Tree Species Traits Determine the Success of LiDAR-Based Crown Mapping in a Mixed Temperate Forest

Round 1

Reviewer 1 Report

General: This article deals with delineate individual tree crowns using remote sensing data. Five automated LIDAR-based individual tree crown delineation methods were applied across 15 plots . Various factors at plot and tree level were analyzed under the primes of evaluating its influence on  the crown delineation.

Introduction:

The introduction is well written and explains difficulties and different approaches of ITDC techniques and LIDAR technology. Crown shape is especially affected by neighboring trees (competition). Literature and lengths are sufficient for the introduction. However, the research objectives at the end / or hypothesis could be stated more clear. Currently the research objectives sound almost descriptive: “…applied five automated LIDAR-based ITCD methods…” “…We identified tree- and plot-level…”, but you actually conducted quite an analysis on that. You could also  list here already groups of the analysed factors to better prepare the reader on main objects of the analysis.

Materials and Methods:

The research area seems to be suitable and sufficiently large and complex in terms of species composition and traits  for such a study.

However, from my perspective this section is currently not optimal.

Figure 1: looks a bit simple, especially for a remote sensing journal. Basically it just shows where approximately the Megaplot is situated. I suggest to improve this figure and provide more details (e.g. an additional zoom into the megaplot where the distribution of conifers vs broadleaf’s becomes visible). But it is up to you. Maybe you have your cause why it is kept that way.

This section currently covers 5 pages. Can you write this section shorter?  I think some parts are too detailed and can be written way shorte (e.g. line 183-190).

Since you want to link your results later with ecology the section on the indices is for my understanding a bit short (a short part of 2.5). Also when I look at the level of the overall manuscript I expected a bit more sophisticated or ecological valid indices. This is a small weakness of this study. To some extend you just use the very indices (especially with regard to diversity). Diversity experts nowadays do not take the results of these indices without any critique. I recommend you to read for example the publication of Jost in Oikos from 2006: Entropy and diversity. You may reflect on his argumentation in the discussion.  Shannon and Pielou are quite easy to calculate but in theory (as in practice) there are examples where they actually failed to reflect the level of diversity.

You also used an updated version of the spatial Clark and Evans index, right? Okay. It I think it is right to look at the spatial sphere in your study on delineation as the crowns are mingled and neighborhood of trees (as you wrote above). But I guess there would be some other indices even more interesting for your study. A tree-level based index that takes into account the spatial arrangement of species diversity (neighbouring trees to a tree) is presented in Hui et al. 2011 (published in the journal Forest Science) : https://academic.oup.com/forestscience/article/57/4/292/4604193

I guess it is too complicated to consider such additional indices now in afterwards in the analyses of your study. But I think it would add to your study to reflect on it in the discussion part.

 

Results

Interesting and well structured. Interestingly especially the diversity aspects seem to count (line 335-339)

Discussion:

Overall well written as well. As mentioned above I would consider to also include a bit more on other diversity indices, most likely fitting into 4.2.

 

Overall I recommend minor revisions before publication

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The paper is well structured and clear, with a coherent flow between premises, experiment, analysis and results. The final discussion can be useful even for readers not specialized in the topic.

The only thing open to question is, in my opinion, if a so high number of literature items is really so necessary to support  purposes and outcomes of the paper (but this is only a personal opinion).

Thus, the paper requires only minor checkings, see the attached file for details.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

In the paper entitled Tree Species Traits Determine the Success of LiDAR-based Crown Mapping in a Mixed Temperate Forest, the authors compared the performance of five different tree crown delineation algorithms on different forest stands with varying canopy structures and species compositions. Their main conclusion is that more than the segmentation method, the canopy structure is responsible for the performances of the segmentation method, conifer being overall better segmented. The study brings interesting insights into the canopy segmentation problem, especially by showing that no matter the method used, broadleaves trees are hard to segment and may need additional information to be efficiently segmented (spectral for instance). The authors linked the performances of the segmentation methods with tree species traits like the shape and size of the crowns, that vary from species to species but also with environment.

