Classification of Rainfall Types Using Parsivel Disdrometer and S-Band Polarimetric Radar in Central Korea
Round 1
Reviewer 1 Report
This manuscript compares and illustrates the existing methods of rainfall type classification based on disdrometer and radar, and verifies them with data. The methodology used in this study is overall sound, aside from some minor issues.
However, the language of introduction is more obscure, and the logical coherence needs to be improved. In addition, some parts of this paper are not concise enough. Below are some comments towards the text.
Major comments:
The dataset in this paper was selected from 12 days from June 2015 to March 2016.Why only 12 days? Is it because there's only 12 days of rain? Or is it just the 12 days with effective data? Are the possible non-liquid precipitation excluded? More detailed data sources are expected. The variables selected by the fuzzy logic method are: ZH, ZDR, KDP and AH. Why these variables? On what grounds? Explain the physical meaning of the differences in these variables. Where do the criteria "30.02" and "39.96" in Figure 5 come from and what is the basis? Please elaborate in detail.
Minor comments:
Parsivel and rain gauge should be added to Figure 1, and units should be added to the horizontal and vertical coordinates. There is a problem with unit in Figure 2. Please correct it. Line 192: What does “VPRs” stand for here? Please specify. Line 218-220: Method BR03-VPR is not described clearly enough. Is it means the result after the classification of BR03, and then determine the rainfall type according to the presence or absence of the bright band and the reflectance threshold? Line 245: Why “The goal of this method is to retrieve b from ZH, ZDR, and KDP”? Is it not for the purpose of inverting DSD? Please specify the method. Line 257-277: Fuzzy logic algorithm is described in too much space, which is a mature algorithm. I suggest that the author can simplify the content and cite relevant references. The size of the letters in Eq. (19) shall be uniform. Line 295-296: Modify this sentence. Line 303: D here is D0, right? Figure 6: Titles of this figure is wrong. Figure 7: The legend “2<R<6” is wrong. Line 323-341: The descriptions are suggested to simplify and add physical meaning. How to obtain the Jincheon lines (Eq 21-23) in Figure 9? The font sizes in Table 3 should meet journal standards.Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
General comment
The paper investigates the classification of rainfall in stratiform and convective rainfall using data from disdrometer and S-Band Radar in central Korea. The paper present interesting results because (1) rainfall classification methods are useful (2) the study investigate the performance of previously published classification methods (3) the authors proposed a new modified classification method for disdrometer that is well suited for radar observation.
The data sampling and thesingle location used may limit the validation of the method, but it can be considered as a good case study. The paper is relatively well structure and Figures are of good quality. However, the paper could potentially improve in quality if the clarity of the framework is enhanced (i.e. abstract and in the connections between sections).
It could be done by adding sentences that clearly explain that there are 2 steps (1) developing a new modified disdrometer classification method (using Jincheon data) and then (2) Compare that new method with the previous one on radar data at Yongin. A schematic summary could be helpful to summarized what is used and for what purpose. It may give a fast understanding of the study.
Specific comments
Abstract
Line 1
Precise the dataset you use constsit of 12 days spread between June 2015 to March 2016
Line 3
Add a sentence that explain the general idea.
Something like: First, we classiffied Stratiform and convective using Parsivel disdrometer measurement and secondly we verified the method against well establised radar classifictaion method. Moreover, a new modified (BR03-VPR) method was developed for disdrometer.
Main part
Missing line numbering, please arrange it.
p9 Line 243 is a block of line
p14 …. Line 353 is a block of line
Line 359 idem
Line 364 idem
Line 369 idem
p 17 Line 378
p 18 Line 382
The Figure 10a illutrates ZDR relation agaisnt D0 . It seems to me that the new classification method (Jincheon) presents the lowest performance compared to BR02 and YOU16. Would you comment on it explaining it potential impact when used with radar data.
Details comments:
Tables 1 p102 Unclear caption …
“Table 1. Rainfall types from the 12 days from June 2015 to March 2016 by the VPR method”
→ ?
Data sampling used in the study: 12 days from June 2015 to March 2016
Figures
Fig 2.
“A modified method, BR03-VPR, is a combination of the boxes I and II. “
→?
The Modified method BR03-VPR, is a combination of the boxes I and II.
Fig 6.
“Top: Droplet number concentration distributions for the stratiform, convective,” ….
->
“Top: Droplet number concentration distributions versus Droplet size for the stratiform, convective, ” ….
