Skip to content

Volcanic unrest detection using trans-dimensional McMC: application to dike intrusion events at Mount Etna

Erica De Paolo 1, Nicola Piana Agostinetti2, Flavio Cannavò1

  • Affiliations: 1 Istituto Nazionale di Geofisica e Vulcanologia Osservatorio Etneo, Catania, Italy. 2 Università degli Studi di Milano Bicocca, Milano, Italy.

  • Presentation type: Poster

  • Presentation time: Monday 16:30 - 18:30, Room Poster Hall

  • Poster Board Number: 284

  • Programme No: 2.4.52

  • Theme 2 > Session 4

Abstract

Volcanic monitoring is conducted at observatories worldwide through the analysis of continuously updating time-series from the geophysical monitoring network. The detection of upcoming unrests, typically characterized by anomalous signals from geodetic and seismic stations, is crucial for the early warning management. However, due to the presence of noise of different nature, difficult to suppress in near real time, the time gap between the beginning of the anomaly and the issuance of the unrest warning is challenging to be minimized. Several approaches spanning from traditional statistics to machine learning have been tested in the literature, given the need for ever more accuracy. We test the performances of trans-dimensional Markov-chain Monte Carlo methods to detect sudden changes in multivariate time series. The case study of Mount Etna dike intrusion event of the 24th December 2018 is used to simulate the near real time data arrival. The main advantage of this method is the full data-driven change point selection, thanks to the changing dimensionality of the problem. We are able to avoid false alarms thanks to the joint inversion of multiple GNSS stations and other geophysical signals. Multiple independent chains in a computationally efficient framework, reveals peaks of sampled models frequency at the location of the target events in both synthetic and real datasets. We compare the timing of our warning outputs to those obtained with classical methods. We prove that the proposed approach constitutes a highly sophisticated and efficient tool for monitoring activity support.