Scattering Properties of Non-Gaussian Ocean Surface with the SSA Model Applied to GNSS-R
Round 1
Reviewer 1 Report
This manuscript has implemented the SSA model with both Gaussian and non-Gaussian sea surface slopes’ PDF to analyze GNSS-R scattering. It is unclear to me whether the formalism is the same as Bourlier 2004 [29] or there are some derivations by the authors. It has to be noted that Bourlier 2004 finally focused the study on C- and Ku-bands, higher bands in the microwave spectrum. Furthermore, it is broadly accepted that L-band forward scattering is little affected by short wavelengths, and one can typically define a cut-off wavelength under which ocean waves of such short wavelengths do not have impact in the final scattering process. These aspects were neither taken into account nor discussed in the manuscript. These aspects would not be an issue if the results were convincing, but Figure 8 clearly shows that the data do not follow the implementation of these models. Moreover, perhaps SSA was not necessary to inspect the effect of non-Gaussian sea surface roughness: in the past the KO-GO with Gram-Charlier slopes’s PDF was used, with good match with observational GNSS-R data (a 2008 article not cited in this manuscript, see below).
Other comments:
After diagonal reading of [29], my impression is that this study applies the formalism developed in Bourlier [29] to L-band and forward scattering geometries. Is my interpretation correct or the authors have done some of the development? Please clarify in the rebuttal letter as well as in the body of the manuscript.
Intro, general references lines 22-24:
- add e.g., most of them are only examples (many more have been published);
- references on wind direction [4-6], missing a former one (2008) about wind direction and sense (up- /down-wind): doi:10.1016/j.rse.2008.02.003;
- references ocean surface altimetry [8,9]: the 9th is NOT on surface altimetry, but surface waves (so it should be listed as a separate application). Regarding ocean suface altimetry, it is important to highlight that the only spaceborne precise ocean surface altimetry has been reported in doi:10.1109/JSTARS.2019.2952694, with results one order of magnitude preciser than [8];
- besides sea ice detection, there are other sea ice and ice shelf applications, for example, sea ice concentration, type and altimetry (doi:10.3390/rs11212565, doi:10.1016/j.rse.2019.05.021, doi:10.1002/2017GL074513)
Line 35: “environmental factors.[14].”
- Remove the point before [14];
- Perhaps [14] is not the best reference to illustrate the “sensitive to wind speed and other environmental factors”. Perhaps better a tutorial or more general overview?
Line 40-41 “ In addition, the single polarization method limits the acquisition of phase information” What do you mean by ‘phase information’? Phase can be the carrier phase, the code phase, or a given phase of a process/fenomena. In any case, the single polarization does not limit the acquisition of phase information (neither carrier not code phase).
Lines 63-69: “However, to improve the accuracy of the cross-polarization, the SSA-2 model was used to calculate the full polarization scattering coefficient for the bistatic case, taking into account surface Bragg scattering, and it was found that the change in wind direction had little effect on the NBRCS for the circular polarization case and that there was still ambiguity and symmetry in the up/down or crosswind direction, but that there was some sensitivity to wind direction if the bistatically scattered signal is received in a non-pure specular direction[25]”
- doi:10.1016/j.rse.2008.02.003 did notice up-/down-wind sensitivity, both in the models and data. For modelling them, the key point is to introduce a non-Gaussian PDF of the slopes (note that doi:10.1016/j.rse.2008.02.003 used the simple KA-GO).
Lines 77 a 83: these two references describe back-scattering processes, the statements that apply to backscattering do not necessary apply to forward scattering. Remember that for L-band bistatic scattering, non-Gaussian roughness was studied in doi:10.1016/j.rse.2008.02.003 (2008). Again, a key element is the roughness description, there following the Gram-Charlier distribution of the surface slopes (Cox and Munk 1954), including skewness and peakedness factors.
