Identification of Snowfall Riming and Aggregation Processes Using Ground-Based Triple-Frequency Radar

Wang, Danyang; He, Wenying; Bi, Yongheng; Xia, Xiangao; Chen, Hongbin

doi:10.5194/egusphere-2025-4233

Preprints

https://doi.org/10.5194/egusphere-2025-4233

Preprints

05 Sep 2025

| 05 Sep 2025

Status: this preprint is open for discussion and under review for Atmospheric Measurement Techniques (AMT).

Identification of Snowfall Riming and Aggregation Processes Using Ground-Based Triple-Frequency Radar

Danyang Wang, Wenying He, Yongheng Bi, Xiangao Xia, and Hongbin Chen

Abstract. Riming and aggregation are critical ice-phase microphysical processes in winter clouds, but their overlapping signatures and dynamic transitions pose challenges for conventional single-frequency radar detection. We introduce a novel gradient-based identification method using ground-based triple-frequency dual-polarization radar observations. By analyzing vertical gradients of triple-frequency radar variables, rather than their absolute values, we discern these microphysical processes through physically based thresholds that reflect particle growth regimes. This approach captures subtle spatiotemporal variations in riming and aggregation that conventional threshold methods would miss, particularly in resolving layered riming-aggregation transitions. The dynamic gradient-based method demonstrates the enhanced physical consistency and adaptability near process boundaries, which obviously improve the tracking of ice-particle evolution. These advances provide a pathway to refine microphysical parameterizations and enhance high-resolution snowfall forecasting.

Received: 29 Aug 2025 – Discussion started: 05 Sep 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Danyang Wang, Wenying He, Yongheng Bi, Xiangao Xia, and Hongbin Chen

Status: open (until 02 Jan 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2025-4233', Anonymous Referee #1, 02 Oct 2025 reply
The manuscript presents a gradient-based identification method to distinguish riming and aggregation processes using triple-frequency ground-based radar observations. The authors combine both traditional threshold-based diagnostics and a newly developed gradient-based multi-parameter method, applying these to a well-documented snowfall case during the TRIPEx-pol field campaign. The work is carefully written, well referenced, and demonstrates that the authors have carried out a thorough literature review of the state of the art.
The scientific motivation is clear: distinguishing riming and aggregation is a long-standing challenge, and improvements in radar-based diagnostics can directly benefit the representation of microphysics in numerical models. The authors build directly on the foundation of Planat et al. (2021), who introduced a gradient-based approach for single-frequency polarimetric radar data. Here, this idea is extended to triple-frequency radar, which increases the sensitivity to particle density, shape, and size evolution.
A weakness of the current study lies in the lack of independent validation. Without in-situ ground-based hydrometeor observations (e.g., particle imaging or disdrometer measurements), the conclusions cannot be fully verified. Because the melting level was above the surface during this event, the results must be considered a proof of concept rather than a definitive validation of the method. Especially that Mason at al. have shown that the triple frequency (DWR-DWR) signatures can be also modulated by the shape of the PSD. Future work should attempt collocation with in-situ particle imagery or hydrometeor classification to substantiate the gradient-based classifications.
Another important point relates to interpretation. The gradient method identifies the altitude regions where riming and aggregation are most active, but it should be expected that observational signatures of large aggregates (e.g., enhanced DWRX–Ka relative to DWRKa–W) will appear below the regions diagnosed as aggregation-active by the gradient approach. Clarifying this causal relationship would strengthen the physical interpretation.
Overall, this is an innovative application of an existing idea that extends it to triple-frequency radar and demonstrates the advantages of gradient-based methods for identifying transitions. The work is rigorous, clearly presented, and worth publishing after minor revisions.
In the revised version please address these aspects:
Emphasize more clearly how this method complements or improves upon the approach of Planat et al. (2021).

Discuss potential strategies for validation (e.g., future campaigns with particle probes or disdrometers).

Expand slightly on how these results could be generalized to other sites or incorporated into retrieval/parameterization schemes.

Clarify in the discussion the expected vertical offset between regions of “active aggregation” and observed aggregate signatures.

discuss what temporal averaging was applied to the data and how the vertical gradient was computed (I suggest using Savitzky-Golay method)

Reply
Citation: https://doi.org/10.5194/egusphere-2025-4233-RC1
- AC3: 'Reply on RC1', wang danyang, 30 Nov 2025 reply
  
  We sincerely thank you for the thorough and constructive review of our manuscript, and we are grateful that you acknowledged the innovation of our work and the value of its extension to triple-frequency radar methods. In response to your comments, we have thoroughly revised the manuscript, including enhancing our methodology (for example, using a Savitzky–Golay filter to smooth the data and reduce noise), strengthening the theoretical explanations (with an added clarification of the vertical offset mechanism), and expanding the discussion of future validation strategies and application scenarios (such as independent validation and exploring integration with advanced retrieval schemes).
  Reviewer Comment 1: “Emphasize more clearly how this method complements or improves upon the approach of Planat et al. (2021).”
  Response: We sincerely thank the reviewer for this insightful comment. In the revised manuscript, we have expanded the Introduction and Section 2.2.2 to more clearly explain how our method builds upon and improves the PIVS approach of Planat et al. (2021).: “Given that vertical gradients of radar observables contain rich information about precipitating particle evolution, Planat et al. (2021) introduced the Process Identification based on Vertical Gradient Signs (PIVS) method using a single-frequency dual-polarization radar. Their pioneering work demonstrated that vertical gradients of Ze and its polarimetric profiles can effectively indicate zones where riming and aggregation processes coexist. This provided an important foundation for radar-based snowfall microphysical classification. While PIVS successfully identified coexistence regions, it could not distinguish riming-dominated from aggregation-dominated layers. Precipitation particles in winter clouds exhibit pronounced spatiotemporal variability in their microphysical properties (density, size, and shape). This variability makes it particularly challenging to distinguish rimed from aggregated particles using any single radar measurement alone (Karrer et al., 2022; Planat et al., 2021; Thompson et al., 2014). To overcome these difficulties, we developed an improved identification method using ground-based triple-frequency radar observations. Building upon the vertical-gradient concept of Planat et al. (2021), our technique exploits the vertical gradient signatures of multiple radar parameters. In particular, we incorporate Ze, DWR, MDV, SW, and LDR into the analysis. By simultaneously tracking changes in particle size, density, and shape with height, this multi-parameter, triple-frequency approach provides a more robust physical framework for process identification. As a result, it offers enhanced sensitivity to subtle microphysical process transitions and greater flexibility in capturing regions where riming and aggregation processes transition or coexist—capabilities beyond those of conventional threshold-based methods or the original single-frequency PIVS approach.” Likewise, Section 2.2.2 now explicitly notes that the new method “To further improve the flexibility and robustness of our identification method, we introduce a gradient-based multi-parameter vertical‐gradient (VG) method that extends the single-frequency PIVS concept of Planat et al. (2021) to a triple-frequency radar framework. This new VG approach leverages combined information from multiple radar observables to more comprehensively discriminate between regions dominated by riming and those dominated by aggregation. In particular, we jointly analyze the vertical gradients of various radar parameters (e.g., DWR, MDV, SW, and LDR). By exploiting these multi-parameter gradient signatures, the method becomes more sensitive to microphysical changes with height and more adept at capturing process transitions or coexistence of riming and aggregation than earlier single-parameter approaches.” These additions clearly emphasize how our method complements and advances the previous approach.
  
