Contrast classes and agreement in climate modeling

Abstract

It's widely argued that agreement-or "robustness"-across climate models isn't a useful marker of confirmation: that the models agree on a hypothesis does not indicate that that hypothesis should be accepted. The present paper argues against pinning the failure of agreement-based reasoning on the models. Instead, the problem is that agreement is a reliable marker of confirmation only when the hypotheses under consideration are mutually exclusive. Since most cutting-edge questions in climate modeling require making distinctions between mutually consistent hypotheses, agreement across models is unlikely to help answer these questions. Because the problem here is agreement (and not the models), we should expect that there are other ways of using the models that are more informative and reliable.