Deep Learning Based Spatial Distribution Estimation of Soil Pb Using Multi-Phase Multispectral Remote Sensing Images in a Mining Area

Round 1

Reviewer 1 Report

In this paper entitled “Deep learning based spatial distribution estimation of soil Pb using time-series multispectral remote sensing images in a mining area". The idea is interesting, however the core point of the article is missing and not convincing (Section 3). Please find my attached comments.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

In manuscript, a deep learning based model using 3D convolution is proposed to estimate the Pb content from the constructed time-series multispectral remote sensing images. Time-series multispectral remote sensing images used to estimate Pb distribution. In addition, the spatial distribution map was produced using 3D-ConvNet from the constructed time-series multispectral image. The study can provides a new approach to heavy metal pollution monitoring. From my literature review, it has been observed that no similar study has been conducted in the study area before. The manuscript is fluent and well-structured in terms of its treatment of the topic. The methods used in the study are considered professionally appropriate. The figures and tables presented in the study have sufficient detail and are original. Therefore, I think, the study will contribute to the reader of the journal. Also, I evaluate it as original research. The manuscript can be published.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The manuscript still needs much improvement, and I think the current version is unsuitable for publication in this journal.

 

Abstract:

The logical relationship between front and back is confusing, and the expression is not concise and clear enough, so reorganization is necessary.

 

Introduction:

1.     The summary is not comprehensive enough, and the method introduction is too cumbersome, which needs to be further condensed and supplemented.

2.     Almost all the space is used to introduce 4 traditional machine learning algorithms and deep learning algorithms.

3.     In contrast, the use of remote sensing data that the author focuses on does not make any statement.

4.     The last paragraph seems to be highly repetitive with the abstract, and the organization needs to be rethought.

 

Materials: Only the introduction of the study area and the acquisition of soil heavy metal data, and the introduction of remote sensing data is missing.

 

Method:

1.     Only a brief introduction of the principles of the four machine learning algorithms is given, but the reasons for choosing the four machine learning methods listed in the article, as well as their data input, parameter settings, etc., are lacking.

2.     From the author's introduction, it can be seen that in addition to focusing on modeling methods, the selection of remote sensing data seems to be the focus of the author's attention. However, the author of the data part has not done any further comparative analysis work, such as the comparison with the single-phase remote sensing data modeling results mentioned above by the authors, etc.

3.     The reason why the authors chose the time series imagery to use the Landsat series data, as well as the processing of the Landsat-7 data banding problem, the processing of the number missing in the time series due to cloudy, etc., and the processing of the unbalanced interval of time phases, etc. have not been introduced in the Methods section.

4.     For the study of soil heavy metals in mining areas, what is the biggest difference between single-temporal and multi-temporal?

 

Results and analysis:

1.     The content arrangement is unreasonable. The introduction of specific remote sensing data should not appear in the results section. In addition, there is no comparative discussion on the results of each method.

2.     In addition, judging from the table given by the author in the results section, a lot of scene images of the time-series images seem to be removed during the period from April to December due to the cloud influence, so can it still be called a time-series? Or is multi-phase more appropriate?

3.     The contents of the diagrams and table descriptions in part 4.2 are completely repetitive, and in my opinion, it is unnecessary, just choose one of them to present the results.

4.     There is an error in the description of the horizontal and vertical coordinates in Figure 4. In addition, the results in the figure shows that each model seems to have an obvious underestimation of the heavy metal concentration of some samples in the range of 19-23. I am curious about the reason for this phenomenon.

 

Besides, in my opinion, it is also very important to point out the advantages and disadvantages of the proposed method and give an outlook for future work.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

This study introduces a model that utilizes deep learning and 3D convolution to estimate the amount of Pb in multispectral remote sensing images over time. The authors used a complete and conventional computational model to simulate and analyze the data of heavy metal pollution in soil. The research content is novel. It provides a new method for monitoring heavy metal pollution in soil. I think the manuscript can be published in this journal after minor revisions.

 1. It is suggested to add some information about the representative heavy metal pollution problems, especially lead contamination in the INTRODUCTION to highlight the necessity of this study.

2. Please check all the graphs and tables again (e.g. Figure 4 shows an inappropriate "vertical line" in the lower right corner.)

3. The methodological highlights of the manuscript (ease and accuracy compared to other methods) should be highlighted in the CONCLUSION.

none

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

authors have addressed all comments.

Reviewer 3 Report

The authors have made detailed revisions and responses to the suggestions of the previous manuscript. The revised version can be accepted for publication in this journal.