Quantifying the Effect of River Ice Surface Roughness on Sentinel-1 SAR Backscatter
Round 1
Reviewer 1 Report
The manuscript discusses the alignment of Sentinel-1 VV and VH backscatter with ice roughness features over a 0.25 km² reach of the Yellowstone R., USA. The overall manuscript objective was to quantify the correlation between surface roughness metrics, calculated as a standard deviation of elevation (h) within a 0.09x0.09m moving pixel window, and Sentinel-1 VV and VH backscatter intensity gamma0, dB. Table 1 of the manuscript summarizes predictor variables, but I found it hard to draw a dierct correspondance between this table and the presented figures in the manuscript body. Were they all mostly similar?
From a radiometric point of view, as the manuscript states, height variations should be measured over a distance less than 1cm, which is not achieved in the manuscript, as the authors reiterate on several occasions. I would add here that sub-centimeter scale is also not exactly what interests and informs the ice jam researchers. The median h is from 11mm to 19mm averaged over 0.0081m², while we would rather look at h values in a range of decimeters, in case of the largest boreal rivers, over a surface exceeding 100m² immediately before or during the break-up period. From this viewpoint the study setup is an overkill in the sense that it explores the variable with precision that is only rarely needed in the real-life applications. This is in a certain sense 'parallel' to the discussion hydraulics, where microscale channel topography features are being created by turbulent eddies of respective size but overall channel roughness (at least in sandy rivers) is related to macroscale topography, i.e. larger dunes with linear dimensions compalable to channel size. From the other hand, the C-band backscatter over 100m² pixel area might be only in minor degree affected by microscale scattering, but to an important degree, by the difference in reflectance/backscatter between wet and dry snow or ice surface topography under the dry snow (which is in part discussed in the manuscript). The authors briefyl discuss this effect in the manuscript. Besides overall evaluation of the manuscript, it would have been informative to present both SfM DEM and Sentinel-1 imagery for the respective dates so that the general readership could relate. To conclude, the manuscript presents the correct methodology, two relevant datasets, and a correct statistical analysis, the narrative is neat but unengaging. A certain effort from the authors is needed to explain how microscale roughness is represented in the C-band S1 imagery, and what is the immediate use or implications of their results for the general audience. The point concerning ice thickness effect is highly speculative and might not appear in the Abstract, also more data are needed to support this speculation, at least an idea of what is the possible range of the ice thickness values across the studied reach.
Author Response
Please see attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this paper the authors present a quantitative investigation of the effect of river ice surface roughness on C-band Sentinel-1 backscatter. Also, they used un crewed aerial vehicle-based Structure from Motion photogrammetry to generate high-resolution (0.03 m) digital elevation models of river ice surfaces, from which they derive measurements of surface roughness. They employed Random Forest (RF) as regression models to explore quantitative relationships between ice surface roughness and Sentinel-1 backscatter. A weak relationship between river ice surface roughness and Sentinel-1 backscatter were found. The authors suggest that C-band SAR backscatter is more strongly controlled by structural properties than by the surface roughness because C-band penetrates a significant distance below the river ice surface. I think that the work is well structured and presented. From my point of view the greatest strength of the work is based on the methodology that is very well presented, what makes the work susceptible to be reproduced by other researchers interested in the same topic. The syntax allows its understanding and reading. The figures and tables have high quality and are suitable for illustrating the work. Citations are sufficient, current and interesting for this work. The paper address relevant scientific questions within the scope of Remote Sensing Journal.
The authors make a great effort to characterize the roughness in the field. They develop very high resolution MDts from stereoscopic pairs of their own elaboration and images obtained with a drone. They have also addressed with different strategies the main problem of this work, which from my point of view, is the difference in scale between the Sentinel-1 pixel and the MDts (10m vs 0.030m). From my point of view, a more appropriate strategy would have been to address the problem on a large scale using several training areas of homogeneous and different characteristics between them instead of focusing the work on an area of 0.25 km2. However, this observation does not affect the assessment of the quality of the work presented.
I think that the work is well structured and presented. From my point of view the greatest strength of the work is based on the methodology that is very well presented, what makes the work susceptible to be reproduced by other researchers interested in the same topic. The syntax allows its understanding and reading. The figures and tables have high quality and are suitable for illustrating the work. Citations are sufficient, current and interesting for this work. The paper address relevant scientific questions within the scope of Remote Sensing Journal.
