General comments

In my opinion this paper gives a good overview of the analysis of the current European windstorm risk based on high resolution global climate model data sets. Especially for insurance and reinsurance companies the resulting PRIMAVERA windstorm event set could get a reference for risk management processes: Due to requirement by the EIOPA (European Insurance and Occupational Pensions Authority) insurance and reinsurance companies have to discuss the impact of climate change scenarios on their business.  As the climate change predictions of the EUROPEAN windstorm risk are still quite vague in many regions as even the sign of change is still not robust further research are needed. The presented paper and the mentioned further work on coupled PRIMAVERA climate projection could reduce this uncertainty.

From a scientific point of view the results and applied techniques of the work are probably not completely new ground, but I like the ‘engineering’ approach to build up the model event set. The presented results on storm clustering, the estimated return periods of the extreme windstorm season 1989/90 and the correlation of windstorm occurrence with NAO phases help to discuss the windstorm risk in EUROPE.  

I believe this paper should be published, provided that some minor revisions are applied. I would prefer a more critical discussion of some assumptions and results.

Specific comments

Introduction:

I think that some hints on the requirements of the EIOPA, which wants EU based insurers to discuss their business in the context of climate change is missing. Perhaps these two links could be helpful:

Consultation on Application guidance on running climate change materiality assessment and using climate change scenarios in the ORSA | Eiopa (europa.eu),  https://www.eiopa.europa.eu/document-library/consultation/consultation-application-guidance-running-climate-change-materiality-0

and

European Commission, Joint Research Centre, Spinoni, J., Formetta, G., Mentaschi, L., et al., Global warming and windstorm impacts in the EU : JRC PESETA IV project : Task 13, Publications Office, 2020, https://data.europa.eu/doi/10.2760/039014

Data:

ERA5 wind gust data is set as a reference for the reality. I am missing a reality check or at least some references on the quality of the gust speed representation of the ERA5 data. Gust speeds are the main driver of the used loss function. If they are of minor quality the results will suffer from it, too.

The authors are using the output of different climate models, which probably differ in the representation of the storm tracks and other wind related issues. I would like to see some general figures on the representation of wind over Europe for each model in the appendix (e. g. storm tracks, high percentiles of wind, ...), . Perhaps some short remarks on the main differences compared to ERA5.  

Loss estimation:

The authors are using a loss index derived by Klawa and Ulrich (2003). Although this index is widely used, there could be other possible formulations for a loss function. For example: power law damage function with higher exponents than 3.  For example:

Prahl, B. F., Rybski, D., Burghoff, O., and Kropp, J. P.: Comparison of storm damage functions and their performance, Nat. Hazards Earth Syst. Sci., 15, 769–788, https://doi.org/10.5194/nhess-15-769-2015, 2015.

Higher exponents could amplify LI differences between the used climated models and could have large impact in estimations of future losses, when wind speeds might get higher. I would not expect a detailed evluation of possible effects, but a least a discussion.

Results

Line 305:

„Figure 5 shows a selection of some of the most damaging storms from the re-analysis and ones of similar strength (as measured by the LI) from the PRIMAVERA models.“

What is „similar strength“? For example: In the case of Daria the LI is 6x or 2x larger than the PRIMAVERA model footprints.  I think it could be helpful to see a selection of model footprints, which are really close to he Daria´s LI. Furthermore, I would like to see footprints of events, that are close to the 200 year event. Do they look real?  

Please see attached pdf for my full review.

This study creates and analyses a winter wind storm event set for Europe using the PRIMAVERA multi-model ensemble. Therefore, tracks of extra-tropical cyclones are used to generate footprints of windstorm events for the model ensemble as well as ERA5 reanalysis data. Bias adjustment is applied to estimate 3-s wind gusts out of the daily maximum model output, whereas two different bias adjustment approaches are assessed. By means of the calculation of a loss index, the distribution of losses, return periods, clustering, and relationship with NAO are investigated.

The manuscript is well structured and provides an event set which is worthwhile to be analysed for loss estimation of European wind storms.

With respect to two paragraphs the manuscript reads as it had to be finalized in a certain time frame. The authors write that it was not possible to work on issues concerning the tracking method for ERA5 as well as for some PRIMAVERA simulations. This is not a scientific argument. I suggest to better explain why the data is used as it is and why the whole study is not sensitive on the issues of the tracking method which led to a reduction of data.

Nevertheless, I suggest to publish the work after minor revision.

Minor comments

L 40: Can you provide a reference for this definition?

L 58 ff: you are describing the use of dynamical models to generate event sets. Disadvantages are e.g. coarse resolution of the models with all its difficulties and pit falls (too zonal storm track, too small latent heat release, etc). But you are also mentioning the WISC event set and studies using ensemble prediction systems where the horizontal resolution is comparable to the used PRIMAVERA models. Can you explain the reason and advantages of using your model ensemble in comparison to those studies?!

L 124: see first comment. There is no reference for this definition.

L 137: what does it mean if you are writing that track were unavailable at the time? Are the tracks provided by a computing center? Couldn’t you perform the tracking by yourself?

L142: you are mentioning the definition of the footprint again and arguing that you want to comply with industry standard. Reference would help

L 143: what is the reason for exactly this definition of the domain?

L 146: what does „central day of the 72h period“ mean? Is it 36h before and 36h after this day? If this is the case, this information is very important. Please use this shortly, when mentioning the 72h period and the connection to a cyclone track the very first time. I was wondering before how to connect a 72h period to a cyclone which lives for a couple of days.

L 185: I am wondering about the time frame you had to finalize your study. In how far the results are influenced or biased by this reduced set of 12 winters?

L 192ff (Fig 2): I do not understand how the mask is calculated you are using to separate footprints. Since you are writing to consider gridpoints less than 1500km away from the cyclone track, I expect an area (tube shape) around the track. That seems not to be the case. Can you explain why?

L 206: does this reference define footprints as 3-s gust over 72h? This would be important to use earlier as commented before.

L 209: At L 133 you are writing to use hourly maximum gusts of ERA5. Are those representative for 3-s gusts since you are using this as „observations“

L 268: What is the source for your population density?

L 292ff: The comparison is done on the same grid, isn’t it? Model wind speed is statistically downscaled to the ERA5 grid. Can you explain the mechanism why the coarse resolution underestimates LI?

Do you have an explanation why the resolution effect is cannot be seen anymore for severe footprints?

L 335: it is hard to compare the distribution of ERA and PRIMAVERA especially for high LI values. Would it be beneficial to use CDF? Additionally it has the advantage to be independent of bin width.

L 343: Return periods for ERA are not calculated with a GPD fit but empirically, isn’t it? That means that the most intense season (which is 89/90) has return periods of the length of the time series, i.e. 35 years. Maybe it is worth to shortly explain this just to avoid misunderstanding.

L 345: I do not understand the return periods of the unrealistically extreme events (open circles).

L 455ff: there are different uncertainties for the use of the empirical method or the GPD for the bias adjustment. Is it possible to take this into account when calculating return periods in Fig 7? The confidence interval in Fig 7 is just due to the GPD fit of the PRIMAVERA LI data but the uncertainty to calculate the LI is not taken into account. At least it would be worth to discuss it.