Intra-eruption forecasting using analogue volcano and eruption sets

Abstract

Volcanic eruptions can consist of multiple phases, separated either by quiescence, or by a switch to a different style of activity, e.g., from explosive to effusive. We present a data set of 6871 phases from 2670 historical eruptions, extending a previously curated set of 698 eruptions. We update a probabilistic model for forecasting eruption progression through multiple states. The new data and model are then used to investigate if ongoing eruptions can be better forecast by using only a subset of the data. In particular we consider using (a) only data from volcanoes with similar composition (e.g., basalts), (b) only data from volcanoes of the same type (e.g., stratovolcanoes), (c) only data from volcanoes with the same composition and type, (d) only data from the volcano concerned. Except for shield volcanoes, none of these limitations outperform the entire data set, although using data from only the target volcano may be effective if there is enough of it, or the behavior of the target volcano is relatively unique (e.g., Stromboli).