Before the First Earthquake Alert Message Exists, the Shaking Has Moved Eleven Miles

An earthquake alert that arrives while the floor is already moving reads like a broken alerting system. It is closer to a statement about distance. The USGS ShakeAlert system publishes a summary report for individual West Coast earthquakes, and each report prints the number of seconds between the earthquake's origin time and the system's first message. Across nine of those reports, for events between June and September 2026, the interval ran from 3.6 seconds to 26.3 seconds. The ground does not wait during those seconds.

Shaking travels outward at a speed the agency publishes, so every second spent detecting, estimating, transmitting and delivering converts into ground already shaken by the time a phone makes a sound. What follows is that chain, what the nine published summaries measure, and the radius those seconds imply.

The Chain

Seven steps sit between a fault slipping and a phone screen lighting up. One is the earthquake itself and two are wave travel. The remaining four are data handling, arithmetic and telecommunications, and they are where the seconds accumulate.

1. The rupture begins. Every number in the reports is measured from this instant, printed on each summary as “ANSS origin (UTC)”.
2. The fastest waves reach the nearest instruments. USGS states that “The fastest-moving seismic waves (primary or P-waves) travel about 3.7 miles per second and generally do not produce strong shaking.” How long this step takes depends on how far the nearest seismometer sits from the rupture, which is a fact about the network rather than about the earthquake.
3. The data reach a processing center. In the agency's description, “The ShakeAlert sensor network detects earthquakes quickly, and immediately transmits data to a ShakeAlert Processing Center, where estimates of the location, size, and shaking intensity of the earthquake are determined in a matter of seconds.”
4. The first message is issued. Each summary prints this under the heading “ShakeAlert Messages Issued (after origin time)”, with an Initial, a Peak and a Final entry in seconds.
5. The message is handed to a delivery pathway. An app over Wi-Fi, an app over a cellular connection and the Wireless Emergency Alert system are separate paths with separate delays.
6. The phone alerts. The message now exists for a person rather than for a machine.
7. The damaging waves arrive. USGS states that “P-waves are followed by slower moving, and generally more damaging waves (secondary or S-waves) and surface waves that travel about 2.5 miles per second.” This step is running in parallel with steps two through six, and it does not stop for any of them.

What Nine Published Summaries Measure

The reports are posted event by event and carry, in the listing page's words, “information about earthquakes detected by the USGS component of ShakeAlert Earthquake Early Warning System”. They are generated automatically when the system's estimated peak magnitude is 3.5 or larger, and a person reviews them before posting when that estimate is 4.0 or above. The same page adds what happens below that line: summaries for earthquakes under M4.0 are produced “to test software readiness, but the results are generally not reviewed for quality or accuracy”. Each one prints the origin time, the time of the first message, the seconds between them, the magnitude the system estimated at three stages, how far the estimated location sat from the location seismologists later settled on, and how many instruments were reporting. Both timestamps are printed too, and subtracting them can miss the printed seconds figure by a tenth: the Eastvale summary shows an origin at 12:37:12.7 and a first message at 12:37:16.4, a difference of 3.7, against a printed 3.6.

Nine reports, ordered by the seconds from origin to the first message, with the count of instruments the report lists within 100 kilometers of the epicenter:

• 3.6 seconds — M 3.4 northwest of Eastvale, 2 September 2026. 172 instruments.
• 3.7 seconds — M 4.1 north of San Leandro, 13 August 2026. 100 instruments.
• 4.1 seconds — M 4.0 east of Gonzales, 11 August 2026. 31 instruments.
• 4.4 seconds — M 3.8 north of Ridgecrest, 23 August 2026. 26 instruments.
• 4.8 seconds — M 3.2 east of Cloverdale, 12 August 2026. 36 instruments.
• 5.5 seconds — M 5.6 southeast of Willits, 24 June 2026. 13 instruments.
• 5.8 seconds — M 2.9 northwest of Calipatria, 11 August 2026. 60 instruments.
• 18.9 seconds — M 4.9 offshore west of Fortuna, 29 August 2026. 2 instruments.
• 26.3 seconds — M 4.9 offshore west of Gold Beach, 6 August 2026. No instruments.

Two of the nine reports list two or fewer instruments within 100 kilometers, and both of those events sat offshore. The other seven list between 13 and 172, and their first messages came between 3.6 and 5.8 seconds after origin. The median of those seven is 4.4 seconds, which is this article's calculation across the seven published figures rather than a number the agency prints anywhere. Four of those seven — Eastvale, Ridgecrest, Cloverdale and Calipatria — settled below the M4.0 line quoted above, and the median itself is the Ridgecrest figure.

The Willits report also prints what those seconds meant on the ground. It lists Willits itself at 12 kilometers from the epicenter with about zero seconds of warning and Strong shaking, Ukiah at 24 kilometers with about 2 seconds and Moderate shaking, and Santa Rosa at 112 kilometers with about 26 seconds and Light shaking. That single column is the argument: warning time is a function of distance from the epicenter.

From Seconds to Kilometers

The conversion needs one constant and one published speed. A mile is 1.609344 kilometers, and USGS gives the damaging waves a speed of about 2.5 miles per second, which works out to 4.02 kilometers per second. At that speed the shaking front covers 17.7 kilometers, or 11.0 miles, during the 4.4-second median of the seven reports that list more than two instruments within 100 kilometers. That distance is this article's calculation, and it describes the state of the ground at the moment the message exists inside the system, before anyone has been told anything.

Delivery is a second budget. USGS Open-File Report 2021-1026, on expected warning times, sets out the assumption it uses: “Values of 1 second for alerts delivered via the internet or to cell phone apps and Android phones via Wi-Fi, 4 seconds for alerts delivered to cell phone apps or Android phones via cellular connections, and 8 seconds for WEA are within the range of latencies reported to ShakeAlert by various partners during testing in 2019 and 2020 and our own WEA tests.” The report then qualifies the numbers directly: “However, individual alerts may be faster or slower than these values.”

