Can We Predict Earthquakes? The Current State of Science

Earthquake prediction is often called the Holy Grail of seismology. Despite billions of dollars in research and decades of effort, scientists still cannot reliably predict when and where the next big earthquake will strike. Here is why, and what we can do instead.

The Prediction Problem

A scientifically useful prediction requires three specific things: the location of the future earthquake, its magnitude, and the time window in which it will occur. All three must be accurate enough to be actionable. We can identify fault zones where large earthquakes are likely (location) and estimate their maximum possible size (magnitude), but the timing remains elusive.

The fundamental challenge is that earthquakes occur deep underground, often 10-100 km beneath the surface. We cannot directly observe the stress buildup or the rock deformation that precedes rupture. Everything we know about pre-earthquake conditions comes from indirect measurements at the surface.

What Has Been Tried

Scientists have investigated dozens of potential precursors: changes in radon gas emissions, electromagnetic anomalies, groundwater level changes, animal behavior, foreshock patterns, and crustal deformation measured by GPS. None has proven reliable enough for operational prediction.

China successfully predicted the 1975 Haicheng earthquake (M7.3) based on foreshocks and unusual animal behavior, evacuating the city before the main shock. But the very next year, the 1976 Tangshan earthquake (M7.8) struck without any detected precursors, killing an estimated 242,000-655,000 people. This demonstrated that precursors, when they exist at all, do not appear consistently.

Earthquake Early Warning Systems

While prediction remains out of reach, early warning systems are a proven technology that saves lives. These systems exploit the speed difference between P-waves (fast, compression waves, ~6 km/s) and S-waves (slower, shear waves, ~3.5 km/s that cause most damage). When an earthquake is detected, the system sends an alert at the speed of light, which arrives before the damaging S-waves.

Japan's system gives Tokyo an average of 30-60 seconds of warning before strong shaking arrives from offshore earthquakes. Mexico City gets 60-90 seconds of warning from earthquakes on the Pacific coast. The US West Coast ShakeAlert system provides seconds to tens of seconds of warning.

Even a few seconds of warning is enough to: automatically slow trains, open fire station doors, shut down industrial equipment, alert hospital surgeons to pause procedures, and give people time to drop, cover, and hold on.

Machine Learning and the Future

Researchers are applying deep learning to continuous seismic data, looking for subtle patterns invisible to human analysts. Some promising results include faster characterization of earthquake size (useful for early warning) and detection of slow-slip events that may precede large earthquakes. However, no AI system has yet demonstrated the ability to predict earthquakes with actionable accuracy.

For now, the best approach remains: identify hazardous zones, enforce strong building codes, educate the public, and invest in early warning infrastructure. Track current seismic activity on our real-time monitoring page.