I thank the authors for their detailed responses. My main comments are resolved with the additional discussion included in the new manuscript. I only have a few optional suggestions remaining. Congratulations for this excellent article.
Previous comment 1.1:
My main concern is that the simulated storms are insufficiently resolved to represent real hailstorms, so the study might yield (partially) misleading results. It has been shown that at ≈ 2km resolution as used here, peak updraft speeds are substantially reduced (e.g., Adlerman and Droegemeier 2002 and references therein). This issue might also affect other updraft characteristics like width and depth. Since these factors are extremely important for hail growth (e.g., Lin and Kumjian 2022 and references therein) a strong impact can be expected. I think Hailcast somewhat corrects for these factors (Adams-Selin and Ziegler 2016) but it is unclear how well it performs with the COSMO model and at this particular resolution. How severe these limitations are is not discussed in the article and is thus difficult to say. However, there are two reasons that at least point towards a strong bias in the simulations. (a) The sister-study that goes more into the verification of the model (Cui et al. 2024) shows that the hail distribution in the tested cases is not well covered. As far as I know, this preprint is still under review, but I’m not a reviewer on it so I don’t know if it will be accepted. Since the methodology is the basis for all follow-up articles, like the present one, I’m somewhat hesitant to recommend publications, at least not without emphasizing these limitations. (b) Lines 323-326: the obvious discrepancy between CAPE and the maximum w in the simulations is not discussed. For example, Fig. 10 at least roughly suggests that average CAPE values of for example 1500 J kg−1 leads to maximum updrafts of 20ms−2 in your simulations, while from the wmax equation (line 324) one would expect 55ms−2. Yes, the latter is a theoretic value not necessarily realized due to entrainment, but especially hailstorms (often supercells) can realize most of their CAPE (e.g., Peters et al. 2019, 2020a). Also, vertical velocities > 50ms−2 are likely common in supercells (e.g., Peters et al. 2020b) while only extreme outliers reach this range in your simulations (your Fig. 10). This is consistent with the expected negative bias in updraft intensity at 2.2km resolution (Adlerman and Droegemeier 2002). So far, the authors only briefly discuss that “fine-scale processes influencing hailstorm development” might not be represented (line 460). I think this insufficiently describes the problems of the simulations for the reasons outlined above. A much deeper discussion is necessary and the reader should be made aware which results must be taken with a grain of salt because of the biases in updraft characteristics (see, e.g., comment 12 below, but I could see that there are other less obvious implications). I think it is still worth to publish these important results but to me there is some uncertainty to all of them, which must be made clear.
Reply 1.1: We thank the reviewer for raising the important issue of resolution-related uncertainties in simulating deep convective processes relevant for hail formation. We fully agree that kilometrescale models, including the 2.2km COSMO simulations used here, operate within the so-called “grey zone” where small-scale turbulence and convective dynamics are only partially resolved. Nevertheless, this resolution represents the current state of the art for convection-permitting climate simulations over continental domains and has been widely used in recent studies (e.g., Prein et al., 2021). In particular, Prein et al. (2021) demonstrated that while some characteristics (such as updraft core structure and vertical mass transport) continue to improve with grid spacing < 2 km, most climate change signals in mesoscale convective systems (MCSs) are already robustly captured at 4km grid spacing compared to reference large-eddy simulations with a grid spacing of 250 m. Currently, a further refinement to sub-kilometre grid spacings remains computationally unfeasible for multi-year simulations over large domains. The resolution-related limitations and associated impacts on hail-relevant processes are discussed in lines 458 and following; we agree that this could be clarified further and have added appropriate text passages (see addition b below). Concerning point (a): We mention that the accompanying study (Cui et al., 2024), which provides a dedicated verification of the same modeling framework, was accepted on 31 May for publication in J. Geophys. Res.––Atmospheres. Concerning point (b): We acknowledge the discrepancy between CAPE and modeled updrafts and will elaborate on its implications in the revised version, including selected references you suggested.
Manuscript addition – (a) reference to Cui et al. (2024), insertion on L96: “Despite limitations associated with the native grid resolution, this validation indicates that the overall spatial and seasonal patterns of hail activity are reasonably well captured.”
