Hidden patterns in volcanic seismicity: deep learning insights from Mt. Etna’s 2020–2021 activity

Understanding the temporal evolution of volcanic activity is crucial for eruption forecasting and hazard assessment. We use an unsupervised machine learning method, Deep Embedded Clustering, to classify daily seismic spectrograms of Mount Etna between November 2020 and November 2021, a period that includes two major lava fountain sequences and quiescent phases. Using data from the horizontal components at two summit stations, we identify four clusters corresponding to distinct seismic regimes associated with different volcanic phases: (1) quiescence or non-dominant seismic features related to fluid dynamics, (2) fluid pressurisation indicated by elevated Long Period (LP) events, (3) preparatory phase, and (4) eruptive lava fountain episodes. These clusters closely match expert-defined volcanic phases and are validated against independent volcanic state indicators, including LP event catalogues, RMS amplitude trends, and eruption logs. Notably, a preparatory phase is observed before the lava fountains of February 2021, likely linked to the volcano’s recharging phase. After the first eruptive sequence, a cluster dominated by LP events emerges, which may reflect fluid pressurisation within the volcanic system. The approach also identifies ambiguous days that reflect mixed behaviour. These results demonstrate the potential of unsupervised learning as a reliable and supportive tool for volcanic monitoring and eruption forecasting.