GNGTS 2021 - Atti del 39° Convegno Nazionale

GNGTS 2021 S essione 1.2 100 AN UNSUPERVISED MACHINE LEARNING MULTI-PARAMETRIC APPROACH TO CLASSIFY MAJOR EXPLOSIONS AND PAROXYSMS AT STROMBOLI VOLCANO (ITALY) USING RADAR AND OPTICAL SATELLITE IMAGERY C. Corradino 1* , E. Amato 1,2 , F. Torrisi 1,3 , S. Calvari 1 , C. Del Negro 1 1 Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania, Osservatorio Etneo, Catania, Italy 2 Dipartimento di Matematica e Informatica, University of Palermo, Palermo, Italy 3 Dipartimento di Ingegneria Elettrica Elettronica e Informatica, University of Catania, Catania, Italy Abstract Stromboli volcano has an activity almost exclusively explosive, characterized from events with high magnitude, which are the main hazards for the inhabitants of the island. Here, we propose a methodology to discriminate between paroxysms and major explosions by only using satellite- derived measurements. Three discriminative features are taken into account, namely the ther- mal trend over the summit area, the presence of large pyroclastic materials ejected during each explosion and the height of the volcanic plume produced during the explosive event. We use optical satellite imagery to compute the Land Surface Temperature (LST) and the plume height, while the presence of large pyroclastic deposits is retrieved by using radar satellite imagery, i.e. Sentinel 1-SAR intensity data. Once we identified the input feature vectors from the previous esti- mates, we designed a k-means unsupervised classifier to group the explosive events at Stromboli volcano based on their similarities in two clusters, namely paroxysm and major explosion. Corresponding author: claudia.corradino@ingv.it