Earthquake early warning explained: the ideas that matter

01 Idea one: warning is not prediction

Prediction would tell us before rupture begins that an earthquake is likely at a particular place and time. Early warning begins after the rupture has started. Sensors detect the first arriving signal, estimate what is happening, and warn locations that the stronger motion has not reached yet.

The difference is more than a vocabulary lesson. It explains both the power and the limit of the system. EEW can exploit a known event in progress, but it cannot provide tomorrow’s earthquake calendar or guarantee a warning for the nearest point to the rupture.

02 Idea two: P waves are the first clue

Earthquakes send several kinds of seismic waves. The faster P wave often arrives before the slower S waves and surface waves that cause much of the damaging motion. A nearby sensor can therefore provide information about an event before a more distant community experiences its strongest shaking.

The useful signal is not a magic alarm hidden in the earth. It is an early, partial observation. The system must decide how much confidence that partial observation deserves and update the estimate as more stations and more waveform arrive.

03 Idea three: the alert is about your site

A magnitude describes an earthquake at its source, but people need to know what is likely at their location. Distance, depth, rupture direction, local soil, and building characteristics all influence experienced shaking. Early-warning services therefore convert network measurements into an estimate of expected intensity or arrival time for a target area.

That local framing also explains why two people can receive different messages—or why one person receives no useful lead time while someone farther away does. The alert is a forecast of approaching motion at a site, not a universal label attached to the earthquake.

04 Idea four: speed and accuracy trade places

Waiting for every station can improve an estimate but reduce the time available to act. Issuing an alert from the first strong evidence preserves time but leaves more uncertainty. Systems balance those goals with staged updates, thresholds, confidence measures, and different policies for different automated actions.

The U.S. Geological Survey describes the purpose of ShakeAlert as delivering alerts potentially seconds before strong shaking and supporting both public and automated responses. “Potentially” is doing important work: lead time depends on geography, network conditions, and the event.

05 Idea five: the action must be simple

An alert is useful only if the next step is already understood. For people, Drop, Cover, and Hold On is a short instruction that remains valid when a warning arrives or shaking begins. For infrastructure, the action may be to slow, isolate, shut, pause, or switch to a safe state.

Complex instructions consume the very time the system created. Good alert design puts the protective action first and leaves detailed explanation for later, when the immediate hazard has passed.

06 The limits belong in the explanation

Japan’s Meteorological Agency warns that the system can provide valuable seconds but also documents limitations. Strong shaking can arrive before an alert, especially near the source; estimates can change; and a warning does not mean every site will experience the same intensity. These are not footnotes. They define responsible use.

A credible explanation therefore includes blind zones, uncertainty, false alarms, communication delays, and the difference between a notification and a guarantee. Understanding the limits helps people respond to an alert without expecting an impossible level of certainty.

07 The compact mental model

Remember a five-step loop: rupture, detect, estimate, deliver, act. The first step has already happened when early warning begins. The middle steps race against the waves. The final step determines whether the information changes risk.

That model is enough to avoid the two most common misunderstandings. EEW is neither clairvoyance nor useless noise. It is a real-time estimate that can buy a few protective seconds for some places, under some conditions, when people and systems are ready to use them.