A New Approach of Ensemble Learning Technique to Resolve the Uncertainties of Paddy Area through Image Classification

Round 1

Reviewer 1 Report

The authors develop an interesting and useful idea, creating models of uncertainty in a context of data from different sources (different resolutions, time series, textures, different indexes, auxiliary information, etc., and using different learning approaches to solve classification problems.

From the point of view of the publication of the article, it would be welcome, not to comment on the methods that have been used and the results in the introduction, it also happens in the method section where results are advanced.

The description of the methods is excessive, it is a good reminder of the characteristics of the various procedures used, but it could be synthesized much more by not incorporating known aspects of the methodologies and by concentrating on aspects related to work.

A key aspect such as the true field data, it would be important to go a little deeper, it is mentioned that “the ground truth is for the results of the 2015 paddy area interpretation and the latest version of the status map of the cultivated photo by the Agriculture Administration ”, but the interpretation procedure should be detailed.

The style of writing the results obtained with the different methods is a bit confusing and redundant with the data provided in the tables. The results of the tables should probably be left and in the text limited to the discussion of them.

Comments for author File: Comments.docx

Author Response

see attached file.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript is addressing a missing part of the image classification problem to enhance the overall accuracy. The theory behind the methodology was well explained, and the introduction was also well written to formulate the research gap and objectives. However, I have the followings suggestions/comments which can be helpful to improve the manuscript to publish in this journal.

1. It is not clear how the test and train data were divided.
2. How did authors deal with the unbalanced class distribution in the dataset for classification (more '0' class patches than class '1' patches)?
3. Did authors use mean values of spectral bands, VIs, and texture per patch?
4. Overall figures can not be visualised properly (part of the most figures were cut out due to page margins).
5. In figure 6, instead of the total area, it is good to see one zoomed region, which gives a clear idea about patches.
6. Captions in both tables and figures should be rewritten (they should be self explainable).
7. Addition to Figure 9, it would be great to see a table with example single patches classification output (class and probability) and how it improves after FDS.
8. Add a paragraph in the discussion, to review the pros and cons of this methodology in terms of technical/computational intensity against the accuracy with and without FDS.

Author Response

see attached file.

Author Response File: Author Response.docx