A research team has uncovered dozens of previously hidden slow slip events beneath California's San Andreas Fault using artificial intelligence and highly sensitive strainmeter data, revealing that these silent fault movements systematically trigger increased low-frequency earthquake activity. The findings, published in Nature Communications, fill a key observational gap in understanding how stress accumulates and releases along active faults, with implications for earthquake hazard assessment.
Parkfield, California, sits on the San Andreas Fault and is one of the most intensively monitored fault zones in the world. Despite decades of study, some fault processes have remained difficult to observe because they produce only subtle signals. Slow slip events, where faults move silently over hours or days without generating noticeable shaking, are notoriously hard to detect with conventional earthquake monitoring methods.
AI uncovers what traditional methods missed
The research team, led by Dr. Zahra Zali of the GFZ Helmholtz Centre for Geosciences, analyzed continuous data from borehole strainmeters-instruments that measure extremely slow, small deformations in the Earth's crust. While these sensors generate enormous data streams, subtle transient signals can easily be lost among long-term trends and environmental noise. To overcome this, the team developed a deep-learning workflow that exemplifies AI for Science & Research, automatically grouping similar deformation patterns together without relying on predefined signal templates.
The neural network learned a compact representation of the strain data and then used unsupervised clustering to separate deformation signals from background noise. This approach identified a previously unrecognized population of short-duration slow slip events that release stress within a few hours. "These events are difficult to identify by conventional methods because they are small and often hidden within complex background signals," Zali said. "Artificial intelligence allowed us to recognize their patterns that would otherwise have gone unnoticed."
Linking silent slip to seismic activity
The team compiled the first catalog of short-duration slow slip events at Parkfield derived directly from strainmeter observations. Independent data from nearby creepmeters supported the existence of these events. By estimating the location and direction of slip, the researchers found that the events occurred at shallow depth and were consistent with the right-lateral motion of the San Andreas Fault.
The connection became clearer when the researchers compared the timing of these slow slip events with recordings of low-frequency earthquakes (LFEs)-a special class of weak seismic signals associated with fault slip processes. They observed that LFE activity consistently increased following the occurrence of slow slip events. "Our results show that these slow fault movements are not isolated phenomena," said Patricia Martínez-Garzón, working group leader at GFZ and professor at RWTH Aachen University. "They appear to be linked to changes in seismic activity, which suggests that slow slip may play an important role in how stress evolves along active faults."
Filling a gap in earthquake science
Slow slip events have been studied extensively in subduction zones, where one tectonic plate dives beneath another, but comparable observations on transform faults like the San Andreas have been limited-especially for short-duration events. The new Research helps bridge this gap. The detected events also follow the same scaling relationship between size (seismic moment) and duration as regular earthquakes, supporting the idea that fault slip occurs along a continuum from silent deformation to destructive quakes.
The findings highlight how machine-learning approaches can reveal signals hidden within large geophysical datasets. The researchers expect that similar short-duration slow slip events may exist on other faults worldwide, and that future studies using dense monitoring networks could uncover additional examples.
Why this matters for General, Science and Research
For professionals in geoscience and related fields, this study demonstrates that AI-driven analysis of continuous geophysical data can expose fault behaviors that traditional detection methods overlook. The ability to automatically identify subtle deformation patterns without predefined templates opens new possibilities for monitoring active faults and assessing earthquake hazards. The approach can be adapted to other tectonic settings and data types, offering a pathway to more complete catalogs of fault activity. As strainmeter networks and other high-resolution monitoring tools expand, integrating machine learning into routine analysis workflows will become essential for extracting actionable insights from the data.
Your membership also unlocks:
