Résumé
Rainfall extremes are analyzed by local variables such as annual or seasonal maximum of rainfall for various durations between one hour and one day, the excess over a high threshold, or the temporal series of storm events. Models from univariate or bivariate extreme value analysis or stochastic rainfall generators have been presented to describe the asymptotic behaviour of these variables. In the case of temporal storm events series, the temporal persistence has been modeled through a Markovian process. The models uncertainties have been analyzed too, with Bayesian or frequency analysis methods. Models have been validated with long series of daily rainfall, hourly series and storm events series, provided by Météo-France and Cemagref. In particular, it has been often noted that the distribution of the extremes has no bound and a heavier tail than the Gumbel or exponential distribution.