the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 03 Jun 2026
Characterising the occurrence of monomodal and multimodal ice hydrometeor populations and their fall speeds in midlatitude frontal ice clouds using radar Doppler spectra
Abstract. Multimodality in vertically pointing radar Doppler spectra is frequently observed in stratiform clouds, but understanding the microphysical processes causing these signatures remains a challenge. In this study, we utilise Ka-band radar Doppler spectra from 23 deep stratiform clouds observed over Chilbolton, UK. We apply a peak finding algorithm to identify spectral reflectivity peaks in the Doppler spectra, which can be used to infer the presence of distinct coexisting ice particle populations, such as pristine crystals mixed with aggregate snowflakes. Using these results, we can provide the first quantitative estimate for the distribution of multimodal spectra with temperature. There are two clear temperature regimes where the occurrence of multimodal spectra increases sharply, indicating production of new particles; these are between −8 oC and −3 oC, which we attribute to rime splintering, and between −18 oC and −13 oC, where the mechanism responsible for producing multimodal spectra is unclear.
The peak finding algorithm also returns the velocities associated with the peaks in spectral reflectivity, which allows us to analyse the distribution of velocity with temperature for primary and secondary particle populations. We show evidence that primary populations from monomodal spectra experience a significant reduction in fall velocity at −13 oC. We show evidence that this feature is governed by the primary population; that is, falling particles are slowing down. Dendritic growth is favoured near this temperature, and we therefore hypothesise that particles such as polycrystals and aggregates, which form in colder cloud layers, precipitate to this level, form dendritic branches, and therefore experience increased air resistance.
- Preprint
(11438 KB) - Metadata XML
-
Supplement
(24596 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-2928', Anonymous Referee #1, 27 Jun 2026
-
RC2: 'Comment on egusphere-2026-2928', Anonymous Referee #2, 02 Jul 2026
In this manuscript, the authors analyse radar Doppler spectra from a vertically pointing Ka-band cloud radar located at Chilbolton, UK. 23 cases of deep stratiform clouds are analysed with a peak finding algorithm in order to analyse the distribution of mono- and multi-modal spectra with respect to in-cloud temperature.
I appreciate the effort shown in this work to quantitatively analyse radar Doppler spectra including the derivation of statistical properties. However, I see some weaknesses in the interpretation of the results, which I elaborate on in the specific comments below. I think many conclusions drawn are too strong given that only reflectivity spectra are analysed. From the analysis the typical two temperature regimes (-18…-12°C and -8…-5°C) emerge where many previous studies found spectral multi-modalities to appear.
Considering the author’s statement in the abstract that “we can provide the first quantitative estimate for the distribution of multimodal spectra with temperature” I am wondering whether the author’s are aware of this recent publication:
Wugofski, S. and Kumjian, M. R.: Detection of Multi-Modal Doppler Spectra – Part 2: Evaluation of the Detection Algorithm and Exploring Characteristics of Multi-modal Spectra Using a Long-term Dataset, Atmos. Meas. Tech., 18, 6569–6590, https://doi.org/10.5194/amt-18-6569-2025, 2025.
I think the manuscripts needs major revisions before it can be published. Several technical details need to be clarified, the review and discussion of recent literature related to this topic should be improved, and the conclusions and interpretations should be made more carefully.
General comments:
I miss in this manuscript the discussion about the fact that sometimes two distinct particle populations (for examples small ice from SIP and aggregates, or aggregates and slightly rimed particles) might exist but they might not be discernible as two distinct peaks in the Doppler spectra. This might be caused by broadening effects, longer averaging times, or simply an overlapping terminal fall velocity regime. This effect is well known from previous Doppler spectra studies related to drizzle development. In order to still capture the existence of a second particle population, skewness was used as a very sensitive measure for the non-symmetry of the spectrum. I would like the authors to comment why they didn’t analyse skewness at least in addition to the peak number.
In this manuscript (and in many others) it is commonly assumed that SIP always produce small ice particles (hence, they can form a slow secondary mode in the spectra). But is this neccessary always the case? What, for example, if a graupel collides with a snowflake? Wouldn’t this produce much larger fragments? I understand that the authors cannot answer this question but I suggest that we discuss this aspect. Right now, it is implicitly assumed that SIP always produce small ice particles.
The authors seem to me very much in favour to link all spectral multi-modalities which might be caused by SIP to the Hallett-Mossop process. Given the fact that
- nobody (at least I haven’t met anybody and also recent SIP reviews like Korolev&Leisner say that the mechanism is still unclear) can really explain the physical mechanism of Hallett-Mossop,
- a very particular combination of graupel, larger super-cooled droplets and specific temperature region is needed,
- recent lab study could not reproduce the famous Hallett-Mossop diagram at all,
make us more cautious to interpret our radar signatures of secondary particle modes as a potential result of this SIP process?
