A Hierarchical evaluation of IPCC AR4 coupled climate models with systematic consideration of model uncertainties

Abstract

The capability of reproducing observed surface air temperature (SAT) changes for the twentieth century is assessed using 22 multi-models which contribute to the Intergovernmental Panel on Climate Change Fourth Assessment Report. A Bayesian method is utilized for model evaluation by which model uncertainties are considered systematically. We provide a hierarchical analysis for global to sub-continental regions with two settings. First, regions of different size are evaluated separately at global, hemispheric, continental, and sub-continental scales. Second, the global SAT trend patterns are evaluated with gradual refinement of horizontal scales (higher dimensional analysis). Results show that models with natural plus anthropogenic forcing (MME ALL) generally exhibit better skill than models with anthropogenic only forcing (MME_ANTH) at all spatial scales for different trend periods (entire twentieth century and its first and second halves). This confirms previous studies that suggest the important role of natural forcing. For the second half of the century, we found that MME_ANTH performs well compared to MME _ ALL except for a few models with overestimated warming. This indicates not only major contributions of anthropogenic forcing over that period but also the applicability of both MMEs to observationally constrained future predictions of climate changes. In addition, the skill-weighted averages with the Bayes factors [Bayesian model averaging (BMA)] show a general superiority over other error-based weighted averaging methods, suggesting a potential advantage of BMA for climate change predictions.