Earthquake Detection and Early Warning Multi-Source
by Nick Clark | Published April 25, 2026
Earthquake early warning systems—USGS ShakeAlert in the western United States, Japan's JMA EEW, Mexico's SASMEX, Taiwan's CWB, and Turkey's AFAD—operate against a physics deadline measured in single-digit seconds. They fuse strong-motion seismometers, broadband stations, GNSS displacement, structural accelerometers, and increasingly smartphone and consumer-IoT sensors. The environmental-disruption primitive provides multi-source corroboration, multi-medium sensing, governed active probing, and signed observation lineage as the architectural substrate that EEW operations now require.
Earthquake Early Warning Domain
USGS ShakeAlert went fully public in California in October 2019, Oregon in March 2021, and Washington in May 2021, operating on roughly 1,675 seismic stations across California, Oregon, and Washington as of the 2023 ShakeAlert Annual Report. The system distributes alerts through the Wireless Emergency Alert (WEA) channel, the FEMA IPAWS infrastructure, the MyShake and QuakeAlertUSA apps, and partner Earthquake Early Warning licensees including transit operators (BART), utility operators (PG&E), and elevator OEMs. JMA EEW, operational since October 2007 in Japan, draws on more than 690 seismic stations supplemented by Hi-net borehole sensors and the K-NET/KiK-net strong-motion networks operated by NIED.
Mexico's SASMEX, operational since 1991 and one of the oldest public EEW systems, covers the Guerrero, Oaxaca, Michoacán, Jalisco, Colima, and Puebla seismic gaps and broadcasts to Mexico City over a dedicated 162–163 MHz radio link with audible sirens in over 12,000 public locations. Taiwan's CWB system, deployed in 2014, achieves first alerts in approximately 8–10 seconds after origin time using a P-wave on-site algorithm. Turkey's AFAD operates a national strong-motion network that contributed to rapid intensity estimation during the February 2023 Kahramanmaraş sequence (M7.8 and M7.5).
The newest layer is consumer-grade sensing. The MyShake project at UC Berkeley uses smartphone accelerometers, with an installed base above 1.5 million users as of 2024, and detected the July 2019 Ridgecrest sequence and the 2021 Tokyo M5.9. Google's Android Earthquake Alerts System extends the same approach to billions of devices globally, broadcasting alerts in countries without a national EEW. The Quake-Catcher Network, the Raspberry Shake citizen-seismograph network (over 2,000 stations), and structural-health-monitoring sensors in modern buildings round out the multi-medium picture.
The Architectural Requirement
Earthquake detection has hard latency budgets. The S-wave, which carries the destructive ground motion, propagates at roughly 3–4 km/s in the upper crust; the P-wave, which arrives first and carries the warning signal, propagates at roughly 6–8 km/s. For an event whose epicenter is 50 km from a population center, the warning window between P-wave detection and S-wave arrival is on the order of 5–8 seconds. Every architectural decision—source authentication, sensor fusion, false-positive rejection, alert distribution—must complete inside that window.
The architectural requirement, then, is not a single highly trusted source but a federation of differently-credentialed sources whose corroboration can be evaluated in milliseconds. A USGS ANSS-class broadband station, a NIED Hi-net borehole sensor, a Raspberry Shake citizen station, and a smartphone accelerometer report in different units, with different sampling rates, different latencies, different noise profiles, and different trust levels. The detection system must accept observations from all of them, weight them by source class, and produce a corroborated event hypothesis whose confidence is itself attributable to the contributing sources.
Multi-medium corroboration adds a second axis. A real earthquake displaces the ground in a way that strong-motion sensors record, that GNSS receivers register as a static displacement after the fact, and that structural accelerometers in nearby buildings record as a coherent shake-front. A spoofed or malfunctioning single-source signal does not produce coherent multi-medium evidence. The primitive must allow the detection system to require corroboration across mediums for high-impact alerts (mass WEA broadcast, automatic train stop, gas-line shutoff) while accepting single-medium evidence for lower-impact alerts (in-app warnings, advisory notices).
Why Procedural Compliance Fails Alone
Operational EEW networks already implement multi-station declaration thresholds. ShakeAlert's EPIC and FinDer algorithms require multiple stations to declare before an alert is issued; JMA EEW uses similar coincidence logic. These are procedural rules encoded in the central message broker and applied to a closed federation of pre-vetted stations. They work well for the legacy network and they fail for the emerging crowdsourced and cross-jurisdiction layers.
A procedural rule that says “four stations within 100 km must declare” cannot distinguish a genuine event corroborated by three USGS broadbands and one Raspberry Shake from a spoofed event corroborated by four compromised consumer devices, unless the rule is rewritten to encode source-class weights. Once those weights are added, the rule becomes a small policy engine; once events span jurisdictional boundaries (Cascadia rupture spanning USGS and NRCan; San Andreas event spanning California and Baja California where SASMEX operates), the policy engine must reason about cross-network credentials it does not natively understand.
