Thanks for the opportunity to review this interesting paper. I find the topic of the paper of high interest for AMT since it presents an innovative atmospheric measurement opportunity that involves new technology and that requires ad-hoc processing.
However, there are, in my opinion, major deficiencies in the presentation quality that need to be addressed in order to make the suggested measurement technique applicable by other researchers.
Moreover, there are some weaknesses in the scientific reasoning presented that I recommend addressing before publishing. Those weaknesses might not affect the results, but it would be unfortunate to leave incorrect statements on published science that can be misused in the future.
Given the fact that the paper is already under review and the main data is already presented I recommend a major revision and I am definitely willing to contribute again in the review process. However, given the extent of weaknesses of the paper that I will try to list I also encourage the authors to take the time to reformulate the study from scratch and submit a new paper. In either case, I hope my comments will help the authors improving their study.
GENERAL COMMENTS:
1) The paper lacks a clear and concise description of the procedure adopted to measure the atmospheric quantities. The concept of baseline and the separation of dry and wet weather is briefly presented at lines 78-82 and does not catch the attention it needs. When I first read the paper I thought that the measuring principle was Eq. 1, but in reality, it is a different formula that involves the concept of baseline (which only partially resembles the components of Eq. 1). The authors have to make a clear definition of dry and wet weather condition (at the moment the reader have to extrapolate from the contest). For dry and wet weather it would be beneficial to have a formula like Eq. 1 that explains how weather conditions influence your measurements (signal loss).
2) The paper title and abstract suggest that the main results of the study are to use CML to measure rain-rate and humidity. However, in the results section, it is possible to find only an attempt to measure the rainfall rate. Regarding humidity, the paper presents an attempt to estimate microwave attenuation from RH measurements and not the opposite. The results section is filled with arguments about the data processing which do not belong to the results section (wet antenna attenuation, k-R modeling, and gaseous attenuation). It is preferable to put all the data processing in one section and leave the Results section for the atmospheric measurement results and the accompanied estimated uncertainties (that will be of course a consequence of the processing).
3) The information about methods, data, and results are scattered around and it is very hard to follow the logic of the paper. The authors made an interesting assessment of the ITU k-R relation using PARSIVEL synthetic measurements. I believe that this is an interesting analysis, but it is logically separated from the atmospheric measurement attempts. It might be useful to separate it from the rest making it an entirely separate section (between data and results) or even better would be to include it in what is now section 3.4 making it a self-contained development of the k-R based retrieval technique. Doing that the paper will emphasize the two main atmospheric measurements, namely the dry-weather estimation of water vapor and the wet-weather estimation of rainfall (which includes wet antenna attenuation estimation as a processing step). It would also help to merge section 4 and 5, putting the discussion of the results close to the presentation of them.
4) The difficulties in understanding the procedure adopted come also from the fact that not all the variables introduced in the paper are properly defined. Some variable names are reused (c and d are power-law coefficients in section 3 and 4, but where first introduced in section 2 as the speed of light and distance between antennas). Frequency f changes measuring units from Hz at line 93 to GHz at line 96. The definition of dry and wet weather is not explicit, it comes just at the end of the paper from practical considerations. The capped Dm (Eq. 11) is not defined (I think it is the assumed threshold between convective and stratiform events). The separation between stratiform and convective is not described anywhere (it was already question 3 from referee 1); after reading the paper several times I am supposing the threshold Dm is given by an imaginary line in between the two of figure 4b, but this is not written in the paper. Finally, the parameters of the DSD are reported in Tab. 3, but they are not explained (already question 5 from reviewer 1).
5) Some comments appear to come out more from wishful thinking than from a proper quantitative evaluation of the results. As an example in lines 442-446 the uncertainties related to WAA and DSD assumptions are discussed only in a qualitative way. The authors missed the opportunity to quantify the uncertainty related to WAA to RR estimation as a function of link length. Alternatively, by analyzing the k-R scatterplot of the Prague data one could potentially make some quantitative assessment of the RR retrieval uncertainties due to DSD assumptions (the underestimation of RR below 2 mm/h is a very interesting aspect related to this).
SOME CONCERNS ON THE FIRST REVIEW:
6) During the review process, the CML sub-link naming scheme changed. This modification makes sense since it simplifies the naming scheme, but I do not see in the author's response a mention to this change. Also, the naming change seems not consistent: link 3008-3009 became link 6 in the map of figure 3, the same link in table 1 became link 3 (but I see that there is a reordering problem here), but again in figure 10 the cluster of data point that was previously from 3008-3009 are now belonging to the subplot of link 3. I suggest mentioning all the changes made to the manuscript in the ``answers to the reviewers" documents, also the ones that have not been suggested by the reviewers.
7) I wasn't able to find the details of the T-matrix simulations in section 3.5 as the authors answered comment 2 by Dr. Guyot. Actually, I wasn't able to find those details anywhere in the manuscript. I suggest to include the T-matrix parameters information and to move it to section 2 (not 3.5 as the authors mentioned), this is because the T-matrix parameters are essential to reproduce the results of figure 2.