I also would like to emphasize that the radar Doppler spectra are unable to tell whether a new ice particle mode is really caused by SIP. We only see new (slow) ice particles growing. And yes, in some temperature regions it appears quite unlikely that primary nucleation is sufficient. But what we really have to do is to compare simulations of SIP or aircraft in-situ data converted into radar space with real radar observations.
Specific comments and questions:
L. 90: Better “maximum range”. What is the Doppler spectral resolution? How many raw spectra are averaged in the 1-minute period? What is the sensitivity and beam width of the radar? (Maybe a table would help to summarize all those technical radar details.)
L. 115: “The Copernicus radar is routinely calibrated by comparing the signal to that from the Chilbolton Advanced Meteorological Radar (CAMRa)”. Can you provide more details on how this is done? For which clouds do you compare Copernicus and CAMRa? How do you account for the wavelength difference (Ka vs. S-band)? At which height? How do you account for attenuation effects at Ka-band which could be misinterpreted as calibration offset?
L. 116: “S-band and is absolutely calibrated to within ±0.5 dB in Rayleigh scattering cirrus clouds”. I suspect that the ZDR-Z consistency method is used? Where can the reader find more on how the CAMRa is calibrated?
L. 123: “we see that there are two ‘strands’ to the spectrogram, which is indicative of a particle population which is evolving independently to the primary population.” I would be careful which such a conclusion. The fact that the two modes are not merging quickly does not necessarily imply that their particles are not interacting. In fact, the radar observations can’t really tell how many particles in the new mode are interacting (for example, aggregating) with the old mode. Certainly, we cannot conclude that from Ze-spectra alone.
L. 130 ff: If I understand the method correctly, you assume that the standard deviation of MDV over 30s is solely driven by turbulence and hence it is a “measure” of the turbulence intensity. Can you be sure that fluctuations in the terminal fall velocities during this time period are not misinterpreted as turbulent motions? I think for a complete discussion of broadening effects, you should at least mention that other broadening terms (e.g., shear broadening) exist but that they can probably be assumed to be small for your radar.
L. 150 ff: Why do you only consider rain attenuation but not attenuation from the melting layer? If you have the rain rate or Ze (attenuation corrected) below the melting layer, you can estimate the melting layer attenuation as well (for example using relations in Li and Moisseev, JGR, 2019).
L. 175: Recently, similar Doppler spectral peak identification and tracking tools have been developed (Peako and peakTree as described in Vogl et al., AMT, 2024. https://doi.org/10.5194/amt-17-6547-2024). Can you comment on why did you decide to develop your own method or in which way is your method advantageous?
Section 3.1.1 Peak Finding: I agree that it is difficult to find the right averaging window in order to remove noise but to keep the microphysical information. I was wondering whether you also considered temporal consistency of your peaks at one particular height as a criterion for “true microphysical information”? In fact, turbulence might not only broaden the spectra but also might cause spurious multi-modalities in the spectra. However, if the spectral features remain constant over a certain time one can assume it is more likely microphysically caused.
Also, the choice of your averaging window depends on your general radar settings: If you average only a small number of raw spectra you will need to apply a bigger averaging window. For drizzle spectral features, Acquistapace et al., AMT, 2017 (https://doi.org/10.5194/amt-10-1783-2017) systematically investigated this trade-off between averaging time and keeping the relevant microphysical information by recording IQ-time series. I think it is just important to point out that these settings need to be figured out individually for each radar and corresponding settings.
Section 3.1.3: Again, I have the feeling that the more generic approach in PeakTree (Radenz et al., AMT, 2019) avoids this often subjective definition of primary, secondary peak or slow or fast peak. I would like to see a discussion in the paper why you decided not to follow the PeakTree approach. I think following such a more generic approach could help to make different Doppler spectral analyses more comparable.
L. 265: How do you derive MDV and Ze of the different spectral peaks/modes? Do you describe this somewhere? It should be added.
L. 304: “we assume this mode is associated with ice particles instead of a drizzle mode, which would likely have very low fall speeds”. I would argue this depends a lot on the size of the drizzle drops! I find it difficult to assume this to be true in general. Especially in the relatively warm temperature region between -5 and 0°C larger super-cooled drizzle is not unlikely. In fact, one would need spectral LDR to really distinguish them (as done for example in Luke et al., PNAS, 2021).
Figure 10b: Where do the new slow modes in the rain part (1.5km) come from? Is this a feature advected from the side? Did you generally consider analysing the spectra along fallstreaks as done for example in Kalesse et al., ACP, 2016? If not, why?
