Apparent limitations in the ability of CMIP5 climate models to simulate recent multi-decadal change in surface temperature: implications for global temperature projections

Observed surface temperature trends over the period 1998-2012/2014 have attracted a great deal of interest because of an apparent slowdown in the rate of global warming, and contrasts between climate model simulations and observations of such trends. Many studies have addressed the statistical significance of these relatively short-trends, whether they indicate a possible bias in the model values and the implications for global warming generally. Here we re-examine these issues, but as they relate to changes over much longer-term changes. We find that on multi-decadal time scales there is little evidence for any change in the observed global warming rate, but some evidence for a recent temporary slowdown in the warming rate in the Pacific. This multi-decadal slowdown can be partly explained by a cool phase of the Interdecadal Pacific Oscillation and a short-term excess of La Nia events. We also analyse historical and projected changes in 38 CMIP climate models. All of the model simulations examined simulate multi-decadal warming in the Pacific over the past half-century that exceeds observed values. This difference cannot be fully explained by observed internal multi-decadal climate variability, even if allowance is made for an apparent tendency for models to underestimate internal multi-decadal variability in the Pacific. Models which simulate the greatest global warming over the past half-century also project warming that is among the highest of all models by the end of the twenty-first century, under both low and high greenhouse gas emission scenarios. Given that the same models are poorest in representing observed multi-decadal temperature change, confidence in the highest projections is reduced.