Hierarchical Classification of Soybean in the Brazilian Savanna Based on Harmonized Landsat Sentinel Data

Round 1

Reviewer 1 Report

This manuscript presents a new method for mapping soybeans based on a hierarchical classification of HLS that operates on several spectral indices, and validated with 192 ground observations from a two-day trip. They report that HLS improves overall accuracy and Kappa significantly over Landsat 8. I think this is a nice piece of progress and I see no major errors. I would have added other machine learning approaches, but that takes time and would probably not add much to the end result. I think the work should be published.   I have just a few minor questions to raise.   The introduction is brief but adequate and motivates the work. The Materials and Methods section describes the data, indices, and classification are very well written and easy for a grad student to follow. I don't really think the accuracy equations are necessary because they are well-known.   Did you consider other ML methods like CART or SVM? why did you choose RF only?   In Figure 7, SAVI Mar 30 is a poor predictor, but SAVI for other months is a much better predictor. NDWI tracks this as a predictor. Did you explore which variables are poor predictors at the end of the rainy season? Can you remove one of these variables due to collinearity?   The section on misclassification starting on line 484 is very useful. But I am wanting some explanation of how the sampling affected the results. I have not used AgroTag and I don't know how the lack of training samples, in particular for coffee (line 205), can be handled for future work. Are we again at the point where long-duration field campaigns are needed, for each crop type? that would defeat the purpose.  

Author Response

Dear Reviewer, please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The ms remotesensing-1817105 with the title of Hierarchical Classification of Soybeans in the Brazilian Savanna Based on Harmonized Landsat Sentinel Data explored the harmonized Landsat 8 OLI and Sentinel-2 MSI (HLS) dataset to detect soybean production in the western state of Bahia, Brazil, using a hierarchical classification system. The ms is well organized but here are some comments to make the ms stronger.

Remove s from soybeans in the title

Can authors add some values of the most important results in the abstract?

Add the Latin name of soybean in the keywords: Glycine max L.

The introduction section: The authors should not focus in the whole introduction on Brazil, they should first talk in one or two paragraphs on the issue in Brazil, then they should move from local issue into global issue.

L37: Please add the Latin name of soybean

L39: what are the names of these grains? Please mention them

L66-67 and other places: Please add the Latin names with English names of all crops in the first mention.

The authors should cite the methods they used in their material and methods.

The results section is well written, but I recommend the authors to move some tables into supplementary.

The most concerning issue for me, I do not see any indication of statistical analysis in the material and methods as well as in the tables of results section! Can issue check this issue?

The discussion was well written.

The conclusion section: The authors should deeply focus on the most important finding to show for the readers. I recommend authors to make it shorter and add future work.

Regards, reviewer

 

 

 

 

 

 

Author Response

Dear Reviewer, please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The ms has been improved