Global sensitivity analysis for optimal climate policies: Finding what truly matters

Climate policy decisions rely heavily on the predictions of climate–economic models. These models are known to be sensitive to their initial assumptions and parameterization. Despite the broad literature exploring this sensitivity, universal, well-established practices are still lacking in this field. This paper endorses a holistic, global approach to sensitivity analysis and advocates it as an indispensable routine in climate–economic modeling. An application of a highly-efficient method of global sensitivity analysis to the seemingly simple case of the DICE model provides two fundamental insights. First, only global and comprehensive—as opposed to local or selective—sensitivity analysis can deliver a full and trustworthy picture of the effect of parameters' uncertainty on the model's solution. Second, a comprehensive decomposition of the uncertainty in the model's output is achievable at modest computational costs. Such decomposition is thus desired and potentially attainable for climate–economic models of higher complexity.