First author

Predicting earthquakes has proved an elusive goal for seismologists, in part because direct measurements from deep inside faults have been lacking. California's San Andreas Fault Observatory at Depth (SAFOD) site now allows scientists deep access to the world's most-instrumented section of fault. There, Fenglin Niu of Rice University in Houston, Texas, and his colleagues fired experimental pressure-generated pulses and monitored the time taken for the resulting seismic waves to travel through a 10-metre section of rock at a depth of 1 kilometre. On page 204, they confirm that stress-induced changes occur at the San Andreas Fault in California shortly before rupture events. Niu tells Nature that if such changes also occur elsewhere, reliable earthquake prediction may be possible.

What efforts have been made in the past to predict earthquakes?

Laboratory studies in the 1960s showed that the speed of seismic waves travelling through a rock varies with the level of stress applied to the rock owing to the opening and closing of micro-cracks. And that micro-crack density increases rapidly before an earthquake. But failure to replicate those findings in the field dampened interest.

What work has been done since then?

Researchers have been trying to precisely measure changes in the velocity of seismic waves at fault depth. Recent technological improvements, including repeatable pulse sources, high sampling rate data-acquisition systems and massive data–processing capability, have made measurements reliable. At SAFOD, earthquake locations can be very accurately determined, making reliable estimates of changes in stress possible.

Were earthquake events a help or a hindrance to your research?

They provided valuable data. We wanted to show that stress-dependent seismic velocity seen at the surface also occurs deep within faults. There were two anomalous spikes in our data. Co-author Tom Daley wondered whether the spikes could be signals of changes in tectonic stress. Our calculations show that the spikes occurred 10 hours before a magnitude-3 earthquake and 2 hours before a magnitude-1 earthquake, which suggests that they relate to pre-rupture stress.

Do you think earthquake prediction will become possible during your career?

These initial data are encouraging, but are just the start. We need to establish whether these signals are pervasive at different sites, and, if so, how the signal timing relates to earthquake size.