A Deep Learning Approach to Estimate Soil Organic Carbon from Remote Sensing

Round 1

Reviewer 1 Report

General Comments:

The main content of the article is the use of deep learning and remote sensing technologies to estimate Soil Organic Carbon (SOC). While the manuscript needs revision before it can be published. (i) The current model predicts SOC values based on a single satellite image, without considering the temporal changes in SOC. This might affect the accuracy of the model's predictions, especially in rapidly changing areas. Future work could include time series data to capture the dynamics of SOC over time; (ii) Although the manuscript provides a detailed description of the techniques and methods used, some parts, such as the specific settings of the model parameters and details of the training process, might benefit from more exhaustive explanation, This would help readers better understand the research methods and potentially replicate the results; (iii) The “Introduction section” can be shorten and simplification, only reserve the contents related to the research topic.

Several comments can be found as follows.

Additional Comments/Questions:

(1)    L217: [], Please add the corresponding reference.

(2)    L256: “[?]”, Please cite the corresponding reference correctly.

(3)    L395: “The architecture is shown in Figure .” You seem to have forgotten to add the figure title

(4)    L571: “[GitHub Repository Link Placeholder] “, Would it be better to add a link?

(5)    Figure 1, Figure 7, Figure 8 would be better if the figure were redrawn for uniformity and standardization.

(6)    In Figure 3, “Unet” should be rewritten as “U-Net”

(7)    Deep learning is a subset of machine learning, and it is essential to provide further clarification in the text to avoid potential confusion.

(8)    When it comes to the methodology, details such as the objective function, the training setup, and the experimental environment need to be described.

Comments for author File: Comments.pdf

General Comments:

The main content of the article is the use of deep learning and remote sensing technologies to estimate Soil Organic Carbon (SOC). While the manuscript needs revision before it can be published. (i) The current model predicts SOC values based on a single satellite image, without considering the temporal changes in SOC. This might affect the accuracy of the model's predictions, especially in rapidly changing areas. Future work could include time series data to capture the dynamics of SOC over time; (ii) Although the manuscript provides a detailed description of the techniques and methods used, some parts, such as the specific settings of the model parameters and details of the training process, might benefit from more exhaustive explanation, This would help readers better understand the research methods and potentially replicate the results; (iii) The “Introduction section” can be shorten and simplification, only reserve the contents related to the research topic.

Several comments can be found as follows.

Additional Comments/Questions:

(1)    L217: [], Please add the corresponding reference.

(2)    L256: “[?]”, Please cite the corresponding reference correctly.

(3)    L395: “The architecture is shown in Figure .” You seem to have forgotten to add the figure title

(4)    L571: “[GitHub Repository Link Placeholder] “, Would it be better to add a link?

(5)    Figure 1, Figure 7, Figure 8 would be better if the figure were redrawn for uniformity and standardization.

(6)    In Figure 3, “Unet” should be rewritten as “U-Net”

(7)    Deep learning is a subset of machine learning, and it is essential to provide further clarification in the text to avoid potential confusion.

(8)    When it comes to the methodology, details such as the objective function, the training setup, and the experimental environment need to be described.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors provided a new deep learning approach to map SOM using remote sensing imagery. Generally, this manuscript had good novelty. Some minor suggestions might be useful for authors. 

Introduction part: suggest authors shorten your introduction part, existing description is too specific, authors should add a paragraph to display your study purpose and listed the following sections content briefly. Mainly is the study aim and your novelty. 

Your manuscript title, i personly think authors should add the study area, detailed remote sensing datasets, and your promoted methods. This could be more specific.

Discussion part: suggest authors add the section title, especifically 5.1 should be the comparsion and validation of your promoted method and results, section 5.2 could be the study limitations and further study.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

I have been requested to review the submission titled “A deep learning approach to remote sensing of soil organic carbon” to the Remote Sensing journal. The subject fits within the scope of the journal; however, there are some issues that need to be addressed by the authors before submission, as detailed below.

 1. I suggest changing the title from “A deep learning approach to remote sensing of soil organic carbon” to “A deep learning approach to estimate soil organic carbon remotely” or “A deep learning approach to estimate soil organic carbon from remote sensing”.

 2. Please, do not repeat keywords that already exist in the title. Two important missed keywords are U-Net and Sentinel-2. Other suggestions: CORINE, LUCAS, and CHLSOC.

 3. The Introduction section is a bit too long with unnecessary details about the results obtained by the previous studies. Some paragraphs can be merged since they deal with the same subject (same for other sections). I suggest ending the Introduction with the objective of the study conducted by the authors rather than beginning the Materials and Methods section with the objective of the study.

 4. The quality of most of the figures needs to be improved. Overall, letters and numbers are too small.

 5. The structure of Materials and Methods section is too confusing. I suggest using the following structure to the manuscript:

1. Introduction

2. Background

2.1. The land use/cover area frame statistical survey

2.2. CHLSOC

2.3. CORINE

2.4. The U-Net architecture

3. Materials and Methods

3.1. Study area (Tuscany)

3.2. Datasets (materials)

3.3. Methods

Note: Put the flowchart (Figure 3) at the end of the first paragraph of this subsection. Please, indicate Stages 1 and Stage 2 in the flowchart. In fact, Stage 1 should be the remote sensing (Sentinel-2) data processing (left part of the figure).

4. Results

5. Discussion

6. Conclusion

6. L493-498 (10 ML algorithms): this should be stated in the Methods subsection. The evaluation of impact of elevation data is not reported in the Methods section either.

 7. The citation of Figure 7 is on page 7, but the figure appears on page 16. Both should be on the same page or on the following page at the most.

Moderate to extensive editing is required.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The second version of the manuscript is much improved; however, I still have the following issues that should be considered by the authors:

1. At the end of Introduction section, please, state the objective and the novelty of the study. In other words, move the first paragraph of the Materials and Methods section to the end of the Introduction section.

2. The quality of Figure 1 needs to be improved. The two figures of Lucas data set should be (A) and (B) and the CHLSOC should be C. Coordinates for the Lucas maps are missing. Please, add more details in the title of the figure.

3. I do not think the structure of the Section 3. Materials and Methods are acceptable in the current structure: (i) this section should start presenting the study area, specifying the main materials and, at last, describe the main steps of the methodological approach (a flowchart would be very helpful); and (ii) Section 4. A ML pipeline for SOC estimation seems to be part of the Methods so that apparently there is no reason to open a new Section.

Moderate English editing is demanded.

Author Response

Please see the attachment.

Author Response File: Author Response.docx