Fig 9 caption (Use one notation or the other but be consistent all over the article in text and figures)
log Nw
→
log10 Nw
You may specify that the Jincheon data is classified with BR03-VPR like using the label Jincheon: BR03-VPR
Text
Line 107
“A detailed description of the disdrometer can found in Löffler-Mang...”
→
“A detailed description of the disdrometer can be found in Löffler-Mang...”
Line 143
“A H (dB km − 1 ) also important and is given in terms of”
→
“A H (dB km − 1 ) is also important and is given in terms of N ( D ) (mm − 1 m − 3 )”
Line 151
Remind here what VPR mean → Vertical Profile of Reflectivity.
Line 212
Is the modified method develop in this study in particular and never before ? It could be called an unified method ?
“A modified”
→
“The modified”
Line 308
Which figure are you refering to when you introduce YOU16 classification ? The previous Table only focused on VPR, BR03 and BR03-VPR.
“YOU16 showed that the mean ...”
-->
“However, YOU16 showed that the mean ...”
→
Line 313
“Furthermore, the DSD of the convective rains (D max = 7.50 mm) was much broader than that of the stratiform rains (D max = 4.25 mm).”
You meant for YOU16 classification ?
You may want to have the paragraph on YOU16 Line 308-313 in a different section as it is already a comparaison between disdrometer and radar classification.
Line 360
Unfortunately (to be avoid, look like a personal comment)
→
However
Line 440
was introduced for rainfall in Korea … based on disdromter data.
Line 442
if the slope and/or intercept of the equations were altered.
Line 455
Finally, it is suggested that the rainfall identification should be conducted using the new modified
BR03-VPR method.
→ ?
Finally, it is suggested that the rainfall identification should be conducted using the new unified
BR03-VPR method.
Line 463-464
because the climate of East Asia is quite different from those of the United States and
Europe
It could be considered as a weak argument … is it important to mention if east Asia is different that Europe or US ? it is very different from Africa as well or Australia or Middle East. It is actually same latitude as US and Europe, just the monsoon (Baiu) period. According to the Koppen climate classification, it is actually relatively close to Europe and US compared to many other regions. There are all humid continental climate group + Hot summer subtype
https://en.wikipedia.org/wiki/Humid_continental_climate
https://en.wikipedia.org/wiki/K%C3%B6ppen_climate_classification#/media/File:K%C3%B6ppen-Geiger_Climate_Classification_Map.png
It may be better to mention that this method can potentially be used to characterise rainfall type in any region of the world (just coefficient may differ) and that is important for radar calibration as well Numerical Weather Forecast validation using radar data.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Overall Comments:
One way the manuscript could be improved is in terms of clarity. There are many difference datasets and approaches discussed here, so care must be taken to make it extremely clear what is being referred to when. For example, DSD parameters are being retrieved from radar, but radar variables are also being simulated from disdrometer data, and it was often unclear to me which was being referred to at any given time. There is a lot being presented here (climatological analysis, a comparison of DSD-based classification methods and development of new ones, and then the comparison with a fuzzy logic approach), so readers should be led systematically through the progression of the logic, which at times seemed to jump around (e.g., specific comment 10). There are also a few sections of the text that seemed repetitive, could be combined, or were out of order (e.g., specific comments 3, 5). The other overall area of concern regarding the manuscript is some of the scientific decisions. I found the lack of bias control of the S-pol data a bit concerning, as well as the subjective analysis (e.g., validation done by visual inspection) of much of the work presented. At times, vague and imprecise validation became an issue (e.g., specific comment 38). At other times, what was said in the text did not match what I was seeing in the figures (e.g., specific comments 29-33). This is not to say the work is bad or necessarily wrong, but care should be taken to defend and confirm the claims being made to make them as precise and defensible as possible.
