GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 1.2 GNGTS 2024 Unveiling hidden volcano dynamics with Artfcial Intelligence (AI) and Earth Observaton (EO) S. Cariello 1,2 , C. Corradino 1 , F. Torrisi 1,2 , G. S. Di Bella 1,2 , C. Del Negro 1 1 Isttuto Nazionale di Geofsica e Vulcanologia, Osservatorio Etneo, Catania, Italy 2 Department of Electrical Electronics and Computer Engineering, University of Catania Volcano hazard monitoring is essental for understanding the behavior of rapidly changing volcanoes and, consequently, for beter forecastng volcanic hazards and related impacts. From this perspectve, several satellite sensors are now available, providing thermal infrared data at various spatal resolutons and revisit tmes. Additonally, future satellite missions are being planned to maintain a near-constant "eye" on thermal volcanic actvity across the planet. This huge volume of data necessitates the development of Artfcial Intelligence (AI) tools to automatcally extract relevant informaton on the volcano's state in a short tme. Recently, we demonstrated the potental of a cascading pipeline for classifying high-temperature volcanic features and quantfying the spatal extension of thermal anomalies in high-resoluton satellite data (Sentnel-2 MultSpectral Instrument, S2-MSI). The ability to combine two separate machine learning models —a scene classifer and a pixel-based segmentaton model (Corradino et al., 2022) —into a "top- down" cascading architecture makes this method highly efectve, achieving an accuracy of 95%. These fndings illustrate how the cascading technique can be used to fully characterize any accessible satellite image in almost real-tme, ofering valuable assistance in the mapping, monitoring, and characterizaton of volcanic thermal features. The model's high level of accuracy enables us to detect thermal signals that are ofen challenging to pick up with current detectors. Indeed, the thermal increase produced by intracrater actvity during unrest phases provides valuable data for understanding volcanic phenomena, allowing the development of more accurate predictve models and a beter understanding of the internal dynamics of volcanoes. While these thermal detectons have already served as possible precursory signals for specifc volcanoes, a comprehensive feld investgaton of such changes has not been conducted yet. In this study, we aim to examine the thermal changes captured by satellite data on Etna and Stromboli, two of the most actve and monitored volcanoes in the Mediterranean region. The objectve is to identfy and assess signifcant changes in thermal anomalies during periods of unrest, utlizing the outcomes generated by the cascading model. Over the last two decades, the eruptve actvity of Etna has been characterized by persistent degassing and a frequent intertwining of explosive and efusive eruptons from its four summit craters. Identfying the actve craters and quantfying intra-crater budget emissions in terms of the areal coverage of thermal anomalies can unveil interestng scenarios associated with the volcano’s