Neural networks and dynamical system techniques for volcanic tremor analysis

R. Carniel

Abstract


A volcano can be seen as a dynamical system, the number of state variables being its dimension N. The state is usually confined on a manifold with a lower dimension f, manifold which is characteristic of a persistent «structural configuration». A change in this manifold may be a hint that something is happening to the dynamics of the volcano, possibly leading to a paroxysmal phase. In this work the original state space of the volcano dynamical system is substituted by a pseudo state space reconstructed by the method of time-delayed coordinates, with suitably chosen lag time and embedding dimension, from experimental time series of seismic activity, i.e. volcanic tremor recorded at Stromboli volcano. The monitoring is done by a neural network which first learns the dynamics of the persistent tremor and then tries to detect structural changes in its behaviour.

Keywords


neural networks;dynamical systems;time series analysis;volcanic tremor

Full Text:

PDF

References


DOI: https://doi.org/10.4401/ag-3967

Published by INGV, Istituto Nazionale di Geofisica e Vulcanologia - ISSN: 2037-416X