Uncertainties in lava flow hazard maps derived from numerical simulations: The case study of Mount Etna

The procedure for the derivation of a hazard map for lava flows at Mount Etna through lava flow simulations is critically reviewed. The DOWNFLOW code is then used to explore the sensitivity of the hazard map with respect to input settings. Three parameters are varied within ranges close to values recently applied to derive similar hazard maps: (i) the spacing between computational vents; (ii) the spatial probability density function (PDF) for future vent opening; and (iii) the expected length of future lava flows. The effect of increasing the spacing between computational vents tends to be compensated at the lower elevations, and a vent spacing smaller than about 500 m warrants an overall difference with respect to a reference map which is smaller than 6-8%. A random subsampling of the elements used to obtain the input vent opening PDF (− 20%, − 40% and − 60%) originates significant but drastically smaller differences in the obtained map with respect to the reference one (~ 10%, ~ 12.5% and ~ 17% respectively, on average). In contrast, our results show that changes in the expected flow length originate, by far, the highest changes in the obtained hazard map, with overall differences ranging between ~ 20% and ~ 65%, and between ~ 30% and ~ 95% if computed only over inhabited areas. The simulations collected are further processed to derive maps of the confluence/diffluence index, which quantifies the error introduced, locally, when the position of the vent is misplaced by a given distance.