Effects of Ensemble Configuration on Estimates of Regional Climate Uncertainties

Internal variability in the climate system can contribute substantial uncertainty in climate projections, particularly at regional scales. Internal variability can be quantified using large ensembles of simulations that are identical but for perturbed initial conditions. Here we compare methods for quantifying internal variability. Our study region spans the west coast of North America, which is strongly influenced by El Niño and other large‐scale dynamics through their contribution to large‐scale internal variability. Using a statistical framework to simultaneously account for multiple sources of uncertainty, we find that internal variability can be quantified consistently using a large ensemble or an ensemble of opportunity that includes small ensembles from multiple models and climate scenarios. The latter also produce estimates of uncertainty due to model differences. We conclude that projection uncertainties are best assessed using small single‐model ensembles from as many model‐scenario pairings as computationally feasible, which has implications for ensemble design in large modeling efforts. Key Points Internal variability at regional scales can be characterized consistently with a large ensemble of one model or multiple small ensembles Estimates of uncertainties due to model disagreement are improved when multiple ensemble members are available for each model and scenario