Despite detecting precursor signals like radon emission and animal behavior, reliable earthquake prediction remains impossible. The complexity may be in fact fundamentally chaotic.
Michael Fiegle, CC BY-SA 3.0
The Parkfield prediction experiment
In the sparsely populated hills of central California, the town of Parkfield is directly on top of the San Andreas Fault. This location made it the center of one of the most ambitious scientific endeavors in seismology: the Parkfield Earthquake Prediction Experiment. Historically, Parkfield experienced moderate, magnitude 6.0 earthquakes with unusual regularity. Quakes were recorded in 1857, 1881, 1901, 1922, 1934, and 1966, suggesting a recurrence interval averaging 22 years. This pattern appeared so consistent that in 1985, the U.S. Geological Survey (USGS) and the California Geological Survey issued a formal prediction. They forecast with 95% confidence that a magnitude 5.5 to 6.0 earthquake would occur on this segment of the fault by 1993.
This prediction turned Parkfield into the most densely monitored earthquake zone on Earth. Scientists blanketed the area with an array of instruments designed to capture the moments before a quake. The network included creepmeters to measure slow surface slip, strainmeters and GPS sensors to detect ground deformation, and a dense network of seismometers to record the tiniest tremors. The goal was to identify reliable, short-term precursor signals—the goal of earthquake prediction, which could one day be used to issue public warnings. The scientific community waited for the fault to move.
A decade of silence
The 1993 deadline came and went without a major seismic event. The predicted earthquake finally struck on September 28, 2004, over a decade late. The magnitude 6.0 quake was the right size and in the right place, but its timing defied the forecast. More importantly, the vast and sensitive network of instruments detected no clear short-term precursors. There were no unambiguous changes in ground strain, creep rate, or local seismicity in the minutes or hours before the mainshock. The lack of a clear signal was a result for deterministic earthquake prediction.
The Parkfield experiment demonstrates the immense challenge. While some potential precursors are studied globally, none have proven reliable. Spikes in radon gas, a radioactive element released from rock, have been observed before some earthquakes, but not consistently. Anecdotal reports of unusual animal behavior, a phenomenon recorded since ancient Greece in 373 BC, are also common. Studies have documented toads abandoning breeding sites and farm animals becoming restless hours before a quake. These behaviors are not consistent enough for reliable forecasting. Many geophysicists now view earthquake fault systems as fundamentally chaotic. In this view, the crust is always in a state of critical stress, where any small rupture has the potential to cascade into a large one—or not. This inherent unpredictability means that finding a simple, universal precursor may be impossible.
💡Fun Facts
The San Andreas Fault Observatory at Depth (SAFOD) project drilled a borehole over 3 kilometers deep directly through the fault zone near Parkfield to install instruments and retrieve rock samples.
The earliest known account of unusual animal behavior before an earthquake dates to 373 BC in Greece, when rats, snakes, and weasels reportedly fled the city of Helike days before it was destroyed.
Some animals may be sensitive to the very subtle P-wave (primary wave) of an earthquake, which arrives seconds before the more destructive S-wave (secondary wave) that humans feel.
Radon-222, the isotope most often studied as a potential earthquake precursor, has a short half-life of just 3.8 days, making it a sensitive tracer for rapid geological changes.
