Skip to content

A machine learning approach to volcanic eruption nowcasting using geostationary satellite-based thermal features

Claudia Corradino 1, Simona Cariello1, Alessandro La Spina1, Ciro Del Negro1

  • Affiliations:  1Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Etneo -- Sezione di Catania, Piazza Roma 2, 95125 Catania, Italy 

  • Presentation type: Talk

  • Presentation time: Thursday 09:15 - 09:30, Room R380

  • Programme No: 3.1.14

  • Theme 3 > Session 1

Abstract

Geostationary satellite thermal imagery has been shown to be fundamental to timely monitor volcanic eruptions. Among satellite-based monitoring signals, Volcanic Radiative Power (VRP) offers a detailed overview of the pre, syn and post volcanic behavior over time. The synergistic integration of satellite data with artificial intelligence (AI), particularly data-driven AI, has shown to solve predictive complex real-world problems positions them as an effective tool for volcano hazard monitoring. This work proposes the use of Machine Learning based on VRP-derived features to classify both the volcanic state of alert providing a probability of having an eruption and the type of volcanic activity taking place. By applying signal processing techniques on VRP records, we extracted different features, which are linked to different phases and types of volcanic activity. The learning process is driven by a vast satellite data archive acquired on Mt. Etna used as study case. This innovation proves crucial for the early identification of subtle patterns that may indicate significant changes in volcanic behavior, ultimately enhancing our ability to predict and mitigate volcanic risks. This approach offers a tool for volcanic eruption nowcasting, transferable to different volcanic systems. Importantly, VRP data used in this study are sourced from Spinning Enhanced Visible and Infrared Imager (SEVIRI) observations on the Meteosat Second Generation (MSG) satellites, providing high temporal resolution with data available every 5/15 minutes. The outcomes of this research offer promising avenues for advancing early warning systems and improving our preparedness in the face of volcanic events.