Rice Planting Area Identification Based on Multi-Temporal Sentinel-1 SAR Images and an Attention U-Net Model
Round 1
Reviewer 1 Report
In this paper, the authors propose a rice monitoring method based on the combination of multi-temporal Sentinel-1 PolSAR data and an attention U-Net model. The proposed rice monitoring attention U-Net (RMAU-Net) model exploits the rich scattering traits of rice at the different growth periods by using Sentinel-1 dual-polarimetric images acquired in specific months. The methodology and data used are adequate to achieve the objective of the paper. The authors showed that the results obtained by the proposed method are better than the results obtained by other methods available in the literature, with a pleasing generalization ability in different years. The paper is clearly presented.
Author Response
Please see the attachment.
Author Response File: 
Author Response.doc
Reviewer 2 Report
Review to Remotesensing-1856820
Rice is one of the most important food crops for human beings. The long-term and inter-annual rice distribution monitoring can provide an important scientific basis for food security. This study developed rice monitoring attention U-Net (RMAU-Net) model with transfer mechanism to improve its generalization ability in different years. However, the writing and analysis are thin, and cannot well reflect the innovation of this work. Thus, this paper is not appropriate to publish.
Major comments:
1. The experiment design is deficient. In my understanding, the highlight of this work is to explore the effectiveness of other polarimetric information other than original SAR data in rice monitoring. However, the experimental design does not fully illustrate this idea. I am not sure if the accuracy improvement comes from the decomposition of PolSAR dataset or the use of multi-temporal data shown in Figure 8. Judging by the results, the advantage of decomposing PolSAR dataset are not obvious and using multi-temporal data seems to help a lot.
2. Are there any improvements to the U-Net model? If so, please explain the reasons and idea more clearly and design appropriate experiments.
3. The transfer mechanism is well applied to different years. Considering the relative flat environment in the selected study area, I am curious about the generalization ability in different areas, especially in Southern China with the fragmented and complex environment.
4. A “Discussion” section is necessary to analyze your results and reiterate your progress compared to the literature.
Specific comments:
1. It would be better to add corresponding intensity images to intuitively show the benefits of decomposed polorimetric data.
2. The selected criteria of selected model used for comparison needs to be stated.
Author Response
Please see the attachment.
Author Response File: Author Response.doc
Reviewer 3 Report
Lines 25-26: “the proposed method can significantly improve the classification accuracy, compared with the 25 traditional methods” – this statement needs support by concrete figures for substantiation of the increased accuracy.
Line 97: The work is devoted not to rice monitoring, in my opinion, but it is mainly directed to the improvement of rice crop identification on the large-scale areas. Rice monitoring makes me think more of phenological observations and growth patterns, so, please, concretize this statement and make it clearer.
Lines 107-110: Do we really need this information? I suppose that it is not necessary and artificially makes the paper unjustified longer.
Figure 1 b-d: I suggest you provide the name of the axis with figures for each plot, because it is not clear enough without explanation what the scales (0-1 and 0-90) are representing.
Line 210: Just five images. Is it sufficient to train the model? If you think so, this statement needs substantiation.
Lines 295-297: The methodology of data augmentation should be explained in more details.
Table 3: I suggest the change of “Memory” into “RAM”
Line 310: If I understood you properly, this finding was through the number of empirical raw runs?
There are no references for the equations 13-16. Please, cite the source or if these formula are of your own design, claim this fact.
I suppose that Conclusions should contain more concrete figures obtained in the study to support the benefits of RMAU-Net accuracy in comparison to other methods used.
Author Response
Please see the attachment.
Author Response File: Author Response.doc
Round 2
Reviewer 2 Report
The manuscript has been largely improved. However, I am still curious about the inspiration for the polarimetric decomposition. Is it a commonly used method? Why did you chose to decouple into the three polarimetric decomposition parameters? How did decomposition calculate and realize, in a brief manner ?The specific process that you listed involved many complex equations. Some more details are needed in the Method section.
Author Response
Dear reviewer, thanks for your comments and questions with regard to the issues of polarimetric decomposition technique. First, the polarimetric decomposition technique and the polarimetric decomposition parameters have been used in the applications of PolSAR image since 1990s, especially widely used in land-cover type classification, such as the classical "Wishart H/A/Alpha classification method". Second, the reason of using the H/A/Alpha parameters is that, as some supplements to the orginal PolSAR data, these parameters can directly reflect the physical scattering mechnisms of the targets and are helpful for the identification of rice planting areas. Third, although the derivations of these 3 parameters are complex as introduced in this paper, we do not have to solve complex mathematically problems or progamming codes to get them, because these parameters can be obtained by the PolSARpro software, which is an open source software that can process PolSAR data and undertake the task of PolSAR image classification using traditional methods. We have mentioned the above issues in the newly revised paper.
