Can a Remote Sensing Approach with Hyperspectral Data Provide Early Detection and Mapping of Spatial Patterns of Black Bear Bark Stripping in Coast Redwoods?

Round 1

Reviewer 1 Report

The article presents a study about assessment of tree damage using high resolution hyperspectral data from UAV carrier. It focuses and analyses trees damaged by black bear, more specifically the bark removal and sapwood damage of these trees and its appearance in reflected spectral characteristics. The paper describes evaluation of the technique by of support vector machine and random forest for classification of trees into 3 classes: healthy tree, fresh damage, old damage.

The results indicates that it is possible to distinguish old damaged trees (more than 4 month after attack) though it is difficult to distinguish fresh damage and healthy trees. Authors correctly mention that the date of damage is very hard to estimate and they utilize only the previously mentioned 4 month window to define fresh damaged trees.

The paper is interesting and bring some new insides and I would recommend it to be accepted in Forests, however I have one major comment and several minors to be addressed before publication:

Major:

• I do not understand why pixel based approached was used for classification.  Then the results are, that some crowns are partially classified as old damage and some as fresh damage etc. Object based classification where objects are crown segments should be used here in my opinion, and if not, very clear explanation why using pixel based approach should be provided

Minor

Division of 90 to 10 in training and validation is not very common. Usually one would expect something like 70 to 30 or 66 to 34 and similar. Authors mention that This approach was repeated 10 times but it is not clear what was repeated 10 times and how.

There are many typo errors (maybe only in mine pdf version) here some Row 185,211,213,246  Error! Reference source not found..and it goes through the text. (but this is only small problem which can be easily solved)

Author Response

The article presents a study about assessment of tree damage using high resolution hyperspectral data from UAV carrier. It focuses and analyses trees damaged by black bear, more specifically the bark removal and sapwood damage of these trees and its appearance in reflected spectral characteristics. The paper describes evaluation of the technique by of support vector machine and random forest for classification of trees into 3 classes: healthy tree, fresh damage, old damage.

The results indicates that it is possible to distinguish old damaged trees (more than 4 month after attack) though it is difficult to distinguish fresh damage and healthy trees. Authors correctly mention that the date of damage is very hard to estimate and they utilize only the previously mentioned 4 month window to define fresh damaged trees.

The paper is interesting and bring some new insides and I would recommend it to be accepted in Forests, however I have one major comment and several minors to be addressed before publication:

Thanks for the valuable comments.

Major:

• I do not understand why pixel based approached was used for classification.  Then the results are, that some crowns are partially classified as old damage and some as fresh damage etc. Object based classification where objects are crown segments should be used here in my opinion, and if not, very clear explanation why using pixel based approach should be provided

Pixel based classification was chosen because of the very high spatial and spectral resolution in the dataset. We wanted to use hyperspectral data to determine if we could discern variation within one species, the redwood. We also wanted to experiment with different thresholds for number of pixels, to identify tree tops better, like other studies (e.g Nasi et al, 2015). The object based approach gets us to objects but then aggregates all pixels with their heterogeneity in brightness and doesn’t help us answer the question of whether we can discern these health classes using hyperspectral imagery. We have indicated the reason behind this choice in line 347-351.

Minor

Division of 90 to 10 in training and validation is not very common. Usually one would expect something like 70 to 30 or 66 to 34 and similar. Authors mention that This approach was repeated 10 times but it is not clear what was repeated 10 times and how.

Lines 324 -327 were changed to better explain the cross validation procedure and the purpose of choosing a 90/10 split.

There are many typo errors (maybe only in mine pdf version) here some Row 185,211,213,246  Error! Reference source not found..and it goes through the text. (but this is only small problem which can be easily solved)

The manuscript was edited to include cross referencing and in-text citations. The updated manuscript should include properly cited in-text citations that reference the correct image/table.

Reviewer 2 Report

The manuscript addresses its topic of black bear damage to redwoods by hyperspectral UAV scanning in a sound and systematic fashion. The result is partially disappointing, since recent damage apparently cannot be so detected, but this conclusion is valid.

The presentation, however, suffers from some misgivings that must be corrected before publication. The most important (and easiest to fix) is correct reference to Figures and Tables. Right now the manuscript is inundated with notes of "Error! Reference source not found" that force the reader to guess what Figure the text refers to. So please fix this at first opportunity.

The caption of Figure 5 does not contain an explanation to the coloring of the "layers" in the spectra. There is some explanation in the text (without proper reference to Figure 5).

There are two Figure 1's. The second one should probably be Figure 6?

The notion of Jeffries-Matusita (JM) distance is central to the conclusions of the paper, but it is not defined in section 2.5, although its impact is described. I feel it necessary to include the definition here as well and not just by referring to another paper. It is used extensively in Results section without naming it, so expressions like "there is no region in the VNIR that approaches 1.4" make no sense to the reader. Instead, it should read "there is no region in the VNIR where the JM distance between healthy and recently damaged canopies approaches √2" instead.

I want to commend the authors on pictures like Figure 8. This is very illustrative!

There are also some minor typos like "enhibited" and "to similar" instead of "too similar" that should be fixed.

Author Response

The manuscript addresses its topic of black bear damage to redwoods by hyperspectral UAV scanning in a sound and systematic fashion. The result is partially disappointing, since recent damage apparently cannot be so detected, but this conclusion is valid.

The presentation, however, suffers from some misgivings that must be corrected before publication. The most important (and easiest to fix) is correct reference to Figures and Tables. Right now the manuscript is inundated with notes of "Error! Reference source not found" that force the reader to guess what Figure the text refers to. So please fix this at first opportunity.

Thanks for the valuable comments. The problem with the figure and table references has been fixed and changed. The newest manuscript has been changed to include correct referencing and table/figure labeling.

The caption of Figure 5 does not contain an explanation to the coloring of the "layers" in the spectra. There is some explanation in the text (without proper reference to Figure 5).

Figure 5 cited in the main text and descriptive Figure caption was given.

There are two Figure 1's. The second one should probably be Figure 6?

Figure 6 has been changed to properly represent the correct figure.

The notion of Jeffries-Matusita (JM) distance is central to the conclusions of the paper, but it is not defined in section 2.5, although its impact is described. I feel it necessary to include the definition here as well and not just by referring to another paper. It is used extensively in Results section without naming it, so expressions like "there is no region in the VNIR that approaches 1.4" make no sense to the reader. Instead, it should read "there is no region in the VNIR where the JM distance between healthy and recently damaged canopies approaches √2" instead.

Thank you for your feedback about the JM statistic. We heeded your advice an changed the section pertaining to that information. Lines 341-348 were changed to incorporate your comments.

I want to commend the authors on pictures like Figure 8. This is very illustrative!

Thank you for the complement.

There are also some minor typos like "enhibited" and "to similar" instead of "too similar" that should be fixed.

General typos and grammatic language has been changed throughout the manuscript.