  Reviewer Comment 2,3: “Discuss potential strategies for validation (e.g., future campaigns with particle probes or disdrometers).” “Expand slightly on how these results could be generalized to other sites or incorporated into retrieval/parameterization schemes.”
  Response: We thank the reviewer for these constructive comments. We fully agree on the importance of both independent validation and discussing the broader applicability of the proposed method. Accordingly, we have expanded the Discussion and Conclusions section to outline possible validation strategies and to clarify how the approach could be generalized and incorporated into retrieval and parameterization schemes. “Direct validation of the radar‑based identifications remains challenging here due to the lack of collocated in‑situ measurements. Future field campaigns can address this by deploying in‑situ probes alongside triple‑frequency radars. Particle-imaging sensors (e.g., a Multi‑Angle Snowflake Camera or a Precipitation Imaging Package) and optical disdrometers (e.g., a two‑dimensional video disdrometer) could capture snowflake size, shape, and degree of riming in real time, providing ground‑truth observations for comparison with the radar data. Synchronizing these in‑situ records with the radar observations would directly test whether riming‑classified regions indeed contain heavily rimed particles and whether aggregation‑identified regions correspond to large, clustered snowflakes. Implementing such validation strategies will build confidence in the classifications and help refine the identification criteria as needed. Because it relies on relative vertical trends rather than site‑specific thresholds, the gradient‑based framework is inherently more generalizable across climates and seasons. It could therefore be applied to snowfall events in other geographic regions (e.g., maritime versus continental snow climates) with minimal re‑tuning, although diverse datasets are needed to verify consistent performance. Demonstrated generality would enable integration into operational retrieval algorithms and numerical model parameterization schemes. For example, weather-radar networks could use the classifications to inform snowfall‑rate retrievals—adjusting for riming or aggregation when estimating snow water content or particle fall speed. In numerical weather prediction models, observed signatures of riming and aggregation could guide improvements to ice-phase microphysics parameterizations – for instance by refining assumptions about riming efficiency or snow density. Bridging these observational insights with predictive tools would improve snowfall‑rate estimates and the representation of ice‑phase processes in models. Ultimately, these advances should yield more accurate snowfall predictions and a deeper understanding of cloud microphysics across diverse environments.”
  
  Reviewer Comment 4: “Another important point relates to interpretation. The gradient method identifies the altitude regions where riming and aggregation are most active, but it should be expected that observational signatures of large aggregates (e.g., enhanced DWRX–Ka relative to DWRKa–W) will appear below the regions diagnosed as aggregation-active by the gradient approach. Clarifying this causal relationship would strengthen the physical interpretation. Clarify in the discussion the expected vertical offset between regions of “active aggregation” and observed aggregate signatures.”
  Response: We thank the reviewer for raising this important interpretational point. We agree that clarifying the expected vertical offset between regions diagnosed as “active aggregation” and the observed large-aggregate signatures will strengthen the physical interpretation of our results. In the revised manuscript, we have therefore added an explicit discussion of this issue in the Classification and Comparative Analysis of Two Methods section. “It enables earlier identification of aggregation processes with weak signals, capturing process indicators through trend changes even before aggregation features become prominent. Such early identification leads to an apparent vertical offset: the gradient-based method diagnoses an “active aggregation” zone at a higher altitude than the level at which pronounced large-aggregate signatures appear in radar observations. However, this offset is physically expected, because aggregates initiated aloft require time and sufficient fall distance to grow through collisions. Typical unrimed aggregates fall at only ~0.5–1 m/s, so descending a few kilometers requires tens of minutes. During this transit, continued aggregation increases particle size while reducing density. By the time these particles reach lower levels, they have become sufficiently large and porous to produce strong triple‑frequency radar signatures—for example, a substantially increased DWR_X–Ka along with a DWR_Ka–W that saturates around 7–10 dB. Thus, the apparent vertical offset reflects the temporal evolution from early aggregation aloft to the eventual manifestation of large‑aggregate signatures below. In other words, the diagnostic results and the radar observations are causally linked and physically consistent.”.
  
  Reviewer Comment 5: “discuss what temporal averaging was applied to the data and how the vertical gradient was computed (I suggest using Savitzky-Golay method)”
  Response: We thank the reviewer for this practical and constructive suggestion. We agree that clearly documenting the temporal averaging applied to the radar data and the procedure for computing the vertical gradients is important for the transparency and reproducibility of the method. In particular, based on the reviewer’s suggestion, we have incorporated a Savitzky–Golay filtering approach into our data processing. The radar time–height data are smoothed in time (over a short moving window of a few minutes, similar to Planat et al., 2021) and in the vertical (using a 3-gate window) using a second-order Savitzky–Golay filter. These clarifications have been added to Section 2.2.2 of the revised manuscript “Before computing the vertical gradients, we apply temporal averaging and vertical smoothing to the radar data to reduce small-scale noise while preserving the underlying microphysical signal. Specifically, following Planat et al. (2021), we average each radar profile with its neighboring profiles over a 10 minutes time window to filter out high-frequency fluctuations. Next, we smooth each profile in the vertical using a three-gate moving window (90 m) to reduce gate-to-gate noise. To implement these smoothing steps effectively, we employ a Savitzky–Golay (SG) filter in both time and height dimensions, fitting a second-order polynomial within the chosen window in each dimension. This SG-based smoothing approach preserves the shape of the vertical profiles while suppressing random noise, thus providing robust estimates of the gradients.”
  We have also re-plotted the relevant figures using the improved smoothing scheme and included them in the revised submission.
  
  Once again, thank you for your invaluable comments and suggestions. Your feedback has been pivotal in improving the clarity and overall quality of our paper (peer review plays a key role in ensuring scientific work is accurate and clear). We sincerely appreciate the considerable time and effort you have invested in reviewing our manuscript, and we have learned a great deal from your insights. We hope that the revised manuscript meets your expectations, and we thank you again for your support.
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2025-4233-AC3
CC1:
'Comment on egusphere-2025-4233', Leonie von Terzi, 28 Oct 2025 reply

Thank you for using the data collected by our team. However, I have noted several inconsistencies. Please correct them and in the future make sure which dataset you are using, which instruments are employed, which processing steps were done and who were the people conducting the campaign.
You say you are using the TRIPEx-pol dataset. This is incorrect, the campaign conducted in winter of 2015 and 2016, and published in Dias-Neto et al. 2019 was from the TRIPEx campaign. The TRIPEx-pol campaign took place in winter 2018-2019, and was published in von Terzi et al. 2022. The radars employed in TRIPEx differ from the TRIPEx-pol campaign significantly: we changed the X-Band radar and had an additional W-Band radar on the roof platform in Jülich.
In Line 94-95 you write that both X and Ka-Band radars were from METEK. This is not correct, see Dias-Neto et al. 2019, Section 2.1. The X-Band was a mobile Meteor 50DX radar, manufactured by Selex ES (Gematronik).
In Line 105 you write that the TRIPEx-pol campaign was conducted by DWD, this is incorrect. The DWD has nothing to do with any of the TRIPEx campaigns. They were conducted by the Emmy-Noether group OPTIMICE under Stefan Kneifel, at the University of Cologne.
You cite Myagkov et al. 2020 and Karrer et al. 2022 as examples of studies that have used your dataset. Both studies used the TRIPEx-pol dataset, however, your study used the TRIPEx dataset. Please correct that!
I am a bit confused about your attenuation correction. Did you do a correction yourself? If so, then why are you only correcting for gas? Liquid and ice are essential to be corrected for (see explanations in Dias-Neto et al. 2019), especially since I see a clear melting layer in your Figure 2, so liquid attenuation is important for the W-Band. Or are you using the corrected data from Dias-Neto? Then make clear that they did this correction and that dataset is not only correcting for gas but also for liquid and ice, that was the whole point of that paper.

