the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 May 2024
Using seasonal forecasts to enhance our understanding of extreme European windstorm impacts
Abstract. Considerable effort is spent at insurance and reinsurance companies to estimate the risk posed by windstorms. Among these risks, strong near surface wind speeds can be particularly damaging, threatening infrastructure, human life, and billions of pounds in insured losses. Here, we use nearly 700 years worth of extended wintertime seasonal forecast output to estimate the impact of extreme European windstorms, with insured losses estimated using a storm severity index (SSI). Using the full integration period of the seasonal forecast model, we follow the UNprecedented Simulated Extreme ENsemble (UNSEEN) method, here applied to windstorms for the first time. After demonstrating that the seasonal forecast model of the UK Met Office represents windstorms with good accuracy, and developing a new method to convert from wind speed to wind gust derived SSIs, the likelihood of occurrence of unprecedented windstorms is quantified for several countries within Europe. The probability that a windstorm that impacts a country will be more extreme than any observed (i.e. an unprecedented or unseen windstorm) is generally between 0.5 % and 1.6 %. The North Atlantic Oscillation (NAO) is shown to influence European windstorms: strongly positive and negative NAO values strongly increase and decrease the likelihood of an unprecedented storm, respectively.. Serial clustering of windstorms within an extended winter is also found to increase the aggregated seasonal impact of windstorms for the countries analysed herein. These results may aid in the prediction of seasonal loss totals, as the NAO, for example, is predictable several months in advance. The analyses presented could be extended to other datasets, thus increasing the sample size of windstorms and allowing for the estimation of very high return period storms and the potential losses insurance companies will be liable to cover.
- Preprint
(5488 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 08 Jul 2024)
-
RC1: 'Comment on egusphere-2024-686', Anonymous Referee #1, 14 Jun 2024
reply
Summary
The authors use ensemble predictions to enlarge the data base to assess extreme storms over Europe. The aim is to better constrain extreme events which is highly relevant for society given their strong economic impact. The authors state that the data set is suitable to be used in this respect and confirmed the relation ship of the NAO to influence European storms. They also highlight that serial clustering of windstorms impacts the impact on the country level.
General
The paper provides some interesting insights in using forecast data to better constrain extreme windstorm impacts. Still the paper suffers from some problems in the method section and structural issues concerning the evaluation and result sections which need to be solve prior to possible publication. Moreover, I got the feeling that something with the bias correction approach is wrong or at least not well explained. Thus, I recommend major revisions.
Comments
Title: Given in content of the manuscript I find the title a bit misleading I rather think the aim of the study is to use seasonal forecast to better constrain and assess extreme windstorm impacts. A mechanistic understanding is not presented in the manuscript.
Abstract
L2-3: I suggest removing the second sentence of the abstract as this is just a repetition of the introduction and fits better there.
L 3: Start the third sentence with “We use nearly…”
L11: Please remove “strongly” before ”increase”.
L12 two points.
L14: I am not aware of any publication which shows predictability of the NAO several months ahead. The authors also do not present any publication on this so avoid such a statement in the abstract and give references in the main text if such publications exist.
L30-34: The authors give the impression that one needs to have several hundred years of data to estimate return periods of rare events, but in principle extreme value theory is developed to this for shorter data sets. It is its strength to use say 30 years of observations and estimate a 100-year return period. Clearly the uncertainty will be high, but it is possible. Please be more balanced here and say that longer time series will lead to a reduction of uncertainty.
L48 and elsewhere in the manuscript: Please change methodology to method.
L66 please change “creation” to “generation”.
L67: The authors noticed a certain problem of the data set used. Unfortunately, they did not assess whether a potential drift may impact their results. They can do this, and I think such an analysis would be beneficial for the publication.
L90: I checked the Degenhardt publication, and the seasonal predictability of windstorms is not overwhelming and over land not existent, so please clarify.
L114, L122 and L123: I am puzzled about that the data is somehow downscaled without saying with which method (dynamically with a regional climate model simulation or statically). I guess that the authors misuse the word downscaled here as they would also say from which resolution they downscale. So please clarify.
Data section: Another issue is that ERA5 is not well introduced and it remains unclear at which resolution ERA5 is used. In section 2 it sounds like that it is interpolated to the resolution of GloSea6 whereas in Fig6 the authors use it in 0.25 degree resolution. Why is this changed (it is maybe a bit unfair to sue ERA5 in higher resolution).
L145 do you use wind speed or wind gust?