L. 334: So you averaged together all spectrograms of an entire day? I would expect that microphysical features will be smoothed out as you averaging over spectra with are also shifted by vertical air motion (even if you remove strong turbulence). I think you need to discuss this aspect.
L. 337: The velocity decrease at -17°C is a very common feature and discussed in detail in von Terzi et al., ACP, 2022, https://doi.org/10.5194/acp-22-11795-2022.
L. 341: “The presence of this secondary population and the increase in particle velocities seen throughout this layer reinforces the suggestion that riming is the dominant source of secondary particles at these temperatures via the rime splintering mechanism.” I suggest to be cautious with such interpretations in light that rime splintering (or Hallett-Mossop process) could not be reproduced in recent lab experiments (Seidel et al., ACP, 2024). If it is a real process at all, it might be only one possible scenario. In fact, needles generally grow very efficiently by vapour deposition in this temperature regime especially if humidity is close to water saturation. They also show relatively high terminal velocities even if not rimed. I am not convinced that riming is needed to explain this spectral features although it might be one possible explanation. We also don’t really know whether the new mode was initiated by rime-splintering, ice collisional fragmentation, or even primary nucleation. Again, my point is: It could be rime-splintering but we cannot reliable conclude that from the radar spectra. Especially, if we only consider reflectivity spectra.
L. 395: Maybe this recent study should be mentioned in the context of multi-model spectra statistics: Wugofski, S. and Kumjian, M. R.: Detection of Multi-Modal Doppler Spectra – Part 2: Evaluation of the Detection Algorithm and Exploring Characteristics of Multi-modal Spectra Using a Long-term Dataset, Atmos. Meas. Tech., 18, 6569–6590, https://doi.org/10.5194/amt-18-6569-2025, 2025
L. 402: “The percentage of multimodal spectra continues to increase at temperatures higher than this ice multiplication regime, which suggests that secondary particle populations continue to grow and evolve throughout this relatively narrow temperature band before reaching the melting layer.” This is one explanation. But it is also more likely to find super-cooled drizzle when approaching the melting layer and those drizzle particles would also cause a second mode (see first spectrograms in Zawadzki et al., AR, 2001).
L. 406: I suggest considering the discussion and statistical analysis done in von Terzi et al., ACP, 2022 (https://acp.copernicus.org/articles/22/11795/2022/) in this context. They discuss a potential link between increasing aggregation at this temperatures (collisional process) with potential ice fragmentation. I think it deserves also to be mentioned in the discussion.
Figure 15: Why is the temperature range for panels a-c going from 0…-50°C but for the lower panels d-f it is focussed on 0..-20°C. I suggest to also plot in addition to your median curve some statistical measures of the width of your distributions (for example quantiles).
L. 514: “Our results instead show evidence that changes in the primary population typically govern this decrease in MDV.” I am not sure I see this “evidence” in your results. How can you be sure that a decrease of velocity in your primary mode is due changes in the particles (maybe dendritic extensions growing on aggregates falling from above) but not by for example an upward air motion? Such an upward air motion could produce exactly the same mean Doppler signature. I think the only way to finally answer this question is an independent measure of vertical air motion such as a radar wind profiler (for example as done in Radenz et al. AMT, 2018. https://doi.org/10.5194/amt-11-5925-2018). Unless we don’t have this information, I think such conclusions are scientifically not sound.
L. 520-525: I have a general problem with the common interpretation that the presence of a secondary mode at -5…-7°C and super-cooled liquid water means “automatically” that we have evidence for rime splintering process. What we often see is that a new spectral mode is appearing at -7°C and that it is quickly accelerating and spectral LDR indicates that this mode is composed of columns and needles (as also aircraft in-situ data confirm). But in my opinion, the radar observations cannot tell anything about the original process (probably SIP) that generates the small new ice particles. Why does it have to be rime splintering? Why can’t also collisional fragmentation (or other SIP) cause the little ice pieces which grow then at -7°C efficiently into needles? It would produce exactly the same spectral signature, right? To me it would be scientifically sound to argue for rime splintering if we could proof that the secondary mode at -7°C appears ONLY in presence of rimed particles and a super-cooled liquid mode in the spectra. But from my own analysis of many Doppler spectra I can assure you, you see this secondary mode (around -8…-5°C) also in cases when the primary mode is not exceeding 1.5 m/s and if there is also clearly no liquid mode. How can we explain this? I think that we should be all very careful that we don’t just “see” processes which we expect but rather carefully proof that it can most likely be only one specific process! Currently, I am not convinced by the “proof” we have. Considering the latest lab findings of nearly non-existence of rime splintering I am even more sceptical about our “common” interpretation.
