A Bayesian Hierarchical Model for Heterogeneous RCM–GCM Multimodel Ensembles

Abstract

Bayesian hierarchical model for heterogeneous multimodel ensembles of global and regional climate models is presented. By applying the methodology herein to regional and seasonal temperature averages from the ENSEMBLES project, probabilistic projections of future climate are derived. Intermodel correlations that are particularly strong between regional climate models and their driving global climate models are explicitly accounted for. Instead of working with time slices, a data archive is investigated in a transient setting. This enables a coherent treatment of internal variability on multidecadal time scales. Results are presented for four European regions to highlight the feasibility of the approach. In particular, the methodology is able to objectively identify patterns of variability changes, in ways that previously required subjective expert knowledge. Furthermore, this study underlines that assumptions about bias changes have an effect on the projected warming. It is also shown that validating the out-of-sample predictive performance is possible on short-term prediction horizons and that the hierarchical model herein is competitive. Additionally, the findings indicate that instead of running a large suite of regional climate models all forced by the same driver, priority should be given to a rich diversity of global climate models that force a number of regional climate models in the experimental design of future multimodel ensembles.