Robust Single-Image Tree Diameter Estimation with Mobile Phones
Round 1
Reviewer 1 Report
This study developed a new approach that utilizes a low-cost smartphone equipped with LiDAR and camera sensors to automatically estimate tree diameter from a single image under real-world field conditions. This is timely and practical research that benefits those who conduct tree inventory in the fields with easy to use smartphones. As the availability of more affordable LiDAR equipped smartphones increases, it is expected that the demand for this kind of research will also increase. This paper is very well written and the methods and findings are clearly explained. I believe this is almost ready to be published as current status.
There are only two parts that I’d like to comment on.
It is unclear whether a “whole bole” approach is scientifically sound or not. Tree trunks have tapered shapes and they are non-linear. So there could be a lot of variations and uncertainties coming in for the estimation (e.g. 1. different species have different taper curves, 2. capturing more higher parts of a trunk in an image will underestimate the tree diameter). In order to use a “whole bole” approach instead of the diameter at a specific height (i.e. breast height), it is better to add some supporting materials to justify the decision.
In 2.3.3. the Thigh and Tlow were determined with a small test data set, but I am not sure whether this set of threshold values works best only for this particular environmental condition, or can be applied for any conditions.
Author Response
Thank you very much for your thoughtful comments. We have responded to your two points below:
Comment 1: "It is unclear whether a “whole bole” approach is scientifically sound or not. Tree trunks have tapered shapes and they are non-linear. So there could be a lot of variations and uncertainties coming in for the estimation (e.g. 1. different species have different taper curves, 2. capturing more higher parts of a trunk in an image will underestimate the tree diameter). In order to use a “whole bole” approach instead of the diameter at a specific height (i.e. breast height), it is better to add some supporting materials to justify the decision."
Thank you for your comment on this aspect of the proposed approach. We believe that our work helps to validate the “whole bole” approach, at least for the types of trees and forests in our sample, as we demonstrate that our diameter estimates have small error and bias relative to the DBH measured manually at exactly 1.3 m. Per your suggestion, we have also added references discussing the choice of fitting a single cylinder to a trunk to estimate DBH in Section 5 (lines 328-330):
“In our work, we take a “whole bole” approach to estimating trunk diameter, similar to cylinder fitting methods used for TLS data [29, 30], by following the assumptions of DBH measurement in a way that is more natural for image data.”
Comment 2: "In 2.3.3. the Thigh and Tlow were determined with a small test data set, but I am not sure whether this set of threshold values works best only for this particular environmental condition, or can be applied for any conditions."
Thank you for your comment. We have added the following to the explanation of the threshold values to clarify that the test set was taken in the Laurel Creek forest and that other environmental conditions may see improved results by resetting these thresholds based on local conditions (lines 178-183):
“To select Thigh and Tlow, we vary these parameters over a small test data set of trees collected from a Carolinian forest in leaf-on conditions (the Laurel Creek location described in Sec. 2.4.2) and choose the thresholds that result in the lowest bias (mean error) metric. It is possible that accuracy could be improved by setting these parameters based on a test sample specific to each study area, but we use the same parameters in all evaluation environments.”
Finally, we also note that you recommended the manuscript for extensive English revisions. Based on the rest of your review, we believe that this may have been a mistake. However please let us know if this is not the case, and we are happy to make any revisions that you require.
Reviewer 2 Report
Dear authors,
It was a pleasure to read your manuscript. The article is well structured and the novelty character emerges from the application you proposed. I appreciate your efforts in this regard and I believe that each such step will contribute to changing the way data is collected in the forest inventory.
Attached you will find a few minor comments. In addition, I would like to emphasize some remarks.
One, is related to the accuracy of estimating DBH using the method you propose. I noticed that the number of trees with large diameters that you analyzed is small (approximately 14 trees with a diameter greater than 50 cm), which means that the evaluation of the application for large trees is not yet determined. Moreover, for large trees, the frequency of estimates with individual errors greater than 5 cm also increases. Considering that the method is applied from the ground level, the accuracy requirement it is not satisfied for large trees or, it must be checked further on a larger number of trees (over 100).
The second remark is related to how you obtained the reference diameter. You measured the girth and deduced the diameter. Why didn't you use a forestry caliper to get the size of the same diameter that you estimated with the proposed method? I believe that the direction of measurement plays an important role in the size of the diameter. I would have kept the direction of measurement. The diameter in the girth has the meaning of an average diameter. Perhaps an analysis of the deviation of the cross-sectional area from the circular shape would have been appropriate. This contributes to the size of the error found in this study and is not due to the method you proposed, but an external factor.
I agree with considering the shape of the trunk as a straight cylinder only for short sections (under 1 meter). At the base of the tree, the shape is that of a neiloid trunk.
I wish you a happy new year!
Comments for author File: 
Comments.pdf
Author Response
Thank you very much for your kind and thoughtful comments. We have fixed all of the line-comments you left on the PDF with one exception:
On lines 415 and in the caption of Figure A4, you noted that there was too much space between a period and the start of the next sentence. Unfortunately, we believe that this is not due to extra space characters in the text, but the formatting template of the MDPI journal adding space to ensure that the text is both left and right justified.
In addition, we respond to your text comments below:
Comment 1: "One, is related to the accuracy of estimating DBH using the method you propose. I noticed that the number of trees with large diameters that you analyzed is small (approximately 14 trees with a diameter greater than 50 cm), which means that the evaluation of the application for large trees is not yet determined. Moreover, for large trees, the frequency of estimates with individual errors greater than 5 cm also increases. Considering that the method is applied from the ground level, the accuracy requirement it is not satisfied for large trees or, it must be checked further on a larger number of trees (over 100)."
We agree that the current evaluation data set does not have many large-diameter trees. Within the data set that we have, we note that while the large trees (> 50 cm DBH) have a higher RMSE, they have almost the same percent error as the smaller trees (7.9 % small trees, 8.1 % large trees) and a lower percent error if the outlier described in the text is omitted. That said, we recognize that the data set does not contain enough samples to fully assess accuracy on large trees, and we added text to the discussion to note this as a limitation (lines 353-358):
“There were other trees in our evaluation data set of similar size (the second-largest tree had a 99 cm DBH) that did not have similar problems. However, we note that only 13% of our evaluation data set has a DBH over 50 cm, and further evaluation is required to assess the algorithm’s accuracy on large trees. Moreover, to consistently handle large and irregularly shaped trunks like the one discussed here, a more flexible first step of the algorithm may be required.”
We have also added a footnote to Supplementary table A1 noting that only 13% of evaluated samples had DBH over 50 cm.
Comment 2: "The second remark is related to how you obtained the reference diameter. You measured the girth and deduced the diameter. Why didn't you use a forestry caliper to get the size of the same diameter that you estimated with the proposed method? I believe that the direction of measurement plays an important role in the size of the diameter. I would have kept the direction of measurement. The diameter in the girth has the meaning of an average diameter. Perhaps an analysis of the deviation of the cross-sectional area from the circular shape would have been appropriate. This contributes to the size of the error found in this study and is not due to the method you proposed, but an external factor."
This is an interesting point. We agree that using a forestry caliper is more analogous to our method because it measures diameter along one direction, and it could result in lower errors by comparing the app estimate to a diameter measurement taken from the same direction.
One concern about using calipers for the evaluation is that it may underestimate our app errors relative to other standard forest inventory methods such as tape measurements. Though calipers are a standard approach to DBH estimation, they may miss variation in trunk diameter due to non-circular trunk cross-sections. Tape measurements are affected by those variations, and so represent a worst-case scenario for the estimation error of our method compared to a traditional forest inventory DBH metric.
