the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Jul 2025
Unravelling the wind impact of clusters of storms, a case study over the French insurer Generali
Abstract. Winter windstorms cause extensive damage to infrastructure and represent the most significant natural hazard for Generali France in terms of insured losses. This study presents a method to systematically link physical storm events with observed insurance claims, enabling a better understanding of which storms, including weaker depressions, drive losses within Generali's portfolio. The proposed association represents a cornerstone for the calibration of insurance and reinsurance processes such as risk assessment, loss modelling and prevention. Beyond analysing individual events, we assess the impact of storm clusters, defined as multiple storms affecting the same region within a 96-hour window, consistent with reinsurance contract definitions. Our findings reveal that 85 % of windstorm-related losses since 1998 are attributable to clustered events. The most intense storms are frequently preceded or succeeded by smaller, yet damaging, depressions. This is illustrated by the case of Storms Anatol, Lothar and Martin in December 1999 and Storm Klaus in January 2009. Furthermore, we find that storms causing damage are more likely to occur as part of a cluster (50 %) compared to the overall population of depressions affecting France (29 %). These findings highlight the importance of accounting for storm clustering in risk modelling and reinsurance strategies.
Competing interests: The contact author has declared that neither of the authors has any competing interests. Authors LH, LB and AP are employed by Generali France.
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.- Preprint
(2705 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-3138', Anonymous Referee #1, 19 Aug 2025
Broadly I like this paper. The findings may be a bit specific to this work and I wonder about the scope for broader applicability but its a nice piece of work which solves a clear problem in working with loss data.
My more specific comments are therefore mostly minor:
17. This is a global figure? Seems very high for Europe, even if Economic rather than Insured loss.
25-29. I Don't entirely understand what this bit is saying.
45-49. Seems well justified.
Figure 1 caption - Should 'different by' be 'defined by'?
120-130 - Interesting discussion of clustering metrics and comments on how a clear definition has not been fixed, or at least not used within the insurance industry. From a loss perspective is it useful to split two clusters affecting France at the same time into separate clusters if they do not overlap?
155-159 Clearly a good dataset with locations recorded. Postcode level is surely fine for ERA5 resolution hazard.
164 - Good observation and an important problem to address.
174 - I can believe this but can you give a bit more explanation/justification. It reads slightly like you argue that the loss data over-represents big storms then remove the little storms yourself. From figure 2a I interpret that lowering N leads to more claims dates than storm dates which is a different problem.
203 - 'Enables the capture'
Section 2 - Overall I find this interesting. Loss data is often focused around extreme events for vulnerability development and this study is an opportunity to work with a sufficiently complete loss history that the impact of smaller events is observed.
246 - How typical Generali's portfolio is of the French market is crucial here. I expect the spread of risks cannot be shown but some commentary could be made as to whether the missing events were very focused on a particular region etc. Some commentary on the exposure comes in 271.
Fig 4 - Is there a relationship between total storm count and cluster counts? The grey bars corresponding to less clustered years appear to group in the 2000s which is also a period of low overall activity based on the red dots.
Section 4.1 seems like the most crucial results, using the dataset and methods discussed in much of the rest of the paper to draw conclusions.
Fig 5. I would expect b) cost per claim to be similar, as observed. But do a) and c) depend on the definition of a storm? If your catalogue contains more small storms the total cost and total claims would be lower, as is observed.
Fig 6. Could we see the spread of losses within clusters for different cluster lengths? For example in a 3 storm cluster is the second ranked loss still 26%?
Section 4.2 shows some value to the method. Often losses from these storms are indistinguishable.
Fig 7/8 what is the resolution of the loss contour maps? The scales go to 6k and 5k respectively but within what area are 5k claims observed? Also there are some white dots that look like they hit Paris that don't seem to be explained.
365-368 Losses are often due to debris, hence consecutive storms may impact losses. This is very difficult for the insurance industry to study and datasets such as this may enable that.
Section 4.3 is also interesting. As stated in 384 Klaus is often treated as a single event.
Citation: https://doi.org/10.5194/egusphere-2025-3138-RC1 -
RC2: 'Comment on egusphere-2025-3138', Anonymous Referee #2, 20 Aug 2025
I read with great interest the paper by Hasbini et al. and found that it provides useful and insightful links between scientific research and industrial applications, particularly in the (re-)insurance sector. Nevertheless, I have several queries, especially regarding the presentation, methodology, and motivation. More specifically, I found it difficult to fully understand the methodological approach, and I have a few concerns about the procedure used to attribute impacts to unique storms. Despite these issues, I find the content very interesting and would therefore recommend a major revision.
Since my concerns mainly relate to the methodological approach, in the following, I focus primarily on the first half of the paper. Addressing these commentss could potentially lead to changes in the results or, at the very least, contribute to a better understanding of the conclusions.