Adding each of those delays to the 4.4-second median produces three different rings around the same epicenter. This article's calculation:

• App over Wi-Fi, 5.4 seconds total. A radius of 21.7 kilometers, an area of 1,483 square kilometers.
• App over a cellular connection, 8.4 seconds total. A radius of 33.8 kilometers, an area of 3,588 square kilometers.
• WEA, 12.4 seconds total. A radius of 49.9 kilometers, an area of 7,819 square kilometers.

The difference between the fastest and the slowest pathway is 6,336 square kilometers, and the slowest ring is 5.3 times the area of the fastest. Two people side by side can receive the same earthquake on different pathways, and the ground between those rings is where one is warned and the other is not.

USGS publishes its own figure for the ring. On the agency's blind zone page, the term is defined as “an area around the earthquake epicenter that cannot be alerted before the strong shaking starts”, and the size is given as follows: “current seismic network and communication limitations result in about a 25-km (15-mi) radius zone centered on the epicenter for which alerts are not possible.” The sentence immediately after that one carries the assumption it rests on: “And, currently, a 2-3 second communication latency for cell phones is not achievable in the U.S.” A 25-kilometer radius corresponds to about 6.2 seconds of total elapsed time at the published wave speed, which is this article's calculation and sits between the Wi-Fi and cellular cases above.

Two rules of thumb fall out of the same arithmetic. Each additional second anywhere in the chain pushes the edge outward by about 4 kilometers. Each 10 kilometers of extra distance from the epicenter buys about 2.5 seconds.

Where the Nearest Instruments Sit Offshore

The two slow reports are the ones where the network had nothing close to work with. The Fortuna event was centered 107 kilometers west of Fortuna, California; its report lists no instruments within 10 kilometers, 2 within 100 kilometers, and 9 instruments used in the final message. The first message came at 18.9 seconds, by which point the shaking front had covered about 76 kilometers. The Gold Beach event was 129 kilometers west of Gold Beach, Oregon, with no instruments within 100 kilometers and 6 used in the final message, and its first message came at 26.3 seconds, or about 106 kilometers of travel.

The instrument count is not a simple dial, and the nine reports say so. Calipatria lists 60 instruments within 100 kilometers and took 5.8 seconds, slower than San Leandro's 3.7 seconds with 100. Willits lists only 13 and produced a message in 5.5 seconds. The count within 10 kilometers does not sort them either: four of the nine reports list no instrument within 10 kilometers, and two of those four produced first messages in 4.1 and 5.5 seconds. What the two slow reports share, and no other report in this set does, is a count of two or fewer within 100 kilometers.

What the First Message Does Not Know

Each summary prints Initial, Peak and Final entries for both timing and magnitude, and the gap between them records the system revising itself while the waves are still moving.

The Willits report is the well-behaved case. Its first message came at 5.5 seconds carrying an estimate of M 5.7; the peak message came at 16.9 seconds at M 6.1; the final message at 31.4 seconds settled at M 5.8. Seismologists later put the event at M 5.6. The location in that first message was 0.5 kilometers from where the earthquake actually was.

The Fortuna report shows the other end. The first message at 18.9 seconds carried M 3.9, the peak message at 24.0 seconds carried M 4.8, and the final message at 30.9 seconds carried M 4.7, against a settled magnitude of M 4.9. The location in that first message sat 64.7 kilometers east of the epicenter. The Gold Beach report shows no revision at any stage: Initial, Peak and Final each read M 4.1 against a settled M 4.9, with the location 48.0 kilometers off at every stage.

This is not a footnote about accuracy, because distribution depends on it. The summaries print the rule that WEA alerts go to the area of intensity 4 and above when the ShakeAlert peak magnitude reaches 5.0 or above. The Willits report states that the alert was distributed over the WEA system; the two offshore reports state that the magnitude was below the threshold. In the Willits case the first estimate of M 5.7 already cleared that threshold, and the entry the report labels Peak arrived 11.4 seconds later at M 6.1, which is the difference between two published entries in one report.

Where This Doesn't Apply

The 25-kilometer figure is a floor, not a ceiling. It is stated with a communication latency of 2 to 3 seconds in view, and the agency's own next sentence says that latency is not achievable in the United States at present. A real ring around a real epicenter, for a person on a real pathway, is wider than that number.

The nine reports measure issuance, not receipt. Everything they print stops at the moment ShakeAlert releases a message. The 1, 4 and 8 second delivery values come from a different document, which describes them as a range rather than a fixed cost.

Constant-speed arithmetic is a simplification. It ignores depth, the path the waves take through rock, and the distinction between S waves and surface waves. The per-city warning times printed in the same reports, including the Willits figures above, come from the agency's own shaking-arrival model rather than from this multiplication, and the two will not agree exactly.

Eight of the nine events were below magnitude 5.0. They measure how quickly the system detected small earthquakes, which is a fair test of the detection chain and a poor sample of the alerts a person is likely to receive. Four of the nine settled below magnitude 4.0, the class the summaries' listing page says is generally not reviewed for quality or accuracy, and that line is drawn on the system's own estimate rather than on the magnitude seismologists later settle on.

Geometry changes offshore and at depth. The Open-File Report puts it plainly: “There will be a region near the epicenter where shaking arrives before the alert, termed the late-alert zone.” Its next sentence is the operating instruction the agency attaches to that fact: “ShakeAlert end users should take protective actions as soon as they feel shaking whether they have received an alert yet or not.” Coverage is also limited to California, Oregon and Washington, so none of this describes what happens anywhere else.