This study uses single-polarisation radar observations and high-resolution insurance data to develop a model relating the Maximum Expected Severe Hail Size (MESHS) diagnostic to crop hail damage. A detailed investigation is performed into the sensitivity of the model performance to data resolution, highlighting important differences between field crops and grapevine, which are linked to their differing spatial distribution. The impact of variations in MESHS threshold and the minimum crop field density used to define exposure are also explored.

I was really impressed with this paper. It is well written and scientifically rigorous, with high-quality figures and appropriate discussion of relevant literature. I particularly appreciated the detailed analysis of how spatial resolution and cropland density impact the different verification metrics. I think this will be an invaluable reference for future studies of hail-related damage to crops and other assets. As such I have no hesitation in recommending it for publication, subject to some minor revisions. There are a couple of issues related to the use of radar data for detecting hail that I would like to see briefly discussed in the paper. These are noted below, together with a few other substantive comments. Beyond this, I have numerous suggests for minor textual and grammatical changes, as well as  adjustments to some of the figures. Rather than listing these all out, I have provided an annotated PDF (see attached). I thank the authors for their efforts in putting this work together and encourage them to reach out to me via email should they have any questions regarding my review.

Substantive comments:

• Several factors can impact the quality of radar-based hail retrievals. Among these are (1) the reflectivity calibration, (2) the methods used to map observations from spherical polar coordinates to a Cartesian grid, and (3) attenuation and resonance scattering effects. The first of these is discussed in my 2020 paper (Warren et al. 2020), the second has been explored by my colleague Jordan Brook (Brook et al. 2022), while the third is addressed in several previous studies (Battan, 1971; Kaltenboeck and Ryzhkov, 2013; Ryzhkov et al., 2013). I think it would be useful to provide some brief discussion on these topics in section 2.1 of your paper. You should note what procedures (if any) are used in Switzerland to ensure an accurate and spatially consistent radar calibration. If no such procedures exist you should consider what impact this might have on your results. Likewise, you should briefly describe the method used to map data from the five individual radars to your common 1km grid and the potential impacts this may have on your hail retrievals. Given that your radars are all C-band, attenuation and resonance scattering effects could both be significant. The former can be corrected for given dual-polarisation measurements (e.g., Gu et al., 2011), although the corrections may be underestimated attenuation due to hail (Borowska et al., 2011). Again, you should note what (if any) attenuation correction is employed in the Swiss radar network.
• When you first introduce MESHS (on L43) it is worth adding a footnote clarifying that MESHS is different from MESH, the Maximum Expected Size of Hail (MESH) originally introduced (as MEHS; maximum expected hail size) by Witt et al. (1998). This is important as MESH is much more widely used (at least outside of Switzerland).
• On L200-201 you note that my study (Warren et al., 2020) achieved an HSS of ~0.5. It might be worth mentioning a couple of other previous studies that have quantified the skill of radar-based hail detection (e.g., Cintineo et al., 2012; Ortega, 2018; Skripniková and Řezáčová, 2014; Kunz and Kugel, 2015).
• When introducing the inflation factor (L260), it would be useful to provide a proper definition of this quantity.
• At the start of section 3.2, you state that CSI and HSS do not exhibit a clear maximum in Fig. 4 (L269-271). I wonder this simply reflects the choice of vertical scale in the figure and the use of 20mm as a lower bound on the MESHS threshold. If you were to plot these metrics separately from POD and FAR, with a maximum y value of 0.4 and a minimum MESHS threshold of zero, I'm sure you would see a more well-defined skill maximum. I was actually curious as to why you didn't consider MESHS values below 20mm. Does this reflect how the metric was defined, perhaps?
• On L273-274 you say "Warren et al. (2020) suggest to additionally constrain the optimal threshold with the condition that B is close to 1 to avoid overforecasting." It is worth noting that Stržinar and Skok (2018) used a similar approach.
• Regarding your discussion of Fig. 5b on L287-290. I note that there is actually a slight increase in CSI going from 1km to 8km resolution at the 40mm MESHS threshold. This is associated with a more pronounced increase in POD and little change in the FAR. However, for the lower MESHS thresholds of 20 and 30mm, the increase in POD is negated by a commensurate increase in FAR, leading to a reduction in CSI. This difference reflects the fact that the points corresponding to the lower thresholds reside in a part of the phase space where CSI shows little sensitivity to POD but large sensitivity to FAR.
• On L309-301 you say that "the choice of the optimal nthresh heavily depends on the chosen spatial resolution." Can you comment on this a bit more? It might even be worth adding an extra figure to illustrate this sensitivity.
• Regarding your discussion on using the Peirce Skill Score (PSS) instead of HSS or CSI (L347-353). I wasn't aware of the Ebert and Milne study, so I don't know if they discuss this, but a key problem with the PSS is that it is insensitive to forecast bias for rare events. This can be seen by examining the equation: PSS = H - F, where H = a / (a + c) is the hit rate (POD) and F = b / (b + d) is the false alarm rate (also known as the probability of false detection, POFD). For rare events, d becomes large relative to the other elements of the contingency table so F becomes small relative to H. As such the PSS is dominated by hits and misses (the POD) and becomes insensitive to false alarms. The CSI and HSS don't suffer from this issue, which is why they tend to be favoured for rare events. As such, I am not sure that it is fair to say that use of the PSS represents an "equally valid verification procedure"  (L347). At the very least, some discussion around this issue with the PSS should be included here.
• I'd be curious to know how the specific model configurations listed in Table 3 were selected. Was some minimum HSS threshold applied? Or are these just intended as representative examples?
• In the conclusions (L391-393), you could also note that your results compare reasonably well with verification of the Severe Hail Index (SHI) and Maximum Expected Size of Hail (MESH) metrics (Witt et al. 1998) in several previous studies (Cintineo et al., 2012; Skripniková and Řezáčová, 2014; Kunz and Kugel, 2015; Warren et al., 2020).

