Spatial Analysis of Dense LiDAR Point Clouds for Tree Species Group Classification Using Individual Tree Metrics
Round 1
Reviewer 1 Report
Review of manuscript forests-2499644 „Spatial analysis of dense LiDAR point clouds for tree species classification by individual tree metrics“ by Martin Slavík et al.
A novel method of tree species group classification based on individual tree parameters derived from unmanned aerial vehicle laser scanning data is presented. Well established statistical models (generalized linear models) and machine learning methods (random forest classifiers) are parameterized and trained on a dataset of 1045 individual trees from four mature stands in the Czech Republic, to automatically assign individual trees into either the coniferous or broadleaved group. The machine learning approach using a random forest classifier and incorporating the Clark-Evans spatial aggregation index turned out to be the best performing technique, yielding an 95.1% classification accuracy.
The manuscript is clearly structured and generally easy to read.
The selection of references in the introduction is somewhat outdated. Methods are clearly described and appropriate. The results are statistically sound, but some extra effort should be spent on a clearer presentation, especially for the random forests. The differentiation between methods, results and discussion is not always strict.
The novel idea to incorporate Clark-Evans spatial aggregation index into the classification is an interesting approach that significantly improved classification accuracy. However, unfortunately, only one coniferous species (Norway spruce (Picea abies)) and one genus of deciduous species (oak (Quercus sp.)) are regarded in the manuscript, and all trees were in the age class of 80-100a. Other age classes, and other species, that are also common in the Czech Republic (e.g. beech (Fagus sylvatica), hornbeam (Carpinus betulus), pine (Pinus sp.) and larch (Larix decidua)) are completely ignored, limiting the scientific merit of the manuscript.
Beside its shortcomings, I think that the manuscript is worth to published pending MAJOR REVISIONS.
Please find attached some specific remarks and suggestions.
L2-3: The title of the manuscript is misleading, in L12-13 it says that individual trees are not classified according to their species, but according to their respective species group (coniferous or broadleaved). Please change the title to reflect the actual topic of the manuscript.
L29-30: "current bark beetle outbreak" seems to be somewhat odd, as the maximum of the calamity was already reached 2019.
L36-121 The whole introduction is based on rather old and outdated references (the most recent one is from 2018). Especially regarding MLS and ALS a lot of research has happened in the last five years, thus the whole section needs a revision and more recent literature must be included.
L125 The differences in tree species composition are elaborated in the following sections, however differences in growth conditions are only mentioned here without any further explanation. What are these differences and how do they effect the study?
L127 “oak sp” seems rather odd, better use “Quercus sp.”
L128 “Spruce sp” seems rather odd, better use “Picea sp.” Are there really more than one species? I guess Picea abies is the only common species of spruce in that area…
L134 What other species than Quercus petraea could be found on these plots? What are percentages of the different species.
L155 There is no reference [231]
L151 “multirotor octocopter” isn’t this a tautology? I mean, obviously an octocopter has multiple rotors, namely eight – otherwise it wouldn’t be an octocopter…
L139 So in fact, the manuscript is only on the classification between Norway spruce (Picea abies) and oak (Quercus sp.).
L158 What is the effective measurement rate and FOV in this application? I mean, the effective FOV is obviously <180°, as only points below the scanner are recorded.
L213-214 I think this sentence belongs to the paragraph before.
L213-221 How many trees did the RF have? How was the number of trees dertermined?
L241-247 This paragraph is partially methods, partially discussion and partially results. Please make a clear differentiation move the respective parts to where they belong.
L251 Do you mean Figure 5?
L280-282 Could you please add significance stars to the p-values in Tab 2 and Tab 3? This would certainly improve the perceptibility of the results.
L316 Labels in Fig. 8 are way too small. In the printed version I cannot read anything, and in the electronic version I must zoom in to at least 200%...
L329-349 y-axes in Figures 9 & 10 are not labeled. I guess it is relative variable importance in percent, but how can relative variable importance be larger than 100%?
L329-349 Figures 9 & 10 are very misleading: First of all, y-axes are not labeled, I guess the label should be ‘relative variable importance in percent’, but how can relative variable importance be larger than 100% on some instances in Fig 10? If I understand it correctly every column represents one predictor variable, while the different colors of the column represent the corresponding variable’s importance for the different target variables. In this case, it simply makes no sense to stack the different colors above each other, as the variable’s importance for the different target variables is not cumulative. For better representation of the RF results, I strongly recommend to have a look at the R-Package “randomForestExplainer”: https://cran.rstudio.com/web/packages/randomForestExplainer/vignettes/randomForestExplainer.html
L372-394 The comparison with the accuracy-levels reached by other studies is quiet meaningless without the information how many species were observed in the respective study. Please add this information.
L396-398 I cannot follow this conclusion, how did you check the completeness and representativeness of the point cloud through the stem profile? You did not make any comparison with TLS data or any other ground truth… I would suggest removing the sentence.
L405-406: Are there any supplementary materials? If so, these should be provided for review!
L450 something went wrong with the character encoding
As a non-native speaker I am not overly qualified to judge English style and language. However, even I noticed several instances of incorrect word order and incorrect use of singular and plural, thus some English editing is certainly required.
Author Response
We would like to express our sincere gratitude for your valuable feedback on our manuscript. Your insightful comments have undoubtedly improved the overall quality and clarity of our work. The reply for all comments is attached.
Reviewer 2 Report
The manuscript, entitled “Spatial analysis of dense LiDAR point clouds for tree species classification by individual tree metrics” is to present a method of tree species classification using individual tree metrics derived from a three-dimensional point cloud from unmanned aerial vehicle laser scanning (ULS). The topic falls within the scope of the Journal.
I recommend a following minor revision before the manuscript is ready for publication.
1. Suggest enhancing the description of existing research problems and difficulties, as well as the current status of individual tree classification research in the study area.
2. Please provide the full name where the UAV first appears. When other abbreviations appear for the first time in the main text, the full name should also be provided. Please carefully check.
3. Please correct the misspelled text in Figure 3 and remove the red curve below the text.
4. Suggest adding the individual tree species classification map of four plots.
Author Response
We would like to express our sincere gratitude for taking the time to review our manuscript and providing us with your valuable feedback. The file with replies is attached.
Author Response File: 
Author Response.docx
Round 2
Reviewer 1 Report
The authors carefully addressed all points risen during the first review and revised the manuscript accordingly. In my opinion, the manuscript is now ready for publication.
