Real-time imaging of a volcano’s internal plumbing to better anticipate eruptions

Internal plumbing “lit up” from the inside

During the preparatory phase of an eruption, a volcano deforms slightly due to the movement of magmatic fluids at depth. Modelling these surface deformations makes it possible to determine and quantify some of the characteristics of the active source that causes them: its location, geometric shape and variation in volume. However, the term ‘tomography’ (which comes from the Greek root ‘representation in sections’) had never been used in volcanology to describe the image obtained from active deformations, even though this method is entirely complementary to other structural imaging techniques (seismic, electrical, gravimetric, etc.).

In a study published in Geophysical Research Letters, researchers from the Institut de Physique du Globe de Paris (IPGP-Université de Paris), the ISTerre (IRD, Université Grenoble-Alpes), the Laboratoire de Biométrie et Biologie Évolutive (LBBE-CNRS-Université Claude Bernard Lyon I) and Gadjah Mada University (Indonesia) have demonstrated that, from an inversion of a map of the Earth’s crust, the deformation of the Earth’s crust can be measured, that, using Bayesian inversion and a relatively simple analytical model, it is possible to obtain a coherent image of a volcano’s internal plumbing from real-time GPS measurements, with each structure used and subjected to magma pressure being activated as it migrates towards the surface.

A revolutionary analytical model

The analytical model used in this study, called the Point Compound Dislocation Model (pCDM), was recently proposed by Nikkhoo et al. (2017). It can simulate any form of source (sphere, ellipsoid, sill, dyke, pipe) in a homogeneous elastic medium and in the far field, with a limited number of parameters. This means that it can replace most of the previous analytical models, which had to be used on a case-by-case basis.

Following the most optimised numerical coding possible of the model’s equations (in C and Matlab languages), the research team implemented the model in an inverse problem adapted to explore the ‘model space’, i.e. the set of probable solutions, through millions of calculations in just a few tens of seconds. The method used is called ‘unsupervised’ because it minimises the amount of a priori information imposed on the calculation: it is the model that determines the set of most probable source locations and shapes, constrained exclusively by the observed GPS measurements and their uncertainties. The result is an advanced and perfectly quantified ‘translation’ of the surface data.