Item

Abstract

The use of structured expert judgment is highly relevant in contexts where epistemic and aleatoric uncertainties are significant. This is particularly important in probabilistic volcanic hazard assessments, where decisions based on uncertain information are often critical.In expert elicitations, participants are asked to provide their uncertainty judgments by suggesting their own 5th, 50th and 95th percentile estimates of numerical values in each question. An advantage of this approach when using numerical models for probabilistic volcanic hazard assessment is that it is possible to obtain probability density functions for each input parameter as well as constraining their uncertainty ranges. More specifically, performance-based elicitations start with “seed” questions for determining experts’ uncertainty quantification skill. The performance scores are thus used to define each expert’s weight to be applied when considering the judgments on the “target” questions, i.e., the actual variables of interest for the case study. In this presentation we describe a new Python tool (‘Elicipy’) which, with respect to existing tools, greatly simplifies the managing of performance-based expert elicitation sessions. This is achieved through the automatic generation of online webforms, which collect all the experts’ answers and check for their consistency, and the analysis using different weighting schemes (Classical Model as default, Equal Weight and others optionally). The workflow automatically produces greatly detailed outputs and assembles them into a powerpoint file available just after the collection of the answers. In this presentation the workflow, from the answer collection to the analysis, is applied to replicate a previous performance-based expert elicitation.