Deep-ocean tsunamis pass almost unnoticed beneath ships. A wave that will run meters up a beach may stand only centimeters above the sea surface in open water. Catching that whisper before it becomes a roar is the central problem of tsunami early warning, and it is a problem of signal against noise.

Tsunami Detection Runs on a Whisper

In the deep ocean, a tsunami wave height is typically measured in centimeters, often less than the daily rise and fall of tides. A wave generated by a magnitude 9 earthquake might be only half a meter high over 4,000 meters of water. The signal is small, slow, and easily buried in the ocean's own restless background.

That background includes tides, which move water up and down by meters over hours; infragravity waves, long-period oscillations that slosh across ocean basins; and microseisms, faint seismic hum from storms and distant coastlines. Together these produce pressure fluctuations at the seafloor that can dwarf a small tsunami.

Classic tsunami sensors, like coastal tide gauges, sit at the shoreline where the wave has already shoaled and steepened. By then, the wave is visible but the warning time is short. Offshore seismic networks detect the earthquake, not the wave, and not every earthquake generates a tsunami. The gap is pressure data from the deep ocean, recorded early and cleanly enough to matter.

Seafloor Pressure Gauges as a Global Array

Bottom pressure recorders (BPRs) measure the weight of the entire water column above them. A tsunami wave adds a tiny amount of extra water, and the sensor records that as a pressure change. The instruments sit on the seafloor, often in deep water, and log data internally or transmit via acoustic link to a surface buoy.

The best-known network is DART, the Deep-ocean Assessment and Reporting of Tsunamis, run by NOAA. As of the mid-2020s, DART comprises roughly 70 stations worldwide, with decades of records at some sites. Each station pairs a seafloor BPR with a surface buoy that relays data to shore via satellite.

Resolution is limited by sensor drift and thermal noise. Pressure sensors change calibration slowly over years, and temperature fluctuations at the seafloor add low-frequency wander. Data gaps are common: instruments fail, buoys break free, and coverage is sparse in the Indian Ocean, South Atlantic, and Southern Ocean compared with the Pacific. The 2004 Sumatra tsunami, for instance, struck coastlines around the Indian Ocean before any deep-ocean pressure sensor in that basin could have detected it; the network there was essentially nonexistent at the time.

New Processing Pulls Waves from Noise

Recent work in ocean signal processing treats the DART array as a distributed sensor rather than a set of independent gauges. By stacking records from multiple stations and beamforming across station pairs, researchers can suppress noise that is uncorrelated between sites while reinforcing a coherent tsunami wave passing through the array.

Stacking works because a tsunami moves at a known speed, roughly the square root of water depth times gravity, which in 4,000 meters of water is about 200 meters per second. That predictability lets processors align records in time and sum them. Tidal and infragravity bands, which move differently, cancel out.

The result, reported in recent oceanographic literature, is detection of seafloor pressure changes equivalent to about 1 millimeter of water height. Signal-to-noise improvement is roughly an order of magnitude, meaning signals previously invisible now stand above the residual noise. A related piece on this site, Cold-Atom Clocks Beat Optical Cavities, explores a similar theme of extracting weak signals from noisy environments.

One demonstration used data from a handful of DART stations in the Pacific over several years, retrospectively detecting small tsunami signals from distant earthquakes that had gone unnoticed in single-station records. The method requires that stations be spaced closely enough that a wave crosses them within a coherent time window; if stations are thousands of kilometers apart, the wave's arrival time differences become too large for simple stacking, and more sophisticated array processing is needed.

What the Method Actually Resolves

The immediate gain is sensitivity to small, slow tsunamis from landslides and volcanic collapse. These events often produce waves that are destructive locally but too small to trigger standard DART alarms. A 1-millimeter detection threshold changes which events are seen at all.

Array timing also gives source direction and speed. If a wave crosses station A before station B, the processor can estimate where it came from and how fast it is moving. That information is independent of seismic estimates, which can be ambiguous for non-earthquake sources.

Cross-checks with deep-ocean tsunami records and coastal tide gauges help validate detections. The limits remain sensor calibration and clock accuracy. A timing error of a few seconds across stations translates to a large error in source location, so the method depends on well-maintained clocks and known sensor responses. A related piece on this site, Hodgkin and Huxley Squeezed the Squid Axon, examines how instrumentation limits shape biological inference, a parallel worth noting.