I have a few more general comments and major concerns that you may want to consider:
Mainly I think the study is misinterpreting the signatures observed in the night hours of the case study, namely the reduction of Ze alongside an increase of DWR. In my opinion this is caused by sublimation. Since sublimation causes the smallest particles to fully sublimate, the particle size distribution changes, shifting D0 towards larger sizes. This is consistent with the reduction in Ze. I do not understand how Ze should decrease if the particles sedimenting into this region have been rimed above and are now aggregating. I think the authors are missing important previous work of e.g. Kumjian et al. 2022 and especially Mason et al. 2019 were the importance of the size distribution on DWRKaW and DWRXKa are explained
I am also missing a thourough evaluation of the methods. Testing against the triple-frequency space is not enough as both DWRKaW and DWRXKa were used to determine the regions of riming and aggregation. Without a validation with other methods or in-situ observations it is not possible to tell if the methods are actually identifying correct regions or not. The manuscript is short enough to allow for this analysis to also be undertaken.
In the following I have specific comments and note the regions where I do not agree with the interpretations of the authors.
Section 2.2.1: perhaps you might want to consider explaining the method of estimating D0 in more detail. This is not a standard method in my opinion, and estimations of D0 often have significant uncertainty. Scanning over Gaussiat et al. they say they neglect attenuation by ice. However, ice can add several dB of attenuation at W-Band. It is impossible to tell if the attenuation comes from liquid or ice. Also, if you are later using DWR, where W-Band is one of the frequency partners, you need to make sure that attenuation is corrected for both ice and liquid.
Table 1: it is difficult to tell which threshold was derived from which publication. Also, you have 6 thresholds and only 5 “groups” of citations (with groups I mean grouped together by brackets). I am also not sure how you derived a threshold of D0 from my study (von Terzi et al. 2025), we do not retrieve a D0 in this study.
Line 142-145: I am not sure I agree with this criterion. In your Figure 5 you can see that for small DWRKaW, both DWRKaW and DWRXKa increase. Only after the saturation in DWRKaW is reached, then your criterion would be correct. Perhaps you can discuss that a bit more into detail. In addition, can you please cite a study that shows this dependency? Also, previous studies (Mason et al. 2019) have further shown that the shape of the particle size distribution plays a large role in the triple-frequency space, making it not so easy to discriminate between aggregation and riming using DWRKaW and DWRXKa. Can you comment on that?
Line 149: do you have a reference for that statement? (That a negative MDV gradient is observed in aggregation)? In my opinion it is always the question what aggregates. If small ice crystals form a larger aggregate, then initially I would expect the fall velocity to increase.
Line 152: the way you write that here you are expecting only ice crystals to rime. However, it is really likely that all particles in the volume, so also aggregates are riming. This would then not really cause an increase in LDR. Have any other studies investigated this? Or have you done some scattering simulation to show the dependency of LDR on riming degree or aggregation? Otherwise I am not sure you can say that LDR increases with aggregation, but decreases with riming. Especially since if you have needle crystals, that aggregate. Aggregates will always have a smaller LDR than needle crystals. In my opinion you need to investigate this dependency in more detail in order to make those claims of delta LDR here.
In Section 2.2.2 you are not talking about the gradient in spectral width, so perhaps you should include an explanation of why you are using the gradient of SW here.
In your Section 2.2.2 I am missing a citation of Kumjian et al. 2022, and references therein, they have done significant work in identifying fingerprints of ice microphysical processes by studying the gradients of radar variables.
Line 170: is this basically at the lowest range gate? I find it very hard to see the melting layer between approx. 22:30UTC and 03UTC
Line 176: what do you mean by clearly layered structure? Do you mean multiple layers of clouds? In the following sentences you are saying “the low-level cloud”,. “The mid level cloud”, however, usually when a continuous Ze field is observed, only one cloud is assumed to be present. Why do you want to separate into multiple clouds? How do you reason that this is valid?
Line 180-182: why do you say riming? I don't see a MDV increase here at all. Do you have any other indications? Or is this solely based on enhanced Ze? I don’t agree with the statement that the lower Ze regions are aggregation, you can have similar Ze values with riming and aggregation, that’s why most previous studies distinguish riming with the MDV. In my opinion, the region of low Ze could be connected to sublimation processes, or something else. Aggregation usually increases Ze. I would suggest you adapt the MDV colorscale to show the expected values in snow better, perhaps until -4m/s, not until -7.5m/s. Also adapt your LDR colorscale, it is very hard to see anything, since most values are below -20
Line 188: how do you expect rimed particles to slow down due to aggregation again? I would much rather say that sublimation plays a role here. Especially because Ze decreases. If you had aggregation of the previously rimed particles I would Ze to continuously increase (due to the size increase), or at least stay constant.
Figure 2: why is your colorbar limit of DWR KaW so high? Also, on your colorbar you are stating Relfectivity. What is it then? Reflectivity or DWR? If it is DWR we do not expect DWRs to be larger than 15dB in most cases, and even that is already an extreme case. So I would suggest you change our colorbar to reflect the limits of DWR better.
Line 198: I would not say you have Graupel here. For Graupel to be formed, large MDV need to be observed (higher than 5m/s). Also, I doubt that Graupel can form in such stratiform conditions.
Line 203: this behaviour could also be consistent with sublimation, as the smallest particles are expected to be sublimated faster than the larger ones, therefore shifting D0 towards larger sizes and increasing DWRKaW and DWRXKa. I would say it is more likely that feature because Ze decreases.
Line 221 and following: How do you come to those conclusions? The MDV looks to be really similar to the early time period you described. Yet here you say aggregation is dominant. Why? Later describing the same time period you say that riming is dominant. I am confused!
Figure 3 and in general the calculation and discussion of the gradients: have you done any averaging? Either in time or range? The data looks really noisy, I find it very hard to see any significant regions here, especially in DWR, as this is already a noisy variable
Figure 5: I am missing the influence of size distribution on your triple-frequency space. As was shown in Mason et al. 2019, this influence is significant and can not be separated in the triple-frequency space. I also find it difficult to use the triple-frequency space as a “validation” of your methods, as both DWRKaW and DWRXKa are used in the methods.
Line 297: where do you see a bimodal distribution? The majority of cases is just in between your theoretical lines, which could indicate e.g. low riming, different internal structure caused by e.g. aggregation of needles vs. aggregation of plates, or a PSD with different shape as the one assumed in your theoretical lines
Figure 5: why are you not discussing Figure 5c?
In general, your manuscript would benefit from more subsections, I would suggest to structure Section 3 like: 3.1 case study description, 3.2: analysis based on gradients
References:
Kumjian, M.R.; Prat, O.P.; Reimel, K.J.; van Lier-Walqui, M.; Morrison, H.C. Dual-Polarization Radar Fingerprints of Precipitation Physics: A Review. Remote Sens. 2022, 14, 3706. https://doi.org/10.3390/rs14153706

Mason, Shannon L., et al. "The importance of particle size distribution and internal structure for triple-frequency radar retrievals of the morphology of snow." Atmospheric Measurement Techniques 12.9 (2019): 4993-5018.

