Use of Drone RGB Imagery to Quantify Indicator Variables of Tropical-Forest-Ecosystem Degradation and Restoration

Round 1

Reviewer 1 Report

This study provides focus on estimation of forest plantations in Chiang Mai University’s Forest Restoration Research Unit (Thailand) with the use of a consumer-grade drone platform. This is worthy goal that could help professional community increase knowledge in estimations of the forest restoration with drone based imagery in this region of the world. Overall, the approach brings together some useful datasets to explore the use of canopy height models and drone-captured RGB images to quantify a range of forest indexes to help guide implementation and monitor of restoration projects. The study corresponds to the scope of Forest journal.

The work has some novelty value, but not highly significant, since drone based RGB imagery have been assessed in the past. However, it does add to what is known regarding the drone surveys of the forest plantations.

Despite this, the paper does have many weaknesses that must be addressed, particularly in the Introduction, Methods and in the Discussion sections. The most significant limitation of the study is that the results are based on only few test sites. A study whose conclusions are based on only five field plots on restored forest plantations can be considered rather as an initial study of the problem. The authors partially point out this fact in the article. I suggest that they emphasize this point in the discussion and conclusions section of the article. It is not clear from the article how the use of drone RGB imagery would improve estimation of degradation levels of the studied forest plantations. The authors should explain this fact in the article or rewrite the objective and scope of the research.

There are many grammatical errors, and only a small proportion of these are listed below. Please check and revise the grammatical errors in the whole manuscript. The paper should be thoroughly reviewed by a native English speaker before it is considered for publication.

I would suggest the following title of the manuscript.  

Estimation of restored tropical forest plantations using drone RGB imagery  

Line 37. Delete comma after “projects”  

Line 38. Better - in Africa (afr100.org/) and 20x20 Initiative in Latin America (initiative20x20.org/)).

Line 40-43. I would not fully agree with this conclusion, because forest restoration often implies the introduction of the ecological functions, including the conservation of biodiversity and carbon storage.

 Line 44-48. Please clarify sentence, as it is a little bit confusing.

Line 56. I would suggest to use “index” instead of “indicator variable”. Variables usually used in modelling or regression equations.

Line 68.  For forest restoration planning

Line 72-85. I suggest to eliminate or rewrite this paragraph, since most of the text has descriptive (general) character, while LiDAR technologies are not relevant to the topic of this work. Meanwhile, there is need more analyses focusing on the previous research findings in the field of study. Therefore, additional review of the scientific articles is required.  

Line 103-106. Please clarify the goal and tasks of the research. So far it is too vague. Also emphasize the novelty of your research based on the new models (patterns) you have developed.

Please add a flowchart to the “Materials and methods” section to make it easier for readers to understand the research flow.

Line 112. Add “sites” after All.  Change to were established

Line 113, 115, 118 Delete comma after (FORRU-CMU), “years”,  “sites”, in Chiangmai

Table 1. Semicolon needed between the text in “Site history”

Please explain “rai”.

Remove “(5 years previously)” and “(2 years before)” - planted with framework tree species in 2007 (5 years previously), burnt in 2010 (2 years before) and replanted in 2012

Substitute ; on comma - limestone quarry floor; all vegetation and top soil previously removed

Add a space after the Table 1.

Figure 2. Increase the site names on the figure.   

Line 145. Perhaps -  per cent of vegetation cover or per cent of on ground vegetation cover

Line 231-232. A mean wood density value of 0.52g/cm2 was applied, from 231 the recent study of Pothong et al., (2021) [41] for forests in northern Thailand

Line 233-234 The above sentence is repeated.  

Line 236. The above ground biomass was calculated….  (delete: from the ground survey)

Line 240. Suggestion: WD is wood density [41]

Equation 6 and 7. Not sure about using ÷ in the formula. Please clarify.

Figure 4. Tree-stoking density (ha)? May be tree/ha?

Figure 5. Height (not heigh). Decrease the size of BMSM, ML, etc. on the images. The lines on the graphs should also be thinner.