Specific Comments:
Line 38: I understand what the authors are saying, but “strong correlation” is a bit confusing here. Perhaps rephrase to something like, “DSD variability can have a strong impact on the relation between Z and R”. Line 55: What is meant by “improving the derivation of precipitation estimation”? Lines 58-60: This needs to come earlier in the paragraph, since gamma distributions are already discussed on line 53. Line 61: Clarify that this peakedness refers to reflectivity and not something else (e.g., DSD) Line 63, 66: It is a bit confusing to mention specific methods (BR09, YOU16) here without actually explaining what they are until later in the paper. Either delay this discussion until the methodology section or explain how the methods compare and contrast. Line 71: This should not start a new paragraph as it still refers to the previous paper being discussed. Table 1 caption: Include a reference for the VPR method, since discussion does not come until later. Line 123: If my understanding is correct, I am very concerned that there was no bias correction done for Z and ZDR. The manuscript states that the QC procedure of You et al. [50] was used, but this does not seemingly include Z and ZDR. This could have significant ramifications for all of the subsequent retrieval work in the paper. Is there a justification for not including any sort of bias correction? Line 143: What is meant by AH being “given in terms of N(D)”? Sections 2.1, 2.2, and 2.3: It would help the clarity of the paper to keep discussions of each dataset together (i.e., first describe the Parsivel, its quality control, and the determination of its DSD parameters, then describe S-Pol, its quality control, and retrieval of DSD parameters) rather than splitting the descriptions, QC, and DSD discussion and bouncing back and forth. As of now it is quite confusing, especially as the Parsivel disdrometer is used to simulate the polarimetric variables using a T-matrix and you’re retrieving DSD parameters from the radar. Lines 145-148: More information is needed about the T-matrix calculations performed on the Parsivel data. What are the parameters of the T-matrix method used here? What temperature was assumed? What radar wavelength? What aspect ratio relation for raindrops? Were the radar variables calculated using the raw binned DSD data or the fit gamma distributions? Line 155: Over what sampling time period is the standard deviation of R being computed? Also, it would help to state explicitly that the BR03 method is a disdrometer-based method applied to the Parisvel data. Line 164: Move this sentence re: VPR to the following VPR section. Similar to the previous comment, it would help to state explicitly that the VPR method is a radar-based method applied to the S-Pol data. Lines 168-177: Much of this should be merged with the following paragraph, which often repeats things (e.g., about brightbands being indicative of stratiform rain). Line 181: I am a bit troubled that the brightband was determined subjectively as many objective criteria for identifying brightbands exist (e.g., Giangrande et al. 2008; Matrosov et al. 2007; Wolfensberger et al. 2016 to name a few). What were the subjective criteria? In addition, why was only reflectivity used and not correlation coefficient (rhohv) or even ZDR? Rhohv would provide a very clear indication of a stratiform melting layer compared to Z where brightbands can occasionally become weak or muddled. Line 183: Is this referring to the maximum height of liquid within storms, or their average cloud-top height? If the former, convective cloud tops can certainly exceed 7-8km in places like the US. Please clarify. Line 185: Should this be a greater-than-or-equal-to sign? Lines 185-190: This section was a bit confusing. What is the “low-convection scenario”? I also don’t understand the maximum height for the “high-reflectivity scenario” being the same as the brightband. Are the authors just stating that the reflectivity threshold should be lower for shallow convection than the threshold used for traditional deep convection? Line 197: The text here says “Yongin” but the Figure 3 caption says Jincheon. Line 214: Which previous studies? Please cite them. Lines 212-225: It is still not clear to me what the BR03-VPR method is. From the description it sounds like first the BR03 method is applied, and then the VPR method is applied, but how are they combined? What if they have conflicting classifications? Line 274: Please cite the studies that have used a fuzzy logic approach for classification of rain. Lines 275-277: I am confused as to how the membership function vertices (a, b, c, and d) were determined. Is the polarimetric data used that from S-Pol or the simulated polarimetric variables from the Parsivel, as suggested on line 281? Which method was used to determine stratiform vs. convective for the formation of the membership functions? Line 281-282: Where did these reflectivity thresholds come from for determining stratiform vs. convective? Why are the other previously described methods not being used? Figure 6: Please add percentile (perhaps 25-75th) bars or shading to each of the datapoints to demonstrate how much variability there is for these mean profiles. Line 308: What process would be responsible for such large drops forming below the ML? Line 308-309: The results of YOU16 for the mean D0 in stratiform vs. convective rain are surprised and a bit unexpected to me, whereas the current manuscript’s results are more aligned with my expectations. Could the difference be whether the convective rain was generated by warm rain processes vs. ice processes? Line 316-17: Please add units to the R ranges. Lines 319-320: I don’t