Mobile Terrestrial Photogrammetry for Street Tree Mapping and Measurements

Round 1

Reviewer 1 Report

The reviews were adequately address and I recommend the paper for publication after a  proofreading.

Author Response

Response: Thank you for your efforts in revising this a second time!

Reviewer 2 Report

This is a well-developed and reported study that should be recommended to accept in present form except for a minor typo error, so I have to recommend after minor revision:

Line 412 has a highlighted blank space. I suspect this is for the addition of a missed citation

A couple of points, from me, for further consideration. These are not essential for publication, I leave to your discretion:

Line 60 ‘in the accessibility ‘of’ advanced’ . the word of seems to be missing.

Line 79: ‘able to accurately identify ??? 88.8% of trees – what attribute of trees; position, species, size, please make clearer.

Line 150 -  methods of DBH measurements could be reported for those less familiar.

Line 170 if ROW has not previously been defined then please define here.

Line 219 – Is it still appropriate to give weblink (or citation) to R? If so then please do.

In the discussion is it appropriate to discuss any limitation associated with the road level being at a different altitude to the soil level in which the measured tree is rooted?  

Author Response

Comments and Suggestions for Authors

This is a well-developed and reported study that should be recommended to accept in present form except for a minor typo error, so I have to recommend after minor revision:

Response: Thank you for your efforts in reviewing this article

Line 412 has a highlighted blank space. I suspect this is for the addition of a missed citation

Response: Removed.

A couple of points, from me, for further consideration. These are not essential for publication, I leave to your discretion:

Response: All changes made as requested.

Line 60 ‘in the accessibility ‘of’ advanced’ . the word of seems to be missing.

Response: of added as suggested.

Line 79: ‘able to accurately identify ??? 88.8% of trees – what attribute of trees; position, species, size, please make clearer.

Response: for this citation, it is as written. Automatic detection of the trees themselves.

Line 150 -  methods of DBH measurements could be reported for those less familiar.

Reponse: this is a very standard technique for those in forestry.

Line 170 if ROW has not previously been defined then please define here.

Response: definition added.

Line 219 – Is it still appropriate to give weblink (or citation) to R? If so then please do.

Response: Weblink added

In the discussion is it appropriate to discuss any limitation associated with the road level being at a different altitude to the soil level in which the measured tree is rooted?

Response: This is a very common scenario for the application, and can be overcome with more cameras, angles, etc. There is little reason to go into depth with this issue here.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

- Please define explicitly what do you mean by X and Y directions (are they relative to vehicle path, or are they just E-W and N-S directions?), in abstract and sec. 3.3

- Table 1, first line (Bauwens). unit lacking; line -5 (Mokros) perimeter rather than peremeter

- please write LiDAR rather than lidar

- can you comment on the issue of obtaining canopy volume by similar technique?

Reviewer 2 Report

This is an interesting paper. There are several points made in the attached manuscript mark up. 

The information that must be supplied is:

Agisoft settings for both sparse and densification steps. These settings will dramatically affect your point cloud outputs and goes to your later discussion regarding artefacts.

The distance from camera to tree will determine overlap when using a fixed interval capture. I don't see any data presented on the range of distances from camera to tree.

Otherwise this is a nice paper and realistic in it's discussion of utility of photogrammetric point clouds for DBH assessment. I am not sure how valuable DBH alone is for management though. 

Comments for author File: Comments.pdf

Reviewer 3 Report

This is another photogrammetry "demonstration" article. It is not under cover of "urban"forestry and car using. Since there were so many reports of the came kind over last 5 years, I am not sure this work has a sufficient novelty to justify its publication. It is already well-known that photogrammetry can be used to measure dbh of isolated trees. There is nothing really new here, well, OK it is a new camera setup and a new "urban" framework, but the "demonstration" is the substantially the same. The proposed camera setup is obviously limited to only trees with clear visible truck at dbh  in the close proximity to the road. I do not see how this experiment extend our knowledge or boundaries of photogrammetry methodology.  

