Skip to content

Identifying dike propagation patterns at Mount Etna through forward-modelling and AI-powered ground deformation inversions

Rebecca Bruni1, Erica De Paolo2, Deepak Garg1, Martina Allegra2, Flavio Cannavò2, Chiara Paola Montagna1, Paolo Papale1, Michele Carpenè3

  • Affiliations: 1Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Pisa, Italy; 2Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania, Osservatorio Etneo, Italy; 3Italian Supercomputing Centre (CINECA), Sezione di Bologna, Italy

  • Presentation type: Talk

  • Presentation time: Monday 11:00 - 11:15, Room S160

  • Programme No: 2.4.9

  • Theme 2 > Session 4

Abstract

Surface deformation at active volcanoes is caused by subsurface magma dynamics, which induces pressure variations within magma sources. These variations cause stress changes in surrounding rocks, propagating deformation signals to the surface, where monitoring networks detect them. To aid in near-real-time analysis of these signals, we are developing a digital twin for dyke propagation-induced deformation at Mount Etna, combining 3D numerical simulations with artificial intelligence (AI).  We simulated 107 realizations of dike intrusions underneath Mount Etna using the open-source multi-physics finite element software GALES on Leonardo HPC (CINECA), solving for the elastostatics induced by overpressurized dykes within a spectrum of dyke geometries mimicking the observed variability at Mount Etna. The 3D computational domain includes the latest DEM topography and heterogeneous rock properties from seismic tomography surveys. The results provide a comprehensive picture of input-output (source - deformation) relationships used to train an AI that reconstructs probability distributions for source parameters from GNSS deformation datasets. Feeding the AI with near-real-time data provides a probabilistic evolution of the source. The inversion procedure will be triggered by another AI module that scans multi-parametric monitoring data streams at Etna Volcano Observatory to identify unrest conditions. This methodology allows to better constrain dike intrusion parameters and to follow the evolution of unrest as deformation patterns evolve. All software will be shared as an open-source tool, providing a valuable resource for civil protection authorities and offering the potential for replication on other volcanoes, representing a significant step forward in volcanic crisis management and monitoring.