L158: The sentence is unclear. Do you only calculate one SSI value per track and what happens if the cyclone track travels over several countries? In the SSI definition the authors use an area around the cyclone center, how is this split if the area covers more than just one country. To me the SSI calculation is rather unclear.
L171-184: Somehow I get the impression that the authors are not aware of the fact that the wind gust provided in ERA5 is a diagnostic variable which is mainly based on surface wind speed. So I expect that if the authors following the 2% criteria of Klawa and Ulbrich (2003) and then check in the wind speed that they find that the threshold is in the range of 96th to 98th. In principle it must be 98th percentile and as gust and windspeed are strongly linked it must be also 98th percentile for wind speed.
Table 1: you show wind speed not gust correct?
L185-199: The description of the bias correction is unclear. It sounds like that the authors apply the quantile mapping method between ERA5 wind speed and ERA5 wind gust and use the coefficients in GloSea6. In this case no bias of GloSea6 is corrected. The authors only made a scaling to obtain from wind speed gust-based SSI, but then name it scaling. IF you would like to do a bias correction you need to estimate coefficients (of regularly spaced quantile intervals) between GloSea6 wind speed and ERA5 wind speed and then scale it to ERA5 gust.
L200: The definition of NAO is not fully clear; do you normalize the mslp difference or the northern and southern center separately. Please clarify which definition you use, is it the one of Hurrell (1995)?
L206-7: The intro sentence is not needed.
Section 4: Here I recommend enhancing the evaluation to the question of whether the 215-day predictions drift and whether such a drift impact the results presented. For this I suggest to analyses the first 1/3 of the 215 days and contrast the results with the last 1/3 of the 215 days. Do the biases, shown in Figure 2, change? What about Fig 5, do we see the increases if we restrict to the first 1/3 and the last 1/3?
L215: I find the track changes quiet string, the storm track is more zonal in GloSea6 compared to ERA5, Changes are in the order of 10%. Please apply a significance test in Fig 2 so the reader concentrates on the important biases.
L234: remove comma after (section 3.2.1)
L235 Remove “here”.
Overall, I find a lot of bracket in the manuscript, maybe the authors can remove a couple or avoid them by making real sentences. It would increase readability.
Section 5: Form me this is more an evaluation section so merge it with section 4 and make it a subsection 4.3
278: Please remove “here”
Fig 4d and 5 d Please change the unit to km/h
L284: I do not see a 36 h increase it looks like a 6 h increase in Fig 5a.
L285: according to Fig 5b I would say it is rather a 20hPa increase of the pressure anomaly.
L289-293: This is an introduction to the results part but only section 6.1 is introduced, please do this with 6.2 and 6.3 (I suggest renaming section 7 to subsection 6.3).
L299-301: I am puzzled here, why do the authors need to apply an additional “bias” correction if the used a bias correction already. To me this is a sign that something in the proposed method is wrong or at least not well explained.
L306-311: The different behavior in mountain regions is again weird as a good bias correction on the region level show be able to correct for this issue.
Fig. 6 Why do the authors suddenly use ERA5 in a higher resolution? I think this is an unfair comparison.
L322-331: To me the interpretation of Fig 6 c with the single event Daria is unclear. The leveling off in Fig 6 c and also in the other examples looks like a resolution issue, I speculate that SSI is somehow restricted to a certain threshold give the resolution of GloSea6. So to me this looks like more of a limitation of the data set.
Section 7: I suggest renaming it 6.3 and avoid 7.1 as only subsection which make structurally no sense.
L358-371: I am puzzled about the paragraph. I thought the entire procedure suggested enables us to see whether season forecasts can be used to enhance the data base so that unseen extremes can be assessed. But part of the paragraph read like the author doubt their own method and biases are too strong.
L370. How do you come to the conclusion that the impact is 1.5 time stronger? How is this estimated.
L375: For me this is a strange result and may show that the bias correction has not worked probably. The Authors state that they find in 672 years 600 events which exceed the strongest of ERA5 (which has a 50 year span). This seems to be very unlikely and hints more that there is a substantial bias over GB.
L384: please remove this line.
Fig 8 and Fig 9. Labels are too small, please revise.
Conclusions: I miss to see a discussion of limitations, e.g. the still rather coarse resolution of the data set (though I agree it is better to most of the climate models used). Then there is also the problem of wind gust parameterization also in ERA5 which could be mentioned here.