281. R² (R2)  There is no need to give both indicators of correlation. R² could be enough.

Figure 6. Please make it a bit sharper and bigger. Did you check other models to adjust the data? For example, exponential or logarithmic? Use “tree” or “stem” throughout the text. There is no difference in provided models in terms of simple and multiple regressions. Please clarify this point.

428. Emphasize the novelty of your research in the conclusion section.

Author Response

Rev#1

This study provides focus on estimation of forest plantations in Chiang Mai University’s Forest Restoration Research Unit (Thailand) with the use of a consumer-grade drone platform. This is worthy goal that could help professional community increase knowledge in estimations of the forest restoration with drone-based imagery in this region of the world. Overall, the approach brings together some useful datasets to explore the use of canopy height models and drone-captured RGB images to quantify a range of forest indexes to help guide implementation and monitor of restoration projects. The study corresponds to the scope of Forest journal.

The work has some novelty value, but not highly significant, since drone based RGB imagery have been assessed in the past. However, it does add to what is known regarding the drone surveys of the forest plantations.

Despite this, the paper does have many weaknesses that must be addressed, particularly in the Introduction, Methods and in the Discussion sections. The most significant limitation of the study is that the results are based on only few test sites. A study whose conclusions are based on only five field plots on restored forest plantations can be considered rather as an initial study of the problem. The authors partially point out this fact in the article. I suggest that they emphasize this point in the discussion and conclusions section of the article.

It is presented as a pilot study – proof of concept.

It is not clear from the article how the use of drone RGB imagery would improve estimation of degradation levels of the studied forest plantations. The authors should explain this fact in the article or rewrite the objective and scope of the research.

There are many grammatical errors, and only a small proportion of these are listed below. Please check and revise the grammatical errors in the whole manuscript. The paper should be thoroughly reviewed by a native English speaker before it is considered for publication.

One of the authors is an English native speaker with more than 30 years’ experience of teaching scientific article writing classes to EFL students in English.

I would suggest the following title of the manuscript.  

Estimation of restored tropical forest plantations using drone RGB imagery  

The suggested new title is grammatically incorrect and is imprecise. FORRU-CMU does not do plantation forestry. In fact the paper focuses on ecosystem restoration as an alternative to plantation forestry.

Line 37. Delete comma after “projects”   DONE

Line 38. Better - in Africa (afr100.org/) and 20x20 Initiative in Latin America (initiative20x20.org/)). – this suggested revision is missing the definite article. The original is more grammatically correct.

Line 40-43. I would not fully agree with this conclusion because forest restoration often implies the introduction of the ecological functions, including the conservation of biodiversity and carbon storage.

It is not a conclusion – it’s part of the introduction and is supported by a citation. We replaced large scale restoration with “tree-planting projects” to be clearer.

Line 44-48. Please clarify sentence, as it is a little bit confusing. – we cut a sentence to ease flow and made the distinction between plantations and ecosystem restoration.

Line 56. I would suggest to use “index” instead of “indicator variable”. Variables usually used in modelling or regression equations. Disagree – an “index” is the value of indicators at any given moment.  At this line …  we list the types of variable not their index values. We are working on an automated degradation index that would combine the variables in this paper with others, but we are not there yet. Therefore, this paper is confined to quantifying individual indicator variables. I think that’s clear from the title.

Line 68.  For forest restoration planning – Done

Line 72-85. I suggest to eliminate or rewrite this paragraph, since most of the text has descriptive (general) character, while LiDAR technologies are not relevant to the topic of this work – shortened and re-written and to emphasize why RGB imagery was used in this study rather than lidar.

Meanwhile, there is need more analyses focusing on the previous research findings in the field of study. Therefore, additional review of the scientific articles is required. – we have already cited 29 references in the introduction and have used them to identify those gaps in literature coverage, which our study seeks to fill. As this is not a literature review, we feel that further literature analyses would lengthen the introduction unnecessarily, without adding to the paper’s central points.