Line 101: ‘subsatellite’, to me, this indicates the point that projects the satellite location on the Earth surface (not other points on the surface), but perhaps I am wrong…
Line 112: ‘However, The spatial resolution’ → “The” should be “the”
Lines 108-113: Careful, ERA-5 in particular, and the global numerical weather prediction (NWP) and reanalysis models in general, saturate at high winds. This is a well-known problem of the NWP, and it can also be seen in the comparisons in Cardellach et al [32].
Equations 12 and 13: please define K
Just after eq (16): “...means the number of waves …” → “… means the wavenumber…”
Equation (18): Delta is not defined. Perhaps it could be simply deleted and direct the reader to Elfouhaily et al., 1997.
Equations (20) and (21): please cite from they are taken, the derivation is not obvious and no reference is given.
Line 174: “Form Equation 15 and 22,” → From instead of Form
Are equations (25) to (30) from Bourlier? Or the authors have expanded them further from those in Bourlier? (not clear from the text).
Equations (25) to (30): these expressions seems derived from more general ones, where, at L-band, some filtering should be accounted for (to lower the weight of short wavelengths). For example, if they were derived for Ka-band or other higher bands in the microwave spectrum, they could not be sufficiently accurate at L-band. Perhaps this explains why at low wind regime the behaviour of this approach is not as expected (Figure 6).
Just before equation (31): “...skewness and peakedness respectively. we have”
- “We” instead of “we”
Line 200: “ Noting that the non-Gaussian NBRCS is on average about 5 dB larger than the Gaussian NBRCS, due to the effect of peakedness correction.” → please justify the sentence, not obvious.
Lines 202 to 209: these statement should be better justified, or simply point to a Figure where they are “visible”. (now re-reading the section, I think these lines are commenting Figure 4, but this was not clear at first because there mention to Figure 4 was in the former paragraph, please merge in a single paragraph or cite again Figure 4 to help understanding the text).
Figure 3 is never commented or cited in the text of the manuscript.
Line 211-212: “It can be seen that the NBRCS is not maximised at the nominal specular scattering point (θ i = θs ) under non-Gaussian ocean surface, the maximum will deviate by 3◦ − 5◦.” → I don’t see this in Figure 5. It’s hard to judge when the sampling of the plot is 5⁰. To me, the Gaussian and non-Gaussian peaks look aligned.
Line 212-214: “ Note that the main difference relating to polarization is also beyond ±20◦ of the purely specular scattering angle, so choosing circularly polarized reception in the purely specular direction has the same effect as linearly polarized reception.” → Taking into account that the RHCP power is 50% H and 50% V, and that the RR component is many dB below, I don’t understand why the NBRC of RL, RH and RV they are all the same around the specular direction. One would think that the linear components should be ~3dB below the circular RL component (you could de-compose the received RL into RH and RL, and the total power should be the same). Why are they all the same?
Line 245: “dow-” → down
Lines 250-251: “In order to verify the performance of the non-Gaussian NBRCS with wind speed, we use the root mean square error (RMSE) and the coefficient of determination (R2) metrics to
compare with least squares, are presented in Table 1.” → please explain better this comparison, as it makes no sense to me that there are rows devoted to CyGNSS (comparing CyGNSS with CyGNSS?!!). Do you mean individual measurements compared with each of the lines in Figure 8? (so, the CyGNSS rows correspond to CyGNSS individual measurements compared to the fit?). Please clarify.
Line 255: “ less than 0” → negative
Lines 253-254: very poor analysis of the results in Table 1 and Figure8. You mention that the Gaussian R is negative in the three angles, but it is also negative in 2 out of 3 non-Gaussian. Similarly, the RMSE have similar ranges of values for Gaussian or non-Gaussian models (slightly better for non-Gaussian, but same orders). Figure 8 shows that the non-Gaussian model works good at ~45⁰ azimuth, but not very well on the other two…
Figure 8: The CyGNSS data appears to me at the same levels in all three plots. For example, in all three plots, the maximum density of points at 7.5 m/s is just below 15 dB, regardless of the wind direction. Similarly, the black fit in all three plots looks very similar, too. However, the SSA models, both Gaussian and non-Gaussian, do shift between plots (as function of the wind direction). This Figure does not validate the model or implementation of the model, but rather the contrary. Perhaps using the KO-GO with modified PDF (Gram-Charlier to introduce skewness and peakedness, as it was done in doi:10.1016/j.rse.2008.02.003, would give more consistent results?)
