The Coupled Model Intercomparison Project (CMIP): Phase 3 vs. Phase 5
Knutti, R. and Sedlacek, J. 2013. Robustness and uncertainties in the new CMIP5 climate model projections. Nature Climate Change 3: 369-373.
In a study devoted to addressing this question, Knutti and Sedlacek performed what they have described as "a first comparison between projections from CMIP3 and CMIP5," in order to see what has happened so far in this regard. And what did that comparison reveal?
The two Swiss researchers report that (1) "the model spread relative to the model mean change for a given scenario is similar or in some cases slightly larger, implying that the models have not converged in their projections," that (2) "the local model spread has not changed much despite [i] substantial model development and [ii] a massive increase in computational capacity," that (3) there is "little evidence from CMIP5 that our ability to constrain large-scale climate feedbacks has improved significantly," that (4) "model mean patterns of temperature and precipitation change … are remarkably similar in CMIP3 and CMIP5," that (5) there is "little increase in model agreement in CMIP5 for precipitation changes," that (6) "robustness over land is slightly higher but also similar in CMIP3 and CMIP5," that (7) "the new models are likely to be better in the sense of being physically more plausible, but it is difficult to quantify the impact of that on projection accuracy," that (8) "if the past is a guide to the future then uncertainties in climate change are unlikely to decrease quickly, and may even grow temporarily," and, last of all, that (9) "it is likely that impact-relevant predictions, for example of extreme weather events, may be even harder to improve."
Clearly, the models still have a way to go, especially when it comes to properly projecting how Earth's climate and weather may behave in the future, as the air's CO2 content continues to rise.