An earthquake early warning system does not predict earthquakes. It detects one that has already started and outruns it, and the size of the head start is set by physics that leaves no room for engineering.
Two kinds of wave leave a rupture at once. The P-wave arrives first and does little damage; the USGS puts its speed at about 3.7 miles per second. The S-wave and the surface waves that follow are what shakes buildings, and they travel at about 2.5 miles per second. Seismometers near the epicenter register the P-wave, and the data travels to a processing center at the speed of a network link, which is far faster than either wave. If the system can classify the event and publish an alert before the S-wave reaches a given city, that city gets a warning.
The USGS operates ShakeAlert on this principle across California, Oregon, and Washington, building toward a network the agency describes as roughly 1,700 sensors. Public alerting began in California in 2019 and extended to Oregon and Washington on 4 May 2021, covering more than 50 million people. Between 17 October 2019 and 1 September 2023, the system issued 41 public alerts for earthquakes of magnitude 4.5 or greater, according to the USGS review of ShakeAlert’s performance over that period.
The gap between two speeds
Everything available to the system is the difference between the two arrival times. At a distance d in miles, the P-wave arrives at d ÷ 3.7 seconds and the S-wave at d ÷ 2.5. The difference grows linearly with distance at about 0.13 seconds per mile.
| Distance from epicenter | P-wave arrives | S-wave arrives | Total gap available |
|---|---|---|---|
| 10 miles | 2.7 s | 4.0 s | 1.3 s |
| 25 miles | 6.8 s | 10.0 s | 3.2 s |
| 50 miles | 13.5 s | 20.0 s | 6.5 s |
| 100 miles | 27.0 s | 40.0 s | 13.0 s |
| 200 miles | 54.1 s | 80.0 s | 26.0 s |
That last column is a ceiling, not a delivery time. Nothing in it has been spent yet on the work the system must do: waiting for the P-wave to reach enough stations to be confident it is an earthquake, estimating magnitude and location, publishing, and pushing the message to a phone. ShakeAlert-powered phone alerts have documented delivery times in the range of 2 to 5 seconds after publication, and that is only the final leg.
Invert the table and the constraint becomes concrete. For the gap to reach 5 seconds, a location must be about 39 miles from the epicenter. For 10 seconds, about 77 miles. For 20 seconds, about 154 miles. Anyone closer than that is inside the region where the damaging wave arrives before, or barely after, the alert does. This is the blind zone, and it is not a gap in coverage that more sensors can close. Adding stations shortens the detection step by a second or two. It does not change the 0.13 seconds per mile that the arithmetic hands out.
Depth compounds it. A rupture 5 miles below the surface is 5 miles away from the point directly above it, so a city sitting on the epicenter is not at distance zero — it is at the depth of the quake, with a gap under a second. The people closest to the strongest shaking are the ones the system serves worst, and the strength of the shaking and the length of the warning fall off together with distance.
What the thresholds are doing
ShakeAlert does not alert on everything it detects, and the two delivery channels have different settings. Wireless Emergency Alerts, the channel that overrides a phone’s silent mode, go to people expected to feel Modified Mercalli intensity IV, light shaking, or greater, for earthquakes of magnitude 5.0 and above. Phone apps and built-in operating-system alerts run looser: MMI III, weak shaking, and above, for magnitude 4.5 and above.
The split reflects the cost of each error. A missed WEA in a damaging quake wastes the whole system. A false or trivial WEA at 3 a.m. teaches millions of people to disable the channel, which also wastes the whole system. The M4.5 threshold sits deliberately below the level where damage typically begins, because the first seconds of data underestimate large ruptures — a magnitude estimate built from a few seconds of P-wave can be revised upward as the fault continues to slip.
What the seconds are for
The lead times in the table are too short for most of what people imagine doing with a warning. At 25 miles, 3.2 seconds is not enough to leave a building, and leaving is the wrong action anyway. It is enough to get under a table, which is what the alert text instructs.
The larger value is mechanical. Machines act in the part of the interval humans cannot use. A 3-second warning is enough for an elevator controller to stop at the nearest floor and open the doors, for a train system to begin braking, for a surgical team to lift instruments, and for an automated valve to close. These consumers of ShakeAlert need no reaction time and lose nothing to a false alarm.
The math says a resident 10 miles from the rupture gets 1.3 seconds of physics, minus everything the system spends, which usually means the shaking and the alert arrive together. A resident 100 miles out gets 13 seconds and can act on it. Any planning built on early warning should therefore be organized by distance from the fault, not by state coverage maps: near the fault, the money belongs in automated shutoffs and in building retrofits that assume no warning at all, and it is only past roughly 39 miles that a human being has time to do anything with the notice.