Fine Classification of Rice Paddy Based on RHSI-DT Method Using Multi-Temporal Compact Polarimetric SAR Data
Ryoichi Sato
Round 1
Reviewer 1 Report
The authors presented the results of a study on the extraction (classification) of rice on radar survey data. In general, conceptually, the article uses modern methods applied to not very modern data. Why the authors do not use Sentinel 1 data to do the same work on a modern day remains a mystery. The set of methods used also seems adequate, except that the authors do not explain the reasons for choosing one or another method. It is also not clear why the Random Forest method, an ensemble method that allows combining several decision trees for a more accurate classification, was not used.
Regarding how to evaluate the results - the authors provide tables comparing different accuracy metrics, Kappa coefficients, which is good, but not enough. The authors need to conduct an additional experiment using available spectral data for the selected period - Landsat or Sentinel 2 programs (if compared with Sentinel 1 data), which will allow to evaluate the adequacy of the results obtained, and also allow the authors to answer the question of which initial data is more preferable.
The next question is how does the developed algorithm perform on territories where not only monoculture - rice - is represented? What would happen if the proposed approach is applied to a different territory? The authors do not give an answer to this question, which raises questions when evaluating the results obtained by the authors.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This article describes a novel methodology for fine clasification for rice paddy, which provides more accurate basic information for growth monitoring and yield estimation of rice.
The methods are appropriate. The results and discussion are clearly presented. The conclusions in this manuscript are supported by the experimental results. The references are based in current papers.
The manuscript is well written and it is easy to comprehend it. Overall, the presented research could be valuable to other researchers in this field.
Considerations:
Authors must follow the specifications of the Microsoft Word template or LaTeX template to prepare their manuscript. Abstract: A single paragraph of about 200 words maximum.
Line 114: Why were transplanting hybrid rice (T-H) and direct-shown japonica rice (D-J) selected for this study? Why has it not also been applied to direct-shown hybrid rice (D-H) and transplanting japonica rice (T-J)?
Line 132: the “Urban” Photo in figure 2 can be improved.
Line 145: Why are twenty-two compact polarimetric parameters extracted from each of the six temporary CP SAR data? Clarify this aspect.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This manuscript presents DT based classification of rice paddy using multi-temporal CP SAR simulation. The reviewer agrees that usage of the temporal data and investigation of CP SAR parameters are very significant in the field. However, there are some unclear points in the manuscript. Please consider to clear them
1 H_i in Eq. (16) is the rice height of the sample field corresponding to each phenological period. However, it seems there is no information how to obtain the information of height. Please make clear whether the height information is obtained by PolSAR data or by local field survey.
If the height information is obtained by the field survey, I am afraid that the proposed method cannot be applied to other rice paddy fields.
Please express your opinion for the robustness of the method if the height information is obtained by the local field survey.
2 In the flow chart of Fig.3, it is shown the training and verification sets are 50% and 50 %, respectively. However, there is no information on the pixel number for training/verification. So, could you clear how many pixel numbers for training is required to obtain accurate classification results as shown in Fig. 12, please?
3. Classification based on DT is compared with that based on SVM in accuracy verification, as in Table 3 and Fig. 12. However, there is less information on the SVM used here. Please explain the detailed information on the SVM for fair comparison.
4 In the field of precision agriculture, classification of rice paddy even in the harvest period is important. However, as shown in Eq. (17) (sum^5_{i=1}), it seems the contribution of the period is excluded in the analysis. Authors had better check the accuracy when the data in the harvest period is included.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Although some points in the manuscript are controversial, the changes made to the text, the detailed answers to the reviewer's questions, and the work done are generally positive. The manuscript may be published after editorial revisions.
