Can Earthquakes Be Predicted?
April 20, 2026
The short answer, from the USGS and every other national seismological agency, is no. “Neither the USGS nor any other scientists have ever predicted a major earthquake,” the agency states plainly, “and we do not know how, and we do not expect to know how any time in the foreseeable future.” But three different things get called “prediction,” and only one of them is impossible.
Prediction vs. forecasting vs. early warning
- Prediction would mean naming a specific window: this fault, this month, roughly this magnitude, with enough confidence to act on. No reliable method to do this exists.
- Forecasting is probabilistic and long-term: “a 72% chance of a magnitude 6.7 or larger somewhere in the San Francisco Bay Area in the next 30 years.” This is real science, based on fault slip rates and earthquake history, and it guides building codes and insurance.
- Early warning detects an earthquake that has already started and races an alert ahead of the shaking, using the fact that data travels faster than seismic waves. Systems like ShakeAlert on the US West Coast, and Japan’s national system, give seconds to tens of seconds of warning — enough to stop trains and open firehouse doors.
What has been tried, and failed
For decades, scientists searched for a precursor — some measurable signal that reliably appears before a large earthquake. The candidates have included foreshock patterns, ground uplift, changes in radon gas in well water, changes in electrical resistivity or groundwater level, unusual animal behaviour, and claimed electromagnetic or ionospheric signals.
None has survived scrutiny. Signals that looked promising after one earthquake did not appear before the next, or appeared just as often when no earthquake followed. The most famous formal test, the Parkfield experiment in California, instrumented a fault segment that had produced regular earthquakes about every 22 years and forecast the next by 1993. It arrived in 2004, with no detected precursor.
Claims of imminent earthquake predictions circulate regularly online. When checked against the record they perform no better than chance, usually because they are vague about time, place or magnitude — or all three.
Why it is so hard
A fault can sit near failure for years. Whether a small patch slips and stops, or grows into a magnitude 8, seems to depend on fine details of stress and friction over the whole rupture area — details we cannot measure kilometres underground. Laboratory and modelling work suggests the outcome may be genuinely unpredictable in principle, not just with current instruments.
What actually helps
Because we cannot predict earthquakes, the effective response is preparedness that does not depend on warning: earthquake-resistant construction, retrofitting older buildings, securing furniture and water heaters, planning for utility outages, and knowing to Drop, Cover and Hold On. A building that survives is worth more than any forecast.
This map is part of the same picture: it shows where earthquakes have happened, which is the best available guide to where they will happen again.
Sources
- U.S. Geological Survey — “Can you predict earthquakes?”, “Earthquake Prediction”, ShakeAlert program pages.
- Field, E. H. et al. (2015). “Long-Term Time-Dependent Probabilities for the Third Uniform California Earthquake Rupture Forecast (UCERF3).” BSSA 105(2A).
- Bakun, W. H. et al. (2005). “Implications for prediction and hazard assessment from the 2004 Parkfield earthquake.” Nature 437, 969–974.