Used in sectors as varied as meteorology, hydrology, demography, insurance, finance, and more, extreme value theory is a branch of statistics dealing with the extreme values in probability distributions. It describes the trajectory of localized physical phenomena over time. It allows, for example, for the study of the extreme values of temperature or precipitation as measured at a particular location.

Currently, no theory allows for the description of the chaotic dynamics of systems that are spatially extended; hence the difficulty of modeling extra-tropical storms (such as Xynthia or Lothar) that sweep several thousand square kilometers.

Seeking to better understand how atmospheric dynamics determines the recurrence of extreme events, researchers from LSCE have defined the conditions necessary for the application of extreme value theory to atmospheric models. They then transformed the theorems of chaos theory into digital tools, which they used to characterize, in terms of chaotic dynamics, the properties of atmospheric pressure fields from 1948 until today.

This approach makes it possible to determine the predictability and scarcity of atmospheric circulation "fields" before producing a weather forecast. Analyzing a daily field takes only one second on a laptop. Results for the North Atlantic are regularly updated and available online.