Prediction vs. forecasting
Scientists draw a sharp line between prediction and forecasting. A true prediction would specify the exact time, location, and magnitude of a coming earthquake within narrow limits, and no reliable method to do this exists today. Forecasting, by contrast, estimates the probability that a quake of a given size will strike a region over years or decades, and this is well established and widely used.
Seismic hazard maps are the practical result of forecasting. By studying fault slip rates, historical earthquakes, and the accumulated strain along a fault, geologists can say, for example, that a major quake has a certain percentage chance of hitting a region in the next 30 years. These probabilities guide building codes and insurance, even though they cannot tell you the day it will happen.
The search for reliable precursors
For over a century researchers have hunted for precursors: measurable signals that reliably appear before a quake. Candidates have included changes in groundwater level, radon gas emissions, ground deformation, unusual animal behavior, and foreshocks. The problem is consistency. A signal may precede one earthquake and be entirely absent before the next, or appear many times without any quake following.
Foreshocks illustrate the difficulty well. Some large earthquakes are preceded by smaller shocks, but most small earthquakes are not foreshocks of anything larger. Only after the main event can seismologists say with certainty that earlier tremors were foreshocks, which makes them useless for issuing a warning in advance.
Early warning systems
What does work is earthquake early warning, which is not prediction at all. Systems like ShakeAlert in the United States, Japan's nationwide network, and Mexico's SASMEX detect the fast, weak P-waves that arrive first and automatically alert people before the slower, destructive S-waves reach them. The warning lead time ranges from a few seconds to over a minute depending on distance from the epicenter.
Even a few seconds of warning has real value. It is enough to stop trains, halt surgeries, shut gas valves, open firehouse doors, and let people drop, cover, and hold on. These systems save lives and reduce damage, but they only work once the rupture has already begun, so they detect rather than predict.
Why exact prediction remains elusive
The physics of fault rupture is deeply nonlinear and sensitive to conditions kilometers underground that we cannot directly measure. Tiny differences in stress, rock strength, and fluid pressure determine whether a small slip stops or cascades into a great earthquake. Because the crust behaves as a complex, self-organizing system near a critical threshold, identical-looking conditions can produce very different outcomes.
Researchers are exploring machine learning applied to continuous seismic data, laboratory quakes, and satellite deformation measurements, and some results are promising for understanding the rupture process. But no approach has yet delivered short-term predictions that beat chance in real-world tests. For now, preparation and building resilience matter far more than waiting for a warning.