L. 536-539: Let’s assume the dendritic structures grow on particle falling into the dendritic growth layer from above. The very efficient depositional growth would release latent heat which could cause (together with the larger cross sectional area of the aggregate+dendritic extensions) a slight updraft and hence a reduction of MDV of the main mode. To your second point: The number of collisions and subsequent fragmentation of ice particles which could produce a secondary mode depends strongly on the concentration and relative velocity distribution in the primary mode. Those are two quantities which we cannot very well derive from the radar observations. I think the only way to better understand these features is combining the radar observations with detailed model simulations. I am also puzzled by the fact that sometimes the second mode at -15°C appears and sometimes it does not. If you argue for riming playing a role (you refer to Fig. 15, is actually not rather Fig. 16?) I think you have to show that the velocity differences you see are statistically significant and not just a result of different samples. As you have the width of your distributions you should be able to apply suitable significance tests.
L. 541ff.: “When we examine the velocity of the primary population in multimodal spectra it is systematically higher than it is in monomodal spectra, which suggests that the processes involved in the formation of the multimodal population are also conducive to increasing the fall velocity of the primary population.” Can you guide the reader where one can see this? If I compare Fig. 15c and d at -15°C, I find roughly MDV of 0.9 m/s for monomodal primary velocity and 1.0 m/s for multi-modal primary velocities. Given the width of your distribution I am very sure that this difference is not significant. In general, I like that we formulate and discuss also more unconventional hypothesis in our publications but here I think too many indications speak against it. For rime-splintering one needs strongly rimed particles (graupel-like) and larger super-cooled drizzle drops. Both of them are extremely rarely observed at -15°C and colder. A slight variation of the MDV of the primary mode around 1 m/s cannot only be explained by riming. Simply a slightly different composition of the aggregates by different monomer types or sizes might cause such a variation. But even if your observation is correct and the primary mode velocity of multi-modals spectra is larger, I think a simple alternative explanation can be given: Let’s assume the new slow mode is initiated indeed by collision-fragmentation (maybe collisions of aggregates with dendritic extensions). The break-off of those dendritic extensions would reduce the cross sectional are of the aggregates and could cause an increase in their terminal velocity. Maybe the collisions even lead to larger aggregates (could be evaluated with DWR spectra) with slightly larger fall velocities. As long as our MDV are below 1.5 m/s many different particle properties can cause variation in terminal velocity. Riming is certainly only one of them.
L. 549-556: I completely agree! I think one way to approach the problem further (in addition to lab studies, which are not trivial to do with snowflakes) is to use detailed modern modelling approaches. This includes aggregation and riming models but also tools such as Lagrangian super-particle models (e.g., Brdar and Seifert, JAMES, 2018) which include secondary ice processes, habit dependent growth, and where the history of particles and their interactions can be traced. Together with extensive scattering databases and radar forward operators (e.g., von Terzi et al., GMD, 2026) we have great tools to link microphysics and radar observables.
L. 591 ff.: In this respect the recent publication by Pfeifer et al., 2025 ttps://doi.org/10.1038/s43247-025-02953-3 should be added.
Typos:
- L. 253: “the the”
- L. 518: “up to -20°C of spectra”?
Citation: https://doi.org/10.5194/egusphere-2026-2928-RC2 -
RC3: 'Comment on egusphere-2026-2928', Anonymous Referee #3, 02 Jul 2026
This is an interesting analysis of radar spectra to analysis differences in velocities and temperatures in mono and multimodal spectra. The authors have some good ideas, but I have several major comments. I have specific concerns about how the authors categorize modes and assumptions about which modes are liquid or ice without sufficient evidence. I have quite a few comments, but want to be clear that this paper is promising and I look forward to seeing it in its final form.
General/Major Comments:
My most general concern lies with the justification for your methods. You use an existing scipy function with specific parameters and make a brief comment that your method is in agreement with Radenz et al. (2019) and Vogel et al. (2024), but do not demonstrate any comparison to existing methods or justify the importance of having a new method. I can assume your method was chosen because of how efficient functions from scipy can be to run on large datasets, however this should be in your paper. More critically, without a comparison to existing methods shown in this paper, it is hard to justify the use of the method and understand how accurate it may be in detecting secondary and tertiary modes. Specifically towards the end of your paper (lines 568-569), you state there is no way to quantify the error of inaccurate peak detection, however if you presented a statistical comparison with any other preexisting methodology (e.g., Peako-Peaktree as you referenced in developing the method), you can establish more confidence in the method and quantify the differences in errors between methods. It would improve the paper to include a one to one comparison of your method. Having 23 cases, depending on the time periods of each used, may be tedious or overdoing it to run the methods from Peako-Peaktree, but at minimum confirming it on the two shown case studies would build much more confidence in the method and results. Any other method you choose to compare against would significantly improve the strength of your conclusions here.
