December is fast approaching, and in much of North America and Europe the crisp days and golden colors of autumn are giving way to a world of sparkling lights, frenzied shoppers, and the sense of merriment that comes with the onset of the festive period. At this time of year, an equally marked transition takes place within the catastrophe risk management community, as the Atlantic hurricane season closes and attention shifts to the onset of the December to February (or, in meteorological parlance, DJF) peak season for European windstorm risk.
However, three notable early season wind events have already impacted Europe during the earlier part of this current windstorm season. These are windstorm Xavier (October 5; Germany, Poland, and Czech Republic), ex-hurricane Ophelia (October 16; Ireland), and windstorm Herwart (October 29; Germany, Poland, Czech Republic, and Austria). The earliest of these, Xavier, formed just one day after the Category Five Hurricane Maria dissipated — at a time when the attention of the global insurance market was firmly focused on the other side of the Atlantic.
Tragically, each of this year’s three early-season events resulted in loss of life. However, across the three events, insured property losses will likely amount to around one billion Euros in total — a relatively gentle reminder that the insurance industry faces significant wind risk on both sides of the Atlantic for several months in autumn.
Although none of this season’s storms have so far been major insurance market events, it is by no means inevitable that early season activity will not lead to major market losses. Both the 2002-3 and 2013-14 windstorm seasons were dominated by billion-Euro events occurring in October (Jeanette and Christian, respectively). Looking further back in history, two of the six most damaging windstorms to have occurred between 1972 and the present (the period for which we have reliable observations) happened before the opening of the peak DJF windstorm season:
November 12-14, 1972: An Atlantic depression, named Quimburga by the Free University of Berlin, intensified over Ireland and U.K. as it moved into a region of sharper temperature gradients produced by a parent cyclone over the Norwegian Sea. By the time it left the east coast of England, it had developed into a significant storm producing some local damage in southern U.K., with a central pressure of 969 hectopascals (hPa).
However, during its short journey over the North Sea to the northwestern German coastline, Quimburga’s central pressure plunged to 953 hPa, fueled by unusually cold air from the parent cyclone and amplified by the seasonably warmer North Sea. Such explosive development led to gusts more than 40 meters per second in built-up areas in northern Germany and the former German Democratic Republic, and gusts above 35 meters per second over a large area including the Benelux region. Lower Saxony got hit hardest, leading to Quimburga’s more usual name: “Lower Saxony Storm”. This event caused more than 50 fatalities, and RMS estimates that if it were to occur today it would cause around €6 billion in insured losses in Europe, with around €5 billion in Germany alone.
October 15-16, 1987: A small disturbance formed in the Bay of Biscay between colder Arctic and much warmer sub-tropical air, then moved quickly to the northeast while rapidly deepening under a particularly strong jet stream. This poorly-forecast system, today known as the “Great Storm of 1987” or “Storm 87J”, attained an estimated minimum pressure of 953 millibars off the Brittany coast and went on to become the deepest measured October cyclone over England in well over 100 years.
It contained maximum onshore gusts of 40 to 50 meters per second in coastal parts of Brittany and southeast England, and over 35 meters per second in the Greater London area. In addition to the great damage caused directly by winds, there was indirect damage from massive tree fall and wind-borne debris, together with unusually high loss amplification as insurance companies struggled to cope with the volume of claims received. RMS estimate that if this event were to occur today insured losses would be around €7 billion.
Beyond reminding us to sharpen our focus on the European region during October and November, the occurrence of such autumnal windstorm activity is also material to our ability to forecast annual windstorm losses, as was outlined in a recent blog post by my colleague Stephen Cusack. Current meteorological research into forecasting North Atlantic seasonal storminess focuses on predicting the DJF-mean of the North Atlantic Oscillation (NAO) index. However, the above historical analysis makes clear that some of the most significant windstorm losses ever experienced occurred outside of this DJF window.
Given this, it should perhaps not be surprising that RMS research (available to RMS clients) shows that while there is reasonable correlation between the skillful forecast variable (DJF-mean NAO index) and insurance losses occurring during the same DJF period, this relationship breaks down when considering losses over 12 month periods most commonly seen for (re)insurance contracts. The correlation between the DJF-mean NAO index and calendar year losses for the period 1972 to present is around 0.13 for Europe (with only Norway and Austria showing correlation above 0.2). It is sobering to think how these correlations represent an upper limit for a hypothetical perfect forecast of the DJF-mean NAO.
