|
Abstract:
|
Atmosphere-ocean circulation models are widely used to understand past climate and to project future climate change. Models within an ensemble (such as the collection of models used by the IPCC AR4) will have varying degrees of similarity for different outputs of interest. Unlike previous approaches for optimal modeling averaging, this presentation considers an alternative view of multi-model ensembles. For use with the North American Regional Climate Change Assessment Program (NARCCAP), multivariate statistical models are employed to characterize modes of similarity within the members of an ensemble. Specifically, we propose a spatially correlated latent variable model which facilitates the exploration of when, where, and how regional climate models are similar, and what factors best predict observed locations of model convergence.
|
