Improving Tree Cover Estimation for Sparse Trees Mixed with Herbaceous Vegetation in Drylands Using Texture Features of High-Resolution Imagery

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

The manuscript with the title “Exploring the potential of high resolution imagery to improve tree cover estimation in the grassland with sparse trees on the Chinese Loess Plateau”, presents a study that uses samples of 0.5-m tree canopy and 30-m tree cover using the combination of unmanned aerial vehicle imagery and worldview-2 (WV-2) imagery to compare against Landsat 8 tree cover estimates and employ different methods to investigate the ability of high-resolution imagery to infer tree coverage. The study's objectives are as follows: “(1) Explore the feasibility of directly inverting tree coverage using high-resolution imagery object-oriented classification in the Loess Plateau region. (2) Investigate evidence of the relationship between high resolution texture information and tree coverage. (3) Compare the differences between direct classification inversion and machine learning modelling estimation of tree coverage, as well as compare the contributions of texture information at different resolutions to tree coverage. (4) Compare the differences between the tree cover products generated in this study and global tree cover products.” At first glance, the paper takes into a worthy topic.  However, the structure is not strong and has some methodological flaws. The main concern of this reviewer is the difficulty that the paper has in highlighting the unique contribution compared to what is already available in the literature, mainly because the objectives cited are hard to follow and do not make clear what research questions or hypotheses they are based on.

I am having problems understanding the study's rationale. It could be straightforward that higher-resolution imagery leads to improved tree cover estimations compared to medium or coarser spatial resolution. The document's title is “exploring the potential of high resolution to improve tree cover estimations in…”, but later objectives and methods are disconnected, leading to confusion about the study's true aim.

In the data set description section (2.3), the authors mention two tree data sets: a) tree and non-tree at 0.5 m scale (643,725 classified sample points), and b) a tree coverage sample data set derived from WV-2, by employing UAV observations on 30 x 30 m sample points (lines 181-183) giving a result of 1132 tree cover samples. Here are some comments: first, the term “scale” is not properly used; it should be 0.5 m x 0.5 m  ( I assume) pixel spatial resolution. If I understood correctly, the authors used the first dataset to validate the second dataset. This is the tricky part: the remote sensing section does not state the spatial resolution of the UAV imagery. Later, in Figure 4, the spatial resolution is set at 0.02 m. Then, suppose the authors have images with 0.02 m and 0.5 m resolutions, respectively. Why create a 30 x 30 m sample data set for later use again as a validation (or “ground truth”) for estimates using different methods? It looks to me like it is running in circles, and it is obvious that the R2 should be high because the source is the same for each tree cover estimate except for that using only Landsat 8 data.

 

The methodology fails to describe how the objectives are fulfilled. More importantly, the overall methodology does not describe the exploration of the potential of high resolution on tree cover estimates; instead, it describes different methods to obtain tree cover estimates (Figure 4).

I understand that high-resolution estimates come from different approaches (visual interpretation, object-oriented classification, etc.), but why combine the high and very high to Landsat-derived features if the aim is to evaluate the performance of high-resolution imagery?

I am going to stop my revision here since the document needs a deep revision regarding the rationale, the objectives, and the method used.

For all the reasons stated above, this reviewer has decided to reject the manuscript in its present form.

 

Particular

Line 87-88. Why the geographical pattern of the Loess Platteu is unique?

Line 154-155 2.24. Tree cover products. What is the purpose of using MODIS-VCF products? Comparative purposes?

Lines 273-275. What predictive models? GLM?

 

 

Comments on the Quality of English Language

The document needs to go to professional editing English services since I found several Grammatical and spelling mistakes.

Author Response

Thank you very much for taking the time to review this manuscript. Please see the detailed changes in the attachment below.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Find my comments in the pdf file.

Comments for author File: Comments.pdf

Author Response

Thank you very much for taking the time to review this manuscript. Please see the detailed changes in the attachment below.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper presents a comprehensive and valuable contribution to the field of remote sensing and environmental modeling, particularly focusing on tree cover estimation in the Chinese Loess Plateau. The study addresses a significant gap in current tree cover products by proposing a novel methodology that combines high-resolution UAV imagery with worldview-2 (WV-2) imagery. By leveraging advanced feature extraction techniques and employing a random forest model, the authors successfully identify optimal spectral and textural features for accurate tree cover estimation, particularly in grassland areas with sparse tree cover. While the manuscript presents valuable insights into tree cover estimation using high-resolution imagery, there are several potential drawbacks that could be addressed:

1. The study focuses specifically on the Chinese Loess Plateau, and while the methodologies and findings are applicable to similar regions, the generalizability to other geographic areas with different environmental conditions may be limited. Discussing the potential applicability and limitations of the proposed methodology in other contexts could enhance the manuscript's impact.

2. The manuscript could benefit from a more thorough discussion of the uncertainties associated with the tree cover estimation process. Assessing and quantifying uncertainties arising from data preprocessing, feature extraction, and modeling could provide a more nuanced understanding of the reliability and limitations of the results.

3. In conclusion section, author highlighted the strengths of the proposed methodology, there is limited discussion of potential limitations or challenges encountered during the study. Addressing potential sources of bias, error, or uncertainty inherent in the methodology would contribute to a more balanced and comprehensive interpretation of the findings.

