Using Deep Learning to Detect the Need for Forest Thinning: Application to the Lungau Region, Austria
Round 1
Reviewer 1 Report
This is a solid study and a well-written paper in general, though there are a few places that need modification or additional information. Please see below.
Research papers present the purpose to be studied/hypothesis to be tested in the introduction. This paper, in lines 93-99, describes the study as though it has already occurred, and it is not presented in the scientific format. It would be more in keeping with research papers to re-write starting with “This paper presents….” something like:
“In this study, we tested a deep learning approach to determine if remote sensing data alone could be used to detect the need for thinning. The two factors investigated were whether the approach could:
1. Detect if thinning is needed, based on deep convolutional neural networks (DCNNs) trained on very high spatial resolution imagery, and,
2. Differentiate between thinnings with different urgencies (i.e. timescales).
An accurate prediction of thinnings from remote sensing data has the potential of delivering information for vast or remote forest areas promptly and thus improving the stability and wood quality of forests.”
The introduction is a bit long, but contains relatively interesting information.
Lines 299-200: Please provide a citation for where the urgency 1 and 2 definitions come from. These definitions are not universally understood.
Lines 332-335: Please explain how you knew that the anticipated information loss would be too high is 1 meter was used instead of making up the 0.2 meter information.
L355-356: please include a citation that defines ‘calamity logging’.
Minor typo:
Figure 2, it appears the box on ‘Tile ceration’ should say ‘Tile creation’
Author Response
We thank the reviewers for their time and suggestions for improvements.
================================================================
This is a solid study and a well-written paper in general, though there are
a few places that need modification or additional information. Please see
below.
================
1)
Research papers present the purpose to be studied/hypothesis to be tested
in the introduction. This paper, in lines 93-99, describes the study
as though it has already occurred, and it is not presented in the scientific
format. It would be more in keeping with research papers to re-write
starting with “This paper presents….” something like:
“In this study, we tested a deep learning approach to determine if remote
sensing data alone could be used to detect the need for thinning. The two
factors investigated were whether the approach could:
1. Detect if thinning is needed, based on deep convolutional neural
networks (DCNNs) trained on very high spatial resolution imagery, and,
2. Differentiate between thinnings with different urgencies (i.e.
timescales).
An accurate prediction of thinnings from remote sensing data has the potential of delivering information for vast or remote forest areas promptly and thus improving the stability and wood quality of forests.”
AUTHOR REPLY: We have revised the paper as suggested.
[Lines 96-104]
================
2)
The introduction is a bit long, but contains relatively interesting
information.
AUTHOR REPLY: We have separated out section "1.1 Previous work"
from the Introduction and made it a new section "2 Previous work".
This reduces the bulk of the Introduction, and also helps address
some issues raised by the other reviewer.
================
3)
Lines 299-200: Please provide a citation for where the urgency 1 and 2
definitions come from. These definitions are not universally understood.
AUTHOR REPLY:
Urgency 1 and 2 are used by the Austrian Federal Forests to plan thinnings.
The division into these three urgency levels, is a practical approach to
disclose "urgent", and "less urgent" thinnings to the forest manager.
[Now lines 249-256]
================
4)
Lines 332-335: Please explain how you knew that the anticipated
information loss would be too high if 1 meter was used instead of making
up the 0.2 meter information.
AUTHOR REPLY:
The average tree crown diameter for a first thinning is 2.5 m. Considering
that the algorithm has to determine the the density of the forest stand by
how densely the trees are standing.
A pixel size of 1 m would imply that a tree crown is represented by
approx. 3 pixels the determination of an individual tree crown is in
that case much harder than if one tree crown is represented by
approx. 13 pixels as in the case of the 0.2 m resolution.
[Now lines 287-292]
================
5)
L355-356: please include a citation that defines ‘calamity logging’.
AUTHOR REPLY:
The right English term is salvage logging, We used an incorrect translation
from German.
Salvage logging is the logging method in forest areas that have experienced \
natural disturbances to regain a part of the economic loss.
* https://en.wikipedia.org/wiki/Salvage_logging
[now line 314]
================
Minor typo:
Figure 2, it appears the box on ‘Tile ceration’ should say ‘Tile creation’
AUTHOR REPLY: corrected
Reviewer 2 Report
The article presents an application of a deep convolutional neural network applied to look for the necessary thinning in a forest. The topic of the article has high interest to the readers, overall, for forest researchers and managers. However, the article needs more effort to be published. It needs to be simplified and focused on the article's topic.
Major comments:
- The title is quite generic, and the article doesn’t collect all the variability of thinning worldwide. I recommend changing the title to something like “Detecting the necessity of forest thinning with deep learning: An application in Lungau region, Austria”
- In my opinion, the structure of the article is confusing, there are many sections and subsections. For example, the first time that I read the introduction, I miss more information and documentation about the thinning, etc., then, I found these in section “1.1. Previous work”. Maybe all this section could be integrated into the introduction, minimizing the repetition of information and avowing simplification of the narrative.
- There are many points in section 1.1 that are out of the article topic. For example: “1.1.3 Change detection”, “1.1.4 Tree species”, and “1.1.5. Tree height and wood volume”. In this way, the authors could mention it as an example of Remote sensing in forestry but explain it in this paper is not relevant. I recommend focusing the content on thinning operations. The section “1.1.6. Forest operations” could be integrated as state-of-the-art in the introduction.
- In section 2, the subsection “2.1.2. Thinnings” is not properly methodology, is discussion. I strongly recommend simplifying the whole section 2 and focusing on the methodology applied. There are several subsections, it could be minimized with narrative.
- Table 4 is not referenced in the text.
