Quantifying and understanding very rare climate extremes using ensemble boosting

In recent years, unprecedented temperature and precipitation extremes have been observed across the world. With further global warming, climate models project extreme events to get even more intense and likely break observational records by large margins. It is challenging to estimate how extreme climate events could get and to quantify the contribution of physical drivers in the future or even in the present climate? To address these questions, we introduce the ensemble boosting method, a model-based method that generates large samples of re-initialized extreme events in climate simulations. In doing so, the method provides physically consistent storylines of climate extremes that can be used to analyse the driving factors and estimate the very high return levels for the event type beyond observational records. We apply ensemble boosting to heat waves in the millennial pre-industrial control run, made with CESM1 and to heavy precipitation in the large ensemble near future simulations, carried out with CESM2. We find that individual members of the boosted ensembles can substantially exceed the most extreme heat and precipitation events over Europe and North America in the respective climatology. Furthermore, we show that estimated upper bounds of heat correspond to the statistical estimates by the generalized extreme value (GEV) distribution and regression models. Therefore, the framework of ensemble boosting might ultimately contribute to adaption and the stress testing of ecosystems or socioeconomic systems, increasing the resilience to extreme climate stressors.