Single-station analysis of Campi Flegrei (Italy) seismic signals using multiscale entropy and unsupervised learning

Seismic activity in volcanic regions such as Campi Flegrei (Italy) provides essential insights into subsurface dynamics and potential hazards. However, high background noise and continuous data volume challenge event detection and classification. Here, we apply a Self-Organizing Map (SOM) approach, combined with Linear Predictive Coding (LPC), STA/LTA ratios, and Multiscale Entropy (MSE), to analyze single-station seismic data. The method successfully identifies uncatalogued events and anomalies associated with fumarolic tremor, and reveals temporal relationships between clustering variation, [Formula: see text] emissions, and rainfall, suggesting environmental modulation. To assess the real-time applicability, the trained SOM was used on independent data from early 2025, confirming its ability to detect tremor intensification and anticipate a major local earthquake (Md 4.4). These results highlight the potential of entropy-based unsupervised learning for rapid seismic characterization and continuous volcano monitoring.