Rob Warren, Bureau of Meteorology, Melbourne, Australia

The manuscript introduces the relationship between a radar-based hail size estimating product and its affectation on different types of crops in Switzerland. It is well-written and easy to follow. The main issue of the research is the lack of a physical or scientific explanation of the results, presenting a simple statistical analysis. This point reduces the work potential, which can be notably improved if the Authors consider some elements associated with the MESHS (Maximum Expected Severe Hail Size).

In the following lines, I present my suggestions/questions/doubts regarding the manuscript:

• Is CLIMADA the acronym of something? 
• L22, 23: The authors should avoid technical reports if scientific references are available (SwissRe, 2021, 2022). Here are some of the multiple possibilities:

Rana, V. S., Sharma, S., Rana, N., Sharma, U., Patiyal, V., Banita, & Prasad, H. (2022). Management of hailstorms under a changing climate in agriculture: a review. Environmental Chemistry Letters, 20(6), 3971-3991.

Gobbo, S., Ghiraldini, A., Dramis, A., Dal Ferro, N., & Morari, F. (2021). Estimation of hail damage using crop models and remote sensing. Remote Sensing, 13(14), 2655.

Bell, J. R., Gebremichael, E., Molthan, A. L., Schultz, L. A., Meyer, F. J., Hain, C. R., ... & Payne, K. C. (2020). Complementing optical remote sensing with synthetic aperture radar observations of hail damage swaths to agricultural crops in the central United States. Journal of Applied Meteorology and Climatology, 59(4), 665-685.

• Add a reference regarding MESHS (L43)
• What is the physical basis of these products? (L44)
• L57: There exists a physical reason for changing the spatial scale linked with the product definition. You need to go deeper into this point.
• L 64: The word "without" appears repeated twice
• Paragraph L 60-68: I found the introduction referring to radar analysis of hail-crop affectation.
• L93-96: From these lines, I understand that the authors did not consider the 5 radars for the whole period, isn't it? Please clarify this point.
• L104: But not avoid it. Do you know if the hailstorms occurring during that period are more or less severe than usual?
• L112: However, this is not consistent with the storm size and the hail path. Please, explain better this point.
• L152: Why 1000 times exactly?
• L158: I assume that you say that two hailstorms occurred over the same place in the same event, is this correct? Clarify.
• L168: What "magnitude" is presented in these fields?
• Section 2.5: Are you considering all the pixels of the full coverage for each event, or only those probably affected by the hailstorm?
• L213: Have you analysed the hailstorm type for each case? Are they similar?
• L227-235: Explain the behaviour of the skill scores depending on when you change the spatial resolution of the radar product.
• L245: Add a map (preferably at the beginning of section 2) of the study region including the location of the places appearing in the text. (The Authors should consider that the reader can provide from a distant place and needs to be as familiar as possible with the geography of the region when he/she reads your research)
• L251: I miss more modern references than 1975 ones.
• Section 3.2: Have you taken into account the fact that if two hailstorms occurred over the same region in a very short time period, the insurance reports would be lower than expected in the second case?
• L330: Do you think there are similitudes with studies referring to the density population relationship with hail affectation?
• The caption of figure 3: Simplify it, indicating only the differences with figure 2.