Uncertainty in climate change modelling: can global sensitivity analysis be of help?

The complexity of integrated assessment models (IAMs) prevents the direct appreciation of the impact of uncertainty on the model predictions. However, for a full understanding and corroboration of model results, analysts might be willing, and ought to identify the model inputs that influence the model results the most (key drivers), appraise the relevance of interactions and the direction of change associated with the simultaneous variation of the model inputs. We show that such information is already contained in the data set produced by Monte Carlo simulations and that it can be extracted without additional calculations. Our discussion is guided by an application of the proposed methodologies to the well-known DICE model of William Nordhaus (2008). A comparison of the proposed methodology to approaches previously applied on the same model shows that robust insights concerning the dependence of future atmospheric temperature, global emissions and current carbon costs and taxes on the model’s exogenous inputs can be obtained. The method avoids the fallacy of a priori deeming the important factors based on sole intuition.