The 2018 false ShakeAlert for an M6.8 off the Oregon coast (later attributed to a station-clock error) and the 2017 SASMEX false alarm during the September 19 Puebla earthquake aftershock sequence both involved single-medium signals that procedural coincidence logic accepted. Adding more stations does not address the failure mode: it only raises the threshold at which a coordinated multi-station fault produces a false alert. The architectural fix is to require evidence across mediums (seismic + GNSS + structural) and across source classes (network + citizen + IoT), with the corroboration policy expressed declaratively rather than baked into broker code.
What the Environmental-Disruption Primitive Provides
Multi-source corroboration is the primitive's first operation. Each contributing observation arrives with a credential identifying its source class (ANSS broadband, NIED Hi-net, K-NET strong-motion, Raspberry Shake citizen, MyShake smartphone, Android AEAS), its station identity, its calibration metadata, and a signed timestamp. The detection layer evaluates a corroboration policy that is explicit about how many sources of which classes are required for which alert tier. The policy is data, not code, and it can be revised under change control without redeploying the broker.
Multi-medium sensing is the second operation. The primitive treats seismic ground motion, GNSS displacement, accelerometer-derived intensity, infrasound (which records the acoustic signature of large ruptures), and even social-signal proxies as distinct mediums whose agreement is independently meaningful. For a Cascadia subduction event, agreement among the seismic medium (ANSS), the geodetic medium (PBO/EarthScope GNSS), and the structural medium (instrumented buildings in Seattle, Portland, Vancouver) is much stronger evidence than agreement among ten seismic stations alone, because a single-medium failure mode (regional clock drift, network partition) cannot manufacture cross-medium agreement.
Governed active probing addresses the cold-start and verification problem. When initial observations are ambiguous—for example, two ShakeAlert stations declaring but no GNSS displacement above noise—the primitive can issue declared probes (request raw waveform from a nearby station, query a structural-monitoring partner for accelerometer status, query an IoT gateway for power-grid frequency excursion) under a policy that bounds probe rate, target diversity, and credential exposure. Signed observation lineage closes the architecture: every alert carries the chain of contributing observations, their credentials, and the policy evaluation that admitted them, so post-event review (and false-alert root cause analysis) operates on evidence rather than reconstruction.
Operational and Standards Mapping
The USGS ShakeAlert Joint Committee for Communication, Education, Outreach, and Technical Engagement (JCCEOTE) requirements for licensee technical partners specify alert reliability, latency, and false-alert rate targets. The primitive's signed lineage and explicit corroboration policy map directly onto JCCEOTE auditability expectations: every alert delivered to a transit operator or utility carries the evidence under which it was admitted. ITU-T X.1303bis (Common Alerting Protocol, CAP 1.2) is the message-format standard for public alerts; the primitive's lineage records nest cleanly inside CAP <info> and <parameter> elements without breaking interoperability.
IRIS/EarthScope FDSN webservice metadata (StationXML, SEED) provides the station-identity substrate the credential layer references, so source-class assertions resolve against an authoritative registry. The IRIS Federator and the ORFEUS European node provide cross-jurisdiction discovery that the primitive's federation policy can consume. For Japan, the JMA seismic intelligence sharing arrangement and the NIED Hi-net data-sharing protocols provide analogous registries; for Mexico, CIRES (the operator of SASMEX) has established data-sharing memoranda with USGS that map onto the federation model.
For the consumer layer, Google's Android Earthquake Alerts System publishes an architecture document describing how aggregated phone accelerometer signals are anonymized and weighted; the primitive's source-class credential treats AEAS as a single high-volume low-trust contributor whose corroboration with one network-class source produces a tier-2 alert and whose corroboration with two mediums produces a tier-1 alert. The MyShake program operates similarly. WEA delivery follows FCC 47 CFR Part 10 requirements; the primitive sits upstream of the WEA gateway and shapes which events qualify for WEA broadcast versus app-only delivery.
Adoption Pathway
Operators of existing EEW networks adopt the primitive incrementally. The first step is to wrap each existing station feed with a source-class credential drawn from the operator's station registry, leaving the legacy coincidence rule in place. The second step is to add cross-medium corroboration as a parallel evaluator whose output is logged but does not yet gate alerts. After a calibration period (typically two to three months covering hundreds of small events) the operator compares the primitive's hypothetical alerts against the legacy alerts and, where the primitive would have suppressed false alerts or admitted true alerts earlier, the corroboration policy is promoted to the gating path.
Cross-jurisdiction federation follows once each participating network has published its source-class credentials. The Pacific Northwest Seismic Network (PNSN), USGS, and Natural Resources Canada have documented mutual data-sharing under the Cascadia Initiative; expressing those agreements as primitive federation policies is a documentation exercise rather than a new technical integration. SASMEX–USGS coordination during M7+ events at the U.S.–Mexico border benefits similarly.
Consumer-layer integration is greenfield in many jurisdictions. National civil-protection authorities adopting Android AEAS or similar (as Greece, the Philippines, and several Caribbean nations have done) can express the integration as a primitive federation between the consumer aggregator and the national seismic network, with the corroboration policy documenting under what conditions consumer-only evidence is sufficient for which alert tier. The result is that EEW operators retain the latency budget the physics demands while admitting the broader sensing fabric the next decade of earthquake monitoring will rely on.