From Detection to Forecast Confidence

Faster magnitude and source estimates can feed directly into warning centers. If a tsunami is detected within minutes rather than tens of minutes, coastal communities gain time to evacuate. The processing also reduces false alarms from seismic-only triggers, which can be costly and erode public trust.

Probabilistic coastal inundation models can ingest the improved source estimates to produce better wave height forecasts at specific beaches. The models still depend on accurate bathymetry and coastal topography, but a cleaner offshore signal reduces one major uncertainty.

Open data and reproducible processing pipelines are part of the push. Several groups have released code and archived DART records for independent replication. This site has argued that replication studies stall where grant cycles end, a dynamic that could slow adoption of new processing methods if funding for validation is scarce.

How to Read a Tsunami Detection Claim

When you encounter a claim about a new tsunami detection method, ask specific questions before accepting it.

• Check the station count and deployment years. A method tested on three stations over six months is not a global system.
• Ask for the detection threshold in millimeters of water. A claim without a number is not a detection claim.
• Compare against independent tide-gauge records. If the signal does not appear in coastal data, the offshore detection may be an artifact.
• Look for peer-reviewed processing code. Reproducibility is the difference between a result and an announcement.
• Demand uncertainty ranges, not single numbers. A source location without an error ellipse is incomplete.

Calibration and Maintenance: The Hidden Cost

Every detection claim rests on the assumption that each sensor reports pressure truthfully. In practice, BPRs drift. A sensor that read 4,000 decibars at deployment may read a few millibars higher a year later, and that offset can masquerade as a slow tsunami. Regular calibration against shipboard pressure standards is expensive and infrequent. Some stations go years between visits.

The problem compounds when stacking records from multiple stations. If one station's clock has drifted by ten seconds, its contribution to the stack is misaligned, degrading the very signal the array was meant to enhance. Clock synchronization via GPS is standard on surface buoys, but the acoustic link to the seafloor introduces its own latency that must be modeled.

Temperature is another confounder. Seafloor temperature varies by a few hundredths of a degree over tidal cycles, and the pressure sensor's response to that variation is not perfectly known. Over long periods, these small errors integrate into low-frequency wander that can mimic a tsunami with a period of tens of minutes. Filtering helps, but it also removes part of the signal.

Coverage Gaps and the Sparse Ocean

The DART array is concentrated in the Pacific, where the U.S. and Japan have funded stations for decades. The Indian Ocean, South Atlantic, and Southern Ocean remain sparsely instrumented. A tsunami from a landslide off a remote island in the South Atlantic might pass between stations without being recorded. The 2004 Sumatra event exposed this gap dramatically: no deep-ocean pressure sensor was in place to detect the wave before it struck.

Adding stations is costly. A single DART station costs several hundred thousand dollars to deploy and tens of thousands per year to maintain. Buoys are vulnerable to vandalism, fishing gear, and storms. In some regions, piracy and territorial disputes complicate deployment. The result is a network that is adequate for the Pacific but incomplete elsewhere.

New processing methods cannot overcome missing data. They can make existing stations more sensitive, but they cannot detect a wave that no sensor recorded. For now, the most vulnerable coastlines are often the least instrumented.

The Path Forward

Improving tsunami detection will require both better processing and better infrastructure. On the processing side, machine learning approaches are being explored to identify tsunami signals in noisy records, though these methods need large, labeled datasets that are still scarce. On the infrastructure side, new sensor designs promise lower drift and longer deployment times, but they are not yet widely deployed.

International cooperation is essential. Tsunamis do not respect national boundaries, and a station in one country can provide warning for another. Data sharing agreements already exist, but they are uneven. Expanding them, and funding the maintenance they require, is as important as any algorithm.

These checks apply to any instrumentation breakthrough. The physics of pressure sensing is well understood, but the path from a clean signal to a reliable warning is long. The recent processing advances are real and useful. They also depend on a network that is aging, unevenly distributed, and expensive to maintain. That tension between sensitivity and sustainability will shape how quickly the improvements reach the coasts that need them.