Emulators of Climate Model Output

Researchers’ and decision-makers’ demand for climate information has outpaced the ability of computationally intensive Earth system models (ESMs) to provide targeted climate projections, particularly when specific output for specific needs is required. Emulators of ESMs—significantly more efficient computationally—aim to produce such information and have seen an accelerated period of development. Emulators’ latest generation greatly varies in method, complexity, requirements, and outputs. Some emulators produce only patterns of average quantities, targeting climate responses to anthropogenic forcings. Others simulate quantities at high temporal and spatial frequency, accounting for the climate system internal variability. We survey and categorize different methods; their advantages and limitations, including statistical approaches of various complexity; and machine learning methods. We discuss how a choice of emulator, based on different methods, inputs, and outputs available, might be or not be fit for purpose, for climate and sustainability science. We identify gaps and research needs informing future developments.