Estimating Return Periods for Extreme Climate Model Simulations through Ensemble Boosting

With climate change, heavy-impact extremes have become more frequent in different regions of the world. It is therefore crucial to further physical understanding of extremes, but due to their rarity in samples, this remains challenging.

One way to overcome this under-sampling problem is through Ensemble Boosting, which uses perturbed initial conditions of extreme events in an existing reference climate model simulation to efficiently generate physically consistent trajectories of very rare extremes in climate models. However, it has not yet been possible to estimate the return periods of these storylines, since the conditional resampling alters the probabilistic link between the boosted simulations and the underlying original climate simulation they come from.

Here, we introduce a statistical framework to estimate return periods for these simulations, by using probabilities conditional on the shared antecedent conditions between the reference and perturbed simulations. This theoretical framework is evaluated in and applied to simulations of the fully-coupled climate model CESM2. Our results show that return periods estimated from Ensemble Boosting are consistent with those of a 4000-year control simulation, while using approximately 5.8 times less computational resource use.

We thus outline the usage of Ensemble Boosting as a tool for gaining statistical information on rare extremes. This could be valuable as a complement to existing storyline approaches, but also as an additional method of estimating return periods for real-life extreme events.