Assessing vent opening location probability at Somma-Vesuvio volcanic complex with uncertainty quantification by using Structured Expert Judgment

Quantifying uncertainty is crucial for producing hazard assessments which civil protection authorities use to mitigate the associated risks. In this study we combine detailed reconstructions of volcanological datasets and inputs from Structured Expert Judgment (SEJ) to produce a first background (i.e. long-term or base-rate) probability map for vent opening location in the next Plinian or Sub-Plinian eruption of Somma-Vesuvius (SV). The SV volcano has, over its history, exhibited a large variability in eruptive styles, and a moderate but significant spatial variability in vent locations. In particular, the vent positions associated with large explosive eruptions, i.e. Plinian and Sub-Plinian, have shown shifts within the present SV caldera. Notwithstanding this moderate shift, the location of a new vent could have a major effect on the run-out and dispersal of pyroclastic density currents mainly due to the presence of the Mt. Somma barrier, as also evidenced by past deposit patterns and illustrated by 3D numerical simulations, and therefore will have important implications for hazard mitigation. Thus far, we have focused on three main objectives: i) the collection and critical review of key volcanological features (position of past vents, distribution of faults, etc.) that could influence the spatial distribution of future vent locations, organized in a specific geo-database where epistemic uncertainties related to feature spatial distributions have been quantified; ii) developing spatial probability density maps with Gaussian kernel function modelling to use with our different volcanological and geophysical datasets, and iii) the production of a background probability map for vent opening position, using weighted linear combination of spatial density maps for the identified volcanological and geophysical parameters, with uncertainties (related to both epistemic and aleatoric uncertainties) explicitly included by using SEJ. Outcomes obtained during two elicitation sessions involving about 15 experts are reported for three expert judgment weighting and pooling models: (a) the Classical Model (CM) of Cooke (1991); (b) the Expected Relative Frequency (ERF) model of Flandoli et al. (2011), and (c) the Equal Weights (EW) combination. The results of combining expert judgements with our spatial modeling of the identified variables illustrate that: a) vent opening probabilities are evenly? distributed around the caldera with a peak in correspondence with the area of the present crater but with about 50% mean probability that the vent will open in other areas of the caldera; b) there is a mean cumulative probability of about 30% that the next vent will open west of the present edifice in the so-called “Piano delle Ginestre” area; c) there is a mean probability of more than 20% that next Plinian eruption will enlarge the present SV caldera and a not negligible probability (of almost 10%) that the next Plinian or sub-Plinian eruption will have its initial vent opening outside the present outline of the SV caldera. Robustness of results have been tested by considering the effects of alternative pooling methods, sub-groups of experts with different backgrounds and experiences and sub-groups of volcanological datasets. Uncertainty analysis also allowed identification of the most controversial issues and to have a first estimate of the associated ranges.