TREMR

The question has been asked for centuries, and studied seriously for over fifty years: can we predict earthquakes? The answer that seismology has arrived at — after decades of research, government programs, and failed forecasts — is honest and clear. Short-term earthquake prediction, meaning identifying when, where, and at what magnitude a specific earthquake will occur, is not currently possible.

This is not a failure of effort. It reflects a deep understanding of how the Earth's crust actually works. This article explains why prediction is so difficult, what the history of attempts looks like, and what scientists are doing instead.

Why prediction is so hard

Earthquakes happen when a fault suddenly slips. The problem is that there is no reliable physical signal that tells us when a fault is about to move.

Intuitively, it seems like there should be precursors. As strain accumulates in rock, shouldn't the surface show detectable changes? If small earthquakes cluster, isn't that a warning of a larger one? Does radon gas seep from the ground, or groundwater levels shift, before a major rupture?

All of these candidates have been studied seriously. The problem is consistency. A phenomenon observed before one earthquake fails to appear before another. The same signal appears during periods with no earthquake at all. After half a century of research, no precursor has been found that predicts earthquakes with statistically meaningful reliability.

The deeper reason is the nature of fault systems. Faults exist kilometers to tens of kilometers underground, where they cannot be directly observed. Information gathered at the surface is indirect. And fault rupture begins at the exact moment strain exceeds a threshold — a threshold that cannot be known in advance, only recognized after the fact.

A history of failed predictions

The history of earthquake prediction is a record of optimism followed by disappointment.

In the 1970s, researchers in the Soviet Union, China, and the United States grew excited about the possibility of prediction. A handful of apparent successes attracted attention. In 1975, authorities in Haicheng, China evacuated residents before a M7.3 earthquake based on foreshock activity and ground deformation — and the evacuation is credited with saving many lives. Seismologists around the world took notice.

One year later, the 1976 Tangshan earthquake (M7.8) struck with no precursors whatsoever and killed approximately 240,000 people. Subsequent analysis showed that Haicheng's apparent success depended on an unusually clear foreshock sequence that is not a general feature of earthquakes. The method could not be replicated.

In the United States, a major prediction experiment was launched at Parkfield, California in 1988. Parkfield had historically produced M6 earthquakes on a roughly periodic schedule, making it an ideal test case. The prediction: another M6 within five years. The earthquake arrived in 2004 — well outside the predicted window.

In 2009, the L'Aquila earthquake in Italy (M6.3) struck six days after scientists publicly assessed the risk of a major earthquake as low. 309 people died. The scientists were prosecuted for manslaughter — and later acquitted. The case became a defining lesson about scientific communication under uncertainty.

Japan's prediction program — and a turning point

Japan runs one of the world's most systematic earthquake monitoring programs. Since 1965, the Coordinating Committee for Earthquake Prediction has designated observation zones across the country, continuously monitoring ground deformation, geomagnetism, and groundwater in areas judged to have elevated seismic risk.

The 2011 Tōhoku earthquake (M9.0) posed a fundamental challenge to this program. Despite Japan's extensive monitoring network, no precursors were detected before the largest earthquake in the country's recorded history. Prediction was impossible right up to the moment of rupture.

In response, Japan's earthquake science community significantly shifted direction. By 2017, the government's Headquarters for Earthquake Research Promotion had effectively abandoned the goal of short-term prediction for the Nankai Trough and redirected resources toward probabilistic long-term assessment and early warning systems.

What science can do instead

Modern seismology offers three things in place of prediction.

Long-term probabilistic assessment — statements like "this fault has a 70% chance of producing an M7+ earthquake within 30 years." Japan's Nankai Trough figure is the most prominent example. This is not prediction — it is a statistical estimate based on historical rupture intervals and strain accumulation rates. It says nothing about when within that window.

Earthquake early warning — systems that detect seismic waves the moment a rupture begins and transmit alerts ahead of the strong shaking. Japan's system, operated by the Japan Meteorological Agency since 2007, can provide seconds to tens of seconds of warning for people at distance from the epicenter. This is not prediction — the earthquake has already started. But even seconds matter: enough time to drop and take cover, stop a train, or halt a surgical procedure.

Anomalous event information — Japan introduced a specific system for the Nankai Trough in 2019: if an M8+ earthquake occurs on the subduction zone, or if unusual plate boundary deformation is detected, an advisory is issued. This is an alert that a larger event may follow — not a prediction that one will.

Animal behavior, folklore, and social media forecasts

After every major earthquake, reports emerge that animals behaved strangely beforehand — catfish thrashed, dogs barked, birds took flight. Are these real precursors?

The scientific assessment is: no evidence that they are. Reports of unusual animal behavior before earthquakes have circulated for centuries, but controlled studies have not confirmed a reliable pattern. Animals display unusual behavior constantly; when an earthquake occurs, people recall the anomalies and not the many other times the same behavior appeared without a quake following. This is confirmation bias, not prediction.

Social media earthquake forecasts operate the same way. If enough people predict earthquakes every day, some predictions will coincide with earthquakes by chance. That is not a signal — it is noise.

What this means for preparedness

Earthquakes cannot be predicted. The damage they cause can be reduced.

This is the foundational logic of modern earthquake preparedness: since warning before the event is not possible, the window for action is now — before any specific earthquake is anticipated. Retrofitting older buildings. Securing furniture. Storing emergency water and food. Knowing where to shelter and where family members will meet. Knowing what to do the moment shaking begins.

The governments planning for the Nankai Trough and the Tokyo directly-below earthquake are doing so explicitly on the premise that the date is unknown. That premise is scientifically accurate. The preparation it demands is real.

Science cannot tell you when the next earthquake will come. It can tell you, with remarkable precision, what happens when one does — and that knowledge, applied before the fact, is what saves lives.