Manuscript addition – (b) discussion of CAPE vs. w discrepancy, addition on L326: “This underestimation is a known feature of kilometre-scale models and has been documented in several studies (e.g., Adlerman and Droegemeier, 2002; Peters et al., 2020a). It results from coarse resolution entrainment, insufficient core thermal contrast, and smoothing of updrafts. While HAILCAST attempts to account for subgrid updraft structure (Adams-Selin and Ziegler, 2016), the magnitude of this compensation at 2.2km grid spacing remains uncertain and is an important limitation when interpreting simulated absolute hail sizes.” Revised discussion in paragraph on limitations (insertion on L460): “The simulation of convective storms at 2.2km grid spacing inherently limits the representation of fine-scale processes critical for hail formation, such as peak updraft speed, entrainment, and hydrometeor interactions within the convective core. As discussed in Prein et al. (2021), even though kilometre-scale models significantly improve upon coarser-grid simulations and reproduce robust climate signals in convective systems, they operate in the “grey zone” where vertical mass fluxes and core dynamics are underrepresented. This likely contributes to underestimated vertical velocities (see also Fig. 10), as compared to theoretical expectations or observations in supercells (Peters et al., 2020b,a). Consequently, hail growth potential may be suppressed in the model compared to reality, particularly for severe events. While HAILCAST includes parameterizations to adjust for updraft strength (Adams-Selin and Ziegler, 2016), uncertainties remain regarding its performance at this specific resolution and model setup. Acknowledging this, we caution that the absolute hail sizes should be interpreted in a relative sense, with a focus on comparative patterns rather than absolute magnitudes.”
Round 2 reviewer comment 1.1: I agree that you used the current state of the art, the comment was not meant as criticism about your approach. Just to expand on this aspect, I still believe you underestimate the limitations with respect to under-resolved updrafts. Adams-Selin (2025) and Fischer et al. (2025) recently emphasized how sensitive 3D hail trajectories are to updraft characteristics. I at least see it as a possibility that trends of certain hail sizes cannot be represented in your study because of the limit in realistic updrafts.
Also, I’m not sure Prein et al. (2021) can be used as a strong support here. As you mention, they found that updraft characteristics continue to improve at resolution <4km and they did not look in detail at hail production or hailstorms in particular, which tend to be non-MSC.
However, that’s just my opinion, I’m ok with the manuscript additions above if you think it makes the limitations clear enough.
Additional comments:
• Line 6 and throughout: Is the italics for numbers intended?
• Lines 57-58: Would it make sense to put this shorter part before (iii) ? At least to me it seems logical to end with the approach you are using. Just a suggestion.
• Lines 303-311: As mentioned in line 123, hailcast does not take horizontal advection of hail into account so the landing position relative to the updraft might not be realistic. Fig. 5a also indicates this as hail mostly falls directly under the track. In other words, the relative horizontal transport of hail and rain (size sorting) is not fully represented. Yet here you seem to take a roughly correct position of hail relative to rain as granted. See radar studies or 3D hail trajectory simulations for more realistic fall locations of hail.
I think you can still include this part given you look at relative differences. However, the caveat should I think be mentioned in the context of this paragraph.
• Lines 497-501: I think it should also be mentioned that you only simulated a 10-year period, so changes in high-end hailstorms might not be represented, especially not at a regional level. |
Review of egusphere-2025-918
Insights from hailstorm track analysis in European climate change simulations
by Killian P. Brennan et al.
Overview: This is an excellent article embedded in a series of articles based on convection-allowing simulations in a current and a warmer climate. This particular study nicely analyzes individual storm properties, mainly showing that hailstorms will produce more severe hail due to stronger updrafts, that the melting of hail does not have a large effect on the hail size in a given storm, and that reduced frequency of hail in some parts of Europe is likely attributed to a reduced frequency of storms and not a reduction in severity. These results are highly relevant.
The figures and scientific language are of high quality and the manuscript has a clear structure. I only have one general major comment. However, since this might affect the robustness of the whole methodology and all studies using these simulations, I recommend major revisions.