The authors make a great effort to characterize the roughness in the field. They developing very high resolution MDts from stereoscopic pairs of their own elaboration and images obtained with a drone. They have also addressed with different strategies, the main problem of this work, which from my point of view, is the difference in scale between the Sentinel-1 pixel and the MDts (10m vs 0.030m). From my point of view, a more appropriate strategy would have been to address the problem on a large scale using several training areas of homogeneous and different characteristics between them instead of focusing the work on an area of 0.25 km2. However, this observation does not affect the assessment of the quality of the work presented.
Author Response
Please see attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper deals with the quantitative investigation of the effect of river ice surface roughness on the Sentinel-1 SAR backscatter in the Yellowstone River watershed. This is a very important topic given the rapid climate change in the sub-Arctic region, which leads to irreversible change to river ice. The paper is generally well written and is nicely presented, as the main conclusions are properly exhibited in figure 7-9. However, there are still some problems in the paper structure and result analysis. Thus, I recommend that this manuscript could be considered for publication after major revision.
My main criticism is that after reading the paper I still don’t have a clear idea about the necessity of conducting ice type classification, since the surface roughness of different type of river ice as well as its connection with SAR backscatter is not fully discussed in the following sections. The spatial and temporal scope of the available research data is small, and I think the findings may not be representative enough, perhaps you need to further explain the universality of the research method.
Please give some quantitative conclusions of the weak relationship between river ice surface roughness and the Sentinel-1 SAR backscatter in the Abstract. The Methods section part is too long and some description is more suitable for the Introduction. The methods used in this study should be introduced as succinctly as possible, mainly focusing on the aspects of improvement and optimization. It is unnecessary to discuss the conclusions of previous studies which utilized the same methods (e.g. Lines 184-189, 217-230). Moreover, as shown in figure 1, there is no need to place the whole watershed as the main figure since you only focused on a small region. However, it is quite important to provide detailed map of UAV-SfM results and the processed Sentinel-1 SAR imagery overlapping research area, so the readers could have a better visual understanding of connections between the two independent data sets. Also, the Discussion part is too long, it has to be further simplified, there is much unnecessary redundancy in the present discussion, please re-organize this part and some descriptions are more suitable in the Results or the Conclusions part. The Conclusion part is rather weak now, the quantitative relationships between ice surface roughness and Sentinel-1 backscatter, as the key point of this research, at least should be better summarized here.
Minor remarks：
Line 44: the presence of daylight?
Line 46: better delete the first ‘and’
Line 55: of -> during
Line 61: of -> among
Lines 73-75: this sentence is too long, please modify it.
Line 121: should be “will affect regression model performance are quantified”
Line 122: higher -> more detailed
Line 261: In my opinion, this label procedure is of vital importance for subsequent data processing, 50% might not seem to be a reliable threshold, why not increase it to 80% or 90%? At least you could conduct related examinations on different thresholds and gave some detailed comparisons or explanations.
Line 281-287: No need to mention the GLCM since you didn’t use it in your research at all. Please keep the text concise and focus on the key points.
Line 364: Is the 0.0011 mm meaningful? Since 0.001 mm is merely 1μm. What physical meaning of ice surface roughness does this value convey?
Line 385: as shown in figure 5 and 6, the February 19 classification results is satisfying, it is better to explain or at least discuss the mechanism here rather simply placing some numbers.
Figure 7-9: Please insert these figures right below the corresponding paragraphs.
Figure 8: Should the bottom line “median roughness” be moved to the top?
Line 413-414: Is there any possible connection between lower classification accuracy and more tightly clustered distribution of March 4 group?
Line 435: the “left tail of surface roughness distributions” is apparent, but it is simply due to the rather small magnitude of minimum and 5th percentile targets. Moreover, the different model performance of using minimum and maximum as the sub-grid size changes is interesting, it is worth conducting further analysis to explain it.
Line 463: remove the first “on”
Line 555: Since you mentioned in the Abstract that C-band SAR could penetrate into the ice layer which reduced the information of surface roughness, the L-band NISAR should further cause more loss of surface roughness information, so is your method still available or compelling with NISAR data?
Author Response
Please see attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors followed the reviewers' comments and significantly improved the manuscript. Now that the motivation of the authors is clearly pronounced, and the importance is stressed of explicitly accounting for microscale effects while using C-band SAR data for large-scale applications is assuming macroscale roughness, I can endorse the publication of this manuscript.
Reviewer 3 Report
the authors have revised the manuscript according to comments. Though I hope more improvements should be addressed, but in fact it is very very difficult to study on the theory of SAR image. It can be accepted after the language should be polished.