Citation: https://doi.org/10.5194/egusphere-2024-686-RC1 -
CC1: 'Comment on egusphere-2024-686', Gregor C. Leckebusch, 24 Jun 2024
reply
The manuscript describes the use of seasonal hindcasts to enhance the understanding of extreme European windstorm impacts. The manuscript claims to go beyond existing literature. Nevertheless, it should be noted that this approach was probably firstly introduced in Walz & Leckebusch (2019) and further related work to this wider topic is published in Walz et al. (2018), e.g. for the link to large-scale modes.
Walz and Leckebusch (2019) investigated the feasibility and added value of using the seasonal hindcasts of the ECMWF System 4 as a hazard event set for European winter windstorms
damage calculations. The windstorms are identified for every ensemble member and every year by an objective windstorm tracking algorithm. The damages are calculated directly from the obtained wind footprints via the open source natural catastrophe damage model CLIMADA for Germany, the UK, France and Spain and compared to the loss from ERA-Interim. The results show that the ensembles of losses in System 4 nicely capture the inter-annual loss variability of the reanalysis. Due to more than 1,500 years of “virtual reality” windstorm data from the hindcasts, the return levels of extreme losses can be estimated fairly accurately. Based on System 4, the losses in the scale of 1990 (January, February, March and December including the prominent windstorm Daria) represent a 20-year event in Germany whereas they represent a 100-year event for the UK. Thus, a considerably shorter return period compared to return periods calculated from ERA-Interim alone.
Further they investigated the link between the annual losses and large-scale drivers derived from mean-sea-level-pressure (MSLP) data in System 4. They could show that within System 4 there is a significant link between increased loss potentials for strongly positive North Atlantic Oscillation (NAO) phases for Germany and the UK as well as a reduced loss potential for Spain. The link between the other analysed indices is weak bar the East Atlantic (EA) pattern index. Thus, if the NAO in System 4 is correct it can be assumed that the windstorms in System 4 are useable. If this premise is given their study shows that the loss estimates and ultimately the return levels of losses from System 4 can be used in an operational way.Also, while Klawa and Ulbrich (2003) introduced the notion of the 98th percentile for loss estimates, the SSI was introduced in Leckebusch et al. (2008), as an objective (!) identification and tracking approach to assess the severity per event without arbitray assumption (e.g. for all days of an event the windspeeds in a state or a fixed radius are used). Details on the differences are especially important for the assessment of return periodes or the wider physical characteristic of the event (e.g. related to the SSI footprint as introduced in Leckebusch et al. (2008)).
The manuscript also comments on the events deduced in Osinski et al. (2016). Here, it seems the authors assume that the pure event-set events identified and used in Osinski et al. (2016) would be similar to observations: "As such, the windstorms produced are not independent from observations and are somewhat constrained to the climatology of the period (e.g. the SSTs)." While the latter is of course correct (but also applying to the seasonal hindcasts in principle), it should be noted that Osinski et al. (2016) developed especially a method to identify and utilise pure ensemle events to seperate from just similar events as apparent in the reality/observations. Please confer their Fig. 10 for details.
These three aspects should be clarified, before publication, idependent from any in-depth review process.
References:
Walz, M.A., M.G. Donat and G.C. Leckebusch, 2018: Large‐Scale Drivers and Seasonal Predictability of Extreme Wind Speeds Over the North Atlantic and Europe.
Journal of Geophysical Research – Atmospheres; 123 (20), 11,518-11,535. https://doi.org/10.1029/2017JD027958Walz, M.A., and G.C. Leckebusch, 2019: Loss potentials based on an ensemble forecast: How likely are winter windstorm losses similar to 1990?
Atmospheric Science Letters, Volume: 20, Issue: 4, Article Number: UNSP e891. https://doi.org/10.1002/asl.891.Citation: https://doi.org/10.5194/egusphere-2024-686-CC1 -
RC2: 'Comment on egusphere-2024-686', Anonymous Referee #2, 27 Jun 2024
reply
As previously noted by other commentators the manuscript overstates the significance and novelty of the work. Past works mentioned in previous comments have considered larger datasets or unprecedented storms and it is not clear enough what new contribution is made here.
I also feel the merits of seasonal forecasts are overstated in comparison to climate models given the track density biases shown which peak in the most relevant areas for loss modelling.The climatological period of 1993-2016 is quite short and so although a large number of forecast winters are developed they are based on a narrow window in climatology terms. This is not an insurmountable issue but is due some consideration particularly when examining the effect of the NAO.