Line 103-106. Please clarify the goal and tasks of the research. So far it is too vague. Also emphasize the novelty of your research based on the new models (patterns) you have developed. – paragraph stating the objectives of the research was revised (now line 93). The novelty of the research is stated in the preceding paragraph (line 86) and at the start of the discussion.

Please add a flowchart to the “Materials and methods” section to make it easier for readers to understand the research flow. – Done

Line 112. Add “sites” after All.  Change to were established – Done

Line 113, 115, 118 Delete comma after (FORRU-CMU), “years”,  “sites”, - Done

Table 1. Semicolon needed between the text in “Site history” – Done, bullets replaced by semicolons.

Please explain “rai” - removed

Remove “(5 years previously)” and “(2 years before)” - planted with framework tree species in 2007 (5 years previously), burnt in 2010 (2 years before) and replanted in 2012 – Done

Add a space after the Table 1.

Figure 2. Increase the site names on the figure.   Done

Line 145. Perhaps -  per cent of vegetation cover or per cent of on ground vegetation cover -  Figure 2. Orthophotos showing size, shape and boundaries of the five sites with unified scale. (BMSM): Ban Mae Sa Mai; (ML): Mon Long; (BPK): Ban Pong Krai; (BMM): Ban Meh Meh; (LP): Lampang.

Line 231-232. A mean wood density value of 0.52g/cm2 was applied, from 231 the recent study of Pothong et al., (2021) [41] for forests in northern Thailand; Line 233-234 The above sentence is repeated.  - CORRECTED

Line 236. The above ground biomass was calculated….  (delete: from the ground survey) - Done

Line 240. Suggestion: WD is wood density [41] - Done

Equation 6 and 7. Not sure about using ÷ in the formula. Please clarify. – regular mathematical notation

Figure 4. Tree-stoking density (ha)? May be tree/ha? – changed vertical axis to Tree-stocking density (stems/ha)

Figure 5. Height (not heigh). Decrease the size of BMSM, ML, etc. on the images. The lines on the graphs should also be thinner. – Height done.

281. R² (R2)  There is no need to give both indicators of correlation. R²could be enough. – Done we’ve deleted all r values and convert all to R² throughout.

Figure 6. Please make it a bit sharper and bigger. - Done

Did you check other models to adjust the data? For example, exponential or logarithmic? – yes

Use “tree” or “stem” throughout the text – stocking density is defined as tree stems per hectare (179) – “tree” is used in all other respects.

There is no difference in provided models in terms of simple and multiple regressions. Please clarify this point. – I don’t understand this comment. Please clarify

428. Emphasize the novelty of your research in the conclusion section.

Added this at the start of the discussion “The novelty of this research lies in the application of a cost-effective off-the shelf drone with RGB camera and open source software to distinguish degradation levels in trial plots of know age since the initiation of forest ecosystem restoration.  

Reviewer 2 Report

Although this research topic is not very new, it is developing rapidly. Many readers will pay attention to the contents of this paper. Some things still need to be improved:

Table 1. The latitude and longitude columns should be made into 1 column

Lines 129-130: Add justification and references.

L. 130: show the positions of the circular sample plots (5-m radius) as part of Figure 2.

L.131-136: In these lines, the authors should explain why this study did not use sufficient ground control points (GCPs). Placing several GCPs in each observation plot in mapping using drones is already a “standard procedure” for high-precision studies. GCPs will be used as a reference to improve mapping accuracy and produce more accurate maps or 3D models.

L.156: add considerations and references to choosing a drone flight path altitude 50 m above ground level.

L.238: The equation is only for trees with GBH>15cm. What about trees taller than 0.5 m (line 141) but having GBH < 15cm?

L.367: Statements/suggestions for using lidar sensors contradict the background of this study. Authors should look for alternatives to improve the accuracy of the models produced by drone mapping.

Author Response

Rev#2

Although this research topic is not very new, it is developing rapidly. Many readers will pay attention to the contents of this paper. Some things still need to be improved:

Table 1. The latitude and longitude columns should be made into 1 column - disagree

Lines 129-130: Add justification and references.- “The BMM site, formerly mixed evergreen-deciduous forest [31] at moderate elevation (601 m a.s.l.), was…” – not sure what justification is needed here. There’s a reference for the forest  type.

