New Structural Complexity Metrics for Forests from Single Terrestrial Lidar Scans

Round 1

Reviewer 1 Report

This is a review on the manuscript entitled "New Structural Complexity Metrics for Forests from Single Terrestrial Lidar Scans" by Batchelor et al. Authors introduced an approach to quantify and assess structural complexity of natural forests using single station terrestrial laser scans. They introduced three structural parameters derived from the TLS scans, namely 'depth', 'openness', and 'isovists'. They highlighted the potential of these metrics to quantify the structural complexity in a more robust manner than commonly applied methods which are often based on expert assessment and are prone to observer bias. The proposed approach is novel and interesting and shows large potential to be applied also by other research groups or forest inventory agencies. Especially for forest inventory agencies I see this approach as interesting as the time required to acquire the data is low (10-15 minutes), but the potential to deliver more robust quantification of the forest structural complexity is high compared to the assessments which are currently often based on expert knowledge. Regarding this point, I found that the potential use of such an approach for forest inventory purposes could have been highlighted a bit more.
In general, the manuscript is very well written and conveys the usefulness of the proposed approach very well. It would have been great to see a comparison with other ways to assess forest structural complexities to see if the proposed approaches would suggest similar assessments as for example an assessment by experts or other metrics. However, I see that such a comparison would be difficult due to differences in definition of structural complexities. Maybe a few words on that could be added in the discussion section.
I specifically liked the discussion section where many potential application of the proposed metrics were highlighted. I am very much looking forward to see these metrics applied to e.g. assessment or description of wildlife habitat. In my opinion, this study builds a great base for many future research applications.
I congratulate the authors for this very well written and interesting study and I am looking forward to reading future studies they mentioned in the outlook. I can therefore suggest this manuscript for publication after only a few minor adaptations as suggested also in the specific comments below. Thank you very much and congratulations again.

# Specific Comments

## Introduction
- L56: ... to get at more ... ? -> Is this the correct wording?
- L58-60: you could add some examples of these rudimentary hand tools here, so we have an idea of what instruments you have in mind here.
- L67: "... by essentially every management organization": This is maybe a bit strong. True, ALS is being used more and more often. But saying that it is essentially used by every management organization may be a bit of a stretch.
- L68: A point is missing here after citations of [30-32].
- L68: However, airborne lidar returns... -> Maybe rather just "airborne lidar can miss much...". The "return" is not actually needed here.
- L72-84: When comparing TLS to ALS, it may be worth to also mention that, whilst being able to retrieve high detailed information from the under and midstorey, TLS can suffer from occlusion effects in the upper storey, depending on the forest structure. Maybe you mention it somewhere afterwards, but I would expect something in the lines of that also here, when you compare TLS to ALS.
- L110: "single point lidar scanning" -> I am not sure if this is the best wording for that. Would "single station lidar scanning" or something similar be better?

## Materials and Methods
L131-132: a closing parenthesis is missing
L189: Why did you exclude herbaceous vegetation ans small woody debris? These belong to the structure as well and excluding them could potentially result in different results for the derived metrics, or not?

L266: where->were


## Results
L272: Explanation does not fit Figure number. Limpy Rock is shown in Figure 5 not Figure 6. please check.
Figure: Probably just an issue that the manuscript was downscaled for creating the review manuscript. But please check on the quality of the figures. The figures are hard to read (especially smaller axis labels).
L342: though? -> through?


## Discussions
L380-397: very nice discussion on the potential and applicability of isovists.
L399-441: Also here, very nice description of potential applications of your proposed approach.

## Conclusions
- in my personal opinion, I would put the conclusions section into a section of its own, i.e 5. Conclusions

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors of New Structural Complexity Metrics for Forests from Single Terrestrial Lidar Scans introduce new metrics, depth (distance from origin) and openness (non-occlusion related gaps) along with a traditionally architectural and design metric, isovist. These metrics are presented as indexes for characterizing and categorizing forest structure and ecotypes. Researchers are rapidly converging on the conclusion that operational deployment of Terrestrial Lidar Scanning (TLS) technology for monitoring and mensuration purposes will need to shift to single-scan sampling, rather than multi-scan registered surveying approaches due to the limited range, difficulty of registration of multiple scans and limited hardware capabilities of existing TLS units. This study is important in that could be one of the foundational demonstrations of the possibilities of mathematically analyzing the point cloud, rather than a more traditional survey/object-based approach analysis. Due to the limited number of studies that use single-scan methods, all analyses such as those presented here that iteratively improve understanding of structural metrics and how they can be used are important.

