No technology can reliably predict exactly when and where an earthquake will strike. The U.S. Geological Survey is clear on this: a true prediction requires a specific date, location, and magnitude, and that remains beyond current capability. What technology can do is detect early signs of seismic activity, measure stress building along faults, and send rapid warnings once shaking begins. Here’s a look at the tools scientists actually use.
Seismometer Networks
The backbone of earthquake monitoring is the global network of seismometers, instruments that detect ground vibrations with extraordinary sensitivity. Modern stations use very broadband (VBB) sensors, often installed in boreholes drilled into bedrock to reduce surface noise. These instruments produce three outputs measuring motion along perpendicular axes (vertical, north-south, and east-west), capturing everything from faint tremors to massive ruptures. Data from thousands of these stations flows into national and international monitoring centers in real time, allowing scientists to locate earthquakes within seconds of their occurrence and track patterns of seismic activity along known faults.
Satellite Monitoring of Ground Movement
Two satellite-based technologies let scientists watch the Earth’s surface shift by millimeters. GPS networks (more broadly called Global Navigation Satellite Systems, or GNSS) have been tracking crustal movement across fault zones like California and Nevada since the late 1980s. Continuous GPS stations record how the ground moves in three dimensions at fixed points, quantifying how much strain is building along a fault over months and years.
Interferometric Synthetic Aperture Radar, or InSAR, works differently. Satellites send radar pulses toward the ground from orbit, then compare the reflected signals between passes. The difference in radar phase between two acquisitions creates an interferogram, essentially a map of how much the surface moved between those two satellite visits. InSAR covers vast areas with high spatial resolution, revealing subtle deformation that individual GPS stations might miss.
The two methods complement each other nicely. GPS gives precise three-dimensional movement at discrete points, while InSAR fills in the gaps with broad spatial coverage in the satellite’s line-of-sight direction. Researchers combine them using statistical methods to build detailed 3-D maps of how the crust is deforming. This doesn’t predict a specific earthquake, but it identifies where strain is accumulating and which fault segments are most likely to rupture.
Fiber Optic Sensing
One of the newer approaches repurposes something already buried under cities and ocean floors: fiber optic telecommunication cables. Distributed acoustic sensing (DAS) technology sends laser pulses through these cables and analyzes the light that bounces back. When the ground vibrates, the fiber stretches or compresses slightly, changing the properties of the backscattered light. By reading those changes, a single cable can function as thousands of individual seismic sensors spaced just meters apart.
This is significant because it transforms existing infrastructure into dense seismic arrays without installing new equipment at every measurement point. Unused “dark fiber,” cables that were laid but never activated for telecommunications, is particularly useful. DAS systems are now being tested in cities, along ocean floors, and near fault zones, giving scientists seismic data from places where traditional instruments would be impractical or too expensive to deploy at scale.
Machine Learning and Pattern Recognition
The volume of seismic data generated worldwide is far too large for humans to analyze manually, so machine learning has become essential. Algorithms like artificial neural networks, support vector machines, random forests, and decision trees are trained on massive datasets to distinguish real earthquake signals from background noise, explosions, or even vandalism vibrations hitting a sensor. Of these, artificial neural networks tend to perform best at classification tasks because their layered architecture can learn complex, nonlinear patterns in the data.
Unsupervised techniques are also in use. Clustering algorithms group earthquake ground motions by their frequency characteristics without being told what to look for, helping scientists identify patterns that might not be obvious. The hope is that machine learning will eventually recognize subtle precursor signals in seismic, GPS, or satellite data that humans have missed, though this remains a work in progress rather than a proven prediction tool.
Ionospheric Monitoring
Large earthquakes and tsunamis disturb the upper atmosphere in measurable ways. When the ground or ocean surface suddenly shifts, it generates waves that travel upward and reach the ionosphere (a layer of charged particles roughly 80 to 1,000 kilometers above Earth) within about 8 to 15 minutes. GPS satellites can detect these disturbances by measuring changes in total electron content, the density of charged particles that GPS signals pass through.
