RMS

Ten Years After – How Windstorm Modeling has Matured since the 1999 European Storms

December 22, 2009

A decade has passed since extra-tropical windstorms Anatol, Lothar, and Martin struck Europe in 1999. On December 3, Anatol caused widespread damage across Denmark, costing insurers €1.4 billion (US$2.0 billion). On December 26, Lothar hit France, Germany, and Switzerland, with winds reaching more than 160 km/h (100 mph), followed 30 hours later by Windstorm Martin hitting southwest France with wind speeds in excess of 180 km/h (112 mph), and high winds extending into northern Spain, Corsica, and Italy. Together the storms caused insured losses of €6.5 billion (US$9.3 billion) in France (FFSA, 2000), with an additional €1.2 billion (US$1.7 billion) in Germany and Switzerland.

Lothar is the second most damaging European windstorm of the modern instrumental era, surpassed only by 1990’s Daria. The year 1999 remains the most damaging year for windstorms in Europe since 1900. A recurrence of the 1999 windstorms would bring insured losses of €1.5-2.0 billion (US$2.1–2.9 billion) in Denmark from a repeat of Anatol, and a combined Lothar and Martin loss of close to €10 billon (US$14.4 billion), three-quarters of which would occur in France.

The 2000 RMS reconnaissance report, Windstorms Lothar and Martin, discusses the impacts of Lothar and Martin on buildings, lifelines, casualties, and the re/insurance market. Ten years later, RMS reflects on the advances in storm modeling since these devastating storms, and offers insight into what the future of windstorm modeling may hold.

During the late 1990s, extreme weather risk in Europe was modeled using an approach that combined historical storm track data and weather observations with simple statistical modeling of the shapes and intensities of storms. Since 2000, fundamental changes in the approach to windstorm modeling have significantly improved the ability to build realistic stochastic event sets of potential intense storms, like Anatol, Lothar, and Martin.

The first of these advances concerns the use of constrained numerical models to model the full space-time structures of historical events. These methods, as used in the current generation of the RMS® Europe Windstorm Model, are similar to those used by meteorological agencies, but performed at a higher resolution. These reconstructions are then used as the basis for deriving the full range of storm characteristics, which are incorporated into the stochastic event set.

Over the past five years, through the relentless reductions in the cost of hardware, supercomputers have become available to catastrophe modeling agencies. Modelers can now harness the power of hundreds of CPUs to conduct free-running numerical simulations capable of generating thousands of years of realistic storm data.

Before the "raw" simulated data can be turned into the stochastic windstorm events in the model, it has to be corrected for its inherent biases. In the beginning, the size of the errors was prohibitive, but as modeling techniques have improved it became possible to remove the biases with aggressive calibration. Getting this calibration right is now one of the biggest challenges in windstorm modeling. However, the benefits of numerical models in producing more realistic storms outweigh the difficulties posed by the calibration.

The corrected results of the numerical simulations can then be used to generate the underlying data for the event sets in a catastrophe model. Event sets generated by free-running numerical models are used in the current RMS® U.S. and Canada Winterstorm Models, and will be incorporated into the 2011 release of the RMS® Europe Windstorm Model.

Significant advances have also occurred around the understanding and representation of windstorm clustering. Clustering reflects the tendency of storms, like Lothar and Martin, to occur closely together in time, and also have very similar tracks and intensities. Modeling clustering requires capturing both its temporal and spatial characteristics. While temporal clustering can be modeled statistically, the inclusion of spatial clustering or the similarity of storm characteristics in certain regions requires understanding the way in which sequences of storms are similar to one another. Both aspects of clustering were introduced into the RMS® Europe Windstorm Model via the RMS® Simulation Platform in 2008.

The fundamentals of how to model windstorm risk, particularly the use of free-running numerical models and the representation of clustering, are now well established. What comes next? Future improvements in windstorm catastrophe models will come from using even higher resolution numerical models that simulate the strongest and smallest storm structure even more realistically. Model uncertainty will also become captured by the use of multiple numerical models. As the ability to define and calibrate the hazard becomes more advanced, modelers will next begin to focus on how to bring the vulnerability assessment of wind damage in Europe up to the same levels of sophistication.