Reviewer 4 Report

Review of “Mobile Terrestrial Photogrammetry for Street Tree 3 Mapping and Measurements” by Roberts et al. 2019

General Comments

This paper’s uses mobile photogrammetry and handheld GPS to estimate tree location and stem diameter remotely in urban settings.  This paper is good because it’s relatively easy to understand and seems to work and I would recommend it for publication pending some major revisions.  With a few improvements and a cursory examination of distance decay functions of point density and accuracy, this paper has the potential to be widely cited because numerous researchers and private companies are attempting similar experiments.  The simplicity is also a weakness because I read it a little unimpressed, partially because photogrammetric methods are compared against handheld GPS which can be of poor quality, particularly near trees but also don’t include any real time assessment of accuracy at the time of collection.  For such methods to be useful, we still need to know how the method matches with targets on the ground in three dimensions and even if this paper doesn’t try and attempt that, it needs to explicitly acknowledge and explain how someone trying to improve on the method would attempt it.  If high resolution GPS and surveyed group points and targets could be tied to the point cloud, then measurements of ‘absolute’ geospatial accuracy/precision are possible.  For the paper to be useful to the average researcher, the process must be more closely detailed for the sake of repeatability. 

My major problem with the paper is that there’s spatial precision error in all data sources, photo interpreted orthophoto points, photogrammetry derived points, and handheld GPS.  It’s difficult to understand the nature of the error and biases if there’s no high precision GPS measurements which can be regarded as close to reality.  Can this be easily added to the study by georeferencing the 88 urban and 52 windbreak trees using precision equipment?  What about precision targets as a reference? 

There are also some simple improvements to the figures which will help communicate the results more effectively, including adding scale bars and statistics.  More care could be taken to clean up the figures, eliminating empty white space, standardizing the sizes of figure elements, etc.  The paper could be vastly improved by providing code and a more detailed description of the field methods, and analysis even though it sounds like it was mostly linear regression statistics in R.

It also would seem this method is highly limited in dense vegetation beyond the initial row of trees, trees occluded from view, even meters away can’t be measured well, but it would be useful to know how far.  For instance, can you answer the question “How far away from the road was the furthest trees you successfully measured?”.  I think it’s also worth mentioning that there are some trees like untrimmed conifers might have occluded stems at breast height so measurements from even the most precise photogrammetric models aren’t possible.  If there is an expanded analysis of the nature of the distance decay of point density in urban versus windbreak environments this paper could be vastly improved. 

Specific comments. 

L27:  Make sure you mention its ‘horizontal’ error you’re measuring here.  In an analysis of LiDAR ground control accuracy (in accordance with ASPRS standards), vertical height in included and if it’s not, it should be mentioned. 

L31: How many sample points are included in the analysis?  N=?  This should be mentioned in the abstract. 

L60: “Computer Technology”  Please be more specific or choose different wording.  “Computer Technology” is redundant and near meaningless, mention something regarding ‘cost’ or ‘monitoring strategies’.

L78:  Is this the only paper which attempts to ID individual trees?  If you want to say ‘current aerial methods are improving’, you should cite other methods or studies. 

L133: How many trees?

L135:  A figure or map would be helpful to see the distribution of the plots.  The one plot along the loblolly pine windbreak is well described, but what about the other sites?  What species compose those sites?  It seems strange to describe 1/10 sites.  How might the site composition impact the study?  Why were these sites chosen?  Please explain why 52 measurements were made from 1 site, while 88 measurement were made from the remaining 9 sites?  This doesn’t make sense to me, please make the rationale clearer. 

L138:  How were stem positions estimated from Quickbird imagery?  Was the canopy digitized and then the centroid was assumed to be the stem?  Was there any attempt to use high precision GPS (not handheld) to measure stem position?  Since the resolution of Quickbird is 65 cm, ground measurements at the same level of precision should correspond to accurate locate the trees.  This process requires more explanation.