During the 2014 Pisagua earthquake in Chile, 33 GPS stations along the coast detected ionospheric disturbances of about 1.25 TECU (a standard unit for electron density). The 2015 Illapel earthquake produced even larger changes of roughly 1.40 TECU to the north of the rupture zone. Scientists identified multiple wave types in the data: fast acoustic-gravity waves traveling at 500 to 700 meters per second from the epicenter, and slower internal gravity waves coupled with tsunami propagation moving at about 300 meters per second.
This technology is primarily useful for detecting events that have already begun rather than predicting future ones. It holds particular promise for tsunami confirmation, since the ionospheric signature of a tsunami can be detected while the wave is still far from shore.
Radon and Subsurface Gas Monitoring
When rock deep underground is stressed and begins to crack before an earthquake, it can release gases trapped in the crust. Radon, a naturally occurring radioactive gas, has been the most studied candidate. Monitoring stations near active faults continuously measure radon concentrations in groundwater and soil. In Turkey’s Denizli Basin, researchers found that radon levels near faults ranged from 0.67 to 25.90 kilobecquerels per cubic meter, with spikes reaching 2 to 13 times the normal background level before nearby seismic events.
The challenge is reliability. Radon levels fluctuate for many reasons unrelated to earthquakes, including rainfall, temperature changes, and seasonal cycles. While elevated radon has correlated with some earthquakes, it has also spiked without any seismic event following. This makes it a potentially useful data point within a larger monitoring system, but not a standalone predictor.
Animal Behavior Research
Reports of unusual animal behavior before earthquakes go back centuries, and there is a plausible biological basis. Animals possess sensory systems that could detect several known earthquake precursors. Rodents can sense low-frequency ground vibrations through specialized acoustic receptors, a capability that likely dates back to the last common ancestor of reptiles and mammals. Sharks and rays can perceive electrical field changes down to the nanovolt level through specialized organs. Many species, including birds and honeybees, have magnetite-based receptors in their heads that detect shifts in magnetic fields, with honeybees sensitive to frequencies below 10 hertz. Spiders and insects have hair-like structures that pick up changes in humidity.
Despite this, no one has built a reliable early warning system based on animal behavior. The signals are inconsistent, and animals react to many stimuli besides seismic precursors. It remains an area of curiosity rather than operational technology.
Early Warning Systems
The technology closest to giving people actionable advance notice is earthquake early warning (EEW), though it works on a fundamentally different principle than prediction. These systems detect an earthquake that has already started and race to alert people before the damaging shaking arrives. The ShakeAlert system on the U.S. West Coast, for example, picks up the fast-moving but less destructive initial seismic waves and sends alerts through wireless emergency messages before the slower, more destructive waves reach populated areas.
Testing shows that ShakeAlert’s wireless alerts reach phones within a median of 6 to 12 seconds. For people located far enough from the epicenter, that translates to crucial seconds or even tens of seconds of warning to drop under a desk, pull over a car, or step away from hazardous equipment. Analysis of historical earthquakes suggests that had this system existed during the 1989 Loma Prieta earthquake, it would have provided meaningful warning time to areas relatively far from the epicenter. Japan and Mexico operate similar systems that have been credited with saving lives.
The Only Successful Prediction
The 1975 Haicheng earthquake in northeastern China remains the only widely recognized case where authorities successfully predicted a major earthquake and evacuated a city before it struck. Scientists had observed a series of precursors, including foreshocks, changes in groundwater, and unusual animal behavior, which a later analysis in Nature attributed to a deformation front that had propagated roughly 1,000 kilometers through northeastern China at about 110 kilometers per year. The evacuation likely saved tens of thousands of lives.
But just a year later, the 1976 Tangshan earthquake killed over 240,000 people with no warning at all. Haicheng was, in many respects, a fortunate combination of clear precursors and decisive action that has never been replicated. It illustrates both the potential and the deep limitations of current earthquake science: the tools exist to monitor faults, measure strain, and detect early shaking, but pinpointing the exact moment a fault will rupture remains one of the hardest unsolved problems in geophysics.