understand what this means. The manuscript states that “the R values for all the stratiform rains were found to be in the lowest category (R < 2 mm/h)”, but that is clearly not true from the figure as there are multiple lines for various rainrate groups up to 20 < R < 40 mm/h. Line 321: Again, I don’t understand what “only the VPR method existed for the convective rains” means given that all three panels exist and vary in Fig. 7b. Line 323-324: It appears to me that the number concentration increases with R in stratiform rain in nearly all size bins, so I don’t understand what this means. For example, drops between ~0.3 and 1 mm clearly increase in concentration with increasing R. Do the authors mean that only for “medium drops” (roughly 1-3 mm) does the concentration increase in every size bin as R increases? If so, that should be stated clearly. The same applies to Lines 324-325 in convective rain. Lines 329-331: See previous comment. To the best of my interpretation the text does not agree with what is shown in the figure here, unless the authors mean uniformly across all size bins. Lines 332-341: Much of this section confused me as I don’t see how it differs from the previous paragraph. By “increase in DSD” (e.g., lines 334, 337), is something other than concentration being referred to? Line 332 references the “average drop size spectra” but that would seem to be concentration in each bin. Lines 323-341 should be re-organized. Lines 353, 359, 364, 368: I am of the understanding that both BR09 and YOU16 also derived their separation lines based on visual inspection, but a bit more information about this process might be value since quantitative verification metrics are being derived from these relations. Is there no quantitative approach for deriving such relations (e.g., finding the line that minimizes the error for both classes)? Also, since verification metrics are being computed, was a line based on visual inspection chosen and went with, or was the line iteratively converged on by minimizing the errors in Table 3 and weighting convective and stratiform errors equally? Even through “guessing”, minimizing the error in this way should at least provide some basis for finding what is truly the optimal relation. Table 4: Mixing BR02, BR09, and BRA04 on the same line is a bit confusing. I would reorganize this so that all references on the left side (as for YOU16 and Jincheon), and if a certain reference doesn’t apply for any given relation just leave it blank (as done for D0-ZDR for YOU16). Figure 10: Add whether this is from S-POL or the simulated variables from the Parisvel. If the former, are these retrieved values of D0? If the latter, are these calculated values of ZDR? Line 378: How does the observed ZDR from S-Pol compare to the simulated ZDR from the disdrometer? What is the impact of below beam effects like evaporation? Line 378-379: More information is needed about how this relation was derived and how it compares with other proposed relations between D0 and ZDR. I also have concerns about the accuracy of the relation, as “the newly derived equations… seem more accurate” is not a scientifically valid statement. Quantitative verification is needed – this cannot be verified just based on looking at the figure. Visually, I do not think the new proposed line fits the data better than YOU16 and BR02 as the relation appears overfit to data for small D0 and doesn’t match the large D0 values at all. To derive a relation equally valid across the full range of D0 and removes the impacts of sampling variability, it is probably better to bin the D0 data, take the mean/median/etc of each bin, and fit the curve to those derived values. This should result in a better fitting curve that is more in line with the existing relations. Lines 394-395: Why are these values different from those proposed on lines 281-283? A table of the fuzzy logic vertex values (a, b, c, and d) should be included. Table 5: To be honest, I am not sure I understand the metrics presented here. How does a misclassification differ from the error? What does the “error” represent if not a misclassification? And how does misclassification differ from the inverse of the accuracy in the other column? Line 434: Overall I agree that the fuzzy-logic approach is best when considering both stratiform and convective rain, but it is not strictly true that S-POL did better for stratiform rain as stated here as SHY95 had a higher accuracy rate and lower misclassification rate (this is stated correctly on line 449). I would add the fact that fuzzy performs the best when looking at both classes (also on line 451).
Typos/Grammar/Errata:
Line 52: Remove “only” and change “are” to “can be” Line 68: Remove “the” Line 72: Change “is” to “was” to keep tense consistent Line 107: Consider changing “channels” to “size bins” Line 145: Change “could” to “can” Line 270: Change “was used” to “is used” Line 294: Remove “seem to” Line 296: Change “bottom” to “bottom left”Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The authors successfully replied the reviewer's comments. The revised version of manuscript was improved,and the reviewer suggest to consider this manuscript for the publication of Remote Sensing.
Author Response
We appreciate the feedback on the revisions and also the valuable comment on our paper. Thank you very much.
Reviewer 3 Report
I have looked over the author responses and am satisfied with most of them. The only remaining uncertainty is #21, describing the BR03-VPR method. It is still unclear to me how the assessments from BR03 and VPR are being combined, and which one takes precedence in any areas of disagreement.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
