Why Can’t We Predict Earthquakes?

In October 2012, Italian courts convicted six scientists and a government official—all members of the National Commission for the Forecast and Prevention of Major Risks—of manslaughter for allegedly downplaying information in the days leading up to the devastating earthquake that struck L’Aquila on April 6, 2009. Tens of thousands of buildings were destroyed, 1000 people were injured, and 308 people died. The courts believed it was because scientists didn't do enough to warn civilians of the risk of a massive quake. (All but one of the convictions were overturned on appeal in 2014, and the remaining defendant had his sentence reduced from six years to two.)

The case caused intense controversy in the scientific community and highlighted a curious fact about earthquakes in general: scientists can’t predict where or when they will occur. But why not?

1. Earthquakes: How Do They Work?
2. The Challenges of Prediction
3. Earthquake Prediction vs. Forecasting
4. Prediction Research

Earthquakes: How Do They Work?

For centuries, people wondered what caused the Earth to shake. In the 1960s, scientists finally settled on the theory of plate tectonics, which posits that the Earth’s surface is built of plates—solid slabs of rock—that move relative to each other on top the hotter, molten material of the outer core. As these plates move around, they slide past and bump into each other; on the boundaries of these plates are faults, which have rough edges and stick together while the rest of the plate keeps moving. When this occurs, the energy that would normally cause the plates to move past one another is stored up, until eventually, the force of the moving plates overcomes the friction on the jagged edges of the fault. The fault unsticks and releases that energy, which radiates outward through the ground in waves, causing an earthquake when the waves reach the surface.

To locate a quake’s epicenter—the place on the Earth’s surface, directly above the hypocenter, where the quake starts—scientists look at the waves produced by the quake. P waves travel faster, and shake the ground first; S waves come next. The closer you are to the epicenter of an earthquake, the closer together those two waves will hit. By measuring the time between waves on three seismographs, scientists can triangulate the location of the quake’s epicenter.

The Challenges of Prediction

Though scientists do create sophisticated models of earthquakes and study the history of quakes along fault lines, no one has enough of an understanding about the conditions—the rock materials, minerals, fluids, temperatures, and pressures—at the depths where quakes start and grow to be able to predict them. “We can create earthquakes under controlled conditions in a laboratory, or observe them close-up in a deep mine, but those are special situations that may not look very much like the complicated faults that exist at depth in the crust where large earthquakes occur,” Michael Blanpied, associate coordinator of the USGS Earthquake Hazards Program, told Mental Floss in 2012. “Our observations of earthquakes are always at a distance, viewed indirectly through the lens of seismic waves, surface faulting and ground deformation. To predict earthquakes, we would need to have a good understanding of how they occur, what happens just before and during the start of an earthquake, and whether there is something we can observe that tells us than an earthquake is imminent. So far, none of those things are known.”

According to Blanpied, the current understanding is that quakes start—or nucleate—small, on an isolated section of the fault, and then grow quickly. “That nucleation can occur anywhere, and even when we have examples of repeated earthquakes, they may nucleate in different places,” he says. “If there is a process that occurs in the seconds—{or} minutes, hours, months?—before an earthquake, that process may be very subtle and hard to observe through miles of solid rock, especially when we don’t even know where to look.”

Another challenge: Big and small quakes might not start differently. “If all earthquakes start small, and some just happen to grow big, then prediction may be a lost cause, because we’re not at all interested in predicting the thousands of tiny earthquakes that happen every day.”

Earthquake Prediction vs. Forecasting

Though pinpointing the exact time and size of an earthquake is currently impossible, scientists can estimate the probability of an earthquake occurring in a region or on a fault over a span of decades. “To do that, we need information about how fast the fault is sliding over the long term—typically a few millimeters to centimeters of slip per year—and how big the earthquakes are likely to be,” Blanpied said. “We calculate how much slip is used up in each earthquake, and thus how often earthquakes must occur, on average, to keep up with the long-term slip rate.”

Knowing the date of the last earthquake helps improve forecasting, because scientists can estimate whether they’re early or late based on the repeat time of earthquakes on that particular fault. At the Hayward fault, east of San Francisco Bay, for example, large quakes happen every 140 to 150 years. The last quake on the fault was in 1868, so scientists think that fault could produce another earthquake at any time. “Note, however," Blanpied says, "that ‘any time’ could mean tomorrow or 20 years from now.”

Scientists learned this the hard way. In the 1980s, the USGS predicted that, within 5 years, there would be a magnitude 6 earthquake on the San Andreas fault near the town of Parkfield. “Many types of instruments were deployed in the area to observe the earthquake and also to try to predict it based on various types of precursory signals,” Blanpied says. “As it turns out, the earthquake didn't happen until 2001, which put cold water on the idea of using the timing of past earthquakes to precisely predict future ones. Also, there were no observed precursors, which dimmed the hope that it would be possible to predict earthquakes from observing the ground.”

For now, forecasting is the best we’ve got, and although it’s imprecise, determining the probability of a quake does help developers make good decisions about where to build and what types of forces those buildings should be constructed to withstand. “If our buildings are strong,” Blanpied says, “then it doesn’t matter so much {if we can predict large earthquakes} because we’ll be safe no matter when the ground happens to shake.”

Prediction Research

Quakes pose a threat to 75 million Americans in 39 states, so despite the challenges, scientists at the USGS are working diligently to figure out how to better predict these events. They create quakes in the lab, have drilled boreholes in the San Andreas Fault Zone to get a look at the conditions at depth, and study ground deformation using GPS sensors to understand how stresses build up on faults.

This research was used to create the ShakeAlert early warning system, which issues earthquake alerts via app and collects data on quake conditions to better forecast future movements. The system gives people some time—a few seconds to a minute, maybe—to get away from the quake’s epicenter and shelter in a safe place, while also helping officials to slow or halt public transportation, clear traffic off bridges, and more. It currently covers California, Oregon, and Washington State.

But there’s no promise that a solid earthquake prediction method will ever be discovered. “What we need is a prediction method that works better than random educated guessing, and despite decades of work on this problem, so far nobody has demonstrated that such a method exists and works,” Blanpied said. “I am dubious that we will ever be able to predict the time of large earthquakes in a useful way. However, we can predict a lot of things about earthquakes that are useful, other than the time of their occurrence, and we can use that knowledge to make ourselves and our communities resilient.”

Discover More Fascinating Ways Earth Works:

A version of this story was published in 2012; it has been updated for 2025.