Earthquake early warning signals detected by machine-learning algorithm

The most violent earthquakes can destroy entire cities and trigger subsequent volcanic eruptions, tsunamis and landslides. Caused by the sudden release of energy in the Earth’s outer crust, they are notoriously difficult to predict.

Some methods, such as using seismometers to monitor vibrations in the ground, can indicate the probability of an earthquake, although there is no guaranteed detection method. Instead, communities of people living along high-risk tectonic plate boundaries plan their disaster responses: building safe shelters and carrying out practice drills.

However, researchers at the US Department of Energy’s Los Alamos National Laboratory in New Mexico have developed a system that could, in the future, forecast earthquakes with greater accuracy than before.

The system employs machine learning, whereby a computer “learns” to recognise patterns – such as faces or movements – by processing enormous amounts of data. In this case, researchers led by Dr Paul Johnson at the laboratory analysed data produced by a laboratory-based fault system. Stone blocks were made to slide across each other, while an accelerator recorded the acoustic emission – creaking and grinding noises – produced as a result of the material dilating and strengthening.

This signal had previously been dismissed as low-amplitude noise. However, the instability causing the changing acoustic emissions concluded with a “labquake”, in which the stone block shook and layers compacted.

“These signals resemble Earth tremor that occurs in association with slow earthquakes on tectonic faults in the lower crust,” said Dr Johnson. “There is reason to expect such signals from Earth faults in the seismogenic zone for slowly slipping faults.”

While the pattern had previously evaded the notice of researchers, Johnson and his team’s machine-learning algorithm was able to detect the connection between the changing noises and the onset of the artificial earthquake.

“At any given instant, the noise coming from the lab fault zone provides quantitative information on when the fault will slip,” said Paul Johnson, who led the research. “The novelty of our work is the use of machine learning to discover and understand new physics of failure, through experimentation of the recorded auditory signal from the experimental setup.”

“I think the future of earthquake physics will rely heavily on machine learning to process massive amounts of raw seismic data. Our work represents an important step in this direction,” said Dr Johnson.

The research could, he suggests, have applications outside the potential forecasting of earthquakes. A similar machine-learning approach could be used to identify hidden signals of avalanches or for the non-destructive testing of materials for brittle failure.