Comparing Machine Learning Algorithms for Soil Salinity Mapping Using Topographic Factors and Sentinel-1/2 Data: A Case Study in the Yellow River Delta of China
Round 1
Reviewer 1 Report (Previous Reviewer 1)
Please, see attached file
Comments for author File: 
Comments.pdf
Author Response
Thank you very much for your constructive comments on our manuscript. We have revised our manuscript based on these suggestions.
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report (Previous Reviewer 2)
The map in Figure 1 should focus on the area where the sampling points are located to make the distance between the different points clearer.
In tables 6 and 7, the column name for references should be just "Reference".
The discussion is still poor. The most sensible and usual way to proceed is to compare your results with the results of the works mentioned in the introduction. After ll,the purpose of the introduction is to explain the situation of the study field and the purpose of the discussion is to put your results into perspective within the field.
Author Response
Thank you very much for your constructive comments on our manuscript. We have revised our manuscript based on these suggestions.
Please see the attachment.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
Please, see the attached file
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This paper presents the use of four predictive models to estimate soil salinity in a small study area in China. Sentinel-2 data and geomorphometric features are used as predictors.
Although the research is generally well conducted and the paper is clearly written, I find that it does not present any clear methodological novelties. Moreover, there is no comparison with previous results in the discussion, just a, correct, explanation of the obtained results. This last issue should be addressed by the authors.
My main concern is with the sampling design. The authors claim that the sampling sites were randomly selected, evenly distributed and at least two kilometres apart. However, Figure 1 shows some linear patterns, with sites very close together, which seem to contradict these claims. Independence of training and validation data is a critical issue in predictive modeling to ensure that accuracy statistics are representative of the foreseeable accuracy across the whole study area. To ensure such independence, sampling points should be distant enough from other points. If this is not the case, k-folds should not be random, but should divide the study area into k sectors, and group all the sampling points in each sector in the same fold. This way, validation points in each fold will be far enough from the sampling points used to train the model.
The paper is generally well written, but there are some expressions that I think should be revised. For example, line 169 should read "must make" not "must made"; line 188 should read "Standard classification", not classification standard". Although English is not my mother tongue, I think a language review could improve the paper.
Specific comments:
There is a tendency to name programs instead of algorithms. In Tables 4 and 7, there is no need to list the programs used, but the algorithms used should be mentioned, with the proper citation. There are, for instance, several algorithms to compute terrain roughness, which one did you use?
Similarly, in table 6, a proper citation, instead of just "FAO", should be used. It would also be useful to explain the rationale of using those (so many) indices. What information does each provide that the others do not? If they all provide the same information, they are redundant, and may have a negative impact on the models.
In Table 8 there is no need to include the ranking and it is not clear what "Band" is. In any case, it is common practice to present the importance of Random Forest features as an ordered barplot, which improves clarity.
Why the 0.05 importance threshold?
Figure 6: The X axis should include the names of the features, not their number.
Table 9 shows some peculiar results. It is quite strange that the lowest R2 coincides with the lowest RMSE. Are you sure that there are no mistakes?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Please, see attached file
Comments for author File: Comments.pdf
Reviewer 2 Report
Point 1.
I don't think that the use of topographic features to estimate soil properties can be considered novelty. There is a huge amount of research on Digital Soil Mapping in which geomorphometric features derived from a DEM and remote sensing data are used to estimate soil properties. In fact, you are using a methodology published by FAO as a technical guide. This is the resaon why I think it is necesary for this paper to compare your results with those obtained by other scholars who have applied similar methods to map soil salinity.
Point 2
619 m is far enough to avoid dependencen between samples. It is well known the low spatial autocorrelation of soil properties. You should include the actual figures in the paper and claim that spatial dependence is not an issue, rather than stating that distances were 2 km. You should also explain why the samples are not random.
Point 3:
Ok
Point 4:
Instead of copying the features used in the FAO technical guide, you should explain why you think such features are relevant to estimate salinity, and refer to a paper or book that properly explain how they are calculated. I could not find citation 23 as there is no journal or book title in the reference.
Similar considerations to point 5.
Point 6.
Ok
Point 7.
You should explain that the criterion is qualitative (a drop in importance between 10 and 11). The threshold of 0.05 is just a coincidence, but it seems that it had been stated beforehand.
Point 8
Ok.
Point .
Ok
