A dependent multimodel approach to climate prediction with Gaussian processes.

Simulations of future climate contain variability arising from a number of sources, including internal stochasticity and external forcings. However, to the best of our abilities climate models and the true observed climate depend on the same underlying physical processes. In this paper, we simultaneously study the outputs of multiple climate simulation models and observed data, andwe seek to leverage theirmean structure aswell as interdependencies thatmay reflect the climate's response to shared forcings. Bayesian modeling provides a fruitful ground for the nuanced combination of multiple climate simulations. We introduce one such approach whereby a Gaussian process is used to represent a mean function common to all simulated and observed climates. Dependent random effects encode possible information containedwithin and between the plurality of climatemodel outputs and observed climate data. We propose an empirical Bayes approach to analyze suchmodels in a computationally efficientway. Thismethodology isamenable to theCMIP6model ensemble, andwe demonstrate its efficacy at forecasting global average near-surface air temperature. Results suggest that thismodel and the extensions it engenders may provide value to climate prediction and uncertainty quantification. [ABSTRACT FROM AUTHOR]