An innovative analysis of seismicity investigated with the aid of machine learning highlights its potential for understanding and managing natural risks
The study Causal processes of shallow and deep seismicity at Campi Flegrei caldera, conducted by a team of researchers from the National Institute of Geophysics and Volcanology (INGV), has just been published in the scientific journal “Communications Earth and Environments” of Nature. The results represent the first application of machine learning (ML) techniques to the analysis of the seismicity of the volcanic system located northwest of the city of Naples.
Seismology in recent decades has begun to employ artificial intelligence algorithms, in particular the so-called neural networks, which simulate the neurons in our brain. If properly trained, they can facilitate some of the seismologist's tasks such as recognizing seismic waves.
“During an earthquake, energy is released through seismic waves P and S (the first and second to arrive at the seismic station, respectively). Estimating their arrival time is essential to understand the distance of the station from the earthquake and calculate its hypocenter,” explains Rossella Fonzetti, researcher at INGV.
“Since the signal recorded by seismographs is very often disturbed, even an expert seismologist can have difficulty recognizing them,” continues the researcher. “For this reason, we decided to use these new artificial intelligence algorithms to quickly extract the arrival times of the P and S waves generated by the earthquakes that occurred between January 2023 and June 2024, a period in which the caldera experienced two episodes of increased seismicity.”
The study, carried out thanks to the data rapidly available on the European Integrated Data Archive - EIDA platform (the infrastructure that provides rapid access to seismic signals acquired by the main European agencies), is configured as a further step towards the development of an integrated monitoring tool for the seismic and volcanic evolution of the Campi Flegrei.
"To better understand the causes of the recent increase in seismicity, we relocated the seismic events using different algorithms and compared the new hypocentral locations with previously developed velocity models", adds Genny Giacomuzzi, researcher at INGV.
One of the main results is the significant spatial correlation between the “ring” distribution of the deepest seismicity and a velocity anomaly, highlighted by previous studies, located at 5 km and associated with a magma accumulation zone.
Although this does not imply the imminence of a volcanic eruption, this correlation suggests a causal relationship between the magma ascent and the accumulation of stress in the overlying area and consequent seismic release, corroborating the hypothesis that the magma ascent itself may represent the cause of the ongoing unrest (i.e. instability).
The analysis of seismicity over the last two years also highlights the activation of two fault structures located at the eastern and western edges of the caldera. This result is in agreement with estimates provided by analogical and numerical models which suggest that pre-existing faults linked to the formation of the caldera can be activated during subsequent episodes of uplift and subsidence of the ground (called, respectively, inflation and deflation).
“The high quality of the data obtained with machine learning can also be very useful in seismic tomography investigations to study the velocity structure of the caldera, a fundamental aspect for real-time monitoring of the evolution of the phenomenon”, continues Giacomuzzi.
“The next step, in fact, involves using this high-quality dataset to perform a new 4D tomography of the caldera, aimed at identifying any areas in which recent changes in seismic velocities may indicate further migrations of magmatic fluids or magma”, concludes Claudio Chiarabba, Director of the Earthquake Department of the INGV.
Link to the study:
Useful links: National Institute of Geophysics and Volcanology (INGV)
