Towards robust community assessments of the Earth's climate sensitivity

Abstract. The eventual planetary warming in response to elevated atmospheric carbon dioxide concentrations is not precisely known. The uncertainty in climate sensitivity (S) primarily results from uncertainties in net physical climate feedback, usually denoted as λ. Multiple lines of evidence can constrain this feedback parameter: proxy-based and model evidence from past equilibrium climates; process-based understanding of the physics underlying changes; and recent observations of temperature change, top-of-the-atmosphere energy imbalance, and ocean heat content. However, despite recent advances in combining these lines of evidence, the estimated range of S remains large. Here, using a Bayesian framework, we discuss three sources of uncertainty – uncertainty in the evidence, structural uncertainty in the model used to interpret this evidence, and differing prior knowledge and/or beliefs – and show how these affect the conclusions we may draw from a single line of evidence. We then propose strategies to combine multiple lines of evidence. We end with three recommendations. First, we suggest that a Bayesian random-effects meta-analysis be used to estimate the evidence and its uncertainty from the published literature. Second, we advocate that the organizers of future assessments clearly specify an interpretive model or a group of candidate models and, in the latter case, use Bayesian model averaging to more heavily weight models that best fit the evidence. Third, we recommend that expert judgment be incorporated via solicitations of priors on model parameters.