2. 130: show the positions of the circular sample plots (5-m radius) as part of Figure 2. – Done.

L.131-136: In these lines, the authors should explain why this study did not use sufficient ground control points (GCPs). Placing several GCPs in each observation plot in mapping using drones is already a “standard procedure” for high-precision studies. GCPs will be used as a reference to improve mapping accuracy and produce more accurate maps or 3D models.

We deal with this point in the discussion LINE 397, which has now been amended. “Therefore, it is recommended that future studies that depend on absolute height values in CHMs should apply geo-referencing using elevation data. Specifically, use of GCPs, with precise coordinates measured by RTK or PPK, is optimal. However, such technologies do not come as standard on off-the-shelf drones and they are expensive. Therefore, in order to minimize errors in positioning of sample plots in drone imagery in this study, we replaced the use of GCPs with the stadia method ……. ”

L.156: add considerations and references to choosing a drone flight path altitude 50 m above ground level.- added “…50 m above-ground level (based on experiments at various heights and photogrammetry results at these sites) …”

L.238: The equation is only for trees with GBH>15cm. What about trees taller than 0.5 m (line 141) but having GBH < 15cm? – just checked the reference and the 15 cm figure is a mistake. Pothong et al.’s minimum size was 1 cm diameter

L.367: Statements/suggestions for using lidar sensors contradict the background of this study. Authors should look for alternatives to improve the accuracy of the models produced by drone mapping.

We compare our technique with lidar at several points in the ms, making the point that although lidar has advantages, it is expensive and not yet standard on consumer drones. So we feel that “This problem may be overcome in the future by use of lidar sensors, which are better at revealing forest understorey structure than RGB cameras [19]. However, at present they are prohibitively expensive” should remain.

Reviewer 3 Report

I read the paper carefully. The authors tried to use UAV aerial images and photogrammetric techniqes to monitor indicator variables of tropical forests degradation. Generally many sections of the paper need to be reorganized, and I have some serious concerns too, such as:

The paper needs to be revised again in the English language as I detect some grammatical errors in abstract (e.g. initial degradation levels, to quantify variables related to initial degradation and etc.) and also introduction. In addition to, I observed a lot of very long or very short sentences (the average sentence length of scientific paper should be around 20–25 words.). Therefore, I think the language of the paper need a significant improvement and a professional revision.

Lines 64-67: present more differences and a deeper comparison between UAV and airborne/space born remote sensing datasets. Such as the cost of data acquisition, the effect of clouds and atmosphere, the spectral resolution and etc.

Lines 87-89: … enables the use of RGB imagery to generate 3D 88 forest models that are almost as good as those achieved from lidar point clouds. I am not sure! How about dense forests? For example, when the bare ground has been covered with canopy layers, making it difficult to detect ground points and making it impossible to generate an accurate DTM. Also, some previous studies reported that there are a lot of problems in generating 3D models using SFM, such as image alignment. Consider more details.

The introduction section should be revised to include more information about this paper in order to provide readers with an overview and conclusions on your topic. Be more specific in your objectives and formulate your research questions and hypotheses. 

Figure 1: add a grid to map and present the coordinate system. Also this map need more information such as neighboring countries and legend.

Table 1: Add more relevant information, such as the mean canopy cover in each flight zone (study site), tree density, the date of the UAV data acquisition and so on.

Lines 129-146: Add a new section, "eq. in situ measurement" or "reference dataset," and replace all field work measurements there.

Lines 147-161: Here, add a new section heading and name it “UAV aerial image acquisition” or ...

Lines 147-161: How about georeferencing UAV aerial images? Did you use ground control points? If yes, present your georeferencing errors in a table. If not, Why? What is the accuracy of your GPS for georeferencing?

Lines 163-168. How did you create DTM and DSM? What kind of point clouds did you use? Did you consider point-cloud classification? How many ground points did you use for DTM generation?

Line 173: pictures? Or Images?

