Comparison of Random Forest and Kriging Models for Soil Organic Carbon Mapping in the Himalayan Region of Kashmir

Round 1

Reviewer 1 Report

Excellent work. 

Line 33 (Abstract), suggests using "better estimation" instead of "precise estimation"

Lines 32-33, elaborate some more on RF performance.

Author Response

All the queries/suggestion proposed by the reviewer has been incorporated.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper evaluated the auxiliary variables like satellite spectral indices and topographical attributes and the methods such as Ordinary Kriging (OK), Regression Kriging (RK), and Random Forest (RF) to estimate the soil organic carbon stock (SOCS) in the Himalayan region of Jammu and Kashmir, India. The topic was interesting, but the paper needs more work and clarification.

General comments:

1.      Line 26 : « Auxiliary variables derived from satellite data were used as predictors. ». Pls, specify the Auxiliary variables and the satellite data used in this work.

2.      Lines 29-32 « The best fit model validation criteria used were coefficient of determination (R2) and root mean square error (RMSE). » and « The semi-variogram analysis of OK and RK indicated moderate spatial dependence. ». Already known, they are useless in the Abstract.

3.      Pls, Add the values of R2 and RMSE obtained for each model.

4.      Abstract : The validation of the models accuracy was based on Cross-validation or an independent test set?

5.      « Figure 1. Wangath watershed study area with sampling points ».

 I don't see the location of soil sampling.

6.      Pls, Precise the input for each algorithm.

7.      Lines 206-207 « All descriptive statistics, and geostatistical methods – OK, RF, and random forest were computed using R software 3.6.1 [63]. »

RF or RK ??? Pls, verify.

8.      Pls, Precise the lag size (distance) used in the Kriging method.

9.      Where is Figure 4?

 

10. Pls, compare the obtained results with the results of other work using other machine learning algorithms and remote sensing data. (for example : https://doi.org/10.1016/j.geoderma.2008.06.011; https://doi.org/10.3390/rs14164080).

Author Response

All the queries/suggestion proposed by the reviewer has been incorporated.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

After my careful evaluation, the paper is well organized and written. However, the paper needs some revision to improve. Therefore, you can find my detailed revision comments in the attached file.

Best wishes,

 

Comments for author File: Comments.pdf

Author Response

All the queries/suggestion proposed by the reviewer in the pdf file has been incorporated.

Reviewer 4 Report

This study applied different approaches:  Geostatistical methods - Ordinary Kriging (OK), and Regression Kriging (RK), and machine learning method - Random Forest (RF) for digital soil mapping to predict and evaluate the spatial distribution of SOCS in the Himalayan region of Jammu and Kashmir, India. Although the study is interesting, it doesn't provide an explicit explanation of how it conducted.

 

 Abstract

Could the authors state the values obtained of RMSE and R2 where they explained that the RF performed better than OK and RK while OK performed better than RK  

 

Method

At what resolution was the mapping done?

How many point data were used for training and for validations? This needs to be clearly stated.

Results

Please explain the spatial distribution of soils in the study area as predicted by the various approaches used in this study.

Figure 4 and 5 are missing in the manuscript.

Discussion

Could the authors discuss their findings more thoroughly, perhaps comparing them to recent literature? This would allow them to explain the findings, their implications, and recommendations for improvement and future work

Conclusion

Line 360: state the RMSE and R2

Author Response

All the queries/suggestion proposed by the reviewer has been incorporated.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

General comments:

1.      The validation of the model's accuracy was based on Cross-validation or an independent test set? To be specified in the Abstract.

 

2.      « 3.3.1. Ordinary Kriging » : Describe the stages of Ordinary Kriging. What is the lag size value (distance) used in the Interpolation method? Explain the choice.

 

3.      « Figure 4: Variable Importance Plot – mean decreasing accuracy for SOCS. » From MDA results, what are the predictors (variables) removed from the models? Explain the choice of the threshold value.

 

4.      « Figure 5. SOCS (Mg ha-1) maps generated by Ordinary Kriging (top), Regression Kriging (left) and Random Forest (right) ».

I am so worried about the quality of the presentation of the maps and the spatial variation of SOCS generated by the models :

-          Different presentation of geographic coordinates values was used (spherical using angular units of measure and planar (UTM) using linear units).

-          Each map is made with a different Color Ramp. How can we compare the spatial variation of SOCS obtained from each algorithm?

-          It lacks the scale and the Map Orientation.

 

 

5.      Figure 5 : Explain this great spatial variation of SOCS observed between these maps.

Author Response

Reply to review report

1. The validation of the model's accuracy was based on Cross-validation or an independent test set? To be specified in the Abstract.

Reply: It has based on cross-validation. Added in abstract

2. « 3.3.1. Ordinary Kriging » : Describe the stages of Ordinary Kriging. What is the lag size value (distance) used in the Interpolation method? Explain the choice.

Reply: Added in the manuscript

3. « Figure 4: Variable Importance Plot – mean decreasing accuracy for SOCS. » From MDA results, what are the predictors (variables) removed from the models? Explain the choice of the threshold value.

Reply: None of the predictors was left out. Threshold value selection added in the manuscript

4. « Figure 5. SOCS (Mg ha-1) maps generated by Ordinary Kriging (top), Regression Kriging (left) and Random Forest (right) ».

I am so worried about the quality of the presentation of the maps and the spatial variation of SOCS generated by the models :

-          Different presentation of geographic coordinates values was used (spherical using angular units of measure and planar (UTM) using linear units).

-          Each map is made with a different Color Ramp. How can we compare the spatial variation of SOCS obtained from each algorithm?

-          It lacks the scale and the Map Orientation.

Reply:  Replaced by modified maps

 

5. Figure 5 : Explain this great spatial variation of SOCS observed between these maps.

Reply: Now discussed in the manuscript 

Reviewer 4 Report

- Could the authors use decimal degrees for all the maps in the paper?

- Authors should note that the discussion section is still too scanty. Could authors comply with the suggestions given by the reviewers?

 

 

Author Response

Reply to Review Report

1. Could the authors use decimal degrees for all the maps in the paper?

Reply: Changed to decimal degrees

2. Authors should note that the discussion section is still too scanty.

Reply: Improved substantially