the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Decadal variations of European windstorms: linking research to insurance applications
Stephen Cusack
Abstract. The insurance sector is affected by decadal-scale variations in annual European windstorm losses amounting to a few billion euros, yet has not applied recent advances in understanding and predicting this variability to their pricing of windstorm risk. This is mainly due to an unknown relation between insured wind losses and meteorological definitions of storminess used in research. This study aimed to reduce this uncertainty.
A history of windstorm insurance losses over the past 72 years was developed from winds in weather reanalyses. Then, typical storm proxies used by researchers, such as the North Atlantic Oscillation (NAO) and the Arctic Oscillation, were compared to the new windstorm loss record. The relationship between the proxies and losses has two distinct regimes: highly consistent from 1950 up to the 2000s, then a divergence in the past 10 to 15 years. The recent separation is large and robust, with high confidence that modern values of researchers’ proxies approach levels last seen 30 years ago, whereas decadal-mean losses are far lower today than in the 1980s and ‘90s.
The cause of this divergence was explored. Storm damages are most closely associated with peak gusts deriving their momentum from winds in the free troposphere, and pressure gradients at the surface used in typical climate indices can only partially describe higher level winds. Based on this reasoning, a new Hemispheric Geostrophy Index (HGI) was defined as the difference in 700 hPa heights between the tropics and the Arctic. It was found to vary coherently with decadal losses in the past, and crucially retains this consistency in the past 15 years too. Breaking down the HGI into component parts revealed that lower storminess in recent times is linked to ongoing reductions in poleward baroclinicity. Further development of loss history and climate indices would help bridge decadal research to insurance applications.
Stephen Cusack
Status: final response (author comments only)
-
RC1: 'Comment on nhess-2022-268', Anonymous Referee #1, 28 Jan 2023
General comments:
This manuscript is devoted to the analysis of the recent mismatch between the interdecadal variability of storm losses in Europe, estimated from wind data using a conventional approach, and of indices of the large-scale atmospheric circulation (teleconnections), such as NAO or AO. This lack of agreement may have critical implications for insurance companies and the general population, thereby being very pertinent and within the scope of NHESS. A new hemispheric geostrophic index (HGI), based on the 700 hPa geopotential height, is then proposed as an alternative to the more conventional indices, showing a higher correlation with the recent changes in storm losses in Europe. It is argued that HGI, being closely related to the low-tropospheric thickness rather than to near-surface conditions, explains this better correspondence. Although these findings are scientifically sounding, I found some parts a bit too speculative, thus deserving a more accurate assessment and demonstration. The text is concise and clearly written. The quality of the figures can be improved. Some revision suggestions are outlined below. Hence, I recommend the publication of this manuscript after some revisions outlined below in the specific comments.
Specific comments:
Section 2.1: please describe in greater detail the datasets and the quality of the data. The average of the two reanalysis products (ERA5 and reanalysis) is also worth explaining, preferably taking into account previous research.
Section 2.2: the use of 11-yr running means without values at the ends of the time series can also be improved using other more advanced methodologies, such as a low-pass filter with a cut-off frequency at 10 years.
Ln 104: Please specify "...to the present day".
Section 3.1: the limitations of the event loss equation are not stated, including their potential contribution to the recent bias. This is a very important aspect to discuss.
Sections 3.2 and 3.3: an assessment of the statistical significance of the trends and divergence is essential. For instance, the statements in Ln 142-143, 161-162 and 207-208 are very vague and need to be proved using robust statistical analysis of trends and inversion points.
Ln 231-232 seems to contradict the use of a zonal/hemispheric index. Please clarify.
The last paragraph of section 4.2 deserves a better discussion, including a more detailed analysis and discussion. Please revise.
Citation: https://doi.org/10.5194/nhess-2022-268-RC1 -
AC1: 'Reply on RC1', Stephen Cusack, 17 Apr 2023
The comment was uploaded in the form of a supplement: https://nhess.copernicus.org/preprints/nhess-2022-268/nhess-2022-268-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Stephen Cusack, 17 Apr 2023
-
CC1: 'Comment on nhess-2022-268', Matthias Klawa, 11 Feb 2023
The author addresses an interesting topic in the insurance world. (Extratropical cyclone) Storm losses in Europe have been relatively low in the last two decades compared to the stormy 80s or 90s. Although standard atmospheric indices such as the NAO or AO, whose positive phase is associated with increased storm activity over the North Atlantic and over Europe in general, have tended to increase again in recent years, recorded storm losses (insured) seem to be below average. The author introduces a new class of hemispheric indices which, on the decadal time scale, provide an interesting explanation for the weak storm damage signal in Europe in recent years.
On first reading of the article, I found the writing style, findings and conclusions reasonable, but on second reading I got the impression that there are still some statistical weaknesses in the manuscript. In my opinion, these weaknesses should be corrected before a decision can be made about a publication.
The author compares a storm loss signal with hemispheric indices on decadal scales. If we regard the loss history in Fig. 1, we can easily see the dominant storm loss of the year 1990, which was caused by a remarkable storm series within just 6 weeks (Daria, Vivian, Wiebke etc.). This single dominant year is probably the main reason why the shape of the loss curve in Fig 4 or 5 (11 year running mean) increases drastically after 1985 and drops down after 1996. I wonder, if we would discuss the findings of the author differently, if we remove the very specific year 1990, or replace the storm loss of that year by an average storm loss. I am afraid that we are discussing a random signal here: If we consider storm losses we should be aware, that the exact position of a storm footprint has an huge impact on the loss amount. If the storm does not cover the densly populated areas of Europe even a severe storm produces a small loss. Perhaps, the author could use number of events above a loss threshold instead of loss amounts.
