Leveraging multivariate geophysical and geochemical time series for monitoring volcanic systems: can we use machine learning?

Project Summary

Monitoring volcanic edifices is central for mitigating volcanic risks and hazards. This involves monitoring and analyzing multivariate data, which often have complex characteristics. But observatories need to communicate the state of the volcano in clear, understandable terms for both the public and decision-makers. The challenge of analyzing large amounts of data could be improved through the use of machine learning. Machine learning may be a key method for conducting multivariate analyses of time series at volcanic observatories. Its application could provide new insights, enhance understanding, and help better anticipate
eruptions.

We investigate whether and how signals from seismicity, ground deformation, and CO2 degassing can be combined to detect and forecast volcanic eruptions at Piton de la Fournaise. We analyze signals from the past twenty-four years using supervised and unsupervised deep learning techniques. Our results demonstrate that using various machine learning algorithms shows significant potential for detecting eruptive precursors and could enable the detection of eruptions several days in advance