A Comparative Assessment of Machine-Learning Techniques for Land Use and Land Cover Classification of the Brazilian Tropical Savanna Using ALOS-2/PALSAR-2 Polarimetric Images

Round 1

Reviewer 1 Report

I think it is a relative local study. The performances of algorithms vary greatly with areas, images and LULC types. Thus, the novelty seems not high enough. Also, details of the machine learning algorithms are not provided.

Author Response

Please see attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors presented a comparison of machine learning algorithms and methods for polarimetric Synthetic Aperture Radar classification. Paper proposed well known methods for classification using ALOS2 data over Brazilian Tropical Savanna. Scientifically, paper does not present any new findings or improvements over the existing methods, therefore, I recommend paper to be rejected.

If authors would like to present an overview of existing methods, they should reorganize paper and add the most effective methods for SAR data classification and present the classification results.

Author Response

Please see attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper reports the comparison results LULC classification by using various classifiers like NB, DT, RF, MLP and SVM. It has interesting computational contents and in my opinion, this manuscript deserves for publication. But I have few comments. 1. It could be better that authors explain reason behind why you choose the five classifiers in this study. Many articles relating your study they applied kNN or K-mean clustering. Why authors did not use this method to constructing a model. 2. Authors did not mention the process of parameter tuning. In the view of machine learning, this process is very important. It means that after getting the proper parameter your proposed method could improve the prediction performance. It is better than using default parameter. For the parameter optimization of SVM and RF, authors can see more details in (PAAP: a web server for predicting antihypertensive activity of peptides, CryoProtect: a web server for classifying antifreeze proteins from nonantifreeze proteins, HemoPred: a web server for predicting the hemolytic activity of peptides.) 3 For performance evaluation, authors mentioned " seven shapefiles were generated: five for training the classifiers (5, 25, 50, 100, and 200 samples per class), one for validation, and one for a classification" - Why authors choose 200 samples to be the maximum number of samples? - Please clarify the meaning of validation and classification - Author do this process only 1 iteration or 7 iterations. If authors do this process only 1 iteration, please prepare this experiment again to protect the bias derived from single iteration. Otherwise, please report all validation metrics with mean and sd. - Please give the reason why author use global accuracy, Kappa index, conditional producer´s, accuracy (PA) and user´s accuracy (UA) as validation metrics.

Author Response

Please see attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I list some literature, which might be helpful to you.

1) Guie Li, et al.  "A comparison of machine learning approaches for identifying high-poverty counties: robust features of DMSP/OLS night-time light imagery", International Journal of Remote Sensing (2019) 40:15, 5716-5736.

2) Hu, Lirong, et al. "Monitoring housing rental prices based on social media: An integrated approach of machine-learning algorithms and hedonic modeling to inform equitable housing policies." Land Use Policy 82 (2019): 657-673.

3) Zhang, Qianwen , et al. "Biophysical and socioeconomic determinants of tea expansion: Apportioning their relative importance for sustainable land use policy." Land Use Policy 68(2017):438-447.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Paper has been improved.

Many different machine learning algorithms have been used for polarimetric classification and physical parameters extraction. Still paper presents comparison of methods and discussion on SAR data.

In my opinion paper is not research oriented, but compares methods. Paper contains all sections of scientific paper, is well organized, but does not present any new research method, but highlights existing methods, and as authors explained, they assessed comparative performances. 

This reviewer recommends to reject this paper. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Thank you for your informative response. There is no further comment for your manuscript. Thanks again for sharing your knowledge on the land use and land cover classification.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf