Paddy Rice Mapping in Thailand Using Time-Series Sentinel-1 Data and Deep Learning Model
Round 1
Reviewer 1 Report
This paper is good enough to be published in Journal of Remote Sensing after the minor revision processing.
- The authors need to discuss the relationship between SAR image and the NDVI (derived from the optical image) for paddy rice mapping, such as pros and cons. Why did you use SAR instead of NDVI for this study?
2. There must be the time difference between the given Sentinel-1 image and the Google earth images as the reference images.
3. The authors need to explain the detail reasons for using VH polarizations and U-Net. Please explains more with the references.
Author Response
Dear Section Managing Editor
Dr. Bob Du Sep.25, 2021
Remote Sensing
Manuscript ID remotesensing-1366996 entitled
“Paddy Rice Mapping in Thailand using time-series Sentinel-1 data and deep learning model”.
We appreciate the thorough reviews provided by the referees and handling editor. We agree with these suggestions and have revised the manuscript accordingly. Below is our response to their comments resulting in a number of clarifications. We hope these revisions resolve the problems and uncertainties pointed out by the referees. In the manuscript, the red parts are revisions suggested by the referee: 1, the blue parts are revisions suggested by the referee: 2, and the green parts are revisions suggested by the referee: 3. The wave lines are other changed contents that are corresponding to the suggestions or intended to improve the expressions.
Best Regards,
Hong Zhang
zhanghong@radi.ac.cn
Author Response File: 
Author Response.docx
Reviewer 2 Report
The authors proposed to exploit Sentinel-1 time-series incorporating a deep learning model to generate a national-scale annual paddy rice map in Thailand. This method extracts temporal statistical features of the SAR time-series to mitigate the intra-class variability due to different management practices and improves the paddock discrimination through a modified U-Net model with the fully connected Conditional Random Field. In the reviewer’s opinion, the organization of the manuscript looks appropriate, and the results sound promising. The topic and methodology of this paper would be of interest to the community and agribusiness. However, there are also some comments to be clarified/addressed. The authors are suggested to consider the following comments/suggestion for improving their manuscript before consideration for publication.
Line 23: better specify two other methods
Line 27: suggest changing is -> has
Line 52: better read “the rainy season and the dry season”
Line 54: suggest changing appropriate -> suitable
Line 59: better read “to be monitored”
Line 60: suggest changing progress -> development
Line 90: please pay attention to the grammar issues here and others. Better have a professional editing and proofreading
Line 93: why using sigma0 but not gamma0
Line 94: carefully use the modal verbs. In scientific articles, normally avoid using could, would or should etc
Line 108: suggest changing “must be” to “are required” or so in this line and others
Line 117: 13 dates of SAR images or 13 scenes?
Line 128: better read “in the whole country”, “and meanwhile”
Line 142: axillary information -> auxiliary information?
Line 162: what is the deep rice?
Line 201: better read “Since 2016, routinely global acquisitions of Sentinel-1 mission have offered…”
Line 204: not 6 descending orbits but 4 descending and 2 ascending orbits in Fig. 2?
Line 209: better mention how many dates or epochs of Sentinel-1 coverage apart from the total number of scenes
Line 224: how to generate the label image in Fig. 4, or a reference? Because the labelled image is used as training data, the labelling accuracy should be higher enough. If so, is it possible to implement the labelling algorithm to the whole mosaic?
Line 247: was wondering whether the refined orbit file and thermal/border noise removal have been implemented or not? better provide the detailed Sentinel-1 image pre-processing steps and the software used
Line 265: suggest using the new COP-DEM instead of the SRTM DEM
Line 283: better use the average of the temporal variance divided by the total number of pixels and modify eq. 1 to mitigate the affect in case of uneven image dates across different frames in Fig. 2?
Line 318: probably “high consistency” is not true in Fig. 7(b)?
Line 320: from eq. 1-3, one can see the variance Sigma0 is in squared, but the linear min and max Sigma0. When displaying in RGB, does it really make good sense?
Line 322: please explain why some paddy rice showing in magenta but others in dark blue
Line 344: what is the 50% overlapping rate here?
Line 362: please provide the mosaicking method and how to deal the overlapping areas in Fig. 11. why are there some variation between frames even acquired from the same orbit?
Line 365: please explain how to use the CRF module to modify the U-net model or a reference?
Line 391: please define epsilon in eq. 4
Line 424: suggest using validation instead of verification
Line 455: better read “result of Thailand in 2019”
Line 500: suggest changing declares -> indicates
Line 514: suggest changing explanations -> reasons
Line 516: why not use the images acquired in 2017 to match the LULC data?
Line 524: better read “… the paddy rice areas have been overestimated in our paddy rice mapping result and the 524 FROM-GLC LULC product”
Line 594: why using these two models for comparison?
Line 637: possible reasons for more commission errors by the SVM classifier and omission errors by the FS-U-Net method in Fig 23?
Line 643: please explain why the FS-U-Net method showing poor performance here compared to the original results in ref [50] with similar extents. Should perform better because of less classes?
Line 674: change Fig 27-> Fig 25
Author Response
Dear Section Managing Editor
Dr. Bob Du Sep.25, 2021
Remote Sensing
Manuscript ID remotesensing-1366996 entitled
“Paddy Rice Mapping in Thailand using time-series Sentinel-1 data and deep learning model”.
We appreciate the thorough reviews provided by the referees and handling editor. We agree with these suggestions and have revised the manuscript accordingly. Below is our response to their comments resulting in a number of clarifications. We hope these revisions resolve the problems and uncertainties pointed out by the referees. In the manuscript, the red parts are revisions suggested by the referee: 1, the blue parts are revisions suggested by the referee: 2, and the green parts are revisions suggested by the referee: 3. The wave lines are other changed contents that are corresponding to the suggestions or intended to improve the expressions.
Best Regards,
Hong Zhang
zhanghong@radi.ac.cn
Author Response File: Author Response.docx
Reviewer 3 Report
Attached
Comments for author File: Comments.docx
Author Response
Dear Section Managing Editor
Dr. Bob Du Sep.25, 2021
Remote Sensing
Manuscript ID remotesensing-1366996 entitled
“Paddy Rice Mapping in Thailand using time-series Sentinel-1 data and deep learning model”.
We appreciate the thorough reviews provided by the referees and handling editor. We agree with these suggestions and have revised the manuscript accordingly. Below is our response to their comments resulting in a number of clarifications. We hope these revisions resolve the problems and uncertainties pointed out by the referees. In the manuscript, the red parts are revisions suggested by the referee: 1, the blue parts are revisions suggested by the referee: 2, and the green parts are revisions suggested by the referee: 3. The wave lines are other changed contents that are corresponding to the suggestions or intended to improve the expressions.
Best Regards,
Hong Zhang
zhanghong@radi.ac.cn
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
- I thank the authors for a thorough editing of the manuscript and provide the requested modifications. This latest version is indeed in a good shape for publication in Remote Sensing, after addressing the following minor comments:
- 1. Line 204~206 : Sentinel-1A and Sentinel-1B launched at 2014 and 2016, respectively. This statement needs to be clearly defined.
- 2. Line 211 : “In total, 871 available SAR images were acquired.” However, the total numbers in Table 1 only 758.
- 3. Line 317 : “especially from Match to October “, March
Author Response
Dear Section Managing Editor
Dr. Bob Du Sep.30, 2021
Remote Sensing
Manuscript ID remotesensing-1366996 entitled
“Paddy Rice Mapping in Thailand using time-series Sentinel-1 data and deep learning model”.
We appreciate the careful reviews provided by the referee and handling editor. We have revised the manuscript accordingly. Below is our response to their comments resulting in a number of clarifications. We hope these revisions resolve the problems and uncertainties pointed out by the referees. In the manuscript, the green parts are revisions suggested by the referee: 3. The underlines are other changed contents that are intended to improve the expressions.
Best Regards,
Hong Zhangzhanghong@radi.ac.cn
Author Response File: Author Response.docx
