On the Reliability of Estimated Return Periods for Climate Extremes

Résumé fourni par la source

Abstract This study reflects on the probability of observing an extreme event of interest within a finite dataset, whether derived from observations or model simulations, to inform risk assessment or climate adaptation efforts. To do so, we adopt the concept of engineering reliability, which is defined as the probability that a system remains in a satisfactory state, to assess the reliability of extreme events inferred from a dataset, whether this is from observations or model simulations. This assessment links the number of available observations or simulations to the low frequency of the event, providing a quantitative measure of confidence in our ability to observe or simulate such events over a given time horizon. This approach offers a fresh perspective on the interpretation of an extreme event, where the rarity of an event is considered not only in terms of its frequency but also relative to the length of the dataset used. Our reflections aim to guide preparedness for future extremes and highlight the scientific challenges inherent in their prediction and projection. We emphasize that while large ensembles are essential to overcome the limitations of historical observations, they should be used with caution to avoid overconfidence arising from underlying modeling assumptions. Finally, we stress that statistical extrapolation, whether it is parametric or non-parametric, is unavoidable, as the link between event frequency and the definition of extremes cannot be eliminated.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.