A Decision-Tree Approach to Identifying Paddy Rice Lodging with Multiple Pieces of Polarization Information Derived from Sentinel-1

Round 1

Reviewer 1 Report

Authors propose a decision tree model to identify rice lodging using the various intensity information before and after rice lodging in Sentinel-1 SAR images, and the in-situ lodging samples. The model combines the backscattering coefficients and polarization decomposition parameters, and quantifies the importance of each feature. The idea is somewhat interesting, but the paper can be improved. I suggest the paper be reconsidered after authors make revisions.

 1.Please give more details of the decision tree methods, any clever thing you have designed besides the classical algorithmic flow.

2.the results should be compared to those from common methods.

3. Quality of figures should be improved.

4.Please give more result discussions on the box plots in Figure 6.

5. We note that Figure 1 in your submission contain [map/satellite] images which may be copyrighted. I suggest authors present written permission from the copyright holder to publish the figure, or remove the figure.

6.Some solid data results can be confirmably reported in the conclusion section.

7.Please refine the English expressions throughout the paper.

Author Response

  Thanks for your valuable comments. We accepted these suggestions and addressed most of the concerns raised by you. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors,

The manuscript presents an interesting research subject with application of SAR data. However, the presentation can be improved with the following as example:

Line 41: remove "crop" to read as ".....food source in Asia, ....."

Line 41-43: Long sentence. Can be broken to small multiple sentences.

Line 44-45: Is the loss in yield exclusively due to lodging or there are other factors involved in it?

Line 46-47: Do natural factors such as wind or frost etc. cause rice lodging or there could be factors such as a growth stage, variety, plant structure, a field's location and exposure etc. contributing to it?

Line 49: Explain "high objective influence"?

Line 56: What is meant by "ability to acquire images under clouds"?

Line 58-60: Rephrase for clarity.

Line 64: Delete crop to read "....the lodging area by....." 

Line 138-139: Check the use of past and present tense, such as "extracted" vs "include". Revise to use same tense for consistency.

Table 2: This is a "result" and be part of the section "Results"?

Line 171: "situ"? Is it "in-situ" or "ex-situ"?

Line 172: What is meant by "inhomogeneity?

Figure 3: Box "Sentinel-1 GRD/SLC: Which polarization?

Box "Data Pre-processing", What is included in this?

Line 186: ".... we used H/....." instead of "...we use ...."?

Line 209-210: Do the authors think other factors such as windy conditions could impact the criteria? 

Line 240: Is it "feature importance" or "feature of importance"?

Line 242: Change to read "....factors used Random Forest and XGBoost algorithms to measure .... 

Lines 275-294: This is probably weak explanation with no scientifically backed arguments. It needs to be improved with further deliberations and investigations.

Figure 4: Legend: Combine the Lower and Upper 95% CI to one as "Upper and lower 95% Conf. Int.".

Figure 8: Is there an exposure of the area to a specific windy conditions, e.g. south to north or east to west winds or others? A possible impact of slope? Or a combination of slope and wind? 

Line 367: "show" instead of showes?

Discussion: Can the authors discuss an impact of water (as paddy rice has high water content" and possible impact on backscatter during lodging? How the impact can be overcome? 

Sincerely, 

 

The reviewer

 

Author Response

Thank you very much for pointing out the problems in our manuscript. Following your thoughtful suggestions, we have revised the manuscript. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Worldwide, rice is the most important food staple. It is grown on approximately 155 million hectares and accounts for one-fifth of the global calorie supply. However, lodging is one of the common abiotic adversities during the growth of rice. In addition to affecting photosynthesis, it can seriously damage crop growth and development, such as the reduced quality of the rice and bringing difficulties to automated harvesting. In such a context, to accurately and timely acquire crop lodging areas for yield prediction, agricultural insurance claims, and disaster management decisions are quite important.

The Authors report the phenomenon of rice lodging in Shazai Island, China, caused by heavy rainfall and strong wind, and stablished a decision tree model to identify rice lodging using intensity information before and after rice lodging in Sentinel-1 SAR images, and the in-situ lodging samples. They used sensitive parameters to build the decision tree model to identify rice lodging problem. The remote sensing monitoring problem and solution proposed is well organized and complete. Besides, the presented result can guide the future use of SAR-based information for crop lodging assessment and post-disaster management, also useful for other agricultural cultures with some customization.