The storms this past October remind us that huge storms have occurred outside meteorological winter in the recent past. These non-winter storms are part of the reason why DJF-mean NAO has very limited practical value for management of insurance industry windstorm risk.
You May Also Like
January 26, 2018
Friederike/David: The Day Our Luck Ran Out?
Little more than a week ago, I signed off my previous blog post, discussing storm Eleanor/Burglind, with the following thought:
As the number of minor windstorms impacting Europe this season grows, we are left to wonder how many more “near misses” can we experience before our luck runs out?
At the time, I further noted that:
One-day-out, the forecasts for Friederike were trending towards lower and lower severities
As an immature and fast-moving system, these forecasts were subject to high uncertainty
Eventual losses would be very sensitive to gust speeds experienced in urban areas near to the forecast footprint, such as Rotterdam and Dortmund
In the week since this post, it has become clear that our lucky streak has already come to an end. Storm Friederike (also named David by Météo-France) intensified towards the upper end of the forecast severities in Benelux and Germany, bringing strong gusts to highly populated areas and producing significant insurance losses.
Friederike: Meteorology and Impacts
Storm Friederike can be traced back to a weak near-surface cyclone which formed off the northeastern seaboard of the United States in the latter part of January 15, 2018. Upon moving into the northwest Atlantic during January 16, it encountered strong westerly winds formed by a deep low near Iceland and a strong anticyclone centered near the Azores. These strong steering level winds caused the system, still at this stage a weak cyclone, to speed across the Atlantic at slightly above 100 kilometers per hour.
Friederike’s location relative to the jet became more conducive for storm intensification during the afternoon of January 17. The system’s central pressure deepened by 15 hPa during the next 12 hours as it passed over Ireland and the U.K., first reaching its minimum pressure of just under 980 hPa while crossing the North Sea on its path toward mainland Europe (see figure 1 below).
Pressure gradients are only part of the wind story for this type of storm, and this very modest central pressure belied Friederike’s strength. The fact that it had recently scraped into the weather bomb category (dropping more than 24 hPa over the previous 24 hours), and the swiftness of its Atlantic journey were both significant contributors to its local winds.
Although Frederike’s translation speed may have slowed to around 75 kilometers per hour as it hit the Netherlands then Germany, this remained sufficient to produce very damaging gusts over areas of dense exposure.
Figure 1: Deutscher Wetterdienst (DWD) Surface Pressure Chart at 06:00 UTC on Thursday, January 18, 2018.Peak gusts of over 30 meters per second were felt widely in a 150 kilometer wide corridor stretching from East Anglia in the U.K. to Saxony in eastern Germany; crossing northern Belgium, southern/central Netherlands, and North Rhine-Westphalia. Critically, the strongest gusts in the storm encompassed several areas of high exposure. Stations close to Amsterdam and Rotterdam registered 34 meters per second; whilst gusts in the low-to-mid-30s meters per second were felt widely in North Rhine-Westphalia, around Münster and Düsseldorf.
RMS have released accumulation footprints and stochastic event selections for this storm (accessible on the Event Page on RMS Owl) to help our clients estimate likely levels of portfolio loss from this event. The stochastic event selections further help to understand the significant uncertainty that inevitably surrounds estimates of loss so soon after a European windstorm event. Despite the outstanding loss uncertainty, our modeling products show that this storm is very likely to cause industry property losses of greater than €1 billion across Europe.
Third-party estimates of loss from this storm also point to this storm being the first billion-Euro loss of the season. The GDV yesterday released an updated loss estimate of €1 billion for Germany alone, consistent with estimates of German loss produced by MSK Actuaries (€800 million) and Aon Benfield (€1 billion).
While Germany is the dominant contributor of loss to this event, the storm also caused significant damage in Benelux. In the Netherlands, the VvV issued an early loss estimate of €90 million for residential and auto lines (which could scale to around €150 million+ for all lines), and for Belgium, AG Insurance issued a loss estimate of €50 million.
