Earthquake Early Warning: The Engineering Behind Those Critical 10–60 Seconds

When a magnitude 7.1 earthquake struck southern California in 2019, residents across a wide area received smartphone alerts several seconds before the shaking arrived. Some people had enough time to drop and take cover. Trains automatically slowed. Surgeons stopped cutting. Those seconds came from one of the most elegant applications of physics in public safety engineering: earthquake early warning.

## The Physics That Makes Warning Possible

Every earthquake generates multiple types of seismic waves. The P-wave (primary wave) travels through the earth at 6–8 km/s. The S-wave (secondary wave) travels at roughly 3.5 km/s and carries most of the destructive shaking energy. Critically, P-waves arrive first but cause little damage; S-waves arrive later but are responsible for most structural failure and human harm.

This velocity differential is the physical basis for earthquake early warning (EEW). A sensor network near the epicentre detects the P-wave, algorithms estimate the earthquake magnitude, and alerts are transmitted electronically at the speed of light — roughly 300,000 km/s — before the damaging S-waves arrive at more distant locations.

The warning time available depends strictly on the distance between the observer and the epicentre, minus the time for P-wave detection and alert processing. For locations near the epicentre, warning time may be zero or negative. For locations 100–200 km away, warning time of 30–60 seconds is achievable. For locations 300+ km from a major subduction zone earthquake, minutes of warning are possible.

## ShakeAlert: The US West Coast System

ShakeAlert, operated by the USGS in partnership with state geological surveys, covers California, Oregon, and Washington. The network comprises approximately 1,500 seismometers, with dense spacing in urban areas and along major fault systems.

The alert processing pipeline runs as follows: when multiple sensors detect P-wave motion exceeding a threshold, a detection algorithm estimates earthquake location and magnitude from the first 2–4 seconds of P-wave data. An alert is issued if estimated magnitude exceeds a threshold (typically M4.5 or higher), and the alert is propagated to end users through multiple channels: the Federal Emergency Management Agency's Wireless Emergency Alerts (WEA) system sends cell broadcast messages to all mobile phones in the shaking zone, Google Android sends alerts natively, and iOS receives them through the WEA system.

Automated actions triggered by ShakeAlert alerts in California include slowing BART and Caltrain trains to prevent derailment, stopping surgeries in hospital operating rooms, and closing emergency valves in gas lines.

## Japan's JMA System: The Global Standard

Japan's Earthquake Early Warning system, operated by the Japan Meteorological Agency (JMA), is the most mature national EEW system in the world. Deployed in 2007, it has issued alerts for hundreds of significant earthquakes. The 2011 Tōhoku earthquake (M9.0) provided the system's most consequential test — alerts were issued tens of seconds before the major shaking arrived in Tokyo, 373 km from the epicentre.

Japan operates approximately 1,000 high-sensitivity seismometers (Hi-net) plus additional accelerometer networks. The sensor density is approximately one station per 20 km², far exceeding US coverage. Alert dissemination uses television broadcast interruption (a distinctive alarm tone, familiar to any Japan resident), mobile carrier alerts, and automated industrial systems.

## The False Alarm Problem

The first few seconds of P-wave data provide limited information for magnitude estimation. Early magnitude estimates can be revised substantially as more data arrives. This creates a fundamental trade-off: issuing alerts early (when warning time is most valuable) requires accepting higher false alarm rates.

The ShakeAlert system has encountered this trade-off in practice. A 2018 algorithm version issued an alert for a M4.7 earthquake in Berkeley that was initially misestimated as much larger. The system has been refined to reduce false alarms, but eliminating them entirely would require waiting for more data, reducing warning time. The engineering target is false alarm rates below 10% for alerts that reach end users.

## 2026 Improvements: Machine Learning and Offshore Sensors

The current generation of EEW systems is increasingly incorporating machine learning for rapid magnitude estimation. Neural networks trained on large historical earthquake catalogues can extract magnitude estimates from the first 1–2 seconds of P-wave data more accurately than traditional parametric methods, potentially extending useful warning times.

Offshore sensor networks are being deployed by Japan and the US to detect submarine earthquakes closer to their source, before P-waves reach land-based sensors. Japan's S-net (Seafloor observation network for earthquakes and tsunamis) places 150 seismometers on the Pacific Ocean floor, giving warning of Tōhoku-type megathrust earthquakes 20–30 seconds earlier than land-based detection alone.

The fundamental physics limit cannot be overcome — if you are at the epicentre, you have zero warning time. But for the majority of populated areas exposed to seismic risk, the combination of dense sensor networks, rapid algorithms, and broadband alert dissemination can deliver the seconds that matter.

Earthquake Early Warning: The Engineering Behind Those Critical 10–60 Seconds

// COMMENTS

ON THIS PAGE