Section 3.1.3: I am particularly concerned with the classification of the identified modes as primary and slow. I am equally concerned about not having enough evidence to declare which modes are drizzle or secondary ice. Lines 267-268 state that the primary mode is determined by reflectivity. However in the next sentence, you state the secondary mode is classified by velocity being faster or slower than the primary mode. First, “fast mode” never comes up again. Secondly, knowing there are tri-modal events in this study, it is confusing and ambiguous how they are being evaluated and compared to the full dataset. A higher reflectivity mode will not always be the fast or slow mode, and a more robust way of naming and categorizing the modes is important for understanding the potential processes being observed. There is no comparison between the reflectivity of each primary, secondary, or tertiary mode (beyond the shown spectrograms of two examples). In the bulk of the analysis, you compare “primary” modes defined by reflectivity to just the “slow mode” which remains ambiguous, despite your assumptions. On a large scale, I am concerned by essentially throwing out the drizzle modes in favor of only showing what you presume to be the primary mode and the ice mode. You make this assumption based on fall speed which concerns me. In the absence of aircraft observations or linear depolarization ratio to make a more confident guess on if a mode is liquid or ice, we need to be able to see each of the three modes to have an understanding of if rime splintering could be occurring in the subset of trimodal events.
Specifically from differentiating between modes in 6 Jan 2017 case study:
“When a trimodal spectrum with two slower falling particle modes is identified, the faster of these slow modes is shown in Fig. 9c as we assume this mode is associated with ice particles instead of a drizzle mode, which would likely have very low fall speeds” - this is a bold assumption, especially without showing results in Figure 9 from all detected modes. We are not shown clearly the individual characteristics of each mode, and the spectra in figure 10 struggle to clearly show the multimodal region clearly because the x-scale is set to show the speeds of rain. What should be clear are velocities of the multi-modal layer from -3 to 0 ms-1.
Differentiating between liquid modes and ice modes:
Generally, there seems to be an assumption before the analysis that most or all multimodal events within -8 to -3 deg C are specifically SIP and rime splintering. I am not sure enough consideration is being given to identifying what the modes are before making those assumptions. Korolev and Leisner (2021) highlights this assumption towards rime splintering and points out the varied results from different studies, which I take as advice to caution assumptions that rime splintering is dominant in the atmosphere. This assumption appears to guide most classification in the discussed cases, rather than relying on something more objective. I had specific comments and concerns on this. The secondary mode could be drizzle, primary or secondary ice, though there is still a chance the mode is also liquid. Liquid and ice modes can be differentiated with LDR (e.g. Oue et al., 2015; Sinclair et al., 2016). There is not a strong enough argument yet to assert this. Similarly in line 368, there is not enough evidence for confidence on the active process and other processes should be considered, and weighed. I think it is quite likely you are correct that it is occurring, but we need to see more for the degree of confidence here.
Specific comment in line 378: Would you say that the drizzle mode persists in the first case, as you’re using the argument that it doesn’t in the second case? I struggle to see this in the chosen times for the first case, which worries me about using the justification confidently for the second case.
Analysis shown in Figure 15: I think it could be beneficial to repeat this analysis with a standardization of the primary mode MDV to better illustrate the typical separation between primary, secondary, and tertiary modes. You capture a spread here, but I wonder if information is being lost by averaging such a large and variable dataset, which could be masking more detailed insights into the typical velocity separation rather than spread.
In general, I have some mild concern for only showing two cases; is there a way to visualize even a snapshot of all 23 effectively? The answer might be no, but it was a recurring concern as I read through the paper because I do not have a way to see more of those cases to understand how they contribute to the averages presented.
Minor Comments:
General minor comment: For the 23 cases, you consider both warm, cold, and occluded front events. Is there any merit to splitting this into two separate datasets and examining differences? Examining them as a subset is hopefully not too arduous, and you may have already examined this and decided there is no difference, but it would be useful to know.
Line 27: You could include additional references beyond the 2026 paper, because it is known and studied by a wealth of publications that accurate representation of these PSDs remains a challenge. Additional citations and examples would strengthen this.
Line 54: “to infer microphysical processes” treads a little close to suggesting we observe processes. I’d use a little caution and suggest a rephrase, because the observations from a radar are observations of properties of the clouds/precipitation rather than of a confirmed process.
Line 57: You cite the review article Field et al. here, but don’t include any of the more foundational papers associated with this. Taking a look at the older papers in the citations in the first paragraph of Field et al. where this claim is restated would strengthen this paper.