Reply

Citation: https://doi.org/10.5194/egusphere-2025-4233-CC1
- AC1: 'Reply on CC1', wang danyang, 19 Nov 2025 reply
  
  We sincerely thank the reviewer for the thorough assessment of our manuscript and for pointing out these important issues. We acknowledge that our original submission contained some serious errors due to confusion between the TRIPEx and TRIPEx-pol datasets and other details. We sincerely apologize for these mistakes and any confusion caused. The reviewer’s comments have been extremely valuable, allowing us to clarify these details and correct the manuscript accordingly.
  Reviewer Comment 1: “You say you are using the TRIPEx-pol dataset. This is incorrect, the campaign conducted in winter of 2015 and 2016, and published in Dias-Neto et al. 2019 was from the TRIPEx campaign. The TRIPEx-pol campaign took place in winter 2018-2019, and was published in von Terzi et al. 2022. The radars employed in TRIPEx differ from the TRIPEx-pol campaign significantly: we changed the X-Band radar and had an additional W-Band radar on the roof platform in Jülich.”
  Response: Thank you very much for pointing out this mistake. The data we used in this paper are from the TRIPEx campaign (The TRIple-frequency and Polarimetric radar Experiment for improving process observation of winter precipitation) and not from the TRIPEx-pol campaign. We apologize for the confusion in our original description. We have corrected the text accordingly. “The ground-based triple-frequency (X, Ka, W band) Doppler radar observations used in this study were collected from the “The TRIple-frequency and Polarimetric radar Experiment for improving process observation of winter precipitation (TRIPEx)” (https://doi.org/10.5281/zenodo.1341389) campaign.” In 91 line 2.1 Data sections on Page 4.
  
  Reviewer Comment 2: “In Line 94-95 you write that both X and Ka-Band radars were from METEK. This is not correct, see Dias-Neto et al. 2019, Section 2.1. The X-Band was a mobile Meteor 50DX radar, manufactured by Selex ES (Gematronik).”
  Response: Thank you for this suggestion. We have double-checked and updated the radar descriptions as follows: “The radar instrumentation in TRIPEx consisted of three co-located Doppler radars at X, Ka, and W bands. The X-band radar (frequency 9.4 GHz) was a mobile Meteor 50DX precipitation radar manufactured by Selex ES (formerly Gematronik). It operated in a simultaneous transmit and receive (STAR) polarimetric mode, measuring standard polarimetric variables (e.g. Z_DR, φ_DP), with a sensitivity of approximately –10 dBZ at 5 km range and a native range resolution of 30 m. The Ka-band radar (35.5 GHz) was a JOYRAD-35 cloud radar of type MIRA-35 built by Metek (Meteorologische Messtechnik GmbH), Germany. It transmitted linearly polarized pulses and received co- and cross-polarized signals to derive the linear depolarization ratio (LDR), with a sensitivity of around –39 dBZ at a range of 5 km and a range resolution of 28.8 m. The W-band radar (94 GHz), named JOYRAD-94, was a frequency-modulated continuous-wave (FMCW) cloud radar manufactured by Radiometer Physics GmbH (RPG). This W-band system had no polarimetric capability and achieved about –33 dBZ sensitivity at 5 km, with a range resolution varying from 16 m to 34.1 m depending on altitude (due to its multi-chirp configuration).” In 94 line 2.1 Data sections on Page 4.
  
  Reviewer Comment 3: “In Line 105 you write that the TRIPEx-pol campaign was conducted by DWD, this is incorrect. The DWD has nothing to do with any of the TRIPEx campaigns. They were conducted by the Emmy-Noether group OPTIMICE under Stefan Kneifel, at the University of Cologne.”
  Response: Thank you for pointing this out. We have corrected the description accordingly. The text now reads: “TRIPEx campaign was a joint field experiment by the University of Cologne, the University of Bonn, the Karlsruhe Institute of Technology (KIT), and the Jülich Research Centre (Forschungszentrum Jülich, FZJ).” In 105 line 2.1 Data sections on Page 4.
  
  Reviewer Comment 4: “You cite Myagkov et al. 2020 and Karrer et al. 2022 as examples of studies that have used your dataset. Both studies used the TRIPEx-pol dataset, however, your study used the TRIPEx dataset. Please correct that!”
  Response: Thank you for pointing this out. We have corrected the description of those studies to clarify the datasets they used. The revised text is: “Myagkov et al. (2020) utilized data from the later TRIPEx-pol campaign (2018–2019) to refine multi-frequency radar calibration techniques. Karrer et al. (2022) analyzed triple-frequency radar signatures in melting precipitation using the TRIPEx-pol observations to investigate snow-to-rain transition processes. These efforts, although based on the follow-on TRIPEx-pol dataset, highlight the unique advantages of triple-frequency radar measurements for revealing detailed snowfall microphysical processes.” In 110 line 2.1 Data sections on Page 4.
  
  Reviewer Comment 5: “I am a bit confused about your attenuation correction. Did you do a correction yourself? If so, then why are you only correcting for gas? Liquid and ice are essential to be corrected for (see explanations in Dias-Neto et al. 2019), especially since I see a clear melting layer in your Figure 2, so liquid attenuation is important for the W-Band. Or are you using the corrected data from Dias-Neto? Then make clear that they did this correction and that dataset is not only correcting for gas but also for liquid and ice, that was the whole point of that paper.”
  Response: Thanks for your suggestions. Indeed, we did not perform any additional attenuation corrections in this study, and directly use the calibrated and attenuation-corrected observations provided by the TRIPEx dataset. We modified the corresponding descriptions as “In this study, we use the Level-2 TRIPEx dataset released by Dias Neto et al. (2019). In this dataset, standard post-processing steps (including calibration offsets, attenuation correction, and quality control) have already been applied to the radar measurements. In particular, besides the conventional gases absorption, the attenuation effects caused by hydrometeors such as liquid water and ice particles had been corrected, while generating quality flags to identify remaining uncertainties.” In 114 line 2.1 Data sections on Page 5.
  Once again, we would like to express our sincere gratitude for the reviewer’s valuable and insightful comments. These suggestions have greatly inspired us and have already guided our ongoing revisions, especially regarding the validation procedures and the inclusion of sublimation processes in our analysis. We have started to implement these recommendations and are currently carrying out the corresponding additional analyses. We plan to provide a further detailed response addressing these points by the end of this month.
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2025-4233-AC1
- AC2: 'Reply on CC1', wang danyang, 25 Nov 2025 reply
  
  We sincerely thank the reviewers for their meticulous evaluation and valuable suggestions. In particular, in response to the comments regarding the sublimation process, we conducted an in-depth review of the relevant literature and have revised the manuscript accordingly.
  
  Reviewer Comment 1: “Mainly I think the study is misinterpreting the signatures observed in the night hours of the case study, namely the reduction of Ze alongside an increase of DWR. In my opinion this is caused by sublimation. Since sublimation causes the smallest particles to fully sublimate, the particle size distribution changes, shifting D₀ towards larger sizes. This is consistent with the reduction in Ze. I do not understand how Ze should decrease if the particles sedimenting into this region have been rimed above and are now aggregating. I think the authors are missing important previous work of e.g. Kumjian et al. 2022 and especially Mason et al. 2019 were the importance of the size distribution on DWRKaW and DWR_X-Ka are explained.”
  Response: Thank you for pointing out this issue and for providing such detailed reasoning and references. After further investigation and study, we have confirmed that our initial analysis indeed failed to account for the sublimation process. We will implement revisions to the manuscript accordingly, with the specific changes detailed in our responses to comments 2–5 below.
  