However, there are small points to be adjusted by the authors and in such a context they should take into account:

1. The Authors should improve the reference review in the state of the art level;

 

2. The Authors should make clear in the paper a better description of the rice agricultural problem and its solution based on the use of remote sensing (strong points and weaknesses), as well as advantages in relation to other possible methods to observe the studied phenomenon;

 

3. The Authors should explain where he has found enhancements in the integrity of the ontology-based models, and what they really brought as a contribution if they have observed any comparison with other published papers or even any patents;

 

4. The Authors should explain the used prior knowledge guides, i.e., the entity alignment and text modalities, in fact in relation to the criteria’s used;

 

5. The Authors should explain better Figure 3 (Methodological flowchart for extracting the distribution map of lodging rice using Sentinel-1 imagery).  Authors should consider additional information related to the developed project and choices carried out. Why those choices? In fact, how should be compatible with the presented theories? A better attention in terms of the concepts and customization as a function of the expected application of their arrangement should be presented (also including the references for the SAR backscattering coefficients and polarimetric parameters presented in Table 3). The Authors should present a better discussion about the results presented in Figure 6 (Boxplots presenting the variation values of VV (a), VV+VH(b), VH/VV(c), and Span(d) for healthy and lodging rice plots). This point should be observed carefully, since it is quite important for the qualification of the developed method;

 

6. In relation to the manuscript, there are very few comments related to the integration of the methods to have an applicable system. The agronomic intelligence aspects could be included in the discussion of the results and they have been missed in the paper. Authors should discuss and improve the manuscript in terms of the presented results and its customization for rice;

 

7. Finally, conclusion must be considered only in relation to the presented results. The Conclusions should be really improved. Also I am suggesting a revision in terms of the future perspectives, since the Authors have informed that the result of their study revealed that the Sentinel-1 data has an excellent separate ability for lodging rice and healthy rice, with an overall classification accuracy of 84.38%, and an area classification accuracy of 93.18%. Do they really believe that those excellent values can still be improved to get better accuracy of the lodging information?

Author Response

We appreciate that you provided the valuable comments for the improvement of the manuscript. Following your thoughtful suggestions, we have revised the manuscript. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

In this study, the authors evaluated the potentials of Sentinel-1 data including the backscattering coefficients, the sum of the backscattering coefficients, polarization ratio and Span for identifying rice lodging. The paper is on an interesting topic and the results are useful for the agricultural remote sensing community and very high accuracies were obtained. I list here below some comments to improve the presentation, but in general I would say that the manuscript requires minor revisions only.

 

2.2.1. Remote sensing data and pre-processing

Why did you also use the Ground Range Detected (GRD) products?

Which backscattering coefficient did you use? Sigma naught or gamma naught?

Please provide more details on speckle noise removal (i.e. window size, methods).

 

Did you use SNAP and PolSARpro? Some references are requited.

 

Table 1

Please add more information such as incidence angle and orbit (descending or ascending).

 

3.1.1. Construction of feature parameters

There are some decomposition techniques. Why did you choose these ten parameters?

 

5.1. The mechanism and explanation for our method

Why did backscatter coefficients and polarization parameters change? Which scatterings were related to these changes?

 

Could you clarify the last sentence of this section?

Anisotropy is the parameter complementary to H and shows poor correlation with biomass. 

 

Author Response

We appreciate that you provided the valuable comments for the improvement of the manuscript. Following your thoughtful suggestions, we have revised the manuscript very carefully. Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I see that authors have made their efforts to revise the paper. But there still are some issues should be addressed before publication.

1. I am confused about the XGBoost method when I intend to redo the calculating experiment. I suppose more detail procedures of the XGBoost method should be present.

2. As I understand, the decision tree model is deduced from the result of Figure 6. Actually authors did not use the function of decision tree. There is no decision tree model indeed.

3. The methods for comparison, such as IsoData, K-Means, should also be discussed with some previous applicant works in relevant fields.

 

Author Response

     Thanks a lot for the valuable comments. We have made some revisions, please see the attachment.

Author Response File: Author Response.pdf