While this study represents a conceptual change in moving from an object-based approach to single scan metric-based approaches, there are some concerning issues with the analyses and presentation of the data that must be addressed in order to support the author’s conclusions. Additionally, there are changes that can be made to improve the clarity of the methods and outputs from the processing that would also improve the manuscript. Finally, there are some important citations that are left out of the manuscript that put this study in context of the state of single-scan TLS forest metrics. Missing other foundational citations and unique ways of processing point clouds without mention of why well-established methods of processing (normalization) were not used, implies the authors are not up-to-date on the current state of TLS. While conceptually, this could be an important paper, major revisions are required. Outside of a convincing arguement for the methods chosen by the authors, everything from preprocessing of the raw point clouds through statistical analysis should be redone for consideration of publication. See details below. 

Major Questions that must be addressed:

The authors are not clear on what advantage there was of not normalizing the point cloud to ground prior to rasterization? There are many advantages to this method including the standardization of all heights of points relative to ground. The authors are vague on the exact methodology, but it sounds like they hand masked the ground points to a horizon in the unnormalized point cloud for the distance/openness analyses and also created a plane from a few measured trees at 1.4 meters prior to rasterization in the isovist analysis. This seems fine in the example on figure 3 where most of the trees are near the horizon, but a quick glance at the supplementary material shows that this is the exception, rather than the rule. There are many examples with trees in the foreground extending well below the wavy horizon. This would result in being off on heights by more than a meter in some cases. In the case of the isovist analysis, what method was used to smooth the plane derived from 3-8 height points? Was there a procedure for determining how many trees to measure and distance from origin (scanner) to measure? If all of your measured trees are close, how does that affect the slope of your isovist relative to the ground or due to micro-topographical variation, etc? At 110 m (55 m radius), how sure are you that all points are in the correct plane? Presumably, there is an advantage to using this method that I am missing. If not, then serious errors are being introduced into the analyses. The authors should compare existing methods of solving these problems in comparison to methods that they propose. Please address these concerns if you can and add the necessary details in the methods section.

The authors use non-metric multi-dimensional scaling (NMS) to illustrate clustering of plots in stands/separate them by stand/show structural changes along axes. The NMS as presented relies on sketchy conclusions without any statistical support of conclusions other than post-hoc structural characteristics. Worse, the actual NMS is obscured to highlight a few of the stands. This is not an acceptable method of analysis. There are several distance-based methods for developing confidence ellipses, which would be far more suitable than free-handing convex hulls. If they are not available to you (PC-ORD does not support them), color each point by stand name. Axes of an ordination should always be labeled with the variance that each axis describes. PC-ORD associates these with each axis in the report. For instance, axis 2 could explain Mary’s Peak variance well while axis 1 does not (Mary’s Peak spans nearly the entire axis). Conversely, North Spit can best be defined in axis 1, but is widest across axis 2. Disconnecting the interpretation from the NMS analysis is not appropriate. The primary advantage of NMS is that it creates a measurable dimension-space. It would be appropriate to use bi-plot analysis to display vectors from the center based on variables that were strongly correlated to directions if the relationship was strong enough with a vector length defined by R2 value. If openness, tree spacing, etc. are significant, only then should you include it in your interpretation. PC-ORD has this functionality. The whole point of using the new method the authors propose is to remove bias.

Specific comments:

Line 51. Please write out diameter at breast height when first used, then the acronym every time following that.

Lines 109-110. Please cite existing literature that uses single scan metrics in forests rather than saying that it doesn’t exist. i.e. Anderson et al. 2021 https://0-doi-org.brum.beds.ac.uk/10.1016/j.foreco.2021.119118; Gallagher et al. 2021 https://0-doi-org.brum.beds.ac.uk/10.3390/rs13204168; Wallace et al. 2022 https://0-doi-org.brum.beds.ac.uk/10.3390/fire5040085; etc.

Line 143. Rather than using the acronym GIS and not writing it out the first time, list the actual software used and its citation

Figure 1. This map has PPI issues and needs higher resolution. Rather than OpenStreetMap, use polygons of the forest and the states with clear labels. This is a blurry mess. You describe all plots in the text as being 300 km from Corvallis, show it on the map.

Table 1. This table should be data driven. Order by ecoregion, then DDI. Consider removing ecological systems descriptions from the table. It would be great if you could use the NMS to separate each scan by this category, but it doesn’t look possible based on the NMS shown in the figure. In this case it is irrelevant without linking it to analysis.