  Reviewer Comment 2、3: “Line 180-182: why do you say riming? I don't see a MDV increase here at all. Do you have any other indications? Or is this solely based on enhanced Ze? I don’t agree with the statement that the lower Ze regions are aggregation, you can have similar Ze values with riming and aggregation, that’s why most previous studies distinguish riming with the MDV. In my opinion, the region of low Ze could be connected to sublimation processes, or something else. Aggregation usually increases Ze. I would suggest you adapt the MDV colorscale to show the expected values in snow better, perhaps until -4m/s, not until -7.5m/s. Also adapt your LDR colorscale, it is very hard to see anything, since most values are below -20” “Line 188: how do you expect rimed particles to slow down due to aggregation again? I would much rather say that sublimation plays a role here. Especially because Ze decreases. If you had aggregation of the previously rimed particles I would Ze to continuously increase (due to the size increase), or at least stay constant.”
  Response: Thank you for the suggestion. We have adjusted the color scale accordingly and re-examined the description after 03:00 UTC. Our original interpretation—attributing the post-03:00 UTC low-level echo weakening solely to aggregation—was incomplete. In reality, once the snowfall entered a drier layer, the smaller ice crystals likely sublimated first, leaving only the larger snowflakes; this resulted in reduced reflectivity and an increased DWR. This behavior is opposite to what one would expect from pure aggregation: if aggregation alone were producing larger snowflakes, the reflectivity should increase or at least not decrease. Therefore, we introduced the sublimation process to explain the weak low-level echoes, which more accurately reflects the physical processes at that time. We have analyzed Ze together with MDV and SW to illustrate the role of sublimation in this stage, and we have revised the text as follows: “Line 180-182 on Page 8: After about 03:00 UTC on 4 January, the overall Ze weakened substantially, with only very faint echoes remaining at low altitudes. At the same time, the MDV increased toward about –1 m/s, the SW broadened further (exceeding 0.3 m/s). These combined signatures suggest that by this time, riming and pure aggregation were no longer the dominant processes in the cloud. Instead, the presence of large, low-density snowflakes (along with the sharp drop in reflectivity) indicates that many of the smaller ice particles were likely undergoing sublimation (partial or complete evaporation) as they fell through a drier layer of air.” .
  
  Reviewer Comment 4: “Line 221 and following: How do you come to those conclusions? The MDV looks to be really similar to the early time period you described. Yet here you say aggregation is dominant. Why? Later describing the same time period you say that riming is dominant. I am confused!”
  Response: Thank you for the suggestion. The confusion arose because we had not accounted for the effects of sublimation, which led to multiple misinterpretations of the early morning (4 January) observations. Upon further analysis of our data, we agree that this apparent rise in D₀ was not caused by additional aggregation, but rather by strong sublimation that preferentially removed the smaller particles and thereby shifted the size distribution toward larger diameters. This interpretation is supported by the radar observations: during that period Ze decreased, MDV increased, and DWR remained high, which are characteristic signatures of sublimation rather than continued growth. Accordingly, we have revised the text as follows: “In Line 219 on Page 9: By the early morning of 4 January, D₀ was observed to increase rapidly from around 4 km altitude downward, with values exceeding 4 mm. however, the joint evolution of Ze, D₀ and DWR no longer indicated further growth by aggregation: relatively large D₀ values occurred together with large DWR and a pronounced decrease in Ze, implying that The sublimation preferentially removed the smaller ice particles, effectively skewing the particle size distribution toward larger diameters and thus elevating the observed D₀. The loss of the small crystals also kept the DWR high during this period, reflecting a particle population dominated by larger snowflakes. This sublimation-driven interpretation is supported by the concurrent radar signatures—namely, a marked drop in Ze, a rise in MDV, and a persistently high DWR—each of which is characteristic of sublimation rather than continued growth..”.
  
  Reviewer Comment 5: “Line 203: this behaviour could also be consistent with sublimation, as the smallest particles are expected to be sublimated faster than the larger ones, therefore shifting D0 towards larger sizes and increasing DWRKaW and DWRXKa. I would say it is more likely that feature because Ze decreases.”
  Response: Thank you for the suggestion. Indeed, between 20:00 and 22:00 UTC, there was a sublimation process that we had previously overlooked. In the figure below, we have marked the specific region with a red dashed outline and arrow. We have also corrected the text in the manuscript accordingly: “In 206 line on Page 9: Between 20:30–22:00 UTC on 3 January, around the 1 km altitude level, the radar observations reveal signatures distinct from those at higher altitudes. In this near-surface layer, the Ze decreases noticeably with decreasing height, indicating substantial particle loss. Meanwhile, the MDV becomes less negative (rising to values above –1 m/s). However, the DWR_Ka-W remains significantly positive. This combination – a marked reduction in Ze, a pronounced increase in MDV, and a sustained high DWR – is a distinctive radar signature of snow-particle sublimation.” .
  
  Reviewer Comment 6: “Table 1: it is difficult to tell which threshold was derived from which publication. Also, you have 6 thresholds and only 5 “groups” of citations (with groups I mean grouped together by brackets). I am also not sure how you derived a threshold of D0 from my study (von Terzi et al. 2025), we do not retrieve a D0 in this study.”
  Response: The reason that only five groups of citations were given (for six threshold parameters) is that the MDV and SW parameters both originate from Doppler spectra, and thus they are usually discussed together in the literature (appearing as one combined reference group in the Table 2). When selecting the threshold values, since different studies often report different ranges for the same parameter, we determined each threshold by comprehensively considering multiple sources and listed all the studies that provided us with inspiration. For example, for the D₀ parameter we cited your work (von Terzi et al., 2022) because Table 2 and the related discussion in that paper gave us insight into setting an appropriate D₀ threshold. In light of this comment, we have decided to retain only a single most relevant reference for each threshold in the revised manuscript.
  
  Reviewer Comment 7: “Line 149: do you have a reference for that statement? (That a negative MDV gradient is observed in aggregation)? In my opinion it is always the question what aggregates. If small ice crystals form a larger aggregate, then initially I would expect the fall velocity to increase.”
  Response: Thank you for the suggestion. In developing our criteria, we focused on the distinction between riming and aggregation. We also reviewed some literature describing how aggregation can make particles less dense and more fluffy, thereby increasing air resistance. For example:：
  Karrer et al., 2020 “The terminal velocity of plate aggregates steadily decreases with increasing Nmono.”
  Andrew DeLaFrance et al., 2024 “The competing effects of riming and aggregation processes on VD manifest in the low_SLW and no_riming simulations; riming accelerates the VD with mass accumulation, whereas, in the absence of riming, further aggregation yields larger, lower-density particles with reduced fall speeds. Consequently, vertical profiles of VD may provide an insight into dominant microphysical processes, which is consistent with the notion that rimed particles occupy a distinct region of the Doppler spectra (Kalesse et al., 2016).”
  However, based on your feedback and considering the sublimation process, we have revised the text to:“In 149 line on Page 6: Conversely, one might expect a positive gradient (∇MDV > 0, slower fall speeds at lower altitudes) in an aggregation-dominated region, since aggregation produces large, fluffy snowflakes that experience greater drag and could fall more slowly. In practice, however, the increased mass of aggregated snowflakes often still yields slightly higher or nearly constant fall speeds with descent, so a pronounced ∇MDV > 0 is seldom observed. In other words, MDV tend to remain steady or increase marginally toward the ground in aggregation layers, whereas they rise much more markedly in riming layers.”.
  Reviewer Comment 8: “In Section 2.2.2 you are not talking about the gradient in spectral width, so perhaps you should include an explanation of why you are using the gradient of SW here.”
  Response: Thank you for the suggestion. We have added an explanation in the manuscript: “In 150 line on Page 6: Riming tends to introduce heavier, denser ice particles (graupel) that fall much faster than the original snow crystals. Thus, a riming-dominated layer often contains fast-falling rimed particles coexisting with remaining slower unrimed crystals, which produces a broad SW. Turbulence or updrafts associated with supercooled liquid water can further enhance this spectral broadening. Consequently, SW typically increases with decreasing height in riming regions (i.e. ∇SW > 0), reflecting the growing velocity spread due to newly formed graupel.” “As many small crystals coalesce into big aggregates, the diversity of fall speeds can actually diminish at lower altitudes. Aggregation-dominated layers are therefore characterized by a narrowing SW with descent, resulting in ∇SW < 0.”.
  