Author Response

Thank you very much for taking the time to review this manuscript. Please see the detailed changes in the attachment below.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Please see my comments in the attached file

Comments for author File: Comments.pdf

Comments on the Quality of English Language

Academic English writing should be revised.

Author Response

Thank you very much for taking the time to review this manuscript. Please see the detailed changes in the attachment below.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Many thanks to the authors for sending the revised version of the manuscript and addressing my revisions. I do acknowledge the effort of the authors. However, the answers did not cover entirely my concerns about the methodological approach followed and the unique contribution the manuscript makes compared to what is already available in the literature.

Response 1.  It is clear now that the two tree datasets were obtained using visual interpretation to train and validate object-oriented classification and random forest model.  But now, the question is Why use two independent datasets to validate two different approaches? My question refers to issues of consistency. I do not see the point of using two "ground truth" datasets for the validation of two cover classifications. In the end, they look like independent studies. 

Point 2. The answer provided does not justify or explain how the objectives are fulfilled. Also, I was expecting a more elaborate response on how the methodology contributes to the objectives.

Point 3.  In the introduction section, the authors stated "we chose the Peijiamao watershed within the Loess Plateau region as the study area; used high-resolution satellite imagery (WorldView-2) and moderate-resolution satellite imagery (Landsat8 OLI) as well as UAV imagery; and applied different methods to explore the ability of high-resolution imagery to invert tree cover" I was assuming this statement as the main objective. Then in the answer to Point 3 is simplify to "explore the potential of high-resolution imagery in improving the estimation of tree cover in sparse grassland" It seems pretty obvious that using higher resolution would improve tree cover estimates. It is expected that the synergy of different-scale imagery (again wrong term, should be pixel size), enhances the accuracy of a single estimation based on OLI. The inclusion of high-resolution imagery would improve the accuracy of estimations based on coarser resolutions. One important thing in Figure 7 highlights that although accuracy is improved, it is not substantial (i.e. Landsat R2=0.45, WV2 R2= 0.77, OLI+WV2 R2=0.73 and Object Oriented Classification R2= 0.67).

Point 4. Here the authors stated "Tree extraction in this region is influenced by the presence of green herbaceous vegetation, making it very difficult to extract tree cover using spectral information" and yet they use nine vegetation indices and OLI and WV2 reflectance bands which I found contradictory.

Point 5. I agree, MODIS VCF is currently a commonly used global tree
cover product but I do not think that is a product that can be comparable to those generated by higher resolutions.  The difference among resolutions is too big.

For the reason stated above I stand for my previous decision since I could not find substantial improvement to the manuscript and the revisions made do not meet my expectations.  I acknowledge the huge effort made by the authors and I will invite them to undertake and extensive revision of the manuscript and resubmit.

 

 

Comments on the Quality of English Language

N/A

Author Response

Thank you very much for taking the time to review this manuscript. Please see the detailed changes in the attachment below.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The authors covered all my questions and concerns. The manuscript is now qualified for the publication with a few minor points below: 

1. Your answer:

“New contributions of manuscripts to the field of remote sensing or forestry：We propose a method for estimating tree cover based on texture features extracted from high-resolution satellite imagery, aiming to overcome the influence of grassland background on the accuracy of tree cover estimation in remote sensing.”

This is a good point to highlight your contribution to the current challenge in tree cover estimation. Please add these arguments into the introduction section with supplied citations.

2. A few lines of GLCM introduction should be provided in the manuscript, says what it is? What is it used for? So the reader can inherit your proposed method?

3. Agree with your answer for this question:

“New contributions of manuscripts to the field of remote sensing or forestry：We propose a method for estimating tree cover based on texture features extracted from high-resolution satellite imagery, aiming to overcome the influence of grassland background on the accuracy of tree cover estimation in remote sensing”

Please provide your answer in the Discussion section and mention them as a limitation of this study.

4. Agree with your answer for this question:

“Point 26: Fig. 7, R2 only reached to 0.77 using the best WV-2 image, which was lower than similar study in the literature (https://www.mdpi.com/2072-4292/12/9/1505). Please clarify.”

Please use these arguments in the Introduction section to clarify the background and the motivation to develop the methods in this study.

5. Agree with your answer for this question:

“Point 31: Fig. 9 at 2 m was more accuracy than Fig. 7b at 0.5 m, right? How do you think about this? And why did you not select the model in Fig. 9 (2 m) as the best model for tree cover estimation in this study?”

But please use your answer in the Discussion section to help the reader more  understanding about the reason of better performance of 2 m than 0.5 m models. Also, use your answer to this question “Point 36: Fig. 9, could you clarify an underperformance of the model at 0.5 m compared to the model at 2 m?” to provide additional argument in the Discussion.

6. The typo “Landsat8”, I mean the number 8 need to be separated from the “Landsat”

7. Line 102, please check space among the characters in the sentence

8. Line 165, “ness temperature bands 20, 31–32” please check format of citation

9. Line 311, “using ENVI (V 5.3).” should be v. 5.3 or v 5.3 or version 5.3

10. Landsat8 should be Landsat 8 or Landsat - 8

11. Table 2, each feature like NDVI, VDVI... Should come with a citation to the source of the formula so the reader can track back when needed

12. Line 376 - 378, seems at different font from other paragraphs

13. Line 395, Table 3, please check the bold format of “059”

 

Author Response

Thank you very much for taking the time to review this manuscript. Please see the detailed changes in the attachment below.

Author Response File: Author Response.pdf