A Novel Dual-Branch Neural Network Model for Flood Monitoring in South Asia Based on CYGNSS Data
Round 1
Reviewer 1 Report
Review of Song et al, A novel dual-branch neural network model for flood monitoring in South Asia based on CYGNSS data, submitted to Remote Sensing
Overall: This paper presents a new method for quantitatively determining flood inundation using GPS related datasets. The methodology appears to be sound and it does show some improvement over existing methods. The abstract is a good representation of the content of the paper, except for the last two sentences as noted below. The figures and table are mostly appropriate and informative, and the conclusions are supported by the paper’s content. It would appear that this paper is acceptable for publication with the consideration of the comments and suggestions noted below.
1. Abstract: CNN and BP should be clearly defined. Also, the last two sentences make reference to a “flood evolution law” that is not discussed at all in the paper. These should be deleted.
2. Throughout the paper, there are acronyms that should be defined on first use: GNSS-R on line 50, CYGNSS on line 53, DDM on line 67, CNN on line 92, BP on line 95, SMAP on line 102, DBNN on line 162, to name a few instances. At the end of the paper there is a table of acronyms, but this reader did not know that it was there until finishing the paper! Plus, there are some acronyms that are defined in both the text and in the acronym table at the end, which is redundant. Either define all acronyms in the text at first use, or introduce a *complete* table of acronyms at the beginning for the reader to refer to.
3. Section 2.2 on the SMAP data is too short to be able to understand this mission, the instrumentation and techniques involved, and the resulting datasets. Most importantly, “vegetation information” is much too vague. What does this represent, how is it retrieved, and what are the units? Both soil moisture and vegetation information should be described in a table similar to Table 1 for CYGNSS, or parameters from both missions could be incorporated into one table. In this way, the units and comments for the SMAP data could be presented as they are for CYGNSS.
4. Figure 3: CYGNSS has a typo “CyGNSS”
5. Section 3.1: There needs to be a justification for using thresholds of SNR>1.5 dB, angle less than 65o, and gains great than 0 and 0.4, in the screening of data. Why were these values adopted, and how sensitive are the results to these choices of threshold values (e.g. how do results change if one uses SNR>1.0 dB instead?)
6. Section 3.3: Why is “probably value greater than 0.5” regarded as submerged? Similar to point 5 above, this value warrants some justification, and discussion of how results might change if one adopts, say, 0.6.
7. Line 305 and Figure 7: “inversion results” requires some definition. The units, “normalized value” are similarly too vague to understand what is being presented.
8. Line 310: reference 28 does not appear to be the appropriate one for using soil moisture from SMAP to infer inundated versus non-inundated areas. This reference should also justify the use of 0.4 cm3 cm-3 threshold to use for defining flooded areas. Also, the formatting for the units for soil moisture need to be fixed (a space between cm3 and cm-3).
9. Caption for Figure 7: It seems that the descriptions for (e) and (f) are reversed.
10. Lines 325-337: This is a good discussion, however the point must be addressed that the DBNN technique used here requires input on vegetation from SMAP, whereas the previous PR method uses only CYGNSS. Since SMAP soil moisture is being used as the standard for comparison, it is clear that if there is *any* interdependence between the SMAP vegetation and soil moisture datasets, then this is not a completely independent check on the accuracy of the two methods. In other words, it would seem apparent that the DBNN method will agree more closely with SMAP since it uses related SMAP data for input!
11. line 346 “were” should be “was”
12. Figure 10, it is not clear what “CYGNSS” and SMAP” in the legend, refer to in connection with the previous data. Is CYGNSS from the DBNN or the PR method? Is SMAP derived from the soil moisture threshold technique defined previously?
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The review is in the attached Word document.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
This is a good paper presenting an interesting method of determining flooded areas. I have a couple of minor, but general, concerns and one significant area in which I would like the authors to expand.
Firstly, there are a lot of acronyms used in the paper. I appreciate the list provided at the end, but it is not complete and should be in alphabetical order. Even with the list, the authors should make sure that each acronym is defined at its first use in the paper.
Secondly, in section 4.2, lines 358-362, and Table 4, the authors talk about the 'increased flood area', it is not clear what this is increased from. (Is this an increase of inundated area observed on the first date examined to the peak inundation? Or is it a percentage of the land covered in each state? Or something else?)
My main request is that the authors expand on the description of the training of the DBNN model mentioned in section 3.2.2. Typically a NN is trained using a certain proportion of the data available, using some kind of alternative, independent data as validation. This kind of ground truthing may be difficult in this case, but the authors use SMAP data as a verification in the results. Is the target label, y in equation 6, what is used as the training data? if so, where is y taken from? I think this needs to be clearer. What is used as validation of the training data, and how much data is used for training compared to the amount used in the results.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
This is a well presented and novel work that will be of great interest to the GNSS-R community. I think the overall results look very encouraging and find this work worthy of publication as is.
However I have only one minor edit:
On Pg. 4, line 149, you have mentioned that the SMAP resolution is 9 x 9 km. Isnt the SMAP resolution 36 x 36 km?
Author Response
Please see the attachment.
Greatly appreciated for your kind efforts in reviewing our work.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Thank you for addressing my concerns. I now approve acceptance.
Author Response
We are very pleased that all your concerns have been well addressed.
Once again, many thanks for your kind efforts in reviewing our work and providing us the constructive suggestions. They are very valuable for improving the quality of this paper.
Author Response File: Author Response.pdf
