AI-driven insights into the volcanic processes and dynamics of explosive episodes inferred by satellite-based SO2 estimates, ground-based gas measurements, and petrological data

Identifying changes in a volcano's unrest and tracking the evolution of its eruptive activity are crucial for effective volcanic surveillance and monitoring. Variations in gas composition and amount can be associated with pre-eruptive changes in the volcano plumbing system. When combined with petrological studies, the emitted Sulphur dioxide (SO2) reflects the amount of magma involved (erupted or degassed), making it a useful parameter for constraining volcanic processes, dynamics, and the volume of magma. This work proposes an Artificial Intelligence (AI) strategy to provide new insights into the volcanic processes and dynamics of explosive episodes using a multidisciplinary approach. Through advanced machine learning (ML) algorithms, we investigate the spatio-temporal relationships among the SO2 satellite image time series (SITS), ground-based gas measurements, and petrological data associated with volcanic pre- and syn-eruptive phases. SO2 emissions are estimated via satellite ultraviolet remote sensing, i.e. TROPOspheric Monitoring Instrument. Both the quiescent/pre-eruptive and syn-eruptive/explosive gas phases are constrained from ground-based infrared remote sensing data i.e Fourier Transform InfraRed (FTIR). Rock compositions and textural features (e.g. crystallinity and vesicularity) of volcanic products are estimated by petrological study. The ML algorithm allows to both discover pre- and syn-eruptive patterns indicative of future eruption and better characterize volcanic processes. Unsupervised ML techniques are considered to explore previously unknown relationships without any external bias. We have tested this approach on recent volcanic activity that occurred on Mt Etna.