SYSTEM Cited by 1 source
Android Earthquake Alerts System (AEA)¶
Android Earthquake Alerts (AEA) is Google's planet-scale earthquake early-warning (EEW) system. Android devices act as a distributed seismometer fleet — their accelerometers detect local ground motion, the aggregated signal is used to detect earthquakes and estimate their parameters (location, magnitude, expected shaking intensity), and the system pushes alerts to users in the expected path of shaking before the damaging S-waves reach them. From the sysdesign-wiki's perspective, AEA is interesting as a global-scale real-time sensing + inference + geo-scoped alerting system running on a commodity consumer-device fleet rather than on purpose-built seismic infrastructure.
Stub page. The wiki currently has one ingested AEA source and it captures only the introduction of the 2025-07-17 post (magnitude-estimation framing + the top-line MAE 0.50 → 0.25 improvement). The architectural substance of AEA — sensor-fleet topology, detection algorithm, alert routing, country-by-country rollout, false-alarm handling, privacy model, partnerships with USGS / national agencies — is in the original post but not in the local raw markdown. This page will expand as more AEA posts are ingested.
Design shape (inferred from the one ingested source)¶
- First-magnitude-estimate as the load-bearing numerical surface. The post frames the hardest part of AEA as estimating magnitude in real time, because magnitude determines how far shaking propagates and who should be alerted. All alerting logic downstream of this estimate inherits its error.
- Explicit speed-vs-accuracy trade-off. The first few seconds of an event carry limited data, but every second waiting for the estimate to converge is a second of warning lost. AEA is a canonical instance of the real-time-decision class where latency and correctness trade off directly.
- Continuous-improvement cadence as first-class property. Three-year trajectory of MAE 0.50 → 0.25 implies ongoing model / algorithm iteration against a ground-truth catalog (likely post-event magnitudes from traditional networks), not a one-shot deployment.
- Comparison baseline is traditional seismic networks. AEA positions itself against purpose-built broadband / strong-motion seismic stations (USGS and equivalents) at the first-magnitude-estimate step, not at the full post-event seismic-catalog level.
Reported metrics (from the intro-only raw)¶
- Median absolute error of first magnitude estimate: 0.50 → 0.25 (moment-magnitude units) over the three-year window ending 2025-07-17.
- Accuracy vs traditional networks: "similar to, and in some cases, even better than" — scoped to the first-estimate step.
- Global deployment scale, alert volume, sensor-fleet size, per-country rollout dates, false-alarm rate, end-to-end latency budget, privacy / telemetry design: not disclosed in the captured raw.
Seen in¶
- sources/2025-07-17-google-android-earthquake-alerts — Google Research post on AEA as a global EEW system; captured raw is intro-only, covers magnitude-estimation trade-off + three-year MAE improvement.