General comments:
1) My main concern is that the simulated storms are insufficiently resolved to represent real hailstorms, so the study might yield (partially) misleading results. It has been shown that at ~2 km resolution as used here, peak updraft speeds are substantially reduced (e.g., Adlerman and Droegemeier 2002 and references therein). This issue might also affect other updraft characteristics like width and depth. Since these factors are extremely important for hail growth (e.g., Lin and Kumjian 2022 and references therein) a strong impact can be expected. I think Hailcast somewhat corrects for these factors (Adams-Selin and Ziegler 2016) but it is unclear how well it performs with the Cosmo model and at this particular resolution.
How severe these limitations are is not discussed in the article and is thus difficult to say. However, there are two reasons that at least point towards a strong bias in the simulations.
(a) The sister-study that goes more into the verification of the model (Cui et al. 2024) shows that the hail distribution in the tested cases is not well covered. As far as I know, this preprint is still under review, but I’m not a reviewer on it so I don’t know if it will be accepted. Since the methodology is the basis for all follow-up articles, like the present one, I’m somewhat hesitant to recommend publications, at least not without emphasizing these limitations.
(b) Lines 323-326: the obvious discrepancy between CAPE and the maximum w in the simulations is not discussed. For example, Fig. 10 at least roughly suggests that average CAPE values of for example 1500 J/kg leads to maximum updrafts of 20 m/s in your simulations, while from the w_max equation (line 324) one would expect 55 m/s. Yes, the latter is a theoretic value not necessarily realized due to entrainment, but especially hailstorms (often supercells) can realize most of their CAPE (e.g., Peters et al. 2019,2020a). Also, vertical velocities >50 m/s are likely common in supercells (e.g., Peters et al. 2020b) while only extreme outliers reach this range in your simulations (your Fig. 10). This is consistent with the expected negative bias in updraft intensity at 2.2 km resolution (Adlerman and Droegemeier 2002).
So far, the authors only briefly discuss that “fine-scale processes influencing hailstorm development” might not be represented (line 460). I think this insufficiently describes the problems of the simulations for the reasons outlined above. A much deeper discussion is necessary and the reader should be made aware which results must be taken with a grain of salt because of the biases in updraft characteristics (see e.g., comment 12 below, but I could see that there are other less obvious implications).
I think it is still worth to publish these important results but to me there is some uncertainty to all of them, which must be made clear.
Specific comments:
1) Lines 32-33: can you add a reference to your statement that CC scaling leads to stronger average updrafts and hail? I agree that it is so on average but as you mention further below these links are not 100% clear, so at least providing a reference here seems warranted.
2) Line 33: Also here you could add a reference showing the rise in the 0°C lvl (e.g., Prein and Heymsfield 2020, Gensini et al. 2024).
3) Line 35: Is the slower fall speed really the main reason why small hail is more affected from melting? I always thought it is that larger hail has a relatively smaller surface area and more mass, so the cooling from latent heat exchange can slow down melting more effectively. Can you perhaps add a reference?
4) Line 46: Several other means come to mind: observational proxies like lightning and overshooting tops (e.g., Punge and Kunz 2017), hail pads (e.g., Manzato et al. 2022), hail reports and radiosonde soundings. You don’t need to mention all possible methods though. Perhaps just rephrase that the means you introduce here are the ones that have so far been used in the literature to study climate trends (perhaps adding hail pad studies).
5) Lines 55-56: This might be confusing to some readers. The European domain is not larger than the US, no? I suggest rephrasing.
6) Line 173: It is not immediately clear what “mean storm maximum hail diameter” is. Is it the average over all maximum diameters occurring within the storms at a gridpoint? Consider defining it more clearly once.
7) Line 179: I’m assuming this refers to the area that is over the 10mm threshold? Consider mentioning this explicitly since area could also refer to other variables like precipitation.
8) Line 257: Consider replacing “drawn in” with “generated”, which seems more accurate.
9) Line 261: Can you add a reference or some more context for why you consider this "the inflow level"? Can’t the inflow be dominated by any layers in the lowest 3 km or so and vary substantially for example between elevated storms and supercells (e.g., Nowotarski et al. 2020)?
10) Lines 326-327: It’s not clear to me what you mean by “sampling bias due to coarse vertical resolution”. Aren’t all model gridpoints used? Then the w_max in the simulated cell is what matters for the simulated hail production. So in other words, it’s not a sampling bias but more a model bias which might significantly impact the whole methodology (see general comment 1). Or did I misunderstand something?