- The discussion section needs to be improved. What is the gain of the method in comparison with the current methodologies? This comparison and/or discussion needs to be supported by references.
- Is recommend a final section of “Conclusion” or “Final remarks” (at least) with the most important point of the work.
Minor comments:
- Delete de “]” in “which is practically usable.]” in the abstract.
- Line 22-24: “Although the primary objective of thinning is to prepare the forest stand for the final harvest, it also provides forest owners with the opportunity of obtaining additional income by selling the removed trees.” need to be supported by references.
- Figure 1 that shows the study area needs to be improved with a more general map that gives more information about where the allocation is.
Author Response
We thank the reviewers for their time and suggestions for improvements.
================================================================
The article presents an application of a deep convolutional neural network applied to look for the necessary thinning in a forest. The topic of the article has high interest to the readers, overall, for forest researchers and managers. However, the article needs more effort to be published. It needs to be simplified and focused on the article's topic.
Major comments:
================
1)
- The title is quite generic, and the article doesn’t collect
all the variability of thinning worldwide. I recommend changing the
title to something like “Detecting the necessity of forest thinning with
deep learning: An application in Lungau region, Austria”
AUTHOR REPLY: A good point. We have revised the title as suggested.
================
2)
- In my opinion, the structure of the article is confusing,
there are many sections and subsections. For example, the first time that
I read the introduction, I miss more information and documentation about
the thinning, etc., then, I found these in section “1.1. Previous work”.
Maybe all this section could be integrated into the introduction,
minimizing the repetition of information and avowing simplification
of the narrative.
AUTHOR REPLY: A new background section 2 has been created with duplications
removed, thus making sections 1, 2 and 3 simpler.
================
3a)
- There are many points in section 1.1 that are out of the
article topic. For example: “1.1.3 Change detection”, “1.1.4 Tree species”,
and “1.1.5. Tree height and wood volume”. In this way, the authors could
mention it as an example of Remote sensing in forestry but explain it in this
paper is not relevant. I recommend focusing the content on
thinning operations.
AUTHOR REPLY:
As suggested, we removed sections: "Tree height and wood volume”,
"Tree species” from the new section 2.
3b)
The section “1.1.6. Forest operations” could
be integrated as state-of-the-art in the introduction.
AUTHOR REPLY: This was duplicate material largely removed, with some merged
into the introduction.
================
4)
- In section 2, the subsection “2.1.2. Thinnings” is not properly
methodology, is discussion. I strongly recommend simplifying the whole
section 2 and focusing on the methodology applied. There are several
subsections, it could be minimized with narrative.
AUTHOR REPLY: Much of this section was accidentally
duplicated and has now been removed.
This background material has been merged with the material on Thinning
in section 2.1 (formerly 1.1.1).
================
5)
- Table 4 is not referenced in the text.
AUTHOR REPLY:
Table 4 is now referenced in the text.
[Line 337]
================
6)
- The discussion section needs to be improved. What is the gain
of the method in comparison with the current methodologies? This
comparison and/or discussion needs to be supported by references.
AUTHOR REPLY:
The discussion contains now a comparison and
a discussion of needs supported by references.
[lines 588-603 and 624-629]
================
7)
- Is recommend a final section of “Conclusion” or “Final remarks” (at least) with the most important point of the work.
AUTHOR REPLY: A brief Final remark summary was added.
[lines 643-649]
================
Minor comments:
8)
- Delete de “]” in “which is practically usable.]” in the abstract.
AUTHOR REPLY: done
================
9)
- Line 22-24: “Although the primary objective of thinning is
to prepare the forest stand for the final harvest, it also provides
forest owners with the opportunity of obtaining additional income by
selling the removed trees.” need to be supported by references.
AUTHOR REPLY: Citations added.
[line 24]
================
10)
- Figure 1 that shows the study area needs to be improved with
a more general map that gives more information about where the allocation is.
AUTHOR REPLY:
We are not sure what the reviewer wants here. We asked the publisher to
clarify with the reviewer. They advised us to make our best guess, which is
the text in red in the Figure 1 caption.
Round 2
Reviewer 2 Report
The authors responded to all my comments and changed the article with the improvements proposed. Personally, the article has been improved and has the quality to be published.
Following I give some personal comments that, in my opinion, could improve the article. All in the way to reduce the sections.
Minor comments:
- Merge sections 3.1.2, 3.1.3, 3.1.4 and 3.1.5. these are all “Remote sensing and ground truth data acquisition.”
- Section 3.2.2 could be included in 3.2.1 as Data preprocessing.
- Change the title “Creation of ground truth” to “Introducing ground truth” or “Including ground truth”, or something like that.
- Sections 4.1, 4.2, and 4.3 could be merged into Methodology.
Author Response
We thank the reviewers for their time and suggestions for improvements.
==============================================
- Merge sections 3.1.2, 3.1.3, 3.1.4 and 3.1.5. these are all “Remote sensing and ground truth data acquisition.”
AUTHOR REPLY: done
========
- Section 3.2.2 could be included in 3.2.1 as Data preprocessing.
AUTHOR REPLY: done
========
- Change the title “Creation of ground truth” to “Introducing ground truth” or “Including ground truth”, or something like that.
AUTHOR REPLY: We revised the section heading to: Sec 3.2.7 Design of ground truth label models
========
- Sections 4.1, 4.2, and 4.3 could be merged into Methodology.
AUTHOR REPLY: we merged 4.1.1, 4.1.2, 4.1.3 to make section 4.1: Experimental Design more cohesive. Perhaps this is what the reviewer had in mind as Sec 4.2 is Training and 4.3 is Evaluation Criteria
