Parkfield, California, sits directly astride the San Andreas Fault and has long been an intensely monitored location for seismologists. The town is situated on a specific segment of the fault that historically experienced a series of moderate-sized earthquakes exhibiting regularity. This unusual pattern of repeating events, referred to as “characteristic earthquakes,” established Parkfield as a natural laboratory for studying earthquake generation. Because of this perceived predictability, the area became the focus of the world’s most elaborate effort to capture and analyze the final moments before a seismic rupture.
The 1985 Parkfield Earthquake
The next moderate earthquake in the sequence after the 1966 event was anticipated to occur around 1988, based on the historical cycle. Despite significant planning and monitoring efforts in the 1980s, no significant event of the expected magnitude occurred during the predicted window (1985–1993). The next characteristic earthquake in the sequence finally struck on September 28, 2004, with a magnitude of 6.0. This rupture provided scientists with an unprecedented amount of data due to the dense instrumentation placed in the area years earlier. The event’s moment magnitude was consistent with the size and location of previous characteristic quakes, fulfilling all aspects of the prediction except for the timing.
Understanding the Seismic Recurrence Pattern
Scientists focused intensely on the timing of the next earthquake because the Parkfield segment of the San Andreas Fault had previously demonstrated a remarkable, quasi-periodic history of moderate earthquakes. The historical record showed six successive magnitude 6 events occurring at intervals ranging from 12 to 32 years, with an average recurrence time of approximately 22 years. Key historical rupture dates included 1857, 1881, 1901, 1922, 1934, and 1966.
This apparent regularity led to the development of the “characteristic earthquake” model, which posited that the same segment of the fault ruptures repeatedly in events of similar size and mechanism. Based on the 22-year average interval from the 1966 event, researchers forecasted that the next earthquake would occur around 1988. This forecast gained official endorsement in 1985 from the National Earthquake Prediction Evaluation Council.
In response to this prediction, the U.S. Geological Survey (USGS) established the Parkfield Prediction Experiment (P.P.E.) in the 1980s. This experimental setup involved deploying an elaborate network of instruments, including over a hundred seismometers, creepmeters, and strainmeters. The goal was to capture the final hours or minutes of the earthquake preparation process, allowing for a short-term warning.
The 2004 Event and Scientific Review
The anticipated earthquake failed to occur within the predicted 1985–1993 window, classifying the temporal element of the long-term prediction as a failure. The next characteristic earthquake did not strike until September 28, 2004, a delay of over a decade past the expected time. This late arrival represented a significant deviation from the previously established 22-year recurrence pattern, undermining the simple time-based prediction model.
The 2004 event still provided a wealth of data, as the monitoring network had been continuously operating for nearly 20 years. Analysis of the rupture showed that the earthquake lacked the distinct precursory foreshock activity that had preceded both the 1934 and 1966 mainshocks. The absence of observable short-term precursors, combined with the major timing failure, forced seismologists to fundamentally reassess their understanding of earthquake mechanics and the utility of simple recurrence intervals.
The scientific review concluded that while the Parkfield segment still produces “characteristic” earthquakes in terms of location and magnitude, the precise timing is far less predictable than initially believed. The experiment demonstrated that simple time-based models are insufficient for reliable earthquake prediction, shifting the focus of research away from short-term warnings and toward better long-term hazard assessments. The realization that the recurrence interval is more irregular than previously assumed, with an “aperiodicity” value now estimated at 0.44, marked the end of the strict prediction phase of the Parkfield experiment.