Woody Plant Encroachment: Evaluating Methodologies for Semiarid Woody Species Classification from Drone Images

Round 1

Reviewer 1 Report

The text of the article is correct, it can be published in the current format.

Author Response

"Please see the attachment"

Author Response File: Author Response.pdf

Reviewer 2 Report

In my opinion, the revised manuscript unfortunately has no fundamental difference than the original version. I can see the authors’ efforts in improving the presentation of their results, at the same time, their reluctance to do additional review and research work that can truly improve the manuscript quality. Seemingly, many changes were made to the revised manuscript, however, they were mostly trivial textual edits that have no extra values, which do not change the mediocrity of the manuscript. The authors’ angle of justifying the study’s uniqueness is forced and awkward. Lack of novelty is still my main reason for not recommending this manuscript for publication in journals like Remote Sensing.

1. Line 14, if you consider “affordable, very high resolution” as part of your research novelty, where is your review and justifications on drone prices and camera resolutions in Introduction?
2. Line 59-60, weak logic makes no sense.
3. Figure 1 resolution is low and looks blurry.
4. Line 107-111 poorly written goal and objectives. What do you mean “if RGB imagery can be used for mapping”? At what level of classification accuracy separates “can” and “can’t”? Shouldn’t Objective 4 a part of Objective 1 and 2? Where did you explain your mapping methodology mentioned in Objective 3 later in the manuscript? Did you confuse your classification methodology as a mapping methodology? Objective 4, develop recommendations for what?
5. I am very much unhappy with the limited reviewing work that the authors did in Introduction. There is a large body of research work that is relevant to the study but not reviewed in the manuscript. I did a simple literature search and found this paper: “Using UAV Imagery to Detect and Map Woody Species Encroachment in a Subalpine Grassland: Advantages and Limits”. Can you explain why you did not review similar studies like that?
6. Line 131, “taking advantage of” species appearance differences in December and developing models using images collected only in December might allow you to achieve nice-looking classification accuracies for the study, but also significantly limits your models’ applicability to images collected at other time periods and reduces the merit of the study. Much research work is still missing to prove “RGB images can be used for mapping”. Models should be developed using images collected all year round instead of only one date.
7. Line 292, too small of a training dataset size. Line 295, many of your subsampled training images will have identical information, which does not help with your low training image number issue.

Author Response

"Please see the attachment"

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have made considerable changes to the manuscript. Thank you for addressing the review comments. 

Changes:

1. Still, the images are not clear. For ex: Figures 1, 2, and 3.
2. Equations 1, 2, and 3 needs to be changed. I suggest not using a complete name in the equations. 
3.  Future work directions need to be included in the conclusion section. 

Author Response

"Please see the attachment"

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Please see attached.

Comments for author File: Comments.pdf

Author Response

Comment 1. I was referring to this sentence in your first version revised manuscript line 59-60:“First, it is characterized by a low diversity of woody plant species, which means that affordable stock RGB sensors can be used (this would be impossible in areas with higher species diversity).” 

Response 1. We agree, the sentence does include redundant information since we described the methodology in the previous sentence. To correct this we deleted the second half of the sentence which now reads "First, it is characterized by a low diversity of woody plant species [6]".

Comment 2. Modify your manuscript based on your response

Response 2. We went through the entire manuscript and changed all instances of the word mapping to classifying, with one exception in the conclusion where we discuss the possibilities of using fusion with NAIP imagery collected every 3 years.

Comment 3. The reviewing work in this manuscript is still weak. The authors did not respond to why they did not review the study that I mentioned. I would speculate that it is because the authors did not do one of the fundamental works before writing the manuscript, a comprehensive literature review. The authors only reviewed two more studies in the revised manuscript. If there are no other relevant existing studies, which I do not believe so, please specify explicitly in the introduction. Otherwise, I’m not sure what else to think besides doubting the authors’ academic integrity.

Response 3. We did not include it initially because we did not see it, the paper was published in 2021 and much of the review work was done prior to its publication, however we rectified this and included it alongside 3 other papers with similar scopes published between 2020-2021. Furthermore, we found a review journal article published late last year (https://doi.org/10.34133/2021/9812624) titled: Mapping Tree Species Using Advanced Remote Sensing Technologies: A State-of-the-Art Review and Perspective. In this paper we found one more paper that used very high resolution RGB imagery captured from a drone to classify tree species in the temperate forests of Germany and included it in our paper. This review cites one more paper published last year, but we already cited it in our review. Furthermore, we added some text explaining that this image acquisition methodology had been used in previous studies published recently, however our novelty lies in its application to semiarid grasslands and savannas, a critically important ecosystem/landscape responsible for supplying much of the world's animal products.

Comment 4. The response is not addressing my comment at all.

Response 4. We respectfully disagree with your point saying our dataset is too small. We believe that 635m2 of fully labeled landscape which includes over 250 individual trees, as well as open landscape is enough data for the pretrained model to learn from. Those 635m2 of labeled imagery were also rotated, mirrored and spliced, an accepted method for data augmentation, providing additional data for the model to learn from.