Experimental research

Precursor research

Experimental research

The datasets people propose as earthquake precursors, what the evidence for each actually shows, and the test any such claim has to pass. None of this can raise an alert, and the module is off unless explicitly switched on.

EXPERIMENTAL RESEARCH — NOT A VERIFIED EARTHQUAKE PREDICTION OR OFFICIAL WARNING.

Nothing on this page is a warning, an input to a forecast, or evidence that an earthquake is coming. Precursor research is included because the questions are asked constantly and deserve a real answer rather than either credulity or dismissal.

Module status

Data unavailable

Disabled by default0 of 6 hypotheses have a live data pipeline.

The module stays off until PRAKRUTI_ENABLE_PRECURSOR_RESEARCH is set and at least one verified, licensed pipeline exists. Today none does, so there is nothing to run — and a module that quietly switched itself on because some data happened to arrive is exactly how an unvalidated signal reaches the public.

Precursor output cannot reach an alert or a forecast. That is structural rather than a rule someone has to remember: this code has no import path into the alerting or forecasting modules, and the forecast engine accepts no precursor input. There is no setting that would let one through.

How a precursor claim is tested here

Historical data

Any proposed precursor can be written as an alarm rule: a condition that declares some fraction of time “alarmed”. Two numbers then settle it — the fraction of total time under alarm, and the fraction of target earthquakes missed.

A rule that alarms all the time misses nothing and has told you nothing. Random guessing traces a straight trade-off between the two. Only a rule that sits clearly below that line carries information. This is the Molchan test, and it is where most precursor claims come apart: they catch earthquakes by alarming most of the time.

Prakruti implements that evaluation so a claim can be checked rather than argued about. What it lacks is data — see the status column below.

Retrospective success is not evidence. Nearly every claim here rests on records examined after the earthquake was already known. A threshold chosen with the answer in hand will always find something. The test has to be fixed in advance and applied to periods with no earthquake as well.

Proposed precursors

Experimental research

VLF/LF subionospheric propagation anomalies

Access restrictednot evaluated

Hypothesis, stated so it could fail

Amplitude or phase of very-low-frequency transmitter signals received over a path crossing an epicentral region changes measurably in the days before a large earthquake.

What the evidence actually shows

Reported in retrospective case studies, chiefly from a small number of groups. Prospective tests with pre-registered thresholds are scarce, and the reported effects have not been reproduced independently at a rate that would support operational use.

Confounds that must be excluded first

Solar-terrestrial activity, transmitter maintenance, seasonal and diurnal propagation changes, and receiver drift all move these signals. Any of them can produce an "anomaly" in a record inspected after an earthquake is already known about.

Data required: Continuous VLF/LF receiver time series along fixed transmitter-receiver paths
Candidate sources: Research receiver networks; no open global archive located

Ultra-low-frequency geomagnetic anomalies

Reachable, not ingestednot evaluated

Hypothesis, stated so it could fail

ULF magnetic field power in a specific band rises above its local baseline in the weeks before a nearby large earthquake.

What the evidence actually shows

The best-known case studies have been substantially revised or attributed to instrumental and solar causes on re-examination. No ULF precursor has demonstrated prospective forecasting skill.

Confounds that must be excluded first

Geomagnetic storms dominate this band and are driven by the Sun, not the Earth. Cultural noise — railways especially — produces signals of the same magnitude. A baseline that does not remove both will manufacture anomalies.

Data required: Ground magnetometer time series at 1 Hz or better
Candidate sources: INTERMAGNET observatories (open, but coverage is sparse near most seismic zones)

Total electron content anomalies

Access restrictednot evaluated

Hypothesis, stated so it could fail

Ionospheric total electron content above an epicentral region departs from its expected value in the days before a large earthquake.

What the evidence actually shows

A large retrospective literature exists and a smaller critical one showing that the reported anomalies are generally within normal variability once solar and geomagnetic activity are properly accounted for.

Confounds that must be excluded first

TEC is driven overwhelmingly by solar EUV flux and geomagnetic activity, with strong diurnal, seasonal and solar-cycle structure. Establishing a baseline that removes all of these is the whole difficulty, and it is usually where these analyses fail.

Data required: GNSS-derived TEC maps
Candidate sources: IGS / NASA CDDIS global ionosphere maps (open, registration required)

Groundwater and soil radon

No open source locatednot evaluated

Hypothesis, stated so it could fail

Radon concentration in groundwater or soil gas changes measurably before a nearby large earthquake.

What the evidence actually shows

One of the longest-studied proposed precursors. Individual coincidences are documented; systematic evaluation has not established a usable relationship, and radon anomalies occur frequently without subsequent earthquakes.

Confounds that must be excluded first

Rainfall, barometric pressure, temperature, groundwater level and soil moisture all modulate radon strongly, and their effects are larger than the claimed signal.

Data required: Continuous radon monitoring at fixed stations
Candidate sources: National monitoring programmes; no open global archive located

Satellite thermal-infrared anomalies

Reachable, not ingestednot evaluated

Hypothesis, stated so it could fail

Outgoing longwave radiation or land-surface temperature over an epicentral region rises anomalously before a large earthquake.

What the evidence actually shows

Retrospective claims are common. Studies applying the same detection rule prospectively, or to control periods with no earthquake, generally find the anomaly rate similar in both.

Confounds that must be excluded first

Cloud cover, surface moisture, land use, and the choice of baseline period dominate these products. Selecting the baseline after seeing the earthquake all but guarantees an anomaly.

Data required: Satellite OLR / land surface temperature products
Candidate sources: NOAA OLR products; NASA MODIS/VIIRS land surface temperature (open)

Anomalous animal behaviour

No open source locatednot evaluated

Hypothesis, stated so it could fail

Measurable changes in animal activity precede nearby large earthquakes at a rate above chance.

What the evidence actually shows

Anecdotal reports are ancient and abundant; systematic evidence is not. The few continuous-monitoring studies have produced no rule that works prospectively, and reporting is heavily biased by recall after an earthquake has occurred.

Confounds that must be excluded first

Weather, predators, human disturbance, breeding cycles, and — decisively — the fact that unusual animal behaviour is only remembered and reported when an earthquake follows it.

Data required: Continuous biologging or sensor-based activity records
Candidate sources: Individual research projects; no systematic open archive

Why this module is not a warning system

Historical data

Suppose one of these signals did show real skill. It still would not be a warning system. A rule that catches half of large earthquakes while alarming a tenth of the time sounds impressive and would still produce many false alarms for every earthquake caught, because large earthquakes are rare. Acting on that has costs — evacuations, closures, and the erosion of trust that follows repeated alarms with nothing behind them.

That is why the bar for an operational precursor is so much higher than for an interesting research result, and why nothing here approaches it.

See the forecasting that does work → Read the scientific limitations →