Colloque / Séminaire

Reinsurance companies use catastrophe (CAT) models in order to predict the statistics of extreme events and their physical and financial impacts. Modern CAT models use climate models, associated to statistical models, in order to predict return period curves, and large samples of impact pattern associated to rare events. The rarest events have the highest impact and account for most of the insurance liability. Such events have never been observed in historical data before they occur and are extremely difficult to sample using climate models. Indeed they are so rare, that the computational time of using the best climate models is often prohibitively costly. This difficulty is a major drawback of current approaches.
Heat wave are the climate events that have the most drastic impact on societies and ecosystems. They are by far the most lethal type of weather phenomenon, overall. For instance the 2003 heat waves over Western Europe caused 70 000 fatalities, and future heat waves may lead to the displacements of hundreds of millions of people in the India–Bengladesh area before the end of the century. We will focus on heat waves as a paradigmatic example.
In the recent past, new theoretical and numerical tools have been developed in the statistical physics community, in order to specifically study such rare events. Some of those approaches are based on large deviation theory for complex dynamical systems. Using a large deviation algorithm, we studied the probability of extreme heat waves in a comprehensive climate model. At a fixed numerical cost, several hundreds more heat waves are observed than in a control run. The thousands of sampled extreme heat waves open the door to their dynamical studies, precursor, and fluctuation paths, in a way that can not be foreseen using conventional tools based on direct numerical simulations. Moreover extreme events that can not be observed in the climate model at a reasonable numerical cost can now be studied. This new tool opens completely new perspectives for the study of climate extremes.
We will discuss the potential interest of our approach for improving the statistics of the most rare events in CAT models, in a near future.

The initial part of this research has been funded by the AXA research grant.