There is some merit in the presentation of the results to be useable by the insurance industry and there is potential for a collaboration to develop a validation dataset which can be more useable than say pure ERA5 but failure to take a holistic view of existing work risks undermining this.
I have some concern about the jump from wind SSI to gust SSI - the end result is surely very sensitive to these values. This is given some discussion in relation to Austria.
Citation: https://doi.org/10.5194/egusphere-2024-686-RC2 -
RC3: 'Comment on egusphere-2024-686', Anonymous Referee #3, 28 Jun 2024
reply
The study uses seasonal hindcasts for the preparation of an event set of European winter wind storms. The manuscript assesses storm events of ERA5 reanalysis and the GloSea hindcast using a storm severity index (SSI) which is motivated by the potential loss caused by the storm. The authors are using the UNSEEN method to analyse unprecedented wind storm events. Both, a return level analysis and the assessment of unprecedented events are done for individual countries over Europe. Finally the influence of the NAO on the storm events is investigated.
The manuscript is a revised version after first minor revision due to the necessity to better explain the novelty and added value of the study. In the current version of the manuscript, the authors highlight aspects where the manuscript goes beyond literature (line 65). I would disagree in this argumentation about novelty of the study and the contribution for further application.
The manuscript does not present new ideas. As mentioned in the manuscript itself, the study of Osinski et al. (2016) already investigated an event set of European wind storms although they are not using seasonal hindacsts. It is also not the first time a seasonal hindcast is analysed focussing on wind storms (Befort et al. ,2019; Degenhardt et al., 2023). It is not the first time seasonal hincasts are used to prepare an event set of winter wind storms (Walz and Leckebusch, 2019). While the first studies are mentioned in the manuscript, it is not the case for the latter. This was already mentioned in a discussion comment (https://doi.org/10.5194/egusphere-2024-686-CC1). Since the authors are using a different seasonal prediction system and a different method to define storms and footprints as Walz and Leckebusch (2019), it is still valid to perform this study. But the advantages/disadvantages and the added value of the study has clearly to be discussed, which is completely not done so far.
The definition of the SSI in sec. 3.2 is not properly introduced and cited in the manuscript. Klawa and Ulbrich (2003) introduce a loss index by means of the cubic exceedance of the 98th percentile of wind speed. The SSI as integrated measure of the severity of an event was introduced by Leckebusch et al. (2008). A loss index combined with population density was used by Pinto et al. (2012; https://doi.org/10.3354/cr01111). All indices use normalization by the 98th percentile which automatically reduces bias, which is not done by the SSI in the manuscript (eq. 1). It is completely valid to do it another way but the argumentation is not clear to me. The authors write it is in accordance to vendors cat models (l. 143). Is there a reference? Is this definition a better description of loss-wind relationship?
The bias adjustment of the SSI is crucial because of the sensitivity on unprecedented storms. The authors start with the same 20m/s threshold and further calculate an individual wind speed threshold for each country (with the argument of different wind distributions and wind-gust relationship). Is there any reference showing the gain of this approach in comparison to the cited literature (Klawa and Ulbrich, 2003) where the distribution (98th percentile) is directly used for normalization?
In sec. 6.1 the authors compare return periods of storms (w.r.t. SSI) within ERA5 and GloSea. They shift the curves in order to match a 10y return level. Of interest is “[…] the right tail of the distribution that identifies the extreme storms, rather than the specific SSI values represented in GloSea6 (which will contain some model bias). […]” (l. 299). This shift and match to a 10y return level seems to be arbitrary. Furthermore, the SSI itself in GloSea is very important for the definition of unprecedented storms. The authors see the problem of model bias of GloSea but argue about the general agreement of the distributions of GloSea and ERA5 (l. 361). For the further usage of the results and application for end users (l. 74), choices in the methods as the way of bias adjustment or not adjustment, shift of distribution, has to be well justified.
Given the arguments about novelty of the study and further the description of the method in the context of state-of-the-art literature, I suggest to do a proper literature research and reconsider or at least explain the reasons for the way of SSI definition and bias adjustment. The chosen way and added value should be explained in the context of current literature. I suggest to reject and potentially re-submit the manuscript.
Citation: https://doi.org/10.5194/egusphere-2024-686-RC3
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 231 | 72 | 16 | 319 | 8 | 9 |
- HTML: 231
- PDF: 72
- XML: 16
- Total: 319
- BibTeX: 8
- EndNote: 9
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1