L145:  What are the accuracy specifications of your handheld GPS unit?  I’m a little skeptical of the RMSE measurements unless I know the precision of the field measurements themselves. 

L179:  Put the “Point Selection” tool in Cloud Compare in actual quotes to indicate the tool is its own entity in the software.  Keep in mind someone trying to replicate your methods.  Were the sections of the trees measured manually in Cloud Compare?   At 1.4 meters? 

Figure 2: I don’t like the layout of this figure because there’s a lot of empty white space in the figure.  Also, label the parts of the ‘set-up’.  Where is the camera, mount, GPS, etc.  Think of it as a technical guide to reproducing your study. 

L183:  Is the R code you used ‘shareable’?  Can you make Github (or similar) repository where you can share your analysis with the community?  This goes a long way to showing transparency and repeatability.  It’s not required for publication, but I would highly recommend it. 

L190:  Are you measuring height too?  In the abstract it sounds like DBH and position are the characteristics being measured.  If you’re measuring tree height, that needs to be described more completely. 

L194:  I’m a little confused why the tree species in the urban sites are described here instead of the methods.  Were they determine from the imagery after the fact?  If so, this should be explained, otherwise the description of the other 9 sites should be in the methods with the description of the loblolly windbreak site. 

L215: “generally acceptable” to who exactly and by what standards.  See the ASPRS standards as a reference.  This is an ambiguous statement which needs to reference accepted standards for 3D mapping. 

L219:  “...relatively low for all classifications”.  Relative to what?

L224:  Is the bias constant?  Does bias change with relative position to the road or distance to the camera?  A brief examination of bias is important in the discussion of the results.  If it’s a linear bias, it can be systematically removed, if it’s not, it might be trickier.  Understanding the spatial nature of the bias is more important than the magnitude of the bias itself. 

Figure 4:  Can you embed the figure legend in the upper right-hand corner of the scatterplot.  It will save space and look cleaner. 

Figure 5: Add the 1:1 line and maybe the residual statistics on the figure itself. 

Table 3:  Is the ‘skewness’ important?  If so, how?  If not, then remove. 

L258: exactly the same to the mm level?  Please double check. 

L 301:  This is a major concern.  The orthoimages can be used as a helpful visual aid, but not as a reference because of these potential errors.  Is there any documentation about how large these errors can be?

L315:  I think one of the major shortcomings of the method outlined in the discussion, could be one of the biggest strengths if analyzed properly.  Since the tree stem is not ‘seen’ on all sides by the cameras, and trees occluded by closer trees cannot be mapped effectively, the most interesting part of the study might be analyzing those limitations to the method.  What is the distance decay function for effective mobile tree mapping?  Can these distances be quantified?  How does point density decay as a function of distance?  I believe answering these questions will increase the value of the paper immensely. 

Figure 6:  This figure needs help. The line around the upper-left should either be a box or shouldn’t exist.  It looks strange the way it’s drawn.  All of the sizes of the images are different, standardize.  Add a scale bar so the reader knows how big the stems are.  Why is this figure in the discussion?  It’s this how the stems were measured?  If so, it should be in the methods. 

Figure 7:  This is an interesting figure/image, but the images should be the same size.  Reduce empty whitespace whenever possible.  Showing scale would be helpful. 

Figure 8:  Small empty whitespace under ‘C’.  Widths of spaces between the vertical columns are different.  Standardize and clean up.  Also, show the scale bars for the images. 

Figure 9:  “Approximate”?  Is this the same place in both images or not?  If it’s not, I don’t think the figure is necessarily worth it, since most people will understand GPS degrades under canopy.  If so, add a scale bar (vertical and horizontal) so that the magnitude of the error is apparent to the reader.  The width of the two images is slightly different.  Clean these little things up, otherwise it looks unprofessional.