SEAS5 Counterfactual Analysis
The previous counterfactual analysis demonstrated how small changes in the location and intensity (e.g., wind speed, orientation, size, etc.) of historical events could have had material changes in insured losses. But it is also important to realise that small changes, or perturbations, in initial atmospheric conditions could have led to the development of different storms altogether during each season.
Ensemble numerical weather prediction (NWP) is a useful tool in quantifying the probability of alternative, or counterfactual, storm histories. As part of a long-standing research collaboration with CoreLogic, the School of Geography, Earth and Environmental Sciences at the University of Birmingham created a catalogue of extratropical cyclone footprints using the ECMWF’s seasonal forecast ensemble system, SEAS52. Each year in the 36-year span from 1981 to 2016 contains 51 simulated ensembles. CoreLogic has exclusive access to this synthetic storm footprint catalogue.
For a given year, each ensemble was initialized by adding perturbations to the ECMWF’s ERA‐interim reanalysis valid on the 1st of August. The reanalysis represents the best guess of the atmospheric state using a combination of observations and model prediction. Storms were tracked covering the peak season for European windstorms, i.e., December to February (DJF). The SSI of each resulting storm was then calculated.
An SSI-based exceedance probability curve was calculated using the resulting 1,836 (36 years with 51 ensembles per year) year-long simulation catalogue of windstorm events. The estimated rate, or frequency, of a given SSI return period (RP) was analysed as a function of year.
Figure 8a shows the rate of storms with a 10-year return period is shown for all land points within the West Europe region, defined by areas between 44°N and 55°N and between -15°E and 15°E which encompasses the key exposure subject to Eurowind™ risk (including the significant cities of London, Paris, Amsterdam, Munich, Berlin and Zurich). The annual estimates of the rate (grey line) and the 5-year centred mean (blue line) are shown. The standard error in this mean value is shown by the blue shading. If the risk was independent of time, the rate at each year would equal 0.1. Instead, the rate exhibits peaks centred on the early 1990s, 2000s, and the end of the period in 2016.
Figure 8: The rate of storms produced by the SEAS5 ensemble over the seasons spanning 1981-2016, for event SSI RP levels of (a) 10 years and (b) 60 years. Source: CoreLogic, 2024
The rate of 10-year RP events was relatively high during the 2006-2007 season, which is consistent with the associated occurrence of Kyrill (estimated to be an approximately 10-year RP event within this region). There is a significant decrease in rates centred on the 2008-2009 season, although it should be noted that the standard error is also relatively high.
The time series for an return period of 60 years (i.e., close to the maximum historical RP when assuming a simple empirical estimate of 63 years) is shown Figure 8b. The peak at 1999-2000 is consistent with the historical maximum recorded in that season, although a peak in not observed for the 1989-1990 season. The significant decrease in rates centred on 2007-2008 is consistent with a lack of extreme events in the historical period up until at least 2010-2011. The return to high values by the end of the sampled period suggests that extreme events could have occurred within the last seven years.
The variability exhibited by these time series can be explained, to some extent, by the NAO. The NAO is known to have a strong influence on storm activity in the North Atlantic-European sector. Its positive phase tends to be marked by a poleward shift of storm tracks over Europe, and its negative phase is marked by a southward shift. Hence, a positive NAOI is correlated with in increased rate in the West Europe region considered, and a negative NAOI is correlated with a decreased rate. Figure 9 shows a time-series of NAOI by season obtained from NOAA, along with a centred 5-season mean.
Figure 9: NAO signal from seasons 1980-1981 to 2016-2017. Data source: NWS SPC, 2024
The 1989-1990 and 1999-2000 seasons had a relatively high positive NAOI values, and the negative NAOI during 2009-2010 and 2010-2011 correlates with the reduced rates for both RP10 and RP60. The large positive NAOI values at the end of the period agree with the increased rates.
While a negative phase of the seasonal NAO is a strong indicator of windstorm activity over southern Europe, it is not perfect and does not preclude significant storms occurring in the northern part of the model domain during this phase. The NAO phase is volatile and can change mid-season, which could influence intra-seasonal windstorm activity.
The results of the SEAS5 ensemble data counterfactual analysis increase the certainty that natural variability plays a large role in European windstorm activity. The utility of this approach is that, rather than performing a counterfactual analysis on storms that did occur like what was studied in the previous analyses, this looks at storms that did not occur but could have occurred based on seasonal forecasting. Viewing the storms through this lens provides additional insights that could not be gained by looking solely at historical events.
The Risk Remains High
European windstorm risk remains a top concern within the industry on par with the traditional primary perils that global (re)insurers model on a regular basis. In this white paper, a series of counterfactual analyses using the CoreLogic Eurowind™ model and SEAS5 ensemble dataset demonstrated that European windstorm risk has not significantly diminished. The relatively low losses of recent years can be explained in terms of natural variability of hazard which dominates over any secular decrease in windstorm rates. Just small perturbations to historical storm parameters or meteorological conditions could produce industry-altering European windstorm events.
Fortunately, the industry is prepared for the occurrence of another windstorm loss on, or exceeding, the level of Kyrill. Nevertheless, such an event will be highly disruptive given the relative inactivity of recent years. Allowing European windstorm risk to slide to an area of lower concern could prove catastrophic.
The CoreLogic Eurowind™ Model demonstrates good skill at as a robust representation of today’s climatology with respect to European windstorms.
One area not studied in this white paper is how future climate conditions may affect European windstorm. This area of research will be featured in a future white paper to be released in Spring 2024. The upcoming white paper will show how Eurowind™ remains fit for purpose under conditions set by the possible pathways of future Climate Change.
Please email your CoreLogic Client Success Manager or HazardRisk@corelogic.com for any questions regarding this white paper, the CoreLogic Eurowind™ Model or European windstorm risk to your business.
Visit www.hazardhq.com for updates and information on catastrophes across the globe.
About CoreLogic
CoreLogic, the leading provider of property insights and solutions, promotes a healthy housing market and thriving communities. Through its enhanced property data solutions, services, and technologies, CoreLogic enables real estate professionals, financial institutions, insurance carriers, government agencies and other housing market participants to help millions of people find, acquire, and protect their homes. For more information, please visit corelogic.com.
©2024 CoreLogic, Inc. All rights reserved. While all of the CoreLogic content is believed to be accurate, CoreLogic makes no guarantee, representation, or warranty, express or implied, including but not limited to as to the completeness, accuracy, applicability, or fitness, in connection with the content or the product and assumes no responsibility or liability whatsoever for the content or the product or any reliance thereon. CoreLogic® and Eurowind™ are the trademarks of CoreLogic, Inc. and/or its affiliates or subsidiaries.