Lines 163-166. Explain this. Was the 10 m distance from plot center the error in the consumer GPS unit? It's not clear in these sentences, but it sounds like the operators were moving the laser from plot center to minimize occlusion? If so, what were the procedures for moving the unit without bias? It seems that moving the TLS even a little bit in a forested unit would have a profound effect on all of the metrics you are using.

Line 164. The city and country of origin should be cited for the manufacture of the laser. Is this not a S120, or is it a model I am not aware of?

Line 165. Not all consumer grade GPS units are created equal. Accuracy of the GPS unit is directly linked to where the plots are set up. Please cite the make model and manufacture location as you do the TLS unit for repeatability.

Lines 172-174. The scans were only 100 m apart? Presumably there is some overlap and hence pseudo-replication if the distance-based metrics weren’t limited besides occlusion and the isovists were cropped to 55 m. There is a high probability that the same parts of the scan were measured twice, again regardless of occlusion or perspective. It is recommended that only the four corners be analyzed rather than 9 with overlap.

Lines 176-180. What are the sources of noise using this laser? Are these noise removal points somehow accounted for in the analysis? If not, then these removed points may be accounting for openness elsewhere in the scan.

Lines 181-183. Cite Stoval and Atkins 2021 doi: 10.20944/preprints202107.0690.v1 to support PTX usage

Lines 190-192. Address the preprocessing questions here. How was this accomplished? By hand masking and editing the raster? Why was the height not set to zero using normalization, then classified ground points removed, then rasterization?

Figure 3. Address the wavy-ness of the raster images. Presumably, this is due to slope? There seems to be different levels of waviness in the scans in the supplemental materials.

Line 220. Which GIS software.

Line 248. Name the software here. Presumably PC-Ord, based on the citation, but what version? I am not really sure that the NMS adds anything to the manuscript. The figure is faded, but it looks like there is not really great separation between clusters.

Line 254. Address the concerns with the NMS analysis above here.

Line 310. Possibly, but this is not backed up statistically by actual correlations in the NMS that you have presented. No actual analysis was done to back this claim and the NMS is obscured by hand-drawing over the top. It is impossible to see if any plots were anomalous, nor can you see what drives the axes in the ordination space. It might be open/closed understory and tree spacing, but there is no analysis to test that. PC-ord has bi-plot functionality to test these hypotheses.

Figure 7. Remove the hand-drawn convex hulls, color plots by location, remove the biased axes labels, include a bi-plot with significant variables with vector lengths defined by r2 value, and label axis 1 and axis 2 with significance values. Essentially, get rid of all of the editing done to this figure. You can barely make out the actual NMS underneath. There appears to be a great deal of overlap from the stands that are not highlighted. For spacing, use DDI or stem maps from the scans, can you even get a qualitative measure of open vs. closed understory to test against other than through field measurements?

Figure 8. Redo this figure so that the depth and openness labels are done in the same software. There are artifacts when the figure is scaled.

Line 357. No way. Based on this line, an uncategorized raw scan could be inserted into this ordination space and its location could be derived by convex hull. In the obscured figure 7, there appears to be significant overlap. Additionally, using the current NMS analysis, evidence of a significant signature based on openness and depth is speculation at best. Please perform statistical analyses to support this claim.

Line 360-362. Current photo point and photo load monitoring methods already provide qualitative methods without the need for expensive specialized equipment, have the advantage of being long established, decades of use by thousands of managers and practitioners and are more easily interpreted due to the waviness of the rasters. Focus your advantages on the ability to quantitatively score your outputs, limiting human bias. 

Lines 387-389. This has been done using a 3-dimensional approach rather than a simple 2-dimensional isovist metric. How is this better than existing viewshed methods developed for point cloud analysis in ethology? Cite some of the many examples of the literature that have used the viewshed3d R package and the paper of the package itself. Lecigne et al. 2020 https://0-doi-org.brum.beds.ac.uk/10.1111/2041-210X.13385,

Line 479. This citation is first mentioned on line 100. Please adjust the references to the correct order, by either including it prior to reference 2, or placing it further down the references and adjust all other reference numbers up to that point.

Line 594. This reference is not cited in the text.

Supplementary material. Supplement A refers to Supplement C, the averaged value for each forest stand. This was missing from the document. Please append.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have sufficiently addressed my concerns. There are still some issues with resolution of figures, but this can be addressed by MDPI editorial staff.