Meteorological Cooperative › Warning systems › REP-KJS › Detection, latency and limits

Detection, latency and limits

REP-KJS works because the first seismic wave to arrive at a station is faster and weaker than the wave that does the damage. Everything the system achieves follows from that difference, and so does everything it fails to achieve. This page sets out the method, the time budget, the blind zone and the performance record, including every occasion on which the system was wrong.

The physical basis

An earthquake radiates two body waves. The P wave travels at about 6 km/s in the crust of the peninsula and displaces the ground along its direction of travel, which makes it weak. The S wave travels at about 3.5 km/s and displaces the ground across its direction of travel, which makes it strong and destructive. Surface waves follow more slowly and prolong the shaking.

A station near the origin therefore records the P wave several seconds before the S wave reaches it, and many seconds before the S wave reaches a settlement further away. The first few seconds of the P record contain enough information to estimate magnitude and depth. Sending that estimate onward at the speed of a network beats the S wave to every place beyond a certain distance.

The method in sequence

From origin to broadcast, ordinary times
Stage	Elapsed	What occurs
Origin	0.0 s	Rupture begins at depth
First detection	1.4 to 3.6 s	P wave reaches the nearest station; the trigger fires when the amplitude over the last fraction of a second rises sharply above the background level of the preceding minute
Confirmation	2.2 to 5.0 s	A second and third station trigger; the location is fixed by arrival times, which suppresses false triggers from a single station
First estimate	3.0 to 6.5 s	Magnitude estimated from the frequency and size of the first 3 s of the P record; depth from the order in which stations were reached
Intensity field	3.4 to 7.0 s	Expected intensity computed for every point through the ground-motion model and the soil amplification map
Broadcast	4 to 9 s	Segment 1 leaves the issuing node on every channel at once
Arrival at a device	+0.2 to 1.8 s	Network and device latency; a television decoder is the slowest ordinary receiver
Revision	12 s onward	Estimates refined as more stations report; segments 2 and 3 follow

The warning time actually available

Warning time at a place equals the S wave travel time to that place minus the time consumed by detection and delivery. Since almost every event originates beneath the Belvacius hills in the south-east, the warning time in the Republic is close to a function of distance from that margin. The peninsula measures about 830 km corner to corner, so the range is wide: Cortius receives seventeen seconds and Lindoma, at the far north-western end, receives above two minutes.

Warning time before onset, event on the Belvacius margin at 20 km depth
Place	Distance from epicentre	S wave arrives	Warning time
Corcina	14 km	7 s	None; inside the blind zone
Cortius	89 km	26 s	17 s
Marendo	111 km	32 s	23 s
Lutoliluri	156 km	45 s	36 s
Solvena	190 km	55 s	46 s
Prevessa	210 km	60 s	51 s
Sornorum	238 km	68 s	59 s
Leriduson	300 km	86 s	77 s
Dromera	358 km	103 s	94 s
Lindoma	478 km	137 s	128 s

The blind zone

The places that receive nothing in time

Within roughly 30 km of the epicentre the S wave arrives before the broadcast does. The warning reaches those devices during the shaking or after it. This is a property of the physics and no fault of the equipment: no arrangement of stations, no faster network and no better algorithm removes it.

The blind zone falls on the Belvacius sub-region and the settlements around Corcina, which is exactly the ground where the intensity is highest. Marendo, on the lagoon 111 km away, is the nearest settlement of any size outside it and receives about twenty seconds. The places most exposed to a Republic earthquake are the places a warning serves least well. Construction standards, fixed under Article 178 (1)(b), remain the only protection there.

Denser instrumentation shrinks the zone by a few kilometres and never abolishes it. The cooperative therefore publishes the size of the zone, since a coverage claim the physics contradicts would mislead the people living inside it.

Estimating from little information

The first estimate uses about three seconds of the P record from three or four stations. That is a small sample of an event that may rupture for twenty seconds, and the consequences are systematic.

Magnitude is underestimated for large events. A rupture still propagating has not yet revealed its full extent, so the initial figure of a magnitude 7 event may read as 6.2 and rise as the record accumulates. The effect is negligible below magnitude 5.5 and grows above it. Every large event therefore produces a rising sequence of revisions, and the broadcast states the figure as an estimate at every stage.

Depth is the least stable of the three parameters early on. A shallow event mistaken for a deep one is under-warned, since depth attenuates the shaking, and this is the error most likely to place an intensity estimate one step below the observed value.

Two nearly simultaneous events are the classic failure of systems of this kind. Arrivals from two ruptures combine into one apparent event of larger magnitude at an intermediate location. REP-KJS applies a cluster test that separates events more than 40 km apart, and separation fails when the two lie closer.

Performance record

Every event since March 2021 for which a warning was issued or should have been issued appears in the register below. The cooperative publishes the failures on the same footing as the successes, since a system whose error rate is unknown cannot be relied upon.

REP-KJS performance, March 2021 to the present
Measure	Count	Note
Events warned, REP 3 or above	41	All reached the whole territory
Warnings correct within one step	37	Estimated intensity within one step of that observed
Intensity overstated by two steps or more	2	June 2022 and November 2024, both from depth mis-estimated as shallow
Intensity understated by two steps or more	2	March 2023, soft ground at Ravelluri amplified beyond the model; April 2025, similar
Events missed entirely	1	September 2021, REP 3 at Prevessa; two stations offline for maintenance in the same window
False alerts issued	3	Two from quarry blasting near Ferrista, one from a station fault at Ilvara; each cancelled within 30 s
Median first-alert time	6.1 s	From origin time to departure of segment 1

Cancellation

A warning found to be spurious is cancelled on the same channels within thirty seconds, in the same fixed wording, and the cancellation states the reason. Quarry blasting at Ferrista produced two of the three false alerts, and the blasting schedule is now filed with the cooperative in advance and applied as a suppression window at the two nearest stations.

A cancellation is itself an event in the register. The cooperative publishes the count of false alerts because a system that cancels nothing is a system whose threshold is set too high.

What the system does not attempt

Outside the method

REP-KJS makes no prediction. It detects an earthquake that has already begun. No component of it forecasts an event before rupture, and no such method exists anywhere.

Aftershocks receive the same treatment as any other event: they are detected when they occur and warned on the same threshold. The system offers no statement about when an aftershock will follow.

The intensity estimate is for ground at the location. Motion inside a tall building differs from motion at its foundation, sometimes by several steps on an upper floor, and the estimate accounts for neither the building nor the floor.