Hierarchical fusion of expert opinion in the Transferable Belief Model, application on climate sensitivity

This paper examines the fusion of conflicting and not independent expert opinion in the Transferable Belief Model. Regarding procedures that combine opinions symmetrically, when beliefs are bayesian the non-interactive disjunction works better than the non-interactive conjunction, cautious conjunction or Dempster's combination rule.Then a hierarchical fusion procedure based on the partition of experts into schools of thought is introduced, justified by the sociology of science concepts of epistemic communities and competing theories. Within groups, consonant beliefs are aggregated using the cautious conjunction operator, to pool together distinct streams of evidence without assuming that experts are independent. Across groups, the non-interactive disjunction is used, assuming that when several scientific theories compete, they can not be all true at the same time, but at least one will remain. This procedure balances points of view better than averaging: the number of experts holding a view is not essential.This is illustrated with a 16 experts real-world dataset on climate sensitivity from 1995. Climate sensitivity is a key parameter to assess the severity of the global warming issue. Comparing our findings with recent results suggests that, unfortunately, the plausibility that sensitivity is small (below 1.5C) has decreased since 1995, while the plausibility that it is above 4.5C remains high.