8) The answers of the authors to reviewer 1 (questions 5 and 6) are not addressing the reviewer's concerns. Probably the authors misunderstood the questions since they briefly refer to Ulbrich (1983) to cover the entire discussion, but the points remain unanswered. Moreover, the phrasing used in both the manuscript and the answer is imprecise and leads to a misunderstanding of Ulbrich (1983) findings.
In Ulbrich (1983) it is assumed that any DSD is well represented by a three-parameter modified-gamma distribution. This assumption leads to the conclusion that every couple of moments of the DSD can be related through a power-law. Because of that, if one can characterize a couple of moments through a power-law it follows that the DSD becomes a 1-parameter only function (the free parameter is lambda) and any other couple of moments will be characterized by a corresponding power-law. For this reason, the reflectivity-rain rate fits of Fujiwara (1965) can be converted into fixed parameters N0, mu for the DSD and power-law coefficients epsilon-delta for the Dm-RR relation.
There are many problems with this approach:
- All the assumptions of Ulbrich (1983) have to be valid. The authors did not test, for example, how good a modified gamma with fixed N0 and mu parameter fit the DSDs measured by PARSIVEL
- The error in the Z-R fit are not evaluated and transferred to errors in the N0, mu, or Dm as computed by the implied Dm-RR relation
- The mathematical foundations of Ulbrich (1983) have been demonstrated to be flawed in logic (Illingworth and Blackman 2002), leading to artificial correlations among parameters.
To the best of my understanding, the theoretical DSD is used only to make a rough evaluation of the stratiform or convective nature of the precipitation in the PARSIVEL dataset. I do not think it will affect the results, but the explanation of how to use the theoretical DSD has to be corrected anyway.
What it comes out from these considerations and might be harder to sustain is the following:
``The k-R function (stratiform DSD) applied to the Prague dataset has been estimated from a fit to synthetic data obtained from the Duebendorf dataset (13 months) whose stratiform-convective classification is based on the distance of the data from the Dm-R curves derived with a mathematically faulty logic (Ulbrich 1983) using reflectivity-rain rate fits obtained by observing 31 storms in Florida (Fujiwara 1965)."
This argumentation looks weak to me and I wonder if the authors excluded any other possible option they had to discriminate between stratiform and convective cases in the Duebendorf dataset.
9) A very minor point on comment 2 from reviewer 1. By looking at the figure and its caption I also get the wrong message that there is a drop in the water vapor attenuation around 60 GHz. The detail of k being defined differently for Oxygen and Water is not clear from the figure. If I just look at it, I see that a certain concentration of water vapor absorbs the plotted amount of energy which depicts a dip around 60GHz that shouldn't be there. One way to make the figure less prone to misinterpretation is to define the a) subplot y axes as k\_moist - k\_oxygen, this reflects the description added to the text and conveys a clear message.
Another option is to plot the attenuation only for the frequencies of interest for the paper 70-90 GHz. The rest of the spectrum is not needed since it is not utilized or even discussed as a comparison to lower frequencies CMLs. In this case, I would also avoid plotting the b) panel altogether since it is not used in the paper.
MORE SPECIFIC COMMENTS:
9) Equation 1 -It may sound trivial, but I suggest to introduce the definition of Lt as tx -rx, so that the equation becomes Lt = tx - rx = Lbf + Lm + ...
10) Line 48 - I do not know if the term resonance peak in parentheses can be considered correct. I would avoid it.
11) Line 73 and following - There is some confusion among the use of the terms loss, attenuation and specific attenuation. Perhaps it is better to clear in the introduction that in general, in the text the term attenuation refers to specific attenuation (dB/km) apart from WAA where it actually means loss (dB).
12) Line 112 - also polarization and orientation of the drop is relevant (if the drop is not considered spherical)
13) Line 114 - How is D defined? From the typical usage of the pytmatrix package and Eq. 6 it only makes sense that this is the equivalent-volume diameter. Does this definition match the size measured by the PARSIVEL disdrometer?
14) Line 115 - The contribution of secondary waves is commonly referred to as "multiple scattering" effects; I think the authors can use this term to simplify the discussion. Anyway, the argumentation on why multiple scattering is negligible is wrong.
Usually, the evaluation if multiple scattering has to be taken into account, is done in terms of optical depth (Battaglia 2006). Optical depth takes into account the scattering intensity through Cext, and particle number concentration Nt. Even assuming Cext to be not relevant what becomes important is not N(D) which is the drop density per size bin, but the total drop density Nt=int N(D)dD.
15) Line 119 (Eq. 5) - I believe there is an error in the formula. If Cext is cm**2 I think the coefficient at the beginning of the formula should be 0.4343 and not 4343.0 (Berne and Uijlenhoet 2007)
16) Line 122 - Saying that R and k are equal to moments of the DSD implies that v(D) and Cext(D) are power-laws. This is a reasonable assumption for small drops for which the Stokes approximation of drag force can be assumed and the Rayleigh approximation for scattering applies, but it is not true in general (as it can be seen from Fig. 2 for drops larger than 1 mm). Also, the authors should change the term "equal" with the term "proportional to". As a matter of fact, if the two quantities would be always proportional to a moment of the DSD the relation between the two would be linear and not a power-law (that is what happens at lower frequencies).