11) Lines 362-364: Agreed! One way to test this would be to repeat the analysis for the far-field environment. The appropriate distance from the storm could be determined based on where you stop seeing storm-induced perturbations in the pressure and wind field (Fig. 5) and on the literature (Coniglio and Parker 2020; in general this and other references within could be added to this paragraph). Would this feasible?
12) Lines 372-376: The underestimation of maximum updraft intensity outlined in general comment 1 could be the reason why no nonlinear behavior is seen in your simulations compared to Lin and Kumjian.
13) Line 380: Here or later in the discussion it might be helpful to clarify what exactly “small” refers to. Gensini et al. argue that melting dominates for hail < 4 cm (their Fig. 1) while in your study “small” is used for much smaller diameters (e.g., lines 439, 426). I think this clarification is important to put your work into perspective because it emphasizes that your results point in a different direction.
14) Section 5.2: I’ve never used HAILCAST but if I recall correctly, accurately representing melting of hail in such a model is not an easy task and subject to heigh uncertainty. If you agree, this point should be discussed, because it could be the reason why melting does not have a strong effect in your simulations compared to the other literature.
15) Line 424: The last sentence here seems out of place and could be removed?
16) Line 440: I’d suggest briefly mentioning what “uncertainties” you are referring to because the references alone leave room for interpretation.
17) Lines 447-449 and 456: I fully agree and these are important results.
18) Line 460: I recommend also citing Adams-Selin (2025) here.
Technical corrections:
Line 153: So you mean “SON” for November?
Line 204: Consider replacing “different” with “spatially heterogeneous” and “similar” with “homogeneous” to be clearer.
Footnote 2: Consider shortening to “The second approach is arguably better as…”
References:
Adams-Selin, R. (2025). The Quasi-Stochastic Nature of Hail Growth: Hail Trajectory Clusters in Simulations of the Kingfisher, Oklahoma, Hailstorm. Monthly Weather Review, 153(1), 67–87. https://doi.org/10.1175/MWR-D-23-0233.1
Adlerman, E. J., and K. K. Droegemeier, 2002: The Sensitivity of Numerically Simulated Cyclic Mesocyclogenesis to Variations in Model Physical and Computational Parameters. Mon. Wea. Rev., 130, 2671–2691, https://doi.org/10.1175/1520-0493(2002)130<2671:TSONSC>2.0.CO;2.
Coniglio, M. C., & Parker, M. D. (2020). Insights into supercells and their environments from three decades of targeted radiosonde observations. Mon. Wea. Rev., 148, 4893–4916. https://doi.org/10.1175/mwr-d-20-0105.1
Punge, H. J., Bedka, K. M., Kunz, M., & Reinbold, A. (2017). Hail frequency estimation across Europe based on a combination of overshooting top detections and the ERA-INTERIM reanalysis. Atmospheric Research, 198(July), 34–43. https://doi.org/10.1016/j.atmosres.2017.07.025
Nowotarski, C. J., Peters, J. M., & Mulholland, J. P. (2020). Evaluating the effective inflow layer of simulated supercell updrafts. Mon. Wea. Rev., 148(8), 3507–3532. https://doi.org/10.1175/MWR-D-20-0013.1
Peters, J. M., Nowotarski, C. J., & Morrison, H. (2019). The role of vertical wind shear in modulating maximum supercell updraft velocities. Journal of the Atmospheric Sciences, 76(10), 3169–3189. https://doi.org/10.1175/JAS-D-19-0096.1
Peters, J. M., Nowotarski, C. J., & Mullendore, G. L. (2020a). Are Supercells Resistant to Entrainment because of Their Rotation? J. Atmos. Sci., 77(4), 1475–1495. https://doi.org/10.1175/jas-d-19-0316.1
Peters, J. M., Morrison, H., Nowotarski, C. J., Mulholland, J. P., & Thompson, R. L. (2020b). A formula for the maximum vertical velocity in supercell updrafts. J. Atmos. Sci., 77(9), 3033–3057. https://doi.org/10.1175/JAS-D-20-0103.1.
Prein, A. F., & Heymsfield, A. J. (2020). Increased melting level height impacts surface precipitation phase and intensity. Nature Climate Change, 10(8), 771–776. https://doi.org/10.1038/s41558-020-0825-x