Fighting big data and ensemble fatigue in climate change impact studies: Can we turn the ensemble cascade upside down?

Climate change impact modellers consider the availability of large ensembles of climate model results more and more as problematic. They experience big data or ensemble fatigue and face computational limits. This study proposes an ensemble design approach based on clustering of the climate model skill, climate change signals and statistical downscaling skill, and investigates its potential for ensemble size reduction. The proposed approach is demonstrated for river and urban hydrological impact studies in Belgium, considering the average winter (summer) precipitation amount and extreme daily winter (summer) precipitation amount with a 10‐year return period. The analysis starts from an original 240 membered multi‐ensemble (48 climate models and 5 statistical downscaling methods) and is reduced to 8 (12) members for the average seasonal winter (summer) precipitation amount and 18 (22) for the extreme daily winter (summer) precipitation amount. The range of the impact results by the original multi‐ensemble is generally preserved. However, in some cases, the reduced ensemble shows biased impact results. The cluster analysis confirms the dependence between statistical downscaling methods and points to the interaction between climate models and statistical downscaling methods. (a) Traditionally, the total ensemble includes GHS, several GCMs and different SDMs. Doing so, the total ensemble size builds up and this build‐up is referred to as the ensemble cascade. (b) By combining validation and inter‐dependence analyses, as proposed in this study, the ensemble cascade is turned upside down and the total ensemble size is reduced.