Lines 173-175: DSMs and 173 DTMs were generated at 2-cm resolution, and orthophotos were generated at 1-cm resolution. Why not in the same resolution?

Lines 200-201: … Minimum crown height was set at 0.5 m above the ground to prevent confusing 200 ground vegetation with crown foliage. How this threshold was decided? Did you use in situ measurements or another reference dataset?

Line 214: … In the orthophotos, ground vegetation and tree-crowns were similar in colour. Use scientific languages and be specific.

Lines 214-221: Why didn’t you use classification techniques such as object-based classification methods? Which accuracy assessment method did you use? You generated a base map for your methodology to identify tree canopies, ground vegetation, and exposed soil and rock. But you didn't consider common classification methods and also accuracy assessment results.

My recommendation is for major revision or rejection with a suggestion to resubmit.

Author Response

Rev#3

I read the paper carefully. The authors tried to use UAV aerial images and photogrammetric techniqes to monitor indicator variables of tropical forests degradation. Generally many sections of the paper need to be reorganized, and I have some serious concerns too, such as:

The paper needs to be revised again in the English language, as I detect some grammatical errors in abstract (e.g. initial degradation levels, to quantify variables related to initial degradation and etc.) and also introduction. In addition to, I observed a lot of very long or very short sentences (the average sentence length of scientific paper should be around 20–25 words.). Therefore, I think the language of the paper need a significant improvement and a professional revision.

As an experienced professor of scientific writing who has instructed EFL students for over 30 years, I find no compelling reason for significant revision of the English language in this manuscript, aside from addressing minor typographical errors during the proofreading stage. The syntax and length of sentences align with those found in other articles accepted by FORESTS. Moreover, given the abundance of errors present in the reviewer's comments, it is difficult to consider their qualifications for assessing language proficiency. It should be noted that the responsibility of a scientific reviewer is to evaluate the scientific merit of a manuscript - not to act as copy editor.

Lines 64-67: present more differences and a deeper comparison between UAV and airborne/space born remote sensing datasets. Such as the cost of data acquisition, the effect of clouds and atmosphere, the spectral resolution and etc.

The purpose of this paper is not to serve as a textbook on remote sensing. The introduction already cites 29 references to substantiate the specific points addressed in the manuscript. Addressing topics beyond the scope of the paper, such as spectral resolution and cost-benefit analyses, would needlessly prolong the introduction and divert attention from its central points. Additionally, another reviewer requested that we cut some of the original text that addressed these areas – so, we cannot satisfy both. We’ve gone with Rev#1’s point on this for brevity and focus. 

Lines 87-89: … enables the use of RGB imagery to generate 3D 88 forest models that are almost as good as those achieved from lidar point clouds. I am not sure! How about dense forests? For example, when the bare ground has been covered with canopy layers, making it difficult to detect ground points and making it impossible to generate an accurate DTM. Also, some previous studies reported that there are a lot of problems in generating 3D models using SFM, such as image alignment. Consider more details.

Your uncertainty is noted. Please note the word “almost” in the sentence. We indeed mention the advantages of lidar particularly for forest structure in the discussion. 

The introduction section should be revised to include more information about this paper in order to provide readers with an overview and conclusions on your topic.

Conclusions are placed in the conclusions section. Overview and conclusions are also in the abstract.

Be more specific in your objectives and formulate your research questions and hypotheses. 

We have revised the objective paragraph at the end of the intro following another reviewer’s recommendation.

Figure 1: add a grid to map and present the coordinate system. Also, this map need more information such as neighboring countries and legend.

The location of the plots is clearly marked within Thailand, so there is no need for a grid system. We added  “White lines are Province boundaries within Thailand.” to the legend. I think it’s safe to assume a rudimentary grasp of geography amongst readers of FORESTS

Table 1: Add more relevant information, such as the mean canopy cover in each flight zone (study site), tree density, the date of the UAV data acquisition and so on.

Details of the flights are in the supplementary materials. Canopy cover is presented in Figure 8 and tree density in Figure 5.