17) In Figure 2 and Eq. 9 it is introduced the concept of extinction efficiency, but this is of no use for the application of the proposed study. The important quantity that goes in Eq. 5 is Cext, not Qext. This analysis leads to another wrong statement at line 139. A single large drop contributes much more to attenuation than a single small one.
On the other hand, it is relevant for the study to analyze the relative attenuation of DSDs with small and large Dm and the same RR. In these conditions, it is true that large drops attenuate less because they produce smaller attenuation per unit mass (not per unit area). Also, larger drops fall faster, meaning for the same RR their volumetric concentration is lower.
18) Line 143. I think this is an important part and would be great to have it formulated mathematically as it has been done for RR attenuation. The combination of sections 2.3 and 2.4 models might result in better estimates of RR due to the consistent adjustment of WAA.
19) Section 3.1 and 3.2 are confusing. Wouldn't be better presenting the Duebendorf and the Prague datasets altogether? I think it makes much more sense to say what is each dataset scope, instrumentation, measuring periods, and available data instead of having these three pieces of information scattered around into two sections and three subsections.
The model to discriminate between stratiform and convective precipitation is a method well suited for the Duebendorf dataset part.
Dry and wet weather discrimination fits well the Prague dataset processing, also WAA estimation belongs to this section.
20) Line 222 - Would be better to be quantitative here. What do the authors mean with mean MSL pressure? Could be an international standard atmosphere, Mid-latitude, mean MSL pressure in Prague, or others. Just reporting the number is sufficient.
21) Line 305 It is either "An are rainfall induced attenuations" or "A is rainfall induced attenuation"
22) Figure 6 - The correlation coefficient (CC is of little use to evaluate the discrepancies between observed and simulated attenuation. What a high CC value tells is that if one quantity is increasing or decreasing, the other is doing the same. It does not provide information on constant biases and drifts of the two quantities.
I am not really sure of what is the information that I can get from the linear fits to the data (not discussed in the text).
In such scatterplots, it is usually more interesting to evaluate the deviations of the data from the 1:1 to analyze systematic biases and trends. The discussion up the correlation coefficients at lines 355-359 doesn't seem relevant to me, it also makes December look like the worst-case (smallest CC) while from a visual inspection it is probably the moth giving the best agreement between observed and simulated attenuation.
It seems that the theoretical attenuation is limited to values smaller than 3 dB while the observed ones go up to 6 dB, I wonder what could cause such discrepancies (uncertainties in the humidity measurements or in the evaluation of the measured gas attenuation perhaps). Comparing the distributions of attenuation values might be informative.
24) Lines 426-430 - It is less than surprising that the results of the ITU and the stratiform-DSD-derived k-R model give similar results given the fact that for low-intensity precipitation they are almost indistinguishable (Fig. 8)
23) Line 537 - It is indeed surprising that the rainfall estimation performance turns out so good. The reason is that as Fig. 8 demonstrates there is already at low intensity quite a huge spread of points derived from the DSD data which should translate in a high uncertainty of the k-R model. Moreover, even a perfect k-R model is very sensitive to the uncertainties in the estimated attenuation. The comparison of panels a and b of Fig 11 show qualitatively how the distance between antennas influences the uncertainty in the measured k, but I would like to see that quantity assessed and the influence on the retrieved RR quantified.
Especially by looking at the 70GHz panels in Fig 11b it is possible to see the points of the scatterplot lining up. This probably indicates that the uncertainty in the measurements of k are specific to each receiving station. The separation into shorter and longer CML helps the qualitative assessment, but again the authors miss the opportunity to evaluate the uncertainty of the k measurement as a function of the distance between the stations.
SOME REFERENCES:
Battaglia, A., M. O. Ajewole, and C. Simmer, 2006: Evaluation of Radar Multiple-Scattering Effects from a GPM Perspective. Part I: Model Description and Validation. J. Appl. Meteor. Climatol., 45, 1634–1647, https://doi.org/10.1175/JAM2424.1.
Berne, A. and Uijlenhoet, R.: Path-averaged rainfall estimation using microwave links: Uncertainty due to spatial rainfall
variability, Geophys. Res. Lett., 34(7), L07403, doi:10.1029/2007GL029409, 2007.
Fujiwara, M.: Raindrop-size Distribution from Individual Storms, J. Atmos. Sci., 22(5), 585–591, doi:10.1175/1520-
0469(1965)022<0585:RSDFIS>2.0.CO;2, 1965
Illingworth, A. J., and T. M. Blackman, 2002: The Need to Represent Raindrop Size Spectra as Normalized Gamma Distributions for the Interpretation of Polarization Radar Observations. J. Appl. Meteor., 41, 286–297, https://doi.org/10.1175/1520-0450(2002)041<0286:TNTRRS>2.0.CO;2.
Ulbrich, C. W.1983. Natural variations in the analytical form of the raindrop size distribution.J. Climate Appl. Meteor.22:1764–1775 |
- Printer-friendly version
- Supplement