Line 5, Line 83: I want to point out a study that also quantifies the temperatures associated with multimodal spectra, Wugofski and Kumjian ( https://amt.copernicus.org/articles/18/6569/2025/ ) which seems to complement some of your findings. The study is not specific to frontal clouds, but has some similar findings (see Figure 8).
Lines 138-149: It is unclear and needs to be clarified if this threshold is based on a single case. You state the turbulent contribution is low here, and then apply the calculated turbulence threshold. Is this recalculated for each case, or is the same threshold used in all cases? Is this checked in the cases that are not shown as case studies but still used for the overall calculations to determine the distribution of multimodality with temperature?
Line 170: I am somewhat concerned at the infrequent availability of temperature data (hourly) when attempting to address the temperatures associated with radar spectra with 1 minute frequency. Are there any periods affected with large changes in temperature between hourly data points? Are these potentially skewing any results?
Section 3.1.1, Lines 193-195, and Table 2: I think the choice to not include a parameter for distance between peaks needs to be addressed further, similar to the choices for smoothing (figure 5) and prominence (figure 6). While not part of the detection, another related parameter that may be useful to retain is skewness. It can be informative and previous work (e.g., Luke and Kollias 2021, though other work by both applies here) characterizes the skewness of radar spectra and relates it to the presence of drizzle. You often assume the presence of a drizzle mode, so generally incorporating skewness into the analysis may help address other comments I have regarding determining what each mode is made up of.
Lines 241-243: This sentence contradicts the previous and suggests that mixed phase processes are not active at temperatures colder than 0 deg C. I would recommend more precise language.
Line 272: You say the proportion of ambiguous cases is small; it would be advantageous to quantify how many were ambiguous. Was it one of the 23 events? Was it a certain number of minutes or hours across the entire dataset? As a reader, I want to know and be shown by the author that the impact is limited rather than just be told.
Line 426: You state that multimodal spectra below -20 deg C are rare, but you do not quantify their occurrence in your study. For example, Wugofski and Kumjian (2025) found about 5% of cases across a three year period to be colder than about -20 deg C. A similar statistic from your work would make this claim stronger. Later in line 568, you state your analysis is the first time someone has quantitatively estimated the distribution of multimodal spectra with temperature; the analysis is different but does quantify multimodal layers/events by temperature so steering away from “first” might be best.
Paragraph from line 526-535: This discussion of the cold modes relates to the above minor comment, a sizable proportion of the events in Wugofski and Kumjian (2025) occurred between -8 to -18 deg C. There may be additional insights therein, however the authors partitioned temperature in discrete categories so the ranges are not identical to yours. I also noted in line 572, your 40% of multimodal spectra from -3 to -8 aligns pretty well with Figure 8 from that paper as well.
Lines 552-553: Are there already papers examining the fall speeds of early aggregates published? Dunnavan (2021) initially comes to mind, and references therein may be able to answer or provide insights into this.
Specific comments regarding figures and tables:
Table 1: Consider if estimating depth “from examination of radar images” (I assume visually?) is justified or using an alternate threshold to quantify cloud top is easily available or able to be calculated. Also consider that “within 5 degrees C” is precise enough to describe these events.
Figure 2 (and in general): Most figures showing radar moment data over time show the entire 12 km depth and 24 h time span, which makes it challenging for readers to easily see the features of interest. Reconsider how you’re visualizing these to ensure readers can easily identify the important features. Additionally, I would recommend confirming if this color scale is colorblind friendly. Related, do you disclose what hours of each event you use in the study? Are all 24 hours considered, including periods that do not contain frontal clouds? If so, I would consider adding information regarding that.
Figures 5-6: Both figures have useful information that is challenging to read in the current format of these figures. The black line overlaying the grey could be dotted or dashed to remedy this. In figure 6, you indicate that small changes have little effect. It is challenging to see the effects it does have when the panels are significantly smaller than in figure 5, which leads to the magenta dots covering the peaks more so than highlighting them. Additionally for both of them it would be advantageous to include a reminder of the units (dB for Pr) in the figure or caption.
Figures 9 and 12: This relates to my major comment above. I think it’s critical to, instead of just showing “slow mode velocities” clearly label the modes as “primary, secondary, tertiary” and show all three. For the second case with fewer tri-modal occurrences, this might not be needed. Additionally as stated above, narrowing the figures to specifically show the time periods of interest would make results much easier to interpret. An additional suggestion would be to plot the difference between the velocities of the primary/secondary mode, as when viewing the figure I had to attempt to do this in my head to understand the separation of the modes.