1. Presentation
I found the use of language generally clear; however, I had some difficulty following the text overall. In my opinion, the introduction lacks a consistent narrative and does not make sufficiently meaningful use of technical language. I would strongly suggest revising it according to the comments below:Line 21: Although I agree that tracking methods tend to show higher agreement on the tracks of well-developed storms, I would argue that tracking algorithms can still diverge in their results even for the most intense cyclones. Including a relatively brief review of tracking-method intercomparison studies would strengthen this point.
Lines 27–29: The terms “explosive cyclogenesis” and “strong jet stream” cannot be straightforwardly characterized as physical characteristics. Please revise this phrasing. Additionally, the purpose of these phrases is unclear. In fact, the second half of this paragraph appears to consist of loosely connected statements rather than a coherent argument. I suggest restructuring it for clarity. Probably discussing cyclone dynamics is not of strong concern for this paper.
Lines 30–31: The type of distribution may indeed vary depending on the tracking method used. However, I am not sure I understand how the distribution type justifies the varying number of ETCs “observed at a given location and for a given period.” Please clarify.
Lines 33–34: The statement is vague. Please be more precise.
Line 34: RWB typically occurs due to external forcing on the waveguide — for example, from the outflow of warm conveyor belts. I assume the authors intended to express a different idea here. Please revise, clarity or omit.
Lines 38–40: Dominate what? Which “dynamics” are being referred to? It seems that the term dynamics is being used as a generalization for “all the above.” Please be more specific and precise when using scientific terminology.
Lines 46–48: Why is a 4-day window necessary? Please provide justification for this choice (seems that this is done in the discussions section and thus it comes too late).
Lines 49–53: This section seems more like an introductory paragraph to clustering. Consider integrating it more clearly with the broader discussion.
Line 61: The meaning of “driving characteristics” is unclear. Please define the term more precisely.
Lines 62–63: I agree with the general point, but in some cases, a strict one-to-one attribution of impacts to individual storms is difficult, if not impossible. For example, if a new storm develops within the frontal region of a mature storm, then attributing damages exclusively to one or the other becomes completely arbitrary.
Lines 83–85: Reading the objectives, it seems that resolution-related issues are not thoroughly discussed earlier in the text. If necessary please do so. Additionally, while you mention the limitations of aggregated wind speed data in attributing impacts to a single storm, the importance of distinguishing between storms is not sufficiently developed (see also comments on methodological approach). I overall recommend formulating the introduction to better align it with the study’s stated objectives.
2a. Storm tracks approach
My impression is that cyclone tracking is naively taken for granted considering every track as "a storm". It is quite clear from many cyclone tracking papers that the number of "storms" can be easily tuned and lead to much different results. In the case of this paper, with some very simplistic calculations and supposing a rather constant number of storms per year, then the authors seem to conclude to 100 storms per 6-month season (4439 storms for 45 years), i.e. ~16 storms per month, or more naively, one storm every other day close to France. The potential conclusion of the important role of clusters in provoking damages is thus "predecided" by the storm tracking approach. I feel that the methodological approach should be better discussed and the reader should be provided with more insights about the tracked features (see specific comments below).
Specific comments
Line 109. The storms seem to be treated as "all of the same". For instance, the same radius of influence is used (1300 km) for all tracked features, but I would argue that this is a rather unrealistic assumption. From the perspectives of a morphological approach, the impacts are expected to take place close to the center and along the fronts. So a circular area with a radius of 1300 km seems to be overwhelming. Maybe not so overwhelming for capturing the extent of the fronts but for a small storm this would be certainly the case. This radius actually compares "rather too big" with the surface of France. This seems to further favor the collocation of storms and the conclusion of clusters' high importance. I would advise the authors either to use a dynamically varying effective area for the storms, e.g. change according to intensity, or better, to detect and attribute specific wind patterns to storm track points.Line 96. I am not sure what is meant by "potential noise". As far as I know, the TRACK algorithm identifies and tracks local maxima of relative vorticity. Relative vorticity is a high frequency field. Even a relatively coarse resolution of 0.25x0.25 would pose a computational challenge due to numerous identified local maxima that need to be tracked in time. So smoothing is necessary (I would actually argue that noise is not even appropriate here). How much smoothing is applied and what is the native horizontal resolution of the input field matters greatly for the eventual number of tracked features. These aspects, plus the arbitrary choice of cyclones duration (line 101), are all important to conclude to a high number of tracks. My impression is that in the trascks dataset might be included features (i.e. local maxima of vorticity) that could be rather small, or nested within greater cyclonic structures (e.g. local maxima of vorticity nested in a cold front). Therefore, several (many?) of the tracked features could be hardly considered as "distinct storms".
