Parkfield, population 18, sits on the San Andreas Fault in central California. Besides a café and grazing cattle, the town hosts a dense array of seismic instruments that measure tremors deep below Earth’s surface. The small quakes repeat every few days and act as a model for similar faults around the world. 

In 2010, David Shelly, a researcher from the United States Geological Survey (USGS), published something puzzling about the Parkfield tremors. Instead of releasing built up stress on a regular schedule, the fault created an alternating waltz-like pattern of tremors, three and six days apart. Not only did the odd rhythm mystify scientists, but the kinds of tremors as well, always shifting between low and high-energy events.

“Even at the time of observation, scientists didn’t understand why it was happening. We just said, ‘Wow, there’s this weird behavior going on,’” says Sylvain Barbot, a geophysicist at the Earth Observatory of Singapore. “The Earth is playing music and that's a difficult one to explain!”

A new study in Nature by Barbot proposes a possible explanation for the Parkfield tremors. The results could shake up prevailing ideas on the timing and strength of earthquakes that can happen on a fault.

Current earthquake hazard models assume that earthquakes happen at a relatively constant rate, whether it’s every day, decade or century. They also assume that the fault produces the same kind of earthquake each time it ruptures. For example, if a fault has a quiet earthquake that happens over a period of hours, then it won’t have a quick, building-shattering quake in the same place.

However, Barbot says we may be looking at earthquake hazards too simplistically. “We should expect these faults that host earthquakes to have a complex behavior instead of a simple on and off switch,” says Barbot.

Using basic physics, Barbot and geophysicist Deepa Mele Veedu created a model to explain what’s happening below Parkfield. The new model is conservative and may be applicable everywhere, says Barbot. It combines geometric and frictional properties to explain the Parkfield tremors. The model comes eerily close to reality, too—it matches up with four hundred tremor events between 2002 and 2008. “It’s a promising step in the right direction,” says Barbot.

Because great earthquakes happen so infrequently, Barbot says we’re basing earthquake estimates on a limited set of observations. And those observations don’t tell us the entire picture. It’s like getting to know someone new, he says. “If you know someone for three days you can’t assume they never get angry,” Barbot says. “Faults can be nice and gentle and they can be violent. They have more complexity than we realised.”