All the evidence thus points to this being a significant event, likely to drive one of the largest losses seen in Europe over the last decade. However, it is important to note that the last decade has been a period of low windstorm activity in much of Europe (more on this below), and that this storm does not come close to the levels of loss caused by the largest historical European windstorms like 87J (1987), Daria (1990) and Lothar (1999). Friederike losses are also likely to fall a long way short of those experienced during storm Kyrill in 2007. Rather, it would be appropriate to consider this storm alongside moderate but significant loss events from the past decade such as Klaus (2009), Xynthia (2010), Christian (2013) and Mike-Niklas (2015).
2017/18 Windstorm Season: Losses Continue to Aggregate in Germany
Losses from storm Friederike are strongly concentrated in Germany, where three less severe storms have already caused insured losses this season: Xavier (October 5, 2017), Herwart (October 9, 2017) and Burglind (January 3, 2018). Given the level of expected damage from Friederike, aggregate losses for the season so far in Germany now approach the eightieth percentile of modeled storm seasons. Thus, if the rest of the season were to pass without further incident, it would have a return period of around five years in Germany.
Based on our experience this century, the current season perhaps feels more exceptional than a return period of five years. This cognitive dissonance can be resolved by considering the climate variability of European Windstorms.
Observed windstorm losses in Europe show significant variability, over a range of timescales from seasonal to multidecadal. The multidecadal variability in windstorm losses observed over the last 50 years is significantly larger than the level of future variability that can be confidently forecast for the next 50 years due to climate change — making this a critical topic for understanding European Windstorm risk. To assist our clients in this regard, RMS provides an additional rate set which samples this long-term climate variability, and have authored two research papers outlining the importance of short- and long-term climate variability for insurance risk management.
When considering the historical context for the current windstorm season it is natural to compare against our most recent experience (perhaps that period covered by our claims datasets), and to conclude that this season to date may be exceptional in Germany. However, in common with much of Western Europe, Germany was notably stormier in the 1970s to early 1990s than it has been since. This is reflected in estimates of historical insured losses, for instance the average annual loss (AAL) for the last 10 to 20 years in Germany is only 50 to 60 percent of the longer term AAL (1970s to the present).
We have no forecasting skill around this long-term variability of European Windstorm activity, and so we could return to a stormier period at any time. This means that sound risk management decisions must consider the full longer-term historical picture, despite the discomfort which comes from realizing that shorter claims datasets may not tell the full story. When we consider this longer-term picture, we can see that the 2017/18 season to date in Germany has been stormy, but not exceptional.
Over the course of this winter, RMS has been publishing a series of blog posts charting the progress of the 2017/18 European windstorm season. In October last year, we outlined our forecast for a stormier-than-recent season, before checking in with an update at the end of November following three notable autumnal windstorm events. Following a relative lull in activity in December, we used the fiftieth anniversary of the “Glasgow Hurricane” as a reminder of the potential impacts when strong windstorms directly hit major urban areas.
However, at the start of 2018 this lull was broken with the arrival of storm Eleanor (U.K. Met Office/Met Éireann) / Burglind (Freie Universität Berlin), motivating us to once again check-in on the progress of the storm season. Here we outline the meteorology and impacts of this latest storm, and discuss how it fits into the seasonal forecast issued at the start of the season.
The Story of Eleanor/Burglind
Storm Eleanor/Burglind became prominent just into 2018 on January 2, as a wave along the cold front of a strong primary low situated to the southwest of Iceland. Under the influence of a strong westerly jet stream, this wave developed into a rapidly intensifying secondary low which attained a minimum pressure of 966 hPa as it moved east across Scotland. As the storm continued to track to the east across the North Sea to Denmark, it caused strong winds to the south covering a broad swathe of western Europe. Storm winds were experienced in Ireland, U.K., France, Netherlands, Belgium, Luxembourg, Germany, and Switzerland; while high waves pounded the northern coast of Spain.