This is not a critical comment, because bottom line, there is no right or wrong when it comes to the number of tracks, albeit the reader needs to have a good statistical knowledge of the characteristics and nature of tracked features. For instance, it is not rare for tracking algorithms to perceive a "large scale cyclone" with 2 or 3 distinct centers as a set of that many different storm tracks. Whether there is one "big" cyclone or 2-3 "smaller" ones is an arbitrary choice and depends on the tuning of the cyclone tracking method. However, if there are large heterogeneities in the intensities of tracked cyclones, or if many of these cyclones overlap in time, then one needs to adjust their approach when attributing specific impacts to "storms" (e.g. adjusting the radius of influence). Lines 113-115 suggest that cyclone tracking approach was designed to include heterogeneous storms. This needs to be shown, discussed and linked with an accordingly adjusted attribution methodology (see also below).
Section 2.2
except if I missed it, Figure 1 is not referenced in the text. Please also provide dates in the caption or within the figure (if these are the tracks of named storms, that would be also nice to know). Also it would be useful to provide a reference in the map for 7.5 degrees West.Lines 127-128: Distance is 420 km (assuming that 70 km refers to 1 degree..). But is 96 hours a realistic timescale for a cluster of storms hitting the same place? Please elaborate. The 96h criterion is explained in section 5 but this comes too late.
Lines 130-134. While I appreciate the numbers provided here, could we have a visual impression of the frequency of areas affected by storm clusters in France. Also in connection with the above, could we have a figure with statistical information about the storms included in every cluster?
2b. Attribution procedure
Section 3.1 was rather difficult to understand. Probably I get the big picture but the text is rather complex, dense and in some points rather incomprehensible (at least to me). I would certainly appreciate having an illustrative example (maybe also a flow diagram?) where each step of the procedure is shown in a bit more detail. Maybe the examples in section 4 could serve this cause.Lines 173-174 are quite cryptic to me. What does it mean "understanding of storm damage and Generali’s exposure"? And what would it mean "few claims"? Please elaborate. How does this affect the results in this study. How representative is the exposure of Generali for the total of assets in France? Can your results be generalized about storm impacts in France? Actually, why the minimum number of claims is important here? Maybe one claim alone in an area where Generalli has a low exposure still insinuates an impactful event(?).
I actually failed to understand several concepts in section 3.1. In a rather naive approach, I would say that since you already have the coordinates of the circular area influenced by every track point and since you already have the information on a claim, then you could simply say that if the claim spatially and temporally coincides with a storm-affected area (+/- a certain amount of time), then the claim is attributed to that track point. Of course this way, the claim could be attributed to more than one tracked features. In this case, the claim could be certainly attributed to a cluster of tracks (or storms). But I do not see the feasibility of the "..ultimate goal to associate each claim with a single storm" as stated in lines 171-172. Supposing that two track points are very close to each other and correspond to two different storms, or maybe they correspond to the tracks of a "bigger" storm with two centers. Then, the high wind gusts somewhere between two track points will be due to the interaction of these two storm centers (e.g. two distinct storm centers close to each other may result in higher pressure gradients and thus higher wind speeds). There is no really a reason to say that the high gusts are due to one or the other track point. Even in such a case, one could e.g. attribute the claim to the closest track point of either storm. Given that claims and track points are pinpointed and temporally well-defined, I am not very sure I follow the complexity of the approach here, the meaningfulness of the "dstorm" variable and of the cost function. As stated before, I would appreciate illustrative examples to better understand their necessity and use.
Few additional comments
In the examples of Dec 1999, Line 350 states that misattribution compromises the vulnerability curves. I am not an expert but if I understand correctly the field, the vulnerability curve links wind speed with losses. So this is indifferent of a storm attribution procedure(?). If not, could you please explain a bit more this point?In your methodological approach, if I understood correctly, "dstorm" is defined by coordinates which are located over the Ocean (7.5W). So choosing Xb = 3 days means that the claim could take place while a storm is really far from France (I guess still within 1300 kms of radius). Could you show examples where evidently high wind speed is relevant to a storm with a far-reached center? Maybe I misunderstood this part. Could you please clarify.
As an additional note, please avoid using the verb "land" for stating that a track point is over continental areas (or a cyclone influences such). First of all, the coordinate of 7.5 W used here is over the sea, second, I presume that the use of this terminology is inspired by the field of tropical cyclones where the usual term is "landfall". But even so, "landfall" has a special weight because the highest wind speed is really close to the cyclone center. Therefore, time and area of landfalling cyclones is directly relevant to impacts, which is not -always- the case in ETCs.
Section 2.2 seems more like a "methods" rather than "Data" as stated in the title of section 2.
Section 3.1 Lines 161-167. This paragraph seems more adequate for the introduction.
Citation: https://doi.org/10.5194/egusphere-2025-3138-RC2 -
RC3: 'Comment on egusphere-2025-3138', Anonymous Referee #3, 26 Aug 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-3138/egusphere-2025-3138-RC3-supplement.pdf
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 917 | 53 | 21 | 991 | 18 | 22 |
- HTML: 917
- PDF: 53
- XML: 21
- Total: 991
- BibTeX: 18
- EndNote: 22
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1