Figure 9: this figure is not illustrative, does not help taking any conclusions. First of all, lines Gaussian-phi 45 and non-Gaussian-phi 45 have the same color and symbol, so it is not possible to distinguish both models. Same for phi-90. Second, all the CyGNSS measurements are combined, no matter the phi-angle. If the goal is to link the variability of NBCR at each given wind speed to the variability the model predicts, based on the phi-angle, then you should plot the ‘envelop’ lines (angles at which the NBRC are minimun and maximum, according to each model). This figure, as it is, cannot be published.
Line 265: “for the two stations” → what are these stations? I don’t understand.
Line 266: “improves its symmetry” → symmetry or wind-direction signature? (with 180⁰ ambiguity, at least)
Line 268: “NBRCS variability is greater for smaller scattering angles” →this is counter-intuitive…
Figure 10: another aspect of Figure 10 that is not commented in the manuscript is the fact that RH has nearly 5 dB more power than RL, suggesting that linear polarization might be better. Could you discuss this result? Do you understand the reason?
Figure 11: What is the wind direction? It is difficult to interpret these plots without this piece of information.
Figure 11: the “pit” in the green line is interpolated from the simulated values, in triangles. The actual shape of this “pit” between the triangles at 35⁰ and 40⁰ is arbitrary, “invented”, perhaps better without any line, only the symbols.
Figure 11 Caption: it first reads theta_i =30⁰, then it reads theta_i=20⁰ and theta_i=45⁰ for a) and b). So, which is theta_i? Please revise carefully and note down the right information, without any missing piece.
Almost all citations to references do not have a space before them, for example “ … developed by Cox and Munk[39]” → there should be a space between Munk and [39]. This happen almost every time a reference is cited in the text.
The English language is generally good, but some sentences here and there would require a revision.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Authors tend to present scattering properties of non-guassian ocean surface with the SSA model Applied to GNSS-R in this manuscript. And comments are listed as follows. There exists 5 major issues.
Major issue 1: Among 43 references, only 7 references are published within 5 years. Authors should review more papers within 5 years and present newer references.
Major issue 2: Authors should summarize main contributions of this manuscript and present main contributions in the part of introduction.
Major issue 3:Authors should compare the proposed approximation model with other sea surface scatter models.
Major issue 4: It is unbelievable that the values of coefficient of determination can present negative value in Table 1. The range of R^2 should [0,1]. In this case, values in Table 1 can not support the conclusion of this manuscript.
Major issue 5: Authors should explain why NBRCS of σRH is much higher than σRR in Figure 10.
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The research investigates Scattering Properties of Non-Gaussian Ocean Surface With the SSA Model Applied to GNSS-R. I have the following comments:
1. Why the authoers proposed proposed Non-Gaussian model? Justification is required.
2. When the non-Gaussian model is developed, how statistically siginificance the model is and at what confidence level.
3. The R-squared is very small and some time is negative which means the data does not fit the proposed model well.
4. Othere distributions could be investigated as well.
The quality of English Language seems fine to me.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I think this paper has been revised carefully according to the comments from reviewers. This manuscript can be accepted for publication in this version.
Minor editing of English language required
Author Response
Dear Reviewer
Thank you for reviewing, revising and helping with this manuscript.
Sincerely,
Weichen Sun
Reviewer 3 Report
The authors considered the comments
None
Author Response
Dear Reviewer
Thank you for reviewing, revising and helping with this manuscript.
Sincerely,
Weichen Sun
Reviewer 4 Report
All my questions have been properly worked out. I think this manuscript can be considered to be published in Remote Sensing now.
