Within the actual climate change context and the natural hazard insurance system reform in France, the question of how to define a natural hazard event plays a fundamental role. Reimbursements have to be taken by the state  (via la Caisse Centrale de Réassurance) whenever the extreme event has a return period greater than 10 years. Today, such a decision is made subjectively. Defining an accurate criterion to classify an event into the natural hazard category is only possible if a network of measurements is sufficiently dense to allow an adequate probabilistic modeling of extreme events. For example, imagine that an insurance system is only responsible for events caused by strong winds (say, greater than 100 km/h). If someone demands to be reimbursed, this request has to be evaluated according to wind measurements recorded at nearby stations. The decision has to be made with an incomplete information and a probability approach is needed to take into account this uncertainty.

The goal of this project is to develop a statistical model to optimize the spatial design of a network of stations according to the distribution of extreme events. Two statistical research fields will be needed. Extreme Value Theory should provide the probabilistic structure to study extreme events. Spatial network design should bring the mathematical structure to optimize the network architecture. This project encompasses three scientific communities (math/stat, climate and economy/insurance) and this interdisciplinary effort should allow to develop novel network design tools in a probabilistic framework dedicated to extreme event analysis.