The idea of the algorithm is well presented, but the pseudocode mentioned by the authors (L144) is missing. The results section is the most critical part: the tables and figures, in general, are well presented in an intuitive way, but the authors did not explore them properly; table 4 require a better clarification; some results are presented in the discussion section and I believe they must be in the results section. The discussion section, however, has too much information about the results and is quite repetitive. Therefore, the discussion section must be reshaped and the authors must use the results from the literature to establish comparisons; the main results can be highlighted but in a summarized way.

My punctual recommendations are described below.

L33-34. Confusing. The sentence must be improved. Maybe: “… can be obtained through nondestructive in-situ measurements, the diameter at breast height (dbh, 1.30 m height) not only helps researchers understand the structure of a forest but also reflects forest growth state.”

L37.The authors mentioned the utility of the dbh measurements to forest modeling but added a reference about PLS, the ref. [3]. I recommend replacing or adding a reference related to the applicability of dbh measurements.

L41. Suggestion: … measurements due to its high efficiency in collecting accurate data.

L44. Replace the dash on “stem–level” by a hyphen, e.g., “stem-level”.

L63-79. Most part of this paragraph has too many details about the PLS/TLS-related works so the introduction became quite lengthy. I strongly recommend relocating this paragraph to the discussion section and adapting it to compare the work’s results.

L89. The identification of the acronym AKHKA R2 and GNSS is missing.

L98. I suggest removing the term “according to prior research”, or replace it by something like “It has been demonstrated that…” or “Previous researches have demonstrated that…”.

L99-100. I recommend removing the Figure 1. Consider exploring it (or this information) in the overview section stead, together with Figure 2.

L112. The authors make a statement about the inefficiency of the algorithm for dbh estimation using PLS but did not use reference. If the previous references (i.e. [3,10, 5,8]) did not allow you to make this inference, please consider adding a proper reference at the end of this sentence too.

L116-119. Suggestion: consider simplify the sentence to something like:

“The ANPDA uses a widely applied circle fitting to reduce the impacts of inaccurately co-registered point cloud fragments by iteratively removing the outermost points from a two-dimensional horizontal projected point cloud.”

L135-140. The information exposed in this paragraph was also summarized in the last paragraph of the introduction, together with the work’s objective, so there is no need to repeat this information here.

L149. Suggestion: consider adding the term horizontal when referring to this slice. E.g.: “PLS-based stem point cloud horizontal slice”, or “horizontal slice of the PLS-based stem point cloud”

L161. Suggestion: consider amplify the image of the point clouds of Figure 2. Maybe reducing the range of the graphic’s axes can help with it.  Format the axes label with the same font size as the text. The same works for all point cloud figures in the text.

L169. Suggestion: consider relocate Figure 3 to Figure 2, so you can have a figure with the following sequence: a) with good co-registration; b) with bad co-registration; c) the identification of the outermost point in bad co-registration.

L177-181. The sentence is too lengthy and hard to follow. Please, rewrite it using more than one sentence.

L183. Reformat the figure 4. The texts, especially the ones from the axes, are giant.

L188-190. This sentence is hard to follow. Please rephrase it.

L220. Explicit that the x-axis is set as the polar axis for the probability distribution.

L224. It is not clear what “too few points” means. What is the rule for that?

L243. Explicit the situation where the difference between the two polar angle distribution of points is not significant.

L246. Suggestion: replace the word “attempted” with “used”.

L254. The authors used the notation N for two different things, to refer to the number of iterations (Figure 6 and L225) so as to the number of groups. Define another notation to avoid misunderstandings between both terms.  

L270. The word “represents” must fit better in place of “presents”.

L273. Add the other summation elements (e.g. i=1) so as was did for the equation (4) on L307. 

L274. “Similarity of a current point”?  Shouldn’t be “point cloud”?

L276. Make the same correction pointed for L273 above. Why using “ln” instead of ‘log”?

L292-294. This sentence is hard to follow. What “inside error points” mean? The proportion of errors due to points inside? This term was also used in L299. Please, clarify it.

L288. Consider using “used” instead “attempted”.

L304. Suggestion: consider this change: “… in which the declining of the similarity value…”

L349. Table 2. “Weight” and “Data” have no meaning in the table.

