A Bayesian spatial factor analysis approach for combining climate model ensembles

Global and regional climate models are large computer models that simulate the Earth's climate system and are used extensively in climate change research. Climate models are based on natural laws of physics, chemistry, and fluid dynamics. However, different inputs and parameterizations lead to different climate projections. To help account for inherent uncertainties in the process, collections of climate models, called ensembles, are often used. This paper proposes a method for combining climate ensemble output using a spatial confirmatory factor analysis model to characterize modes of similarity among ensemble members. The proposed model uses both Bayesian and spatial methods to estimate a common climate factor as well as unique factor loadings for each ensemble member. These spatial factor loadings indicate the degree of agreement for each member with the common climate factor. The model is applied to the North American Regional Climate Change Assessment Program ensemble, using both winter surface temperature and winter precipitation. In both cases, the spatial confirmatory factor analysis model finds areas of disagreement among ensemble members where no suitable consensus can be obtained. Copyright (C) 2014 John Wiley & Sons, Ltd.