Seconds before the wave

Earthquakes kill. Tsunamis kill more. The difference between a survivable event and a catastrophic one is often measured in seconds or minutes — time to drop, cover, hold on; time to reach higher ground; time to shut down trains, pipelines, and surgical suites before shaking or inundation arrives.

This note is about earthquake early warning (EEW): systems that detect an event in progress and alert people and infrastructure before the most damaging waves arrive. It is not about predicting earthquakes before they begin. That distinction matters. Confusing the two wastes public trust and misallocates resources when lives are at stake.

What early warning actually does

When a fault ruptures, seismic energy radiates outward in waves. P-waves arrive first; they are faster but usually less destructive. S-waves and surface waves follow; they carry most of the damage. An EEW network detects the initial rupture, estimates location and magnitude, and issues alerts while there is still a gap before strong shaking reaches distant sites.

Japan's JMA system, Mexico's SASMEX, and the ShakeAlert system in the western United States are the reference implementations. They have delivered real warnings — stopping bullet trains, triggering automated industrial shutdowns, pushing alerts to phones. They have also failed, been too slow, or reached the wrong people. Both outcomes belong in the same analysis. Deployment is not a solved problem.

The 2004 Indian Ocean earthquake and tsunami — magnitude 9.1+, more than 230,000 deaths across fourteen countries — remains the defining case for why oceanic coverage and tsunami-specific pathways cannot be treated as optional add-ons. Seismic detection alone is insufficient when the deadliest hazard travels across basins with minutes to hours of lead time that only instrumentation in the water can anchor.

Where the hard problems are

Latency. Every stage costs time: sensor capture, telemetry, processing, authority review, last-mile delivery. For near-field events, the warning window may be seconds. Systems must be engineered for worst-case paths, not demo conditions.

Coverage gaps. Land-based networks are uneven globally. Subduction zones offshore — where the largest tsunamigenic events originate — remain under-instrumented relative to the risk they carry. Distributed Acoustic Sensing (DAS), which uses existing fiber-optic infrastructure as a dense sensor array, is a promising path toward broader coverage without building entirely new networks from scratch. It is not a substitute for maintained, validated, publicly accountable sensor programs.

False alarms and missed events. Both erode trust. In EEW, trust is a safety-critical asset. A population that ignores alerts after repeated false positives is a population left unprotected when a real event arrives.

Equity. Warnings that reach smartphones but not rural communities, migrant workers, hospital wards, or non-dominant languages are not neutral technical failures. They are distributional harms. Civilian disaster systems must be evaluated on who receives how many seconds — not only on whether an alert was technically issued.

Where machine learning helps — and where it does not

Deep learning has materially improved phase picking and event detection from raw waveforms. Models such as PhaseNet and EQTransformer can process continuous data faster and more consistently than manual analyst workflows at scale. That matters operationally: faster, more reliable detection converts directly into warning time.

What these models do not do is repeal seismology. They do not replace physics-based wave propagation models, validated magnitude estimation, or institutional protocols for issuance. They do not, by themselves, close ocean coverage gaps or fix last-mile alert delivery. Treating AI as a substitute for sensor infrastructure or public preparedness is how serious programs lose credibility.

We are interested in human-in-the-loop systems where models accelerate detection and humans retain authority over issuance, escalation, and post-event review — especially where false alarms carry social cost and missed events carry mortal cost.

Quantum and frontier sensing

Quantum gravity gradiometry and related frontier instruments are discussed in the research literature as possible contributors to earlier signal or denser observational fields. We treat these as long-horizon engineering bets, not near-term operational dependencies. Civilian EEW planning should not wait on breakthrough sensors when existing technology, properly deployed, already saves lives.

The responsible posture is parallel: improve today's networks and delivery pathways while tracking frontier instrumentation with sober timelines and independent validation.

Why Neocortic cares

Disaster detection and civilian risk reduction sit within our applied modelling mandate. We are not building consumer alert apps or selling hardware. We care about:

• Measurement — how much warning time do different segments of a population actually receive?
• Institutional design — who has authority to issue, override, or stand down an alert?
• Model evaluation — where do ML detectors fail on edge cases that matter for megathrust or induced seismicity?
• Coupled systems — how do EEW signals integrate with hospitals, transit, schools, and energy infrastructure under real latency constraints?

This work is morally weighted. People die when systems fail. People also die when hype substitutes for preparedness — when "AI prediction" narratives distract from boring, essential investments in sensors, drills, building codes, and equitable alert delivery.

Open questions we are tracking

• Can DAS and legacy networks be fused into continuously validated coverage maps that funding agencies can audit?
• What governance models prevent alert fatigue without introducing fatal hesitation on large events?
• How should EEW systems perform post-event review when model-assisted detection was wrong — publicly, quickly, and with data?
• Where does faster automated detection create new crumple zones — humans nominally responsible for outcomes they had milliseconds to vet?

We welcome partners and funders working on civilian disaster risk reduction, public infrastructure, and operational seismology — especially where the mandate is measured lead time for all communities, not only for demonstration cities.

If this line of work fits your institution, start a conversation.