I would be more comfortable with the new HGIs if the author could show us scatterplots with HGI vs. loss (on a yearly basis, not decadal). Do these yearly HGIs perform similair compared to NAO or AO? Otherwise the impression could be given that these HGIs do work only on decadal scales and that they are just a result of a random hit, which produced the correct up and downs in the graphs.
Specific comments:
Ln 27: Which are the storms exceeding 20 bn USD?
Section 3.1: Maybe I read it over, but does the loss estimation include the whole of Europe? The losses are compared to the PERILS Data. The PERILS losses cover only selected countries.
Section 3.2: The author recalibrates the loss estimation for recent years, because there seems to be an overestimation of winds in the ERA5 data. Is this a known issue for ECMWF ( https://confluence.ecmwf.int/display/CKB/ERA5%3A+data+documentation#heading-Knownissues) or is there any personal communication with ECMWF?
Citation: https://doi.org/10.5194/nhess-2022-268-CC1 -
AC3: 'Reply on CC1', Stephen Cusack, 17 Apr 2023
The comment was uploaded in the form of a supplement: https://nhess.copernicus.org/preprints/nhess-2022-268/nhess-2022-268-AC3-supplement.pdf
-
AC3: 'Reply on CC1', Stephen Cusack, 17 Apr 2023
-
RC2: 'Comment on nhess-2022-268', Anonymous Referee #2, 13 Feb 2023
The manuscript deals with a important topic, which is the decadal variability of windstorms affecting Europe, and the different perspectives of meteorology vs. insurance. While the idea is good, the execution has unfortunately many shortcomings, both in terms of the assumptions and analysis, and the reasoning and conclusions are very speculative on some parts. Therefore, I must suggest the rejection of the manuscript in its present form. However, I think that a strongly reworked version could be an important contribution to NHESS. Below you can find a list of the major caveat of this study, and I hope these help the author to reformulate the manuscript
Major points:
1) datasets, line 75ff - it is not understandable to me why the AVERAGE between two different reanalysis is taken here. This will flatten the fields without necessity. Please do two separate analysis, one for each Reanalysis. To look into the relationship between the large-scale patterns / pressure gradients and storms, a sub-monthly time scale would be preferable (e.g. Fink et al., 2009, doi:10.5194/nhess-9-405-2009)
2) same section - it not clear to me why the wind gust variable from ERA5 was not used, but rather the 10m winds, as it is the former that is responsible for the damage.
3) same section - The version perils data used by the author give extremely limited information and are heterogeneous in space an time, thus hampering the analysis. I would recommend to use the commercial version of the data, to which the author should have access to.
4) data processing - this section is badly written and lacks an lot of details, and thus the methodology is not understandable. For example, I do not unterstand how "monthly data is processed into storm seasons", and the quantification of the storms and their impacts must be done at the sub-daily scale, optimally 3h for ERA5.
5) 3.1. basic method. The metholody by Klawa and Ulbrich is well established and has been used by a large number of publications since. However, it is not clear for which area the data is calculated, or many other details
6) same section - Given that it is not clear for which area the index was calculated, and the PERILS data is only available as a single value for a subset of countries affected by a storm, the reasoning regarding Figure 1 is not understandable. A lot is section 3.2. is quite speculative. In particular, the conclusion in line 161 is not justified.
7) section 3.3. it is not understandable for me how the loss variability of a single, small country like the Netherlands can be used to make assumptions for such a large area. Small countries have either a full hit or not hit (and thus a steeper loss curve), while larger countries like Germany or France have often partial hits, leading to a flatter curve (e.g. Karremann et al. 2014, doi:10.1088/1748-9326/9/12/124016) Thus I totally disagree with the statement in lines 170-171.
8) section 4.1 the variability of the storm activity over Central Europe is partially associated with the mentioned large-scale patters, but is best associated with a eastern shifted NAO pattern (see e.g. Fink et al., 2009). Regarding the NAO pattern I think the author is overstating the results, because a correlation of 0.60 indicates an explained variance of 36% only.
9) I strongly disagree with section 4.2. - to achieve a better skill for storm impacts over Europe, a more specific index should be chosen, like a shifted NAO index, and not an even larger-scale pattern.
10) same section - the author is assuming a linear relationship between winds above the boundary layer and surface gusts, which is a oversimplification as the main factor is actually the turbulence in the boundary layer (see e.g. Born et al., 2012, doi:10.3402/tellusa.v64i0.17471). Thus is particularly true for when strong convective is enbedded in the cold fronts of the storms, see again Kyrill, Fink et al., 2009).
11) Conclusions - given that the analysis has many weaknesses, the conclusions are unfortunately highly speculative.
Citation: https://doi.org/10.5194/nhess-2022-268-RC2 -
AC2: 'Reply on RC2', Stephen Cusack, 17 Apr 2023
The comment was uploaded in the form of a supplement: https://nhess.copernicus.org/preprints/nhess-2022-268/nhess-2022-268-AC2-supplement.pdf
-
AC2: 'Reply on RC2', Stephen Cusack, 17 Apr 2023
Stephen Cusack
Stephen Cusack
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 444 | 173 | 18 | 635 | 3 | 5 |
- HTML: 444
- PDF: 173
- XML: 18
- Total: 635
- BibTeX: 3
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1