Lines 129-146: Add a new section, "eq. in situ measurement" or "reference dataset," and replace all field work measurements there.

Regrettably, I am uncertain as to the specific reference denoted by "eq." (an equation perhaps?), as well as what it is that you want us to  “replace”. However, after making my best effort to comprehend this remark, I surmise that you may be suggesting the inclusion of a table detailing the ground-survey data. We added table S2 to the supplementary materials and referred to it in the text.

Lines 147-161: Here, add a new section heading and name it “UAV aerial image acquisition” or ...

DONE

Lines 147-161: How about georeferencing UAV aerial images? Did you use ground control points? If yes, present your georeferencing errors in a table. If not, Why? What is the accuracy of your GPS for georeferencing?

We used the stadia method to deal with positioning of plots in images, as explained in the text and with more details in the Fig S5.

Lines 163-168. How did you create DTM and DSM? What kind of point clouds did you use? Did you consider point-cloud classification? How many ground points did you use for DTM generation?

We created the DTM and DSM using ODM. The point cloud was generated by the SfM algorithm within the software. Point cloud classification did not work well enough to be used in this software. We used the stadia method to replace GCP’s as mentioned above.  

Line 173: pictures? Or Images?

Rewritten …. “Between 250 and 400 photos, captured, at each site (depending on site size), were processed.”

Lines 173-175: DSMs and 173 DTMs were generated at 2-cm resolution, and orthophotos were generated at 1-cm resolution. Why not in the same resolution?

The quality of DSMs and DTMs created at 1-cm resolution was poot (spiky and rough pixels), but for orthophotos 1-cm resolution could be achieved with better quality. In general, differences in resolution between the DEMs and orthophotos did not matter, as the data from the DEMs and orthophotos were analyzed independently.

Lines 200-201: … Minimum crown height was set at 0.5 m above the ground to prevent confusing 200 ground vegetation with crown foliage. How this threshold was decided? Did you use in situ measurements or another reference dataset?

In the ground survey, we counted and measured trees taller than 0.5 m. So, the same threshold was applied to detect tree boundaries from the CHMs. Furthermore, in a  pre-analysis (comparing threshold settings of 0.5m, 1.0m, 1.5m) the 0.5m setting worked best. The 0.5 m minimum size is standard for surveys when monitoring tropical forest ecosystem restoration. Elliott, S.D.; Blakesley, D.; Hardwick, K. Restoring Tropical Forests: A Practical Guide.; Royal Botanic Gardens, Kew, 2013; ISBN 978-1-84246-442-7

Line 214: … In the orthophotos, ground vegetation and tree-crowns were similar in colour. Use scientific languages and be specific.

I really don’t understand the reviewer’s confusion over this sentence or how many languages are required. However, we have rephrased it in a single language (English) as follows: “In the orthophotos, ground vegetation and tree-crowns were difficult to distinguish in terms of their colour alone.”

Lines 214-221: Why didn’t you use classification techniques such as object-based classification methods? Which accuracy assessment method did you use? You generated a base map for your methodology to identify tree canopies, ground vegetation, and exposed soil and rock. But you didn't consider common classification methods and also accuracy assessment results.

I am not sure as to the specific “classification” techniques the reviewer is referring to, although the term “object-based classification” in forestry typically alludes to the use of spectral signatures to identify tree species. However, as this paper is a preliminary study, aimed at determining whether drones can aid in the recognition of broad levels of degradation, there was no requirement to classify trees within the images, according to their type or species. Nevertheless, we acknowledge the potential of such techniques and indeed are currently collaborating with a UK university to employ AI for identifying tree species from drone images. The accuracy of the drone-based methods was evaluated against ground survey data, and the results of the correlations are presented. It should be noted that neither of the other two reviewers raised this issue.

My recommendation is for major revision or rejection with a suggestion to resubmit.

Round 2

Reviewer 1 Report

      The manuscript has been improved after the modifications. Although not all recommendations were accepted by the authors, it is good to see the most of them have been accepted and revised. I think this version of manuscript has been enough to be published.

Reviewer 3 Report

The paper is now ready for publishing.