Figure 1: Weather chart for 00:00 UTC on January 3, 2018 (from Institut für Meteorologie, Freie Universität Berlin)The RMS Event Response process aims to provide our clients with near real-time insights into ongoing catastrophes, and to provide strong modeling deliverables to assist during major events. Our first job following an event is to assess the likely magnitude of insured damage to ensure that we respond appropriately to meet the needs of our client base. This is often a difficult process, subject to significant uncertainties, as in the immediate aftermath of a storm we must work with partial information — before all weather station data is available for review, and long before the full suite of insurance claims has been counted. However, in the case of Eleanor/Burglind, it quickly became apparent that the observed wind gusts were not indicative of a major loss-causing event.
The footprint for this storm was large and noisy. Peak gusts were typically below 30 meters per second (m/s), though some parts of Switzerland, eastern France, and southwest Germany experienced gusts slightly above 30 m/s. For comparison, gusts well in excess of 40 m/s impacted areas of significant exposure during some of the most extreme historical windstorms. Gusts of up to 48 m/s were measured around Paris and neighboring areas during storm Lothar in 1999; 44 m/s was observed near London during storm 87J; and gusts of 46 to 47m/s were observed throughout central Scotland during the 1968 “Glasgow Hurricane”. Gusts across the full extent of the Eleanor/Burglind footprint were also generally lower than those observed during recent storms where losses exceeded €1 billion, such as Emma and Xynthia.
Based on the early weather station data, we concluded that Eleanor/Burglind would cause widespread minor damage, with some pockets of more significant damage, and would not lead to the kind of aggregation of claims needed to form a major European windstorm loss. Assessment of news reports after the event confirmed that this was indeed the nature of the damage being observed on the ground, and RMS clients can refer to the RMS Event Summary on January 4 for more details of observed damage.
Reflecting on this event almost two weeks later, we now have additional tools at our disposal when trying to understand the impacts of this storm. Insured loss estimates have been issued by national insurance associations, which are collectively in line with the RMS assessment of insured loss from this event. In Germany, the GDV expect insured losses to be significantly below €500 million, consistent with other estimates of €200-300 million for that market; in France, the FFA issued a combined loss estimate from the Carmen and Burglind storms of €200 million; in Switzerland, media reports suggest losses of €60-90 million; and in the Netherlands, the Dutch Association of Insurers figure is €10 million. As these figures include flood losses, as well as storm Carmen losses in France, we believe that a fair wind-only estimate arising from these reports is around €400 million.
Furthermore, the RMS windstorm modeling team, based in major markets in Europe, has had wide-ranging interactions with the market since this event, revealing a consensus view that is broadly in line with these insurance association loss estimates.
The latest available information thus supports the early RMS view that, like storms Xavier and Herwart earlier in the season, Eleanor/Burglind was a notable storm without being a major event from an insured loss perspective.
Half-time Analysis of the 2017/18 Europe Windstorm Season
In line with the RMS season forecast, the current season is so far proving to be stormier than the recent average. However, as was outlined in the blog, the relationship between storminess and loss is an imperfect one. Thus far, despite being a fairly stormy season, estimated aggregate industry losses in Europe to date remain low compared to what might be expected given the long-term average annual loss (AAL) — although in Germany they may have aggregated to around half the long-term AAL at this point.
We are currently halfway through the peak December-January-February period of the 2017/18 season, and it only takes one major event to change the whole nature of the season from an insurance perspective. Over the last few days we have been monitoring another depression (named Friederike by Frei Universität Berlin) due to cross Europe in the next 24 hours. This is now forecast to be a relatively shallow depression moving quickly over Europe, reminiscent of storm Xavier (October 5, 2017).
Whether this storm produces slightly higher or lower losses than Xavier will be sensitive to the gust speeds experienced in urban areas within the forecast footprint, such as Rotterdam and Dortmund. It is interesting to note however that predictions of the severity of this storm have abated over the past couple of days, reflecting the high levels of forecast model uncertainty affecting the very mobile, developing systems responsible for most major loss events.
As the number of minor windstorms impacting Europe this season grows, we are left to wonder how many more “near misses” can we experience before our luck runs out?…
Peter is a Product Manager in the Model Product Management team, focusing on RMS Europe Windstorm Models. He joined RMS in 2012 and spent several years as an Account Manager, supporting RMS reinsurance intermediary clients, before taking up the role of Product Manager during the Version 16.0 Europe Windstorm clustering project. Peter joined RMS upon completion of his PhD in Earth Science from the University of Bristol.