Unexpected Expansion of Rare-Earth Element Mining Activities in the Myanmar–China Border Region

Round 1

Reviewer 1 Report

Reviewer comments:

·         There is no ground control point or field-visit to validate the results. How do you interpret this point?

·         Please explain more details about the sample you used from USGS Library, and it’s origin; and explain why you have chosen Monazite?

·         Did you use Sentinel-2 data only to define the type of mineral at the selected mines? And why didn’t you use for the rest of your work (Sentinel-2 has a spatial resolution of 10-20m, so better than your images)?

·         What is the novelty of this work?

·         In the results, it can be seen that water area (probably acid mine drainage) is inside the class of “mine-area”. Why didn’t you remove the water index? Can you define if there is a lake or a mine-dumps covered by water?  

Author Response

Dear Reviewer

Thank you so much for the insightful comments provided to our work.

Please receive the attached responses.

Author Response File: Author Response.docx

Reviewer 2 Report

1. This study uses satellite images to interpret rare earth mining activities with the random forest method, which is necessary in the research fields of surface resource management and land development suitability. The research content is biased toward applied research, so novel image preprocessing and classification algorithms have not emerged. For technically oriented journals, readers may wish to have some new technical processing methods and detailed discussions, so it is suggested to increase the selection of satellite image bands, the correlation between image resolution and interpretation results, and the operation of machine learning algorithms Details and other content, otherwise this manuscript is basically except for the issue of rare earth elements, and it may be difficult to attract readers at the technical level.

2. There are many related studies on the application of random forest algorithm and other machine learning algorithms in mineral mining detection, so there is no significant difference in the content of the research method description from previous studies. Based on this, suggestions can be made on the relationship between image resolution and ground objects, the possible impact of terrain effects and cloud coverage on the research results, and whether new indicators can be combined with bands and applied to machine learning algorithms. If the above discussion is added to the manuscript, it will have more research value.

3. The article lacks a description of the production of ground-truth data. It is suggested to explain the process of data production and possible technical problems.

4. At present, only 30m resolution images are used for operation and discussion in the use of satellite images. Can better interpretation and classification be achieved by importing high-resolution multispectral images? (such as worldview-2 or other hyperspectral images)

5. The presentation of change detection is very clear and can show obvious changes in the mining area. However, for image researchers, we would like to know if the same image data and methods are applied to areas with more complex land cover in the future, can the same application effect be achieved? (Will the number of surface classification types affect the detection results? If there will be an impact, is it more reasonable to switch to the object detection method?)

6. At present, image recognition mostly adopts object detection method rather than pixel classification method. It is suggested that some related discussions can be added in the manuscript.

Author Response

Dear Reviewer

Thank you for the insightful comments on our work.

Please receive the attached responses below.

Best

Author Response File: Author Response.docx

Reviewer 3 Report

This paper is a good and exciting piece of work built around an accurate methodology. It is so interesting because it’s focussed on mining for rare earth elements which is rapidly increasing, due to current and projected demands for information and energy technologies. This make also the originality of the paper.

The authors access this unexpected expansion in Myanmar China Border region through the processing and analyses of satellite images Landsat and Google Earth of the area of investigation.

The reviewer appreciate the way the authors present geological and geographical settings of the region concerned. The same with the explanations about the importance of rare elements for the needs of our contemporary world.

The methods is developed step by step beginning from the choose of the Landsat images as well as orthophotos from Google Earth,  through their processing with visual assessment and sampling and change detection. The explanation about validation through visual assessment and sampling on Google Earth Pro images

help understanding why they could not go on field, but also the authors have a good skill to observe and validate target on very high resolution images.

The results confirm this efficiency and are satisfactory. Its help understand the new landscape observed in the mining areas throughout the world.

The references are more precise and updated with the more recent pubications on the topics.

Author Response

Dear Reviewer

Thank you for the insightful comments on our work.

Please receive the attached responses below.

Best

Author Response File: Author Response.docx

Reviewer 4 Report

This paper presents a good applied research using time series remote sensing for tracking footprints of mining activities in a border area. I would like to encourage authors to consider revising the paper, addressing issues below:

1) Further information about sampling for reference data should be provided, as probability sampling designs are essential for precision in area estimation as well as in accuracy assessment (both can be based on error matrices). This is even more important for change analysis. It is not clear how area estimates were computed for the research. If area estimation was not based on probability sampling and proper estimation methods, justification should be given.

2) Table 3, class definitions should pay attention to exhaustiveness and mutual exclusiveness. For a binary classification, it is probably more sensible to define mine areas with examples, leaving no room for ambiguity.

Author Response

Dear Reviewer

Thank you for the insightful comments on our work.

Please receive the attached responses below.

Best

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

1. The revised version has obvious additions to the explanation and application methods of the algorithm, and greatly improves the feasibility of applying this method in other regions.

2. It is still recommended to supplement the process of producing ground truth data by visual interpretation, because this is the most direct impact data to verify the accuracy of the model. If this aspect cannot be determined, the credibility of this article will be greatly reduced.

3. It is recommended to optimize the arrangement of pictures and text to increase readability.

4. The basis of category classification and the determination of the number of categories are important factors affecting the accuracy of automatic machine learning calculations. It is suggested that the article can be discussed in more detail. If possible, the feasibility of hierarchical subdivisions can also be supplemented.

5. If similar operations are to be carried out in the future, whether the recommendations for the selection of satellite images and RGB aerial photos (resolution & band selection) can be clearly defined or SOPs proposed in this manuscript.

Author Response

Dear Reviewer

Thank you for the second round of comments our work.

Please receive the attached responses below.

In the revised manuscript, our revisions are highlighted in blue color.

Best

Authors.

Author Response File: Author Response.docx