Figures 10 and 13: This is addressed above, however I strongly would reconsider taking daily/24 h averages of radar spectra. For example, 0-4 UTC 6 Jan 2017 being averaged with the period from 12-24 UTC does not sound meaningful. Reconsider if you’d like to add additional figures showing hourly (or 15-30 minute periods) of averaged spectra limited to the period of interest instead. The benefit of spectral data is in those fine details that are lost in such a large temporal average.You point this out specifically in the second case study, which should indicate that this specific panel of your figure is not capturing something clearly for your audience. This could be much more meaningful with a smaller window of averaging, perhaps 1 hour (or 15-30 minutes).
Figure 14: It is not clear why (b) is needed, as this is easily visible and well shown in panel (a) already.
Figure 15: It is initially unclear if this is an analysis from all cases or a single case - I assume it’s aggregated from all cases, however that should be clearly stated. Additionally, reformatting panel (f) would make the variations easier to see. Moving or modifying the legend to accommodate a better x-scale could help.
Figure 16: This figure could benefit from adding a panel similar to 15(f), particularly because the changes in median are small. Visualizing those medians across a small x-scale would help tell the story more so than just reading the accompanying paragraphs.
Typographical errors:
Line 357: Figure 2 is referenced instead of figure 12.
Line 365: Check the velocity units
Line 374: Figure 10 is referenced instead of figure 13.
Line 518: I assume 20 deg C should be 20%?
Line 539: I believe “then” should be “them”
Citation: https://doi.org/10.5194/egusphere-2026-2928-RC3 -
CC1: 'Comment on egusphere-2026-2928', Jialin Yan, 21 Jul 2026
I suggest briefly mentioning an additional possible interpretation of the enhanced multimodality between −18 and −13 °C.
Within this temperature range, relatively fast-falling ice particles formed in colder cloud layers aloft may descend into the dendritic growth zone, where they coexist with more slowly falling dendritic particles formed locally or undergoing rapid depositional growth. Because the particles originating aloft have higher terminal velocities, they may catch up with the slower dendritic population, allowing two particle populations with distinct fall velocities to occur within the same temperature layer and radar sampling volume. This vertical co-location could produce two distinct peaks in the Doppler spectrum without necessarily invoking secondary ice production. This interpretation also appears consistent with the case presented in the preprint.
I therefore suggest that the authors briefly acknowledge this possibility when discussing the enhanced occurrence of multimodal spectra between −18 and −13 °C. A similar interpretation has been discussed in previous radar-based studies, including Yan et al. (2025).
Yan, J., Oue, M., Kollias, P., Luke, E., and Yang, F.: A radar view of ice microphysics and turbulence in Arctic cloud systems, Atmos. Chem. Phys., 25, 16479–16490, https://doi.org/10.5194/acp-25-16479-2025, 2025.
Citation: https://doi.org/10.5194/egusphere-2026-2928-CC1 -
RC4: 'Comment on egusphere-2026-2928', Anonymous Referee #4, 28 Jul 2026
Review of the manuscript “Characterising the occurrence of monomodal and multimodal ice hydrometeor populations and their fall speeds in midlatitude frontal ice clouds using radar Doppler spectra” by Mammatt et al.
The study of Mammatt et al. 2026 addresses the relevant topic of the occurrence of multiple hydrometeor modes in stratiform cloud systems.
I found the manuscript worth reading and the presented approach well implemented. Thus, basically there are no major concerns which would hinder a publication of the study.
Nevertheless, I list below a few minor comments which I would consider helpful for an improved representation of the work.
- Frequent usage of ‘evidence’. I kindly ask the authors to review their manuscript with respect to this strong impression. Apparently, the word is inspired by the title of an earlier very relevant study of one of the co-authors of the study. There are some examples, where ‘evidence’ truly does not apply. E.g., line 325, where Fig. 10b is discussed. It might be likely that there is a drizzle mode present. But there’s definitely no evidence of it. Basically, I kindly ask the authors to only use the word evidence, where there’s actually a way of proving their statement. Interpretation is not an evidence.
- Section 2.1.1:
- The definition of the range correction is wrong. It has nothing to do with the beam width. It has to do with isotropic expansion of the radiation. The transmitted power P_t scales on a sphere defined by 2*Pi*R²à P_u=P_t/(2*Pi*R2)
- P_u needs to be defined. It should be included in the beginning of the section, together with P_noise.
- It should be noted that Z_c is in dB scale, same for P_u and C_calibration
- Line 90: “it takes observations of the Doppler spectra every minute”. This should be clarified. Is there a spectrum (of which duration) every minute? Or are the spectra averaged over 1 minute? If the second item applies, I kindly suggest to the authors to elaborate a bit more about temporal smoothing effects. E.g., Radenz et al. 2019 or Vogl et al., 2024 are based on spectra smoothed across 2-3 seconds. Both study contain evidence for the reason why such short averaging is required. Thus, a discussion is needed about the suitability of longer averaging times.
