Sensitivity Analysis of the DART Model for Forest Mensuration with Airborne Laser Scanning

Round 1

Reviewer 1 Report

Sensitivity Analysis of the DART Model for Forest Mensuration with Airborne Laser Scanning

In this study, the authors test the use of the DART model to simulate a forest stand and use it to evaluate the sensitivity of ALS acquisition parameters on forest attribute estimation. The paper, although presented with a very technical focus, is very well written and has a wide relevance to all users of ALS for forest attribute estimation. I recommend this paper for publication pending a few specific changes.

Overall comments:

There are a few figures and tables that are presented before they are mentioned in the text. Typically, tables and figures are presented after they are referenced in the text. Please double-check this and change where necessary.

In the results section, there are a lot of discussion points listed throughout (e.g. line 310). The results section should only include the results and not their interpretation. Please remove all discussion points from this section.

Specific comments:

25 – What is “object” scale? I would think that these metrics are calculated at the area (stand) or individual tree scales. Either describe what the object scale is or omit it.

80 – There are certainly benefits to using simulated forests, but what downsides are there? I can think of some, but please list some references.

87 – How big do you define “large-footprint waveform lidar” to be? Could small-footprint lidar be large-footprint if the aircraft is high enough?

102 – If we need to use a better model for structurally heterogeneous forests, why are you testing DART on a (seemingly) homogeneous plantation forest? I think that this point is important but not necessarily covered in the study. Either change it or remove it.

104 – This is the first mention of DART in the text (not the abstract), so define it.

117 – At the beginning of this section, you should define radiative transfer. Why do we use it?

150 – Please explain why the DART model was chosen if all of your stands are likely homogeneous.

163 – To me, it seems like there’s a lot lacking in this section. Since the majority of the paper is a comparison to this ALS dataset, some more details of the processing is needed. For example, what was the overlap? Was the ground classified? If so, how? What methods were done for this? I think the ground classigfication raises an issue for me (more on this below). If you’re comparing this to the simulated data, both should have been classified in the same way.

176 – Was the location taken for these trees? Could the locations be determined using a tree detection approach from the ALS?

178 – I’m wondering why you needed to use completely simulated data. Using a tree detection approach from ALS, you could get the approximate locations and heights of most trees in the stand, right? Also – was there any idea of how the simulated stands would have matched up to either the measured stands or those estimated from an individual tree approach (think in terms of the basal area, stocking, DBH or height distribution, etc.). In order to know that these simulations were within the bounds of a natural stand, I would like to see that there was good correspondence in terms of these metrics between the DART simulation and the field data. For example, you mention that the generated trees had a minimum height of 3m and minimum DBH of 7 cm. So was there a tree that had a 7 cm DBH and was 3m tall? Allometrically, does this make sense?

181 – I know from the reference that these are local, but describe that these parameters are specific to your study area.

192 – Wouldn’t you know the general stocking from the field data? Couldn’t you use the numbers from the field data to parameterize this?

203 – How did you choose these ranges? Are these typical of other ALS surveys?

244 – I’m not sure why you used this “rule-based” ground classification and I think this may explain the higher overestimation in the number of ground returns. I would think that another ground classification method would be better for determining the amount of ground returns later in the study. Additionally, if you’re comparing the simulated point cloud to that of ALS, you should be using similar methods for ground detection. Please explain this reasoning or change the method.

Table 2 – How did you choose these metrics? Are they typical of ALS-based surveys?

Figure 5 – Please describe how you made these plots in the caption. Am I mistaken, or do violin plots typically show outliers outside of the lines showing the range?

339 – There’s a lot of discussion points in this paragraph. Please include only results in this section.

395 – Did any of these characteristics (forks, etc.) come up in the field data? If not, then maybe you don’t need to consider them in this study area.

403 – This paragraph is maybe another place to justify your use of or describe the difference in ground detection schemes. How much do you think the ground classification contributed to the error?

463 – Do you know how much wider it is? Can you calculate it? If it’s not much, this increase in distance may be negligible.

474 – I think “most ALS surveys” sounds like it’s downplaying a lot of recent work in this area. Change to something like “Since ALS surveys frequently are not undertaken…”

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Thank you for presenting a well written and thoughtfully prepared manuscript. It was both informative and interesting. I am not an expert on radiative transfer models but found your approach very useful and what would appear to be a cost-effective way of conducting these kinds of studies. Along those lines, I have no choice but to accept some of your assumptions and hold you to a code of ethics that keeps you honest.

With that being said, I have only a few minor suggestions regarding some of your text and figures.

On line 104, in the introduction you state the acronym “DART” without first introducing it. Yes, you follow with a section that clearly describes the acronym and model itself. For clarity, however, I suggest you introduce the term first, before shortening it.

The text beginning on line 111 seems somewhat out of place. It may be best placed on line 108, and perhaps rephrased to some extent.

On line 260 you use the word “understorey”. I gather this is the common usage in the UK. However you use “modeling” instead of “modelling” so I might suggest using “understory” instead of “understorey” just to stay consistent with a particular style.

In general, your figures and tables are well prepared and presented. The only figure I had a little trouble interpreting was Figure 5. I felt it could have been slightly larger to facilitate an enhancement of the vertical scale. Particularly, in 5b the space between values 0-25 is very narrow.

Beyond these minor elements, this was a remarkably clean manuscript and I thank you for submitting it.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

-The study compares the Discrete Anisotropic Radiative Transfer Model to various metrics from small footprint disccrete return ALS. This is an important topic and worthy of study.
-The abstract is well written and clearly explains the paper.
-The introduction is good and includes plenty of citation. The overview of ALS is good and sufficient. My only recommendation is to include some definition of radiative transfer simulations. You first mention it on Line 79 but don't properly define them. Your overview of DART in Section 2 is good, but some more definition in the introduction when it is first mentioned would be good to add.
-The objectives sound good, but on Line 107 you mention uncertainty analyses but I don't quite see how that fits into the rest of the study. Your explanation of the sensitivity analysis is good, but not the uncertainty analysis.
-Figure 2 - This figure looks neat, but just make sure it's clear what we're looking at exactly. Is this an example output from the DART model? Because it's obviously not an RGB point cloud from ALS. What were these renderings created with? And based on what data?
-Table 2 - Your ALS metrics look good. A few questions though. First, you mention canopy cover % is at 0.5 m resolution, but what is the resolution of the other metrics? Second, I see a bit of redundancy between some of your metrics, like canopy cover, gap fraction, etc. What's the reason for the redundancy? Third, Your definition of LPI is not what I am used to seeing. Usually LPI is defined as (Number of Ground Returns) / (Number of Total Returns). See example reference below:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.222.1600&rep=rep1&type=pdf
-Figure 7 is very good but deceptive. The Y Axes for each row should be consistent. Right now you're skewing the results and showing some of the slopes as more dramatic compared to the others in a row.
-Overall the results and discussion look solid.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf