The Effect of Topographic Correction on Forest Tree Species Classification Accuracy

Round 1

Reviewer 1 Report

Congratulations to the authors for the new version of the manuscript previously submitted. I just wanted to express my consent to the new version of the manuscript and accept it in the present form.

The authors have addressed all the comments and suggestions included in previous rounds of revisions and they could make it compatible with the reviews of the other reviewers.

Author Response

Thank you very much. There is no comment here

Reviewer 2 Report

Dear Authors,

You have to improve the language of your paper; You have to reduce the number of words in the article; Avoid general statement, e.g., L42 - 57; Verify and unify the terminology, e.g., 'ground shape,' 'terrain fluctuations,' etc.; Figure 5 should be a bar chart; How do you know that the accuracy of DEM used to calculate the topographic correction has a significant impact on the topographic correction?

Good luck.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript titled “The Effect of Topographic Correction on Forest Tree Species Classification” presents a study examining the effect of four different topographic correction techniques on the subsequent classification of specific forest species in China’s Mount Taishan, in Shandong Province. I believe the work is novel, and contributes to current understandings of the effects of topographic correction on subsequent image analysis in forested areas. However, I have suggested major revisions be undertaken to address a number of concerns. These largely relate to clearer or more detailed explanations of portions of the authors’ methods, improved writing where explanations or statements are very unclear or confusing, and the expansion of the authors’ discussion to address a number of notable patterns in their results.

 

Introduction Comments

 

The introduction needs to flow better, and follow a more logical order of concepts (e.g., start with describing airborne remote sensing efforts in the forest species classification, and then describe what is done with satellite imagery, while also explaining the advantages and disadvantages). I also recommend the authors expand somewhat on their description of past work – who did the work? What region or area of the globe did they work in? How good were the resulting classifications (how successful were they)? In addition, I believe the authors’ use of the subjunctive verb tense (“could”) should be replaced with the active tense (“can”) wherever possible, and that they examine their use of “although”, “however”, and “therefore” – it is not always used appropriately.

 

Line (Ln) 47: what do you mean by “artificial investigations of ground sample plots”, and by “3S technology”? These need to be explained.

Ln 49-51: this is where I recommend you expand your descriptions of previous work, for example.

Ln 61: It’s good to explain that topographic and terrain correction are both terms used for this concept, but I would strongly suggest the authors pick one of these terms (topographic correction or terrain correction) and stick with it throughout the manuscript! There are a number of places one or the other is used, and it is inconsistent and could be confusing.

Ln 69: what do you mean by “the exponential value”?

Ln 86: what do you mean by “being unable to explain the classification principle.”? What is the “classification principle”, and why don’t these approaches explain it?

Materials and Methods:

Ln 100: please explain what the Mount Taishan forest farm is. Is this the reason that your study area is not one contiguous/spatially exhaustive unit and has islands or pockets of study area separate from the main area? There needs to be an explanation of why this is.

Ln 103: what do you mean by “changes vertically”? Do you mean changes with elevation? If so, please use that term instead.

Ln 107: stating “the proportion of pure forest is large” without explaining what you consider as “large” is too vague. Please clarify.

Figure 1: This caption could be more descriptive, and the figure itself needs 1) some titles or labels on the inset maps, and 2) a better way of displaying the distribution of tree species (the colored lines are very hard to distinguish). Also: where did this information on the distribution of tree species come from? What is its source? In your methods section, does this play a role in your methods? If so, that needs to be very clear. And if not, please explain why not.

Section 2.2.1 (Satellite data): please explain why you chose to use Landsat 8 imagery (instead of, for example, Sentinel-2). If you are classifying tree species, it would seem that Landsat’s spatial resolution would/could lead to a lot of mixed pixels, which would be less likely with a higher spatial resolution sensor such as Sentinel.

Ln 137: the illumination conditions (IC) layer is important in the authors’ analysis, and needs to be fully described here (how it is generated and what the data contain)! There are not explanations of this in the paper, even though this layer is a critical component of the analysis.

Ln 138: you mean “were calculated using the DEM data”. Please list what “other data” aside from slope were calculated from the DEM. This needs to be clear.

Section 2.2.3 (Subcompartment data): this section is very confusing and very unclear. First, what is “subcompartment data”? What is a “subcompartment”? What is “indoor subcompartment division”? Is this section describing the field data that was used for training and validation of your classifications?  The collection of field data needs to be very clear here. Where was data collected? How was that decided? What sort of data were collected? How many field sites were sampled? Etc.

Ln 167-168: please explain what variables the linear regression is calculated on. This is not clear here. Is this the same as the regression mentioned on Ln 183? What are the band’s spectral values regressed against?

Ln 185: I would just use “evaluation of the topographic corrections” here; I would not use the term “effect” here.

Ln 187: again, just use “topographic correction” not “quantitative correction”. Using different terms for the same thing gets confusing. Also, I would suggest that you are evaluating the “efficacy” or “effectiveness” of these corrections (i.e., how well they work), not their “effect”

Ln 200-201: “the training data were produced based on IC data and subcompartment data” – what does this mean? Please explain the procedure in detail. The quality of training data is critical to the success of any classification, and must be explained clearly. Also, you need to explain the difference between the “full coverage” and “shadowless” training data sets and why you chose to do this. Why is this important to your analysis and what question(s) does it answer?

Ln 204-206: please explain how dividing the training data into the two groups helped with overfitting with the random forest classification. Also: does the “full coverage” (I assume that’s what the “including shadow regions means” and suggest you clarify that by only using one term to refer to it) share training samples with the “shadowless” training data set? Finally: did you split your reference data set into training vs. validation? What proportion of the data did you allocate to each?

Figure 2: this is not a sufficiently detailed caption for this figure (also: “technology roadmap” does not accurately describe the purpose of the figure)

Ln 210: this section heading does not describe what you mean. I believe you mean “evaluation of the effect of different topographic correction models on tree species classification”

Ln 220-222: I don’t know what you mean here when you say “superimposing the calculation results with IC data”. Also: by “inconsistency rate” do you mean “classification accuracy”?? If so, then use the latter term as it is an established term in the literature. Finally: I believe you mean that you “evaluated the effect of tree species on the relationship between topographic correction and classification accuracy.” Is that what you mean? Please be clearer!

Results Comments

Ln 224: to me the section heading for 3.1 doesn’t describe the section very well. I think you mean “Evaluation of topographic correction models”. You don’t need to include “analysis of indicators” in there – that is assumed given your described methods.

Ln 239: how did you evaluate the color depth of the corrected images? This should be described in your methods

Ln 241: when you write “All images better retained the texture features of objects”, 1) what do you mean by “texture features of objects”, and 2) better than what exactly?

Ln 246: be consistent and use “topographic correction” rather than “illumination correction”

Section 3.1.2: while it’s true that a reduction in SD of the band’s reflectance values after topographic correction is a good sign, is it possible to have too much of a reduction (i.e., you lose variability that you want to maintain in the image)? If so, how do you determine when over-correction has happened? Please address this issue.

Table 2: again, needs a more descriptive caption, and to explain what TOA and SR mean

Ln 261: I think you mean to say that the NIR band responds well the changes in vegetation. Saying that it is closely related to vegetation is inaccurate/misleading

Ln 262: you are describing the effects of topographic correction on NIR reflectance here, but you refer to “red band correction” here. Is this a mistake?

Ln 263: there is no need for the “However” in this sentence

Ln 269: to me it appears that the correction models compressed the original data (and reduced variability as shown by SD results) rather than stretching them. Please clarify.

Ln 277: I think you need to explain here why the results you’re describing mean that these two particular models produced more successful topographic corrections. E.g., what would the ideal correction look like in the resulting histogram? How do you know these are ‘better’, exactly?

Figure 4: again, needs a more detailed caption. Captions should be detailed enough for someone glancing at the paper to be able to understand the figure/table without having read all of the paper

Ln 282: how exactly did you compile the data to compare changes in SD for these specific tree species? Please explain in your methods.

Ln 287: I would reword this to: “The SD of the SCS+C model-corrected image…” I suggest you be very clear in your text when you are talking about the models themselves, or the topographic model-corrected images.

Figure 5: expand this caption to be much more descriptive

Ln 295-296: what do you mean by “satisfying the water body reflectance…”? And please clarify how a change in water body reflectance (not water band…) reflects the validity of the correction model (rather than ‘algorithm’ – be consistent in your terms).

Section 3.1.4: I like this analysis, I find it very informative. I do have some questions: did you look at water bodies in shaded and unshaded contexts, separately? Or did you look at all of them at once? Or did you only use one? Also, how did you define and sample water bodies? Did you use a pre-existing database of boundaries? Or something else? Please clarify in your methods

Ln 324: the term “density relation” doesn’t make a lot of sense. Perhaps try: “By constructing density plots between the IC dataset and individual bands, we can…”

Ln 336-337: does this mean that Figure 7 shows the results only for pine? This is not clear in the text, or in the figure caption. It needs to be made explicitly clear.

Figure 7: it would be incredibly helpful to have labels on each individual plot, rather than relying solely on the caption to figure out what’s what.

Ln 351-355: I am confused by the classifications from “training sample points covering shadowed areas” vs. those from “training sample points without shadow coverage”. Those with non-shadow training points are classifications wherein the training data was taken only from sample points that were not in shadow according to your IC layer. Is that correct? And the classifications wherein the training data was taken from points covering shadowed areas actually include training samples from all levels of shade (i.e., shaded and non-shaded together). Is that correct, too? If so, these should be referred to as “full coverage” perhaps. What you have written here suggests the training samples came from only shaded areas. This needs to be more explicitly explained in the methods, and referred to here in clearer terms.

Ln 369-371: in these sections of the results, you’ve focused on discussing results from only the two best models. But here, you all of a sudden bring all four models back into your descriptions. That is very confusing, and inconsistent.

Figure 8: it would be nice to see the pre-corrected classification for comparison!

Table 5: these numbers are very large, and hard for a reader to compare quickly/easily. Perhaps showing the % change from SR would be helpful?

Section 3.2.2.: Some of this section is not very clear; I would suggest rethinking this section and how it is presented. A reader does not know in Figures 9 and 10, for example, what the ‘correct’ classification should be or how much better one is over the other – perhaps adding in accuracy results to remind the reader would be helpful.

Section 3.2.3. The heading for this section is the same as the heading for overall section 3.2. I recommend changing that.

Figure 11: please explain what the standard IC and the disagreement rate values shown in these graphs represent, and please explain more clearly how these graphs are to be interpreted. For instance: disagreement between what, exactly?

Discussion Comments

Ln 429: I feel like this first sentence can be worded so much better, and more clearly. Perhaps something like: Topographic correction can reduce the effects of varying topography/terrain surface and subsequent shadowing on spectral reflectance.

Ln 446-450: there was no description of methods or results that compare the different DEMs and these effects on classification accuracies. This must be part of the methods and results if it is something you are going to address here!

Ln 449: you mention “the IC calculation”; this calculation is not described in your methods and very much needs to be.

Ln 457: what is the illumination coefficient, exactly? Also something that should be explained in your methods if it is going to be mentioned here

Ln 464: I would use “spectral band”, not “wave band”

This discussion section is very short, considering all the results and interesting patterns that should be addressed. I believe the discussion needs to be greatly expanded to address at least some of the results/patterns described in the results section. For example: it is obvious that the topographic corrections affected different bands differently (and different models produced different patterns here, too). Please discuss this, and speculate on reasons for this, and how this would influence one’s choice of topographic correction method given the particular spectral bands they would be interested in. Also: why are results so different when using training samples from only unshaded locations vs. all locations? This is a very interesting result, and should be further discussed. In addition: I would like to see some discussion on the differing results from different tree species and potential reasons and implications of this. The use of topographic corrections for classifying particular species is novel in the literature. This deserves some discussion! I would also enjoy seeing some speculation by the authors on how their results might differ if they performed this analysis on spectral indices (like NDVI, or others).

Conclusion Comments

Ln 470 – 481: I don’t understand what you mean by this sentence: “The inconsistency rate [what is the inconsistency rate?? Is it accuracy?] and the IC histogram [which one?] showed that the relationships between….” Please reword this sentence to make it comprehensible.

Ln 496: Please re-iterate what the “optimal model and classification process” are; I’m not sure this is actually made very clear in the paper.

Finally: there are so many different terms used to refer to the topographic correction models, to classification accuracies, and to some other concepts. It is important to choose one term to refer to one concept (particularly if they are critical to your paper), and be consistent in using it. This will make things much clearer.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

I wish to thank the authors for the obvious effort they have put in to improving their manuscript. It is more readable, more consistent in its terminology, and more clearly explained in many sections. The following are my comments on the revised version of the manuscript:

Introduction:

Thank you for adding more background into the introduction. I personally believe this background on the remote sensing-based identification of forest species could be expanded, but I understand the desire to not increase the length of the manuscript too much.

Ln 46: should be “multiple spectral bands” (rather than “multi bands”)

Ln 53: the term “classification effect” does not make sense. Do you mean: “its effectiveness/efficacy in species classification” perhaps?

Ln 55: please explain what this new approach is. It does not make sense to state that previous literature has presented a new, improved approach but not describe what that approach is.

Ln 70: use “can” instead of “could”. Many of the “could”/”would” words in the manuscript have changed to more appropriate “can”/”will” words, but some still remain. Please use “could”/”would” only in appropriate contexts (e.g., speculating on a possible outcome, not a demonstrated or shown outcome).

Ln 85-86: “Many models have good effects on…” is very awkwardly worded. Again, the word “effect” doesn’t really fit in (as it doesn’t in many places in the manuscript!. Perhaps something like: “Many models are effective for the topographic correction of…”

Ln 93: the statement “being unable to explain the classification mechanism” is still to vague and the meaning is unclear. Do you mean that the results of these models (the models themselves) are not easy to interpret?

Ln 103-104: I find the last sentence here misleading. It suggests that at the end of this work you recommend one particular approach above all others. Perhaps you mean “This study explores effective methods for accurate tree species….”?

Methods:

Ln 120: the term Quercus is removed here, but not replaced by something else. I would personally suggest using the common tree species term “oak” instead of “Quercus” throughout the manuscript, just as is done for “pine” species.

Figure 1: this is a great improvement – thank you. Much more legible.

Ln 138-140: this sentence is awkward. Perhaps something like: “We chose Landsat 8 bands 2 through 7 for our analysis, but focused our efforts on bands 4 (red) and 5 (near-infrared, NIR) as these bands are shown to respond most strongly to surface vegetation conditions.” The statement “best reflect the vegetation situation” does not make a lot of sense and is very vague.

Ln 141: I would use “accessed” rather than “obtained”.

Ln 150: again “topographic correction effect” is not a meaningful term. I think “topographic correction effectiveness (or efficacy)” is a good alternative.

Section 2.2.3.: Thank you for adding a more detailed description of the “subcompartment”. These appear to be polygons representing individual, homogeneous forest stands. Is that correct? Can you please explain that in your description in the manuscript?

Ln 160: include a reference to the original source for this map (e.g. a government agency?).

Ln 161-162: what do you mean by “the small class map of the research area”? What is the small class map? Is it something extracted from the subcompartment data? Please explain in the manuscript.

Ln 173: what do you mean exactly by a “forest plot”? This is a term often used to describe a specific area that is sampled on the ground, but I don’t believe that is your meaning here. Do you mean that you chose point locations you assumed had pure forest species compositions, based on the subcompartment data? Please clarify in the manuscript.

Ln 178: I would write out IC (illumination conditions) the first time it is mentioned in this section (because this is the section where it is defined).

Ln 191: should be reworded; something like: “the slopes and intercepts of the linear regression calculated between the IC data and this particular band”

Ln 196: should be “Topography affects….”

Ln 206: should be “were calculated for” instead of “were calculated by”

Ln 213: please be more specific in the manuscript about how “the water area” was selected

Ln 253: it is unclear what you mean by “shadowless training data avoid the algorithm covering up the effect of topographic correction.” Please briefly explain how/your reasoning here in the manuscript.

Ln 250: I would specify “tree species” rather than “tree type” (type could be taken to mean deciduous vs coniferous, for instance)

Ln 252-255: I’m afraid the explanation of “inconsistency rates” is still quite unclear here. Do you mean that for each pure forest subcompartment area (for each species), the proportion of misclassified pixels within the area is calculated (i.e., 1 – proportion of correctly classified pixels)? And that this was done for each reclassification region? What are the IC data reclassification regions?? These are not explained anywhere in your methods, but need to be explained clearly.

Results:

One thing that is not clear in this section is whether the results you are presenting here regarding evaluating the topographic corrections themselves pertain to one particular Landsat scene (of the four you use in your analysis)? If so, which one (e.g., in Figures 3 & 4, please specify what image scene/date you are showing here)? Or do these results reflect results from all four scenes? Did you compare the results between Landsat scenes (particularly since they represent different phenological periods)? What were your results? Or if you didn’t, why not?

Section 3.1.: I would use “Effectiveness of the topographic correction models” since “effects” is not the right word

Table 2: please specify that SR refers to “surface reflectance” and represents pre-corrected conditions. Same for other figures/tables comparing ‘before’ and ‘after’, and showing SR.

Ln 328: do you mean “forest species composition information in the pure forest data”? Or do you mean structure (e.g., height, etc.)?

Ln 330: I believe you mean “when we examined a single tree species…”, not counted ; same on Ln 332

Ln 334-346: This first sentence is awkward and somewhat unclear. Do you mean something like this: “Changes in water body reflectance can be used to evaluate the validity of a topographic correction model by statistical comparisons of this reflectance before and after the correction has been applied to the study area (Figure 6).”

Figure 6. I think by this caption you mean “Percent change in band reflectance over water bodies after topographic correction by each of the four models”. The current caption is unclear.

Table 3. Use “spectral band” instead of “waveband”

Ln 451: by “classification degree”, do you mean “classification accuracy”?? Please use correct terminology.

Section 3.2.3. Previously I commented that this section heading should not have the same title as the larger section itself. However, I do not believe that ‘terrain’ should be substituted here, in the interest of keeping with the consistency of terms in the manuscript. I would stick to the original heading if there is not a better alternative. Same with Ln 465 – I would not replace “topographic” with “terrain” here.

Discussion:

Ln 495 – 496: again, the “effect of topographic correction” or “topographic correction effect” is not meaningful. Please use “effectiveness” or “efficacy” here. Same thing on Ln 521.

Ln 505: what do you mean by “completely explained”? Do you mean that they are not able to remove all illumination differences resulting from topographic variability?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

General comments:

The submitted paper is an interesting approach to the assessment of topographic correction on forest species classification accuracy in China. The methodology design is appropriate, the statistical analysis is well grounded and the results are significant. Nevertheless, I consider that the results presented are too local and the paper has critical deficiencies that should be amended in future versions. Extensive editing of English language is required, and it cannot be accepted for a high impact international journal.

The structure of the manuscript is polished and well written. However, there are many some lacks in the presentation of the manuscript, which would need extra amendments.  

 

 

Structure of the manuscript

Title

The title is very proper for the presented study.

 

Abstract

I found the abstract adequate and concise.

Line 16: Change the verb “realize” to a more appropriate verb.

Line 21: Change “After” to a more appropriate verb word.

Line 27: Specify what “It” means in this sentence.

 

Introduction

This section needs to improve with the inclusion of more scientific background related to the research topic.

Lines 33-38: Improve redaction or remove. It is not necessary.

Line 43: What do you mean with “traditional field investigation”?

Lines 45-46: What do you mean with “It is one of the important directions of forest resources investigation”?

Lines 47-52: The breaks between sentences are very abrupt. Please, improve style and redaction.

Lines 63-64: This sentence would be better located in the discussion section.

Lines 67-69: This sentence would be better located in the discussion section.

Line 73: Change “In order to research” to a more appropriate redaction.

Line 75: Change “Through comparing” to a more appropriate redaction.

I suggest to state clearly which is the scientific gap of the study.

 

Material and Methods

Surface (ha) of the study area is missing.

Line 81: which located. Please check.

Line 83-84: Please remove the coordinates, they are not needed as you are showing them in Figure 1.

Line 84: Citation needed.

Line 86-90: The breaks between sentences are very abrupt. Please, improve style and redaction.

Lines 91-83: Include scientific names for the tree species listed.

Figure 1: Please make zoom to the extent of China in the lower left frame.

Lines 105-106: Please check English grammar and spelling.

Line 110: Include “(SRTM)”.

Lines 117-124: Please provide detailed information about number of images, number of bands and dates for the Sub-compartment data.

Lines 160-161: How do you perform visual comparison? Please, provide more information about this.

Line 178: Figure 2: More details needed for the explanation of this chart.

Lines 181-182: Please check English grammar and spelling.

Line 240: Why exactly 240 sampling points? Please, justify it.

Line 185-186: Please check English grammar and spelling.

 

Results and discussion

There is no discussion section that relates your findings with relevant scientific background. Once you prepare a real introduction section, it will also help you to make up a much better discussion section.

Line 193: You cannot start a sentence with a mere explanation of what you display in a figure. Please, improve redaction style and cite the Figures at the end of the sentence. This problem is repeated along the text.