- Line 197: What is meant by “effects of a sampling a finite number of independent realisations of the hydrometeor population”?
- Line 314: which velocity is meant in ‘we see that velocity increases’? Can you specify?
- Line 370: Why couldn’t the slow-falling peak at -0.25 m/s correspond to a liquid layer?
- Line 375: Figure 13, not Figure 10
- Vertical velocity vs. fall velocity. The authors should spend some space in the discussion/conclusion about the differences between vertical velocity and fall velocity. The study makes the impression that both terms are the same for stratiform clouds, but this is by far not evident.
- Being a bit more critical with respect to the applied smoothing. It is indeed the case that the smoothing affects the number and location of peaks. E.g., also the MDV might be affected if a peak ‘moves’ due to averaging/smoothing. Providing some considerations about this would complement the discussion section.
- Line 491: The reference to Fig. 3 refers to the study of Westbrook and Illingworth 2009, right? Please specify.
- Code publication. It should be aspired to publish the code of the retrieval
- Polarimetry: It would be great if the authors could discuss a bit about the possible advantages of using polarimetry as additional constraint. Polarimetry would, e.g., help to constrain the microphysical properties of the individual fall streaks. Other studies, such as Radenz et al., 2019 or Vogl et al. 2024 apply polarimetric methods, which aids in an improved classification.
- Further points for the discussion: I missed to see some references to basic observational techniques which would indeed help to provide evidence of the presence of certain processes. E.g., the detection of liquid water could be utilized by using lidar techniques or machine learning techniques such as the VOODOO approach (Schimmel et al., 2021)
- Oue et al., 2018, Wugofski et al. 2025 are also studies which use and interpret peak-separation techniques. The authors might consider to include theses references into the introduction and/or discussion.
- Spelling and grammar: I identified several typos and grammatical errors. But I leave the correction of them to the numerous native English-speaking co-authors of the study (preferred) or an AI.
References:
Oue, M., Kollias, P., Ryzhkov, A., & Luke, E. P. (2018). Toward exploring the synergy between cloud radar polarimetry and Doppler spectral analysis in deep cold precipitating systems in the Arctic. Journal of Geophysical Research: Atmospheres, 123, 2797–2815. https://doi.org/10.1002/2017JD027717
Radenz, M., Bühl, J., Seifert, P., Griesche, H., and Engelmann, R.: peakTree: a framework for structure-preserving radar Doppler spectra analysis, Atmos. Meas. Tech., 12, 4813–4828, https://doi.org/10.5194/amt-12-4813-2019, 2019.
Schimmel, W., Kalesse-Los, H., Maahn, M., Vogl, T., Foth, A., Garfias, P. S., and Seifert, P.: Identifying cloud droplets beyond lidar attenuation from vertically pointing cloud radar observations using artificial neural networks, Atmos. Meas. Tech., 15, 5343–5366, https://doi.org/10.5194/amt-15-5343-2022, 2022.
Vogl, T., Radenz, M., Ramelli, F., Gierens, R., and Kalesse-Los, H.: PEAKO and peakTree: tools for detecting and interpreting peaks in cloud radar Doppler spectra – capabilities and limitations, Atmos. Meas. Tech., 17, 6547–6568, https://doi.org/10.5194/amt-17-6547-2024, 2024.
Wugofski, S., Kumjian, M. R., Oue, M., and Kollias, P.: Detection of multi-modal Doppler spectra – Part 1: Establishing characteristic signals in radar moment data, Atmos. Meas. Tech., 18, 6233–6249, https://doi.org/10.5194/amt-18-6233-2025, 2025.
Citation: https://doi.org/10.5194/egusphere-2026-2928-RC4
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 188 | 55 | 17 | 260 | 61 | 14 | 13 |
- HTML: 188
- PDF: 55
- XML: 17
- Total: 260
- Supplement: 61
- BibTeX: 14
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This is a very interesting paper that is clearly written and logically structured. I enjoyed reading it. Even though the presented signatures are well known and have been reported in several previous studies; the presented statistical analysis is interesting and valuable. It would be helpful for the authors to clarify the selection criteria used for the analyzed cases. Is it possible that these criteria influence the resulting statistics, and a brief discussion of potential selection biases.
The multimodal signatures observed in the dendritic growth zone remain an open question. Although such signatures are frequently reported, the underlying mechanisms responsible for ice particle production in this region are still not fully understood. It would be interesting to know whether there the authors have observed any connection between the number of modes and the properties of the cloud above, such as cloud-top height and temperature, etc. Additionally, particles descending from higher levels likely pass through relatively dry layers; could sublimation play a role in shaping the observed distributions?