Slow slip modulates low-frequency seismicity on the Parkfield segment of the San Andreas Fault

Article https://doi.org/10.1038/s41467-026-74095-9 Slow slip modulates low-frequency seismicity on the Parkfield segment of the San Andreas Fault Received: 12 August 2025 Check for updates 1234567890():,; 1234567890():,; Accepted: 28 May 2026 Zahra Zali 1 , Patricia Martínez-Garzón1,2, David Mencin3 & Gregory C. Beroza 4 Understanding how slow slip events (SSEs) influence fault behavior is essential for characterizing the fault slip spectrum and its role in earthquake generation. Here, we show that deep learning applied to strainmeter data can detect shortduration SSEs on the San Andreas Fault near Parkfield, enabling an SSE catalog. SSEs are coherently observed across instruments, with evidence from nearby creepmeters. Location analysis indicates shallow depths and slip consistent with right-lateral motion. They follow a cubic moment–duration scaling law, similar to earthquakes and consistent with both subduction zone observations, and linear scaling as an upper bound. Low-frequency earthquakes increase following SSEs, suggesting that slow aseismic slip modulates seismicity. Detecting these SSEs fills an observational gap in slow earthquake studies and highlights their broader relevance. These findings support a continuum between aseismic and seismic slip, where transient deformation in creeping segments perturbs stress in adjacent locked areas, potentially promoting seismic activity. Faults release tectonic stress through both fast (seismic) and slow (aseismic) slip1,2. Aseismic slip includes steady fault creep and transient episodes such as afterslip and slow slip events (SSEs), which can last from minutes to months. SSEs release accumulated strain along faults and contribute to long-term moment release3,4, potentially reducing the elastic energy available for large earthquakes5,6. They can influence seismic hazard by occurring before7,8, during9, or after earthquakes10–13, and in some cases may trigger seismic events8,14–18. The physical mechanisms that control the occurrence of SSEs remain incompletely understood. Whether a fault slips seismically or aseismically depends on several factors, including physical properties (temperature and pressure), fault zone structure, material properties19,20, mineralogical composition of the fault gouge21, and the presence of geometrical heterogeneities22. SSEs tend to occur in specific depth ranges, typically either in the shallow upper crust, such as along the selected segments of the San Andreas Fault in California, where steady creep is common23, or at greater depths within subduction zones, for example, the Cascadia24 and Nankai margins25, where slow slip occurs in the transition zone between locked and creeping segments. These regions may promote slow slip due to a combination of elevated pore fluid pressures, which reduce effective normal stress26,27, and velocity-strengthening frictional behavior that favors stable sliding over seismic rupture20,28. Such conditions are often inferred near the base of the seismogenic zone or within weak fault materials. In transform fault settings such as the San Andreas Fault, SSEs have also been reported, though their small magnitude and short duration are challenging to detect using either geodetic or strainbased observations29–31. Borehole strainmeters (BSMs) are highly sensitive to strain changes in the surrounding Earth, enabling them to capture subtle deformation that may be missed by high-precision GPS, thereby bridging the measurement gap between seismometers and GPS32. Their ability to capture deformation across timescales ranging from seconds to weeks makes them valuable for detecting transient aseismic phenomena, including short-duration SSEs. Despite their potential, BSM data are often dominated by environmental and instrumental noise, which can obscure subtle tectonic signals33. As a result, confirmed SSE detections using strainmeter data are rare and typically rely on visual 1 GFZ Helmholtz Centre for Geosciences, Potsdam, Germany. 2RWTH Aachen University, Aachen, Germany. 3EarthScope Consortium, Washington, DC, USA. Department of Geophysics, Stanford University, Stanford, CA, USA. e-mail: 4 Nature Communications | (2026)17:5137 1 Article inspection or simple threshold-based criteria13,34, which are time-consuming, prone to subjectivity, and often ineffective due to noise levels in the data31,35. The identified SSEs in tectonic fault settings have been mostly observed on individual strainmeter stations near the San Andreas Fault36, the Marmara region in Turkey13,34, and the Alto Tiberina Fault in Italy37, highlighting both the potential of these instruments for studying aseismic slip and the necessity to develop methodologies that enable their systematic detection in noisy environments. In this study, we present a deep learning-based algorithm to detect SSEs embedded in continuous strainmeter data. We employ a multi-step workflow that includes wavelet-based signal representation, dimensionality reduction via a neural autoencoder, and unsupervised clustering. This approach allowed us to detect SSEs on up to three independent strainmeter stations in the Parkfield section of the San Andreas Fault, with the associated slip also observed on a nearby creepmeter. The spatial coherence of these events enabled source modeling, revealing that the SSEs are shallow, exhibit slip consistent with the right-lateral motion of the San Andreas Fault, and follow a cubic moment–duration scaling law similar to regular earthquakes. Previous studies either could not detect SSEs in strainmeter data due to their small amplitudes and high noise levels, or identified them only on individual stations. The SSEs observed across multiple strainmeter stations provide robust evidence of the spatial coherence of these signals and enable analysis of both the signal characteristics and the spatial extent of the events that cause them. This allows us to explore the moment–duration scaling analysis of short-term SSEs in a transform fault setting, extending earthquake-like scaling observations to shallow aseismic slip. SSEs have been studied on several strike-slip faults in California38,39. Seismic manifestations of ongoing slow processes such as SSE commonly include tremor, very-low-frequency earthquakes, earthquake swarms, and low-frequency earthquakes (LFEs)40–42, all of which may occur in response to aseismic stress transients41. In many regions, small seismic events, including LFEs, are viewed as indirect indicators of slow aseismic slip, because stress changes associated with gradual fault slip can promote brittle failure on small asperities embedded within the slipping region43. This spatial and temporal association is often interpreted as reflecting related slow-slip processes acting within the same fault zone. The temporal correlation between LFEs and SSEs may arise from localized brittle failure within a broader region undergoing slow aseismic slip, as LFEs are generally interpreted to originate f (...truncated)