I have 2 main concerns:

One of the result of the study is that different methods do not vary greatly in performance. This may be due to the method tested. First, I would like to see a bit more discussion about the differences between the methods used. Second, it is not surprising that a method based on the CHM segment conifer better (as we can see in Figure S4, conifer tops are clearly visible from the CHM). The only method using the entire point cloud (Li2012) has been developed for conifers of the Sierra Nevada, and according to the authors, this method takes advantage of the relative spacing between trees. Other methods based on the point cloud and making no assumption on the crowns shape may work better to segment broadleaf trees. Williams2019 for instance: Three-dimensional Segmentation of Trees Through a Flexible Multi-Class Graph Cut Algorithm (MCGC). Or the mean-shift algorithm: see Xiao2019 for details about the best mean-shift settings to segment tree crowns (Mean Shift Segmentation Assessment for Individual Forest tree Delineation from Airborne Lidar Data), or Aubry-Kientz2019 for a comparison between segmentation methods where the mean-shift algorithm performs well (A Comparative Assessment of the Performance of Individual Tree Crowns Delineation Algorithms from ALS Data in Tropical Forests). Moreover, the tree_detection step is important for watershed and region growing methods, the same tree tops are used for the different methods used, which for sure tend to make the results similar.

My second comment is about the validation method, which relies on manually segmented crowns. As mentioned by the authors, only crowns “visually distinguishable” are manually segmented. This probably biased the validation data set toward big emergent tree crowns that are clearly visible from the top. One may be interested in the performance of the algorithms for smaller trees, especially for the algorithm using the entire point cloud that takes advantage of the 3d structure of the point cloud to segment crowns that can be partially covered by others. A comparison with the data acquired during the census can complete the comparison with the manually segmented crowns. At least, the author may discuss the limits of the accuracy assessment method based on manually segmented crowns.

 

I also have some minor comments:

In the method part “Parameter Tuning and Accuracy Assessment”, line 180: “automated delineation were paired to manual delineations”, how was it realized?

Lines 194-202, 4 different cases are presented, the same 4 cases are presented in the legend of figure 3, but in the figure, they are labelled a,b,c,d, which makes it a bit confusing.

Figure 7 is interesting and I would like to see it for other segmentation methods (in the supplementary materials).

Line 413: reference problem.

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

The work “Tree Species Traits Determine the Success of LiDAR based Crown Mapping in a Mixed Temperate Forest”, authors provide a valuable analysis and contribution to the active area of research that is individual tree crown delineation (ITCD). A rigorous comparison of the methods implemented in the R package lidR, developed by Roussel and Auty (2019).

 

I recommend the acceptance of this manuscript after few minor revisions outlined below.

 

General comments:

 

Even it is used an up-to-date R package, I wonder why it was not used the FUSION software of McGaughey (2018).

 

Even if it was used different forest types and species, it was not taking into account other variables as the terrain that alter the forest structure, especially for step catchments.

 

It is typical first call or mention the image and after that place it into the document (as example; figure 4, 6 and 8).

 

In the images scale and label are necessary in order to a better understand of the results, as example it is not distinguished between the samples in Figure S1.

 

The references are mixed between numbers and text along the document, please check it.

 

In general, the document shown a good English language, but it is recommended to carefully review the document (as example paragraphs 314 – 316; 449-450; lines 461, 463).

 

Some specific comments, by line:

 

73: Format.

 

106: Remote sensing data – I t is necessary to describe the UAV sensors used to collect the data, as also the altitude of the performed flight. (resolution of 0.025 m – down sampled to 0.1m). Generally, high resolution is obtained at low altitudes.

 

132: Figure 2 - Scale is necessary.

 

183: The adopted threshold of 50% looks problematic, especially because the MITC and the AITC have a very different spatial resolution.

 

217: There is no explanation of how it is calculated tree metrics of tree-level attributes (DBH, CHM and crown area). Especially problematic the DBH that after is used in the next sections.

 

285: Plot-level accuracy up to 0.9, that it is important to label the plots in order to identify vegetation at plot level, because this values are particularly high compared with values reported for the others 14 plots. Only for plots 9 and 10 the values range between 0.7 to 0.8.

 

303: Figure 6- p < 0.05 is it significant?

 

314 – 316: The sentence needs to be checked.

 

368: Figure S4 - It is needed a legend in order to identify and understand the height of the trees and the canopies identified in the image.

 

384: Figure 7 – Improve the legend of crown area.

 

413: Error in reference.

 

424: It is mentioned Figure 7 for crown shyness, but it looks like the figure not correspond to the mentioned phenomena.

 

461: We found….

 

463: We found that…..

 

459: Some aspects related with LAI-VI relationship (specifically to the NDVI and their calculation), it needs to be described in the Materials and Method section.

 

471: We were able to successfully delineate 62-70% of all trees greater than 40 cm DBH -à this part is not explained in the Methodology or Results section. It is not clear how the DHB was calculated, see also the comment in line 217.

 

519: Include all the supplementary material.

Author Response

Please see the attachment

Author Response File: Author Response.docx