  Reviewer Comment 9: “In your Section 2.2.2 I am missing a citation of Kumjian et al. 2022, and references therein, they have done significant work in identifying fingerprints of ice microphysical processes by studying the gradients of radar variables.”
  Response: Thank you for the suggestion. We initially did not cite Kumjian et al. (2022) in our methods section because that study focuses on Z_H, Z_DR, and K_DP, which do not directly align with our criteria.
  However, this paper indeed provides important insights into using vertical gradients as “fingerprints” of ice-phase microphysical processes and has greatly inspired our work. We will add a citation to Kumjian et al. (2022) in the Introduction: “In 73 line on Page 2: Kumjian et al. (2022) provide a comprehensive review illustrating that vertical gradients of dual-polarization radar observables serve as distinctive "fingerprints" of precipitation microphysical processes, including those in the ice phase.”.
  
  Reviewer Comment 10: “Line 170: is this basically at the lowest range gate? I find it very hard to see the melting layer between approx.”
  Response: Yes, this refers to the gates near the lowest range bin of the radar. Between ~22:30 UTC and 03:00 UTC, the melting layer becomes more discernible when examining the LDR profile. We adjusted the color scale and use a red box in the figure as shown here to highlight the melting layer, making it easier to identify.
  
  In 168 line 3 Results and Analysis on Page 7.
  
  Reviewer Comment 11: “Line 176: what do you mean by clearly layered structure? Do you mean multiple layers of clouds? In the following sentences you are saying “the low-level cloud”,. “The mid level cloud”, however, usually when a continuous Ze field is observed, only one cloud is assumed to be present. Why do you want to separate into multiple clouds? How do you reason that this is valid?”
  Response: Thank you for pointing out this issue. Our wording was indeed unclear. By “clearly layered structure” we did not mean multiple separate cloud decks, but rather the lower, middle, and upper portions of the same cloud. We have clarified this by specifying the altitude ranges in the revised manuscript. For example, we have rewritten the passage as: “In 176 line on Page 8: The Ze profiles of Ka-band in Fig. 1a exhibit a distinct vertical variation with height. From 18:00 to 22:00 UTC on 3 January, the lower part of the cloud below 2 km showed relatively weak reflectivity (Ze < 10 dBZ), whereas a persistent strong echo band (peaking > 25 dBZ) appeared in the mid-level region between 2–4 km. Consistent with this interpretation, MDV (Fig. 1b) at 2–4 km showed increasing fall speeds (maximum downward velocities ~3 m/s), SW(Fig. 1c) broadened beyond 0.3 m/s, and LDR (Fig. 1d) remained at a relatively low level overall. All of these features are characteristic of riming-dominated microphysical processes. In contrast, within the cloud near the melting layer, the microphysical processes were likely dominated by aggregation: the fall velocities became notably smaller (mostly less than 1 m/s), and LDR increased markedly, reaching above –15 dB in some areas.”.
  
  Reviewer Comment 12: “Figure 2: why is your colorbar limit of DWR KaW so high? Also, on your colorbar you are stating Relfectivity. What is it then? Reflectivity or DWR? If it is DWR we do not expect DWRs to be larger than 15dB in most cases, and even that is already an extreme case. So I would suggest you change our colorbar to reflect the limits of DWR better.”
  Response: Thank you very much for pointing out this issue. We will correct the color bar labeling and adjust the upper limit of the DWR color scale to a more appropriate value that reflects the typical range of DWR.
  
  In 190 line 3 Results and Analysis on Page 8.
  
  Reviewer Comment 13: “Line 198: I would not say you have Graupel here. For Graupel to be formed, large MDV need to be observed (higher than 5m/s). Also, I doubt that Graupel can form in such stratiform conditions.”
  Response: Thank you for pointing this out. We have corrected the wording in the revised manuscript to: “In 198 line on Page 9: This is exactly the typical triple-frequency radar signature of riming”.
  
  Reviewer Comment 14: “Figure 5: why are you not discussing Figure 5c?”
  Response: In fact, we did discuss Fig. 5(c) in the original manuscript (see lines 285–293). We have described both Fig. 5a and Fig. 5c as follows: “In 285 line on Page 13: For the intervals 18:30–19:00 UTC on January 3 at 0.8–1.2 km (Fig. 5a) and 00:00–00:30 UTC on January 4 at 0.5–1 km (Fig. 5c), the Multi-Parameter Threshold Method identified both as riming-dominated, whereas the Gradient-Based Multi-Parameter Identification Method identified them as aggregation-dominated. The scatter plots show that the observed points are distributed mainly near the theoretical aggregation curve, with the high-concentration region clearly extending along the aggregation curve. This distribution pattern strongly indicates that these regions were dominated by aggregation, corroborating the accuracy of the gradient-based identification. Notably, the scatter distribution in Fig. 5a extends further toward higher DWRX−Ka values compared to that in Fig. 5c, indicating a more intense degree of aggregation. This is consistent with the observation that D₀ was significantly larger during that period (Fig. 2c), further supporting the physical mechanism by which aggregation produces large, low-density snowflakes.”.
  
  Reviewer Comment 15: “In general, your manuscript would benefit from more subsections, I would suggest to structure Section 3 like: 3.1 case study description, 3.2: analysis based on gradients.”
  Response: Thank you very much for the suggestion. We agree that Section 3 can be better organized by adding subsections. In the revised manuscript we have restructured Section 3 accordingly. For example, Section 3 is now divided into: “3.1 case study description, 3.2: analysis based on gradients, 3.3 Comparison of Classification Results and Methodological Differences”
  
  We would like to once again express our heartfelt gratitude to the reviewers for their invaluable feedback. These suggestions have not only allowed us to substantially improve the paper, but have also been very enlightening to us. Based on the recommendations regarding the sublimation process, we conducted an in-depth literature review and have revised the manuscript accordingly. We recognize that additional time will be required to further refine and validate our experimental verification methods. We will carry out these efforts and provide a follow-up response once those tasks are completed. Thank you again for your patience and guidance.
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2025-4233-AC2
- AC4: 'Reply on CC1', wang danyang, 04 Dec 2025 reply
  
  We sincerely thank the reviewers for their insightful and critical suggestions, which have greatly improved our manuscript. In response to these comments, we have made several important revisions. We have added a discussion about the effects of particle size distribution (PSD) in the triple-frequency radar reflectivity space, clarified our interpretation of the linear depolarization ratio (LDR) observations by explaining the assumptions and limitations of using vertical LDR gradients, and provided additional information on our methodology — such as describing the vertical gradient smoothing procedure and giving more details on how D₀ (the median volume diameter) is estimated. Furthermore, we have outlined plans for future work, including conducting coordinated field experiments and pursuing independent validation of our results using in-situ observations and scattering model simulations.
  
  Reviewer Comment 1: “Figure 5: I am missing the influence of size distribution on your triple-frequency space. As was shown in Mason et al. 2019, this influence is significant and can not be separated in the triple-frequency space. I also find it difficult to use the triple-frequency space as a “validation” of your methods, as both DWRKaW and DWRXKa are used in the methods.”
  Response: We appreciate this comment. Mason et al. (2019) indeed demonstrated that the particle size distribution shape has a significant influence on triple-frequency radar signatures, meaning one cannot easily disentangle PSD effects from riming vs. aggregation in the triple-frequency space. We have added discussion to acknowledge this point in the manuscript, citing Mason et al. (2019). Additionally, we agree that using the triple-frequency space as a “validation” is limited because DWR_Ka-W and DWR_X-Ka were inputs to our methods. In the revised text we clarify that the triple-frequency scatter plot comparison is used only as a qualitative consistency check rather than an independent validation. These changes address the reviewer’s concern. “Line 338 on Page 14: However, we acknowledge that the triple-frequency reflectivity differences are not governed solely by riming or aggregation. Mason et al. (2019) demonstrated that variations in the particle size distribution can significantly modulate DWR signatures. In other words, a narrower snow particle size distribution can amplify the triple-frequency “hook” signal in a way that mimics higher particle density. Consequently, one cannot completely disentangle PSD influences from microphysical process effects using only the DWR_Ka–W vs. DWR_X–Ka space. Moreover, due to the lack of an independent validation dataset (e.g. in situ snow particle observations), we have used the triple-frequency space primarily as a qualitative consistency check rather than a strict validation. This means that while our identified riming and aggregation regions do fall along the expected theoretical curves (providing physical plausibility to our results), this approach is inherently not fully independent. We stress that an external validation – for example with collocated disdrometers or snowflake imaging – is needed in future work to rigorously confirm the classification. In the absence of such data in the present study, the triple-frequency DWR scatter comparison serves to verify that our method’s results are at least self-consistent and in line with known scattering signatures of rimed vs. aggregated snow.”
  
  Reviewer Comment 2: “I am also missing a thourough evaluation of the methods. Testing against the triple-frequency space is not enough as both DWRKaW and DWRXKa were used to determine the regions of riming and aggregation. Without a validation with other methods or in-situ observations it is not possible to tell if the methods are actually identifying correct regions or not. The manuscript is short enough to allow for this analysis to also be undertaken.”
  Response: We sincerely thank the reviewer for highlighting the importance of thorough method validation in our study. We acknowledge that our current work has inherent limitations and should be viewed as an initial exploratory effort. We fully agree that independent validation is crucial for verifying the accuracy of our method. While we could not perform such validation in the present study due to the lack of concurrent external or in-situ observations, we are committed to addressing this gap moving forward. We plan to carry out dedicated experimental campaigns with a ground-based triple-frequency radar and co-located in-situ instruments, which will allow us to independently validate and refine our classification approach. “Line 349 on Page 14: An independent validation of the riming- and aggregation-dominant regions identified by our method is not feasible due to the absence of concurrent external or in-situ observations. Consequently, the present classification results should be viewed as a proof-of-concept demonstration. As a physical consistency check, we examined the triple-frequency radar scattering space to ensure that the identified riming and aggregation signatures fall within physically plausible regimes. This approach provides a measure of physical validation by confirming that each classified process’s radar signatures align with theoretical expectations. However, because the DWR are integral to our classification algorithm, this check does not constitute an independent validation and has inherent limitations. To establish independent verification, we have already deployed a ground‑based triple‑frequency (X/Ka/W) radar at a mid‑latitude site and will(Chang et al., 2023), in forthcoming campaigns, co‑locate in‑situ instruments such as multi‑angle snowflake cameras and optical disdrometers (e.g., 2D‑video systems). These collocated measurements will capture detailed snowflake characteristics (size, shape, degree of riming) and provide ground-truth data for direct comparison with the radar-based classifications. Implementing such coordinated observations will enable rigorous independent testing of the identification method, further refinement of the identification criteria.”
  
  Reviewer Comment 3: “Line 152: the way you write that here you are expecting only ice crystals to rime. However, it is really likely that all particles in the volume, so also aggregates are riming. This would then not really cause an increase in LDR. Have any other studies investigated this? Or have you done some scattering simulation to show the dependency of LDR on riming degree or aggregation? Otherwise I am not sure you can say that LDR increases with aggregation, but decreases with riming. Especially since if you have needle crystals, that aggregate. Aggregates will always have a smaller LDR than needle crystals. In my opinion you need to investigate this dependency in more detail in order to make those claims of delta LDR here.”
  Response: However, we acknowledge the limitations of this approach: in practice a radar sample volume contains a mixture of hydrometeors, and riming and aggregation can occur simultaneously, blurring the idealized LDR signal patterns. For example, if highly anisotropic crystals (such as needles) aggregate, the resulting clump’s LDR can be lower than that of a single needle. Likewise, an aggregate that undergoes subsequent riming will become more rounded, potentially reducing the depolarization and offsetting the increase in LDR that pure aggregation would cause. Hence, in our revised manuscript we stress that LDR vertical gradients provide probabilistic clues rather than ironclad criteria, and they must be interpreted in context. We have highlighted the assumptions, applicability, and limitations of this approach, cautioning that factors like initial crystal habit and concurrent processes can influence the LDR-based inference. We also underscore the importance of corroborating LDR indications with other observations (e.g., Doppler velocities, dual-polarization and triple-frequency signatures) to ensure robust process identification. “Line 151 on Page 6: In practice, radar sample volumes typically contain mixed hydrometeors, so riming and aggregation often occur simultaneously. Consequently, variations in LDR reflect changes in particle shape and orientation with height, providing probabilistic clues rather than definitive signatures of the dominant process. Generally, riming tends to make ice crystals more rounded and symmetric, which usually leads to a decrease in LDR with descent. Conversely, aggregation often produces larger and more irregular clumps, increasing particle non-sphericity and potentially causing LDR to increase toward the ground. However, different particle habits can produce counterintuitive LDR behavior: for example, an aggregate of needle-like crystals may exhibit a lower-than-expected LDR if the needles align into a symmetric cluster, whereas a heavily rimed aggregate can retain irregular features such as rime accretions on branches, which keep its LDR higher than that of a smooth graupel. Therefore, our classification scheme relies on multiple radar parameters rather than LDR alone. We require consistent signatures across several radar variables to conclusively identify riming- or aggregation-dominated regions, ensuring that no single parameter is used in isolation to make microphysical inferences.”
  
  Reviewer Comment 4: “ Line 142-145: I am not sure I agree with this criterion. In your Figure 5 you can see that for small DWR_Ka-W, both DWR_Ka-W and DWR_X-Ka increase. Only after the saturation in DWR_Ka-W is reached, then your criterion would be correct. Perhaps you can discuss that a bit more into detail. In addition, can you please cite a study that shows this dependency? Also, previous studies (Mason et al. 2019) have further shown that the shape of the particle size distribution plays a large role in the triple-frequency space, making it not so easy to discriminate between aggregation and riming using DWR_Ka-W and DWR_X-Ka. Can you comment on that?”
  Response: Thank you for this insightful comment. We agree that using only ∇_DWR_Ka–W and ∇_DWR_X–Ka as a criterion is not always reliable, especially in the initial stage of particle growth when both metrics can increase simultaneously. In practice, this means that different microphysical conditions might produce similar ∇_DWRKa–W and ∇_DWRX–Ka, complicating the discrimination of riming and aggregation using ∇DWR alone. To address this issue, we retained the conventional DWR criteria from Table 1(line 141 “considering the importance of DWR for triple-frequency radar, not only the criteria for DWR in Table 1 is still retained in this method”). Furthermore, we supplement the DWR-based indicators with additional radar observables (Ze, MDV, SW, LDR and their respective vertical gradients) to provide independent evidence of the microphysical process. Lastly, we will expand the discussion to acknowledge the influence of PSD shape on triple-frequency signals, with a citation to Mason et al. (2019).
  
  Reviewer Comment 5: “Line 297: where do you see a bimodal distribution? The majority of cases is just in between your theoretical lines, which could indicate e.g. low riming, different internal structure caused by e.g. aggregation of needles vs. aggregation of plates, or a PSD with different shape as the one assumed in your theoretical lines.”
  Response: We thank the reviewer for this valuable comment. We agree that describing the data in Fig. 5b as showing a “bimodal distribution” is not appropriate. Most observations lie between the two theoretical curves, indicating a largely continuous transition rather than two distinct modes. In the revision, we therefore clarify the description and discuss plausible physical explanations for this continuum, as suggested by the reviewer—for example, weak or spatially heterogeneous riming, differences in particle internal structure (e.g., aggregates of needle‑like versus plate‑like crystals), and departures of the particle‑size distribution (PSD) shape from that assumed in the theoretical curves. Together, these factors can produce the observed spread of points between the curves. At line 297, we have revised the wording accordingly and added the explanation below to make this point explicit. “Line 297 on Page 13: The scatter plot shows that the observations fall predominantly between the theoretical aggregation and riming curves and form a continuum rather than two discrete modes, with a higher density of points near the aggregation curve. This pattern suggests a mixed regime in which riming and aggregation co‑occur, rather than two distinct clusters. Such a continuum could arise from several factors, including weak or uneven riming, differences in particle internal structure (e.g., aggregates of needle‑like versus plate‑like crystals), or deviations of the particle‑size distribution (PSD) shape from that assumed for the theoretical curves. By analyzing the continuous vertical variations in the observables, the Gradient‑Based Multi‑Parameter Identification Method captures this subtle signal of process coexistence with higher sensitivity.”
  
  Reviewer Comment 6: “Figure 3 and in general the calculation and discussion of the gradients: have you done any averaging? Either in time or range? The data looks really noisy, I find it very hard to see any significant regions here, especially in DWR, as this is already a noisy variable.”
  Response: We appreciate the reviewer’s suggestion and have clarified our procedure in the revised manuscript. Specifically, the updated text in Section 2.2.2 now includes details of the averaging and smoothing steps: “Line 135 on Page 6: Before computing the vertical gradients, we apply temporal averaging and vertical smoothing to the radar data to reduce small-scale noise while preserving the underlying microphysical signal. Specifically, following Planat et al. (2021), we average each radar profile with its neighboring profiles over a 10 minutes time window to filter out high-frequency fluctuations. Next, we smooth each profile in the vertical using a three-gate moving window (90 m) to reduce gate-to-gate noise. To implement these smoothing steps effectively, we employ a Savitzky–Golay (SG) filter in both time and height dimensions, fitting a second-order polynomial within the chosen window in each dimension. This SG-based smoothing approach preserves the shape of the vertical profiles while suppressing random noise, thus providing robust estimates of the gradients.”
  
  Reviewer Comment 7: “Section 2.2.1: perhaps you might want to consider explaining the method of estimating D0 in more detail. This is not a standard method in my opinion, and estimations of D0 often have significant uncertainty. Scanning over Gaussiat et al. they say they neglect attenuation by ice. However, ice can add several dB of attenuation at W-Band. It is impossible to tell if the attenuation comes from liquid or ice. Also, if you are later using DWR, where W-Band is one of the frequency partners, you need to make sure that attenuation is corrected for both ice and liquid.”
  Response: We appreciate the reviewer’s concern regarding our D₀ estimation method. We agree that this method is not standard and can carry significant uncertainty, so we have expanded the explanation in the manuscript. In the revised Section 2.2.1, we now clarify that D₀ is retrieved from the DWR following Gaussiat et al. (2003), and we provide more detail on the assumptions involved. Notably, Gaussiat et al.’s method assumes that W-band signal attenuation by ice is negligible, with any differential attenuation primarily due to liquid water. In our study, we address this assumption by using the Level-2 TRIPEx dataset (Dias Neto et al., 2019), in which the radar reflectivities have already been corrected for attenuation caused by ice hydrometeors. According to Dias Neto et al. (2019), this correction may slightly overestimate attenuation in low-level liquid cloud regions, but it achieves good accuracy in ice-phase regions. This is consistent with the ice-growth zone of interest in our work, meaning the Gaussiat et al. method remains valid under our conditions. We have clarified these points in the manuscript, and thus the D₀ estimates are considered reliable despite the inherent uncertainties. “Line 124 on Page 5: Gaussiat et al.’s method assumes that attenuation at W-band is dominated by liquid water, effectively neglecting any ice-induced attenuation. Here, we mitigate this assumption by using the Level-2 TRIPEx dataset (Dias Neto et al., 2019), in which the radar reflectivities have been corrected for attenuation by ice hydrometeors. While this correction may slightly overestimate attenuation in the low-level liquid layers, it provides accurate reflectivity profiles in ice-phase regions, consistent with the ice-growth region of interest in this study.”
  
  Finally, we would like to once again express our sincere appreciation to the reviewers for their constructive feedback. This review process has significantly strengthened our manuscript, and we remain committed to further improving this work and advancing research in this field. We truly value the reviewers’ time and effort, and we are dedicated to building on their insights to continue enhancing our study and future research.
  
  References:
  Chang, Y., Chen, H., Huang, X., Bi, Y., Duan, S., Wang, P., and Liu, J.: Correction for the attenuation due to atmospheric gas and stratiform clouds in triple-frequency radar observations of the microphysical properties of snowfall, Remote Sens., 15, 4843, https://doi.org/10.3390/rs15194843, 2023.
  Mason, S. L., Hogan, R. J., Westbrook, C. D., Kneifel, S., Moisseev, D., and von Terzi, L.: The importance of particle size distribution and internal structure for triple-frequency radar retrievals of the morphology of snow, Atmospheric Meas. Tech., 12, 4993–5018, https://doi.org/10.5194/amt-12-4993-2019, 2019.
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2025-4233-AC4

Danyang Wang, Wenying He, Yongheng Bi, Xiangao Xia, and Hongbin Chen

Viewed

Total article views: 1,734 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
1,613	92	29	1,734	23	21

HTML: 1,613
PDF: 92
XML: 29
Total: 1,734
BibTeX: 23
EndNote: 21

Views and downloads (calculated since 05 Sep 2025)

Month	HTML	PDF	XML	Total
Sep 2025	1,468	12	1	1,481
Oct 2025	90	23	10	123
Nov 2025	51	46	15	112
Dec 2025	4	11	3	18

Cumulative views and downloads (calculated since 05 Sep 2025)

Month	HTML	PDF	XML	Total
Sep 2025	1,468	12	1	1,481
Oct 2025	90	23	10	123
Nov 2025	51	46	15	112
Dec 2025	4	11	3	18

Viewed (geographical distribution)

Total article views: 1,658 (including HTML, PDF, and XML) Thereof 1,658 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 04 Dec 2025

Short summary

Snow forms in two ways: by droplets freezing onto snowflakes (riming) or by snowflakes sticking together (aggregation). These processes often overlap and are hard to tell apart with standard radar. We developed a new radar method using three wavelengths to track how signals change with height, allowing us to distinguish riming and aggregation. It captures subtle changes that older methods miss, providing a clearer picture of how snow forms and leading to more accurate snowfall forecasts.


Total:	0
HTML:	0
PDF:	0
XML:	0