Slow slip events in the roots of the San Andreas fault


The discovery of deep-seated slow slip events (SSEs) was enabled by the establishment of continuous global positioning system (GPS) measurements at the Nankai and Cascadia subduction zones (1, 2). Soon after, tectonic tremors that are temporally and spatially correlated with SSEs were discovered in Japan (3), leading to the recognition of the coupled phenomenon called episodic tremor and slip (ETS) (4, 5). ETS mostly occurs below the transition from brittle to ductile fault zone properties (6), where increasing temperatures and pore pressures due to metamorphic dehydration reactions (7) inhibit fast ruptures. Long-lived tremor signals, in contrast with classical earthquakes, are made of a large number of low-frequency earthquakes (LFEs) that are thought to be due to the activation of small seismic asperities by surrounding slow slip (8). Strain rate transients due to SSEs correlated with tremor bursts are observed for transient durations ranging from minutes to months (9). However, because of geodetic detection limitations, SSEs are not always observed at locations and times of tectonic tremors (10). Recent studies have shown that using the timing of tremor bursts to stack multiple episodes in GPS time series helps to extract an averaged slow slip signal (11). Using transient tremor activity to guide the detection of the SSE’s geodetic signal, (12) demonstrated that many smaller SSEs are occurring during the periods between large SSEs.

On continental strike-slip faults, shallow transient creep events have been captured using strainmeters, creepmeters, and InSAR (interferometric synthetic aperture radar) measurements [e.g., (1316)]. However, spontaneous SSEs at the down-dip edge of the seismogenic zone on strike-slip plate boundaries have not been observed yet with geodetic measurements. Persistent tremor and LFE activity at the deep transition zone have been detected along two segments of continental strike-slip faults in the world: a section of the San Andreas fault (SAF) near Parkfield (17) and on the Alpine Fault in New Zealand (18, 19). The Parkfield segment of the SAF might be different from most strike-slip faults because it overlies the remnants of a subduction zone where serpentine is still actively dehydrating (20). With these processes happening, it is a good analog of young and warm subduction zones, where most of the ETS are occurring. Numerous studies of LFEs on that segment have characterized their temporal behavior since 2003, defining 88 families with a range of temporal behavior from continuous to episodic (21). Here, we extract the first geodetic signature of transient slip associated with LFE bursts on that segment by taking advantage of the redundancy of information provided by more than 10 years of continuous time series from a dense GPS network, 68 stations within 100 km from Parkfield, California (Fig. 1).

Fig. 1 Tectonic context and GPS network.

The right-lateral strike-slip SAF that delimited the Pacific plate from the North American plate is represented by the red line. The circles represent the LFE families color coded by their depths. The white triangles show the network of GPS stations. The black arrows indicate the predicted static surface displacements due to an SSE of Mw 4.9 occurring from 14- to 18-km depth, with the surface projection of the model dislocation shown by the thick red bar. The black squares indicate the cities. LA, Los Angeles; SF, San Francisco; CA, California; NV, Nevada; Pk, Parkfield.

We postprocessed the GPS time series to correct for signals due to earthquakes, postseismic relaxation, seasonal oscillations, and common mode errors (see Materials and Methods). In individual residual time series, only noise can be seen, with an SD of about 1 mm (fig. S1). To account for the timing of LFE bursts, we select the families with strongly episodic behavior, which produce clear transient accelerations in the cumulative LFE count. We assemble them into two groups in which families have episodic transients over the same time periods: the NW (northwest) Parkfield group and the Cholame group (Fig. 2A and fig. S2). To define the central transient event times, we use the maximums in daily LFE counts (Fig. 2B; see Materials and Methods). We then stack 500-day windows of GPS time series aligned on the central event times to increase the signal-to-noise ratio of possible transient slip events. While the stacked GPS time series have a much reduced SD of ~0.15 mm, no transient events can be observed in the individual time series. To detect transient events at the noise level of individual time series, we use a geodetic matched filter technique [see Materials and Methods; (22)] based on calculating the correlation between synthetic templates from a mechanical dislocation model and the residual GPS time series. By discretizing the fault plane into many patches and computing the corresponding synthetic templates and correlations at each station, we can extract the locations that maximize the correlation. We applied this method to the 500-day GPS time series stacked over 20 LFE bursts for the NW Parkfield group (Fig. 3) and 65 bursts for the Cholame group (fig. S2) during the period 2006–2016.

Fig. 2 Temporal behavior of the episodic LFE families of the NW Parkfield group.

(A) The circles represent the location of all the LFE families. The filled circles represent the families of the NW Parkfield group. The gray dots indicate the relocated microseismicity from 1984 to 2011 (47). The hypocenter of the Mw 6.0 Parkfield earthquake is shown by the red star. (B) Cumulative number of LFEs per family (normalized by the total number and offset for clarity) of the NW Parkfield group and summed over the families (blue curve). (C) Daily count of LFEs in the group. The inverse triangles identify the 20 dates of LFE bursts used to align and stack windows of GPS time series. The dashed line indicates the threshold used to select the transient times. SE, southeast.

Fig. 3 Correlation results from the geodetic matched filter analysis for the NW Parkfield group.

(A) Positive part of the correlation functions for all the patches (gray curves). The black curve highlights the correlation function with the highest amplitude. The red dashed line indicates the zero relative stack time. (B) Histogram of the 10,000 maximums of correlation after stacking GPS windows using 20 random dates instead of the LFE burst dates. The red line indicates the amplitude of the observed correlation at the relative stack time zero in (A). The black arrowhead line indicates the 3σ of a Gaussian distribution fit. (C) Amplitude of the correlation at zero relative stack time for each dislocation patch on the fault. The highest correlation values are clustered around the black solid circles showing the location of the LFE families in the NW Parkfield group. The black rectangle and blue square are the SSE rupture areas used to estimate the Mw of the transient event.