L370. Indicate the algorithm (with its reference) used to filter the ground points.

L371-372. The end of this sentence is quite repetitive. Please rephrase it, or remove the last part: “…2 m  above the DEM surface”.

L373. Indicate the cluster approached (with reference) applied in the preprocessing.

L387. Provide the formulas for the Bias, MAE, and RMSE.

L391. Change “double trunk” to “forked stems” throughout the text.

L397-399. This sentence about good and bad slices has been already mentioned in previous sections.

L400-401. This sentence about the circle fitting has been mentioned many times before in the first paragraph of this section so as in the other sections.

L401-402. Suggestion: “… how good a point cloud could be …”. Provide some reference that uses MSE. If it is a very common statistic, make it clear for the reader.

L406. Use a reference related to the measure p. If it was your creation, explains why is the reason for that? It uses a logarithm, so explain how it must be read, what it means if it has a high or low value. What is the expected range of this value?

L408. The term influence is too vague. Consider “impact” instead.

L418-419. There is no need for this sentence. It has been said previously.

L421. Consider changing the word “factors” by others like, e.g., “accuracy indicators” throughout the text.

L424. The information about MAE is missing. Remove the term “error” in “error reduction”.

L425. Use “hyphen” to indicate a variation among numbers e.g. 1.83-9.45 m instead “~”.

L434. A period is missing at the end of the caption.

L437. Explore more about the results regarding the terrain types.

L442. Explore more about the results regarding the accuracies and the different value p. Note that there is a clear pattern on the indicators as the value p increases.

L452. What are the “average error” and “effective rate”? They have not been mentioned before. Again, the table is not well explored, e.g. what a rate of 100% means? Is there any pattern in these numbers that could be highlighted? The same works for the next table in L459.

L460. This description for figure 12 is so short as a caption. Please, explore more your results.

L472. The sentence “ANPDA was effective for improving DBH estimation” was fully repeated (see L470).

L474-475. Change “fluctuation” has “weakened” to “variation…has decreased”.

L475-477. The suggestion presented by the authors is not well connected to the previous statements. Such inference must be better justified.

L479. Artificial forest? Forest plantation is the most common term for that.

L507. Shouldn’t “expect” be “except”?

L512. The meanings about the quality of the p values should be given when introducing this measure, not in the discussion.

L520. The authors mentioned the overestimation of the dbh, but it was not mentioned in the results. The results must lead the reader to such information explicitly and not just show the tables.

L521. Which instruments are you referring to? Give some examples.

L525. This sentence has been repeated many, many times (L470, L472, L481, L571). It is not new information; it just makes the text repetitive.

L526. Which occasions are you referring to? Different forest types? Make it clear.

L532. “This might be because”? It is supposed to use “This is because…”  since it is not a supposition, it is a fact.

L543. What is the meaning of “mature algorithm”, a better one?

L563. Change “program” by “algorithm”.

L571. One more repetition about the efficiency of the algorithm…

Author Response

Dear reviewer

 

We are extremely grateful for all the great comments and suggestions that enabled us to further improve the quality of our manuscript. We have done our best to accommodate all these helpful comments and suggestions in our revision.

 

According to the suggestions of from the reviewers, the manuscript has been totally restructured. It might be a little difficult to identify our revision with Track changes.

We would like to thank you for the great effort you have made for improving this manuscript.

For the point-by-point response please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have addressed all the comments pointed in the first round and I noted their effort in improving the manuscript, especially the discussion and introduction. However, the text still has many problems, especially in the methods descriptions. The authors tried to make a brief presentation of the algorithm, but the algorithm is described again in section 2.4. The first part of the methods has serious issues in its structure so that it cannot be understood, e.g., missing sentences (L123-128 and L177-178) and missing equations and figures (L185, L192, L213-216). Additionally, the reference numbers are unsorted and sometimes doubled (e.g. Brede et al. 2017), so it is hard to follow, and there are some confusing sentences (e.g. L579, L590, L595-596). In this case, the text still requires to be polished.

 

Author Response

Dear reviewer

We are extremely thankful for your comments and suggestions. The manuscript has been revised to address your comments. To avoid the problem of missing sentences, figures and equations while loading the .docx file, I've uploaded a PDF version as a supplement.

 

For details please see the attachment.

Author Response File: Author Response.docx