Line 201: The break between sentences are very abrupt. Please, improve style and redaction.

Figure 3: Compare the area??? Please include EPSG.

Line 214: You cannot start a sentence with a mere explanation of what you display in a figure. Please, improve redaction style and cite the Figures at the end of the sentence. This problem is repeated along the text.

Please check the verb tense and time.

Line 224: You cannot start a sentence with a mere explanation of what you display in a figure. Please, improve redaction style and cite the Figures at the end of the sentence. This problem is repeated along the text.

Line 242: Please check redaction.

Line 244, 259, 272, 284, 309, 334: You cannot start a sentence with a mere explanation of what you display in a figure. Please, improve redaction style and cite the Figures at the end of the sentence. This problem is repeated along the text.

Line 272, 284, 309, 334, 356: Please check the verb tense and time.

Lines 328-329: How different the classifications are? Please provide data about changes observed.

 

Conclusion and outlook

Please rename the section to “Conclusions”.

 

References

Too few references for a research paper in a prestigious international journal.

Reviewer 2 Report

Dear Authors,

below, there are some points you should focus on while working on your manuscript.

Figure 1 must be improved. You have to justify the use of the SRTM data which are some 16 years older than the Landsat data. The SRTM data represent a quasi-canopy elevations or bare earth elevations, otherwise. It is possible that during that time gap (16 years) some new plantations were established changing the terrain elevation data. I suggest to use the latest TanDem-X 90 m. How Well can Spaceborne Digital Elevation Models Represent a Man-Made Structure: A Runway Case Study. Geosciences 9 (9), 387 or AW3D 30m model) model as it is much more accurate and current than the SRTM model. You have to be more specific about the forest inventory data you are describing in Section 2.2.3.as there are fundamental reference data.

L136-138: You have to justify this statement (provide some reference).

L156-157: Justify this sentence.

L158-168. You have to rewrite Section 2.3.2. Focus on the quantitative assessment only.

L162. Justify, why you use SD?

L175. Do you think that the NDVI would produce similar results to the Random forest algorithm? See for example, Conese, C., Gilabert, M.A., Maselli F., Bottai, L., 1993. Topographic normalization of TM scenes through the use of an atmospheric correction method and digital terrain models. Photogramm. Eng. Remote Sens., 59, pp1745–1753; Becek, K., Borkowski, A., and Mekik, Ç.: A STUDY OF THE IMPACT OF INSOLATION ON REMOTE SENSING-BASED LANDCOVER AND LANDUSE DATA EXTRACTION, Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B7, 65-69, https://doi.org/10.5194/isprs-archives-XLI-B7-65-2016, 2016.

Provide a more informative caption for Figure 3. What bands did you used to produce the image, etc.?

L390. I’m afraid that your result, i.e., 4% is too small to draw any reliable conclusion. Similar, 13% better over ‘shadow-less’ fragments appear to be too small for the comfort. There are some other terrain factors that impact the spectral properties of vegetation, e.g., moisture.

I my opinion this manuscript must be thoroughly rewritten before publication.

Reviewer 3 Report

The title of the manuscript can be rephrased to - The Effect of Topographic Correction on the Accuracy of Tree Species Classification. Page 1. Lines 12, 13: The objective of the study can be clearer - ‘…..improve 12 the accuracy of forest survey’ - what exactly is forest survey (classification or measurement of forest parameters, etc.)? Page 1. Line 16: ‘to realize data correction 16 algorithms’ - this part is not clear. Please correct it. Also, abbreviate GEE? Introduction can include additional literature review, specifically on the existing remote sensing studies for tree species classification, why is it important to reduce the influence as topographic correction or terrain illumination correction, why did you chose random forest method for classification, why you wanted to do this research using Google Earth Engine, etc. Some of the sentences throughout the manuscript do not convey the message fully to the readers. Please go through the manuscript and check for any language corrections. In Figure 1, change the color of Chinese Arborvitae and Black Locust; it’s really hard to see those on top of the false color imagery. Page 3. Section 2.2.1: It would be helpful to the readers if you tabulate what bands you used, their wavelengths, etc. Why is the rotation column blank in Table 1? Figure 4 caption can be rephrased. Figure 5, 7, 8, 11 are not very clear. Some text and captions are cut off. Please include a high resolution image. In the Conclusion section, future potential of the study can be included.