Gap Filling Cloudy Sentinel-2 NDVI and NDWI Pixels with Multi-Frequency Denoised C-Band and L-Band Synthetic Aperture Radar (SAR), Texture, and Shallow Learning Techniques
Round 1
Reviewer 1 Report
This study evaluates multi-frequency SAR (C-band and L-band) for the purpose of predicting spectral indices (NDVI and NDWI) to fill in the gaps due to cloudy pixels. Different machine learning techniques were used to calibrate multi-source SAR data to NDVI and NDWI. Several de-noising techniques were applied to remove noise from SAR data. Results are encouraging. The paper is clear, well described and supported by figures, and easy to read.
Major comment comments:
The idea is good. However, the database used is not sufficient. The author uses a single optical image to calibrate the SAR data into NDVI and NDWI. The use of a single image does not evaluate the potential of SAR data to estimate the full range of NDVI (or NDWI) of a single class. SAR data change in behavior with vegetation development and the use of on date image do not allow a complete validation of the proposed method.
Author need to consider a temporal series of Sentinel-2 (to catch all NDVI values for on single class) and Sentinel-1 to address the objective of the study. It is ok if the S2 and S1 are not acquired at the same date. S2 image can be interpolated at the dates of S1.
Minor comments:
Lines 15-16: This sentence is not clear, please be more explicit
Line 18-19: MAE of?
Line 157-160: Please be more explicit when talking about the dates of the images
Lines 225-227: How author deal with this offset?
Lines 243-245: please describe the principle of multi-temporal speckle filtering. Precisely, how you guarantee that this filter do not remove multi-temporal variation that can explain change in vegetation cover.
Lines 245-246: not clear the conjunction use, please explain
Lines 280-281: why author do not simply use all pixels for training and validation?
Lines 282-283: not clear. How the count (number of samples) was evaluated with each of the machine learning regressors?
Line 302: why 500 trees? Please justify the number of trees chosen.
Lines 331-332 : please add references. Please change to something like: penetrate the vegetation and thus have soil contribution to backscatter, whereas the C-band backscatter comes mainly from the vegetation cover, especially when the vegetation is well developed (NDVI > 0.6/0.7)......
Section 3: please be careful when talking about the accuracy as all accuracies seems to be very close
Line 338: ‘when used in conjunction with L-band UAVSAR’ is opposite to what you said before about C- and L-band. Why in this case the use of L-band improves the accuracy.
Table 2: please use the term ‘USAVSAR’ instead of ‘UAV’
Table 2: please use results from the best predictor’s combination and plot estimated NDVI as a function of predicted NDVI using a scatter plot. The same for NDWI. This will make the results more clear.
Figure 4: Please add the filled NDVI.
Figure 4 and 5: it seems that predicted image has higher values than original image for the non-cloudy pixel.
Line 378 – 380: can you please explain why errors are higher on some classes than others?
Lines 398-399: could you please provide a physical explanation.
Lines 405-407: not clear this sentence.
Lines 444: where and what kind of landscape?
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
1. In the introduction, you did not mention what is new in this research. If there is a novelty, please add it to strengthen this paper.
2. Please explain more on what basis do you compare spaceborne and airborne data? For L-band data, I suggest you use spaceborne data too.
3. On lines 141 and 345, please note that the position of the table caption should be above the table
4. Please add an explanation about the influence of the atmosphere on the radar image concerning this study
5. Please re-emphasize your novelty at the conclusion
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
My comments:
1. abstract, all acronyms must be explained.
2. main body of text, all acronyms must be explainedif you used their first time.
3. please check English language in paper.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The author correctly answer my comments
