Reducing uncertainty in climate and Earth system models requires combining advanced uncertainty quantification with optimal use of observations. Challenges remain in identifying dominant sources of uncertainty, calibrating model parameters in the presence of structural error, and designing observations that maximize model constraint. Recent advances in machine learning surrogates, such as perturbed parameter ensembles (PPEs) and statistical emulation, Bayesian inference, and Observing System Simulation Experiments (OSSEs) provide new opportunities for bridging models and observations to improve predictive skill. We welcome contributions spanning large ensembles and sensitivity analysis, statistical and machine-learning-based emulation, Bayesian calibration and history matching, emergent or process-based constraints, structural-error quantification, and optimal observational design, including studies that use multi-model ensembles to advance these aims, as well as integrated model–observation workflows applied from global to regional scales and across the full range of physical climate, atmospheric composition, and coupled Earth system processes.
Ken Carslaw, Hailing Jia