Clustering of eruptive events from high precision strain signals recorded during the 2020–2022 lava fountains at Etna volcano (Italy)

Abstract. Explosive eruption events have been clustered by machine learning techniques applied on strain signal recorded by high-precision borehole strainmeters. We focus on the extraordinary intense and frequent eruptive activity at Etna in the period December 2020 – February 2022 when more than 60 lava fountains occurred. We apply the k-means algorithm on the associated strain variations which are representative of the eruptive dynamics. A novel procedure was developed to ensure a high-quality clustering process and obtain robust results. The analysis identified four distinct groups of strain variations which characterize the events in terms of amplitude, and duration and time derivative of the signal. The temporal distribution of the clusters provides useful insights into the evolution of the volcano activity and reveals transitions in the eruptive style.