Multimodel ensembling has been widely used to improve climate model predictions, and the improvement strongly depends on the ensembling scheme. In this work, we propose a Bayesian neural network (BNN) ensembling method, which combines climate models within a Bayesian model averaging framework, to improve the predictive capability of model ensembles. Our proposed BNN approach calculates spatiotemporally varying model weights and biases by leveraging individual models’ simulation skill, calibrates the ensemble prediction against observations by considering observation data uncertainty, and quantifies epistemic uncertainty when extrapolating to new conditions. More importantly, the BNN method provides interpretability about which climate model contributes more to the ensemble prediction at which locations and times. Thus, beyond its predictive capability, the method also brings insights and understanding of the models to guide further model and data development. In this study, we apply the BNN weighting scheme to an ensemble of CMIP6 climate models for monthly precipitation prediction over the conterminous United States. In both synthetic and real case studies, we demonstrate that BNN produces predictions of monthly precipitation with higher accuracy than three baseline ensembling methods. BNN can correctly assign a larger weight to the regions and seasons where the individual model fits the observation better. Moreover, its offered interpretability is consistent with our understanding of localized climate model performance. Additionally, BNN shows an increasing uncertainty when the prediction is farther away from the period with constrained data, which appropriately reflects our predictive confidence and trustworthiness of the models in the changing climate.

Despite the necessity of Global Climate Models (GCMs) sub-selection in the dynamical downscaling experiments, an objective approach for their selection is currently lacking. Building on the previously established concepts in GCMs evaluation frameworks, we relatively rank 37 GCMs from the 6th phase of Coupled Models Intercomparison Project (CMIP6) over four regions representing the contiguous United States (CONUS). The ranking is based on their performance across 60 evaluation metrics in the historical period (1981–2014). To ensure that the outcome is not method-dependent, we employ two distinct approaches to remove the redundancy in the evaluation criteria. The first approach is a simple weighted averaging technique. Each GCM is ranked based on its weighted average performance across evaluation measures, after each metric is weighted between zero and one depending on its uniqueness. The second approach applies empirical orthogonal function analysis in which each GCM is ranked based on its sum of distances from the reference in the principal component space. The two methodologies work in contrasting ways to remove the metrics redundancy but eventually develop similar GCMs rankings. While the models from the same institute tend to display comparable skills, the high-resolution model versions distinctively perform better than their lower-resolution counterparts. The results from this study should be helpful in the selection of models for dynamical downscaling efforts, such as the COordinated Regional Downscaling Experiment (CORDEX), and in understanding the strengths and deficiencies of CMIP6 GCMs in the representation of various background climate characteristics across CONUS.