Minimal CMIP Emulator (MCE v1.2): a new simplified method for probabilistic climate projections

Abstract

Climate model emulators have a crucial role in assessing warming levels of many emission scenarios from probabilistic climate projections based on new insights into Earth system response to CO2 and other forcing factors. This article describes one such tool, MCE, from model formulation to application examples associated with a recent model intercomparison study. The MCE is based on impulse response functions and parameterized physics of effective radiative forcing and carbon uptake over ocean and land. Perturbed model parameters for probabilistic projections are generated from statistical models and constrained with a Metropolis–Hastings independence sampler. Some of the model parameters associated with CO2-induced warming have a covariance structure, as diagnosed from complex climate models of the Coupled Model Intercomparison Project (CMIP). Perturbed ensembles can cover the diversity of CMIP models effectively, and they can be constrained to agree with several climate indicators such as historical warming. The model's simplicity and resulting successful calibration imply that a method with less complicated structures and fewer control parameters offers advantages when building reasonable perturbed ensembles in a transparent way. Experimental results for future scenarios show distinct differences between CMIP-consistent and observation-consistent ensembles, suggesting that perturbed ensembles for scenario assessment need to be properly constrained with new insights into forced response over historical periods.