Postseismic Relaxation Along the San Andreas Fault at Parkfield from Continuous Seismological Observations

See allHide authors and affiliations

Science  12 Sep 2008:
Vol. 321, Issue 5895, pp. 1478-1481
DOI: 10.1126/science.1160943


Seismic velocity changes and nonvolcanic tremor activity in the Parkfield area in California reveal that large earthquakes induce long-term perturbations of crustal properties in the San Andreas fault zone. The 2003 San Simeon and 2004 Parkfield earthquakes both reduced seismic velocities that were measured from correlations of the ambient seismic noise and induced an increased nonvolcanic tremor activity along the San Andreas fault. After the Parkfield earthquake, velocity reduction and nonvolcanic tremor activity remained elevated for more than 3 years and decayed over time, similarly to afterslip derived from GPS (Global Positioning System) measurements. These observations suggest that the seismic velocity changes are related to co-seismic damage in the shallow layers and to deep co-seismic stress change and postseismic stress relaxation within the San Andreas fault zone.

Information about the stress variations in deeper parts of continental faults can be obtained by studying source properties of microearthquakes (1). Changes in seismic velocities measured by using repeated natural and active seismic sources can also provide information about rock damage and healing at depth after large earthquakes (2, 3) or about stress changes in seismogenic zones (4). The main limitation of these types of measurements, however, is the episodic nature of their seismic sources, which prevents continuous monitoring of crustal properties.

We used continuous measurements of ambient seismic noise to recover continuous variations of seismic velocities within the crust along the San Andreas fault (SAF) near Parkfield, California. With this approach, the cross-correlation function of ambient seismic noise computed between a pair of receivers converges toward the response of Earth between the receivers (the so-called Green's function). Essentially this function represents the seismogram that would be recorded at one of the receivers if a source were acting at the second (5, 6). The temporal evolution of the crust is then tracked by computing cross-correlation functions at different dates for the same receiver pair and measuring the changes between the correlation functions (79).

To monitor variations in seismic velocity along the SAF at Parkfield, we used more than 5 years of continuous seismic noise data recorded by 13 shortperiod seismological stations of the Berkeley High Resolution Seismic Network (HRSN) (10). These stations are installed in boreholes at depths of 60 to 300 m, thus reducing locally generated noise and effects of temperature variations and precipitation (Fig. 1). We analyzed data from January 2002 to October 2007, spanning the times of two major earthquakes that occurred within a 100-km radius of Parkfield: the moment magnitude (Mw) = 6.5 San Simeon earthquake of 22 December 2003, whose epicenter was located 60 km west of Parkfield, and the Mw = 6.0 Parkfield earthquake of 28 September 2004. For every possible pair combination of stations, we computed the daily cross-correlation of seismic noise by using the procedure of (11), yielding 91 × 2140 days = 194,740 cross-correlation and auto-correlation time functions. A reference Green function (RGF) was computed for each station pair by stacking the daily cross-correlations for the entire 2140-day period (12). The velocity changes were then determined by measuring time delays between the RGF and 30-day stacks of cross-correlation functions in the frequency range from 0.1 to 0.9 Hz (9, 12, 13) (Fig. 2B). If the medium experiences a spatially homogeneous relative seismic velocity change, Δv/v, the relative travel time shift (τ/τ) between a perturbed and reference Green function is independent of the lapse time (τ) at which it is measured and Δv/v = –τ/τ = constant. Therefore, when computing a local time shift, τ, between the reference and a chosen cross-correlation function in a short window centered at time τ, we would expect that τ should be a linear function of τ. By measuring the slope of the travel time shifts τ as a function of time τ, we then estimated the relative time perturbation (τ/τ), which is the opposite value of the medium's relative velocity change (Δv/v). The 30-day stacked correlations shown in Fig. 2A exhibit variations because of the seasonal pattern of the location of noise sources (14, 15). Because these seasonal variations mainly affect the direct waves, we did not make differential time measurements for these waves. We also investigated the accuracy of the station clocks by analyzing the temporal symmetry of the correlation functions (16) and correcting for the detected errors (12). Lastly, following (9), we averaged the measured time shifts for each time τ over all station pairs to increase the measurement accuracy.

Fig. 1.

Location of the HRSN (white and black circles) near Parkfield, California, and location of the 2003 San Simeon and 2004 Parkfield earthquakes. The black solid line indicates the surface projection of the 2004 Parkfield earthquake rupture and afterslip extent. The blue circles indicate the epicenters of nonvolcanic tremors detected by (23). The black box on the inset image corresponds to the studied area. The DEM plot was obtained from (27, 28). EQ indicates earthquake.

Fig. 2.

Relative travel-time change measurements (Δτ/τ). (A) Thirty-day stacked cross-correlation functions (CCF) for receiver pair JCNB-SMNB. The black curve represents the reference stacked cross-correlation function. The CCFs are filtered between 0.1 and 0.9 Hz and normalized in amplitude. (B) Time shifts averaged over 91 receiver pairs and coherence measured between the reference stacked and 30-day stacked cross-correlation functions (frequency band, 0.1 to 0.9 Hz).

After the San Simeon earthquake, the seismic velocity along the SAF at Parkfield decreased by 0.04% (Fig. 3). This is consistent with measurements using active sources and fault guided waves that are associated with other earthquakes (2, 3, 17). Creepmeter and Global Positioning System (GPS) measurements show that there was no substantial slip detected along the SAF in the Parkfield area after the San Simeon earthquake (18). This suggests that the velocity change we detected may be related to co-seismic damage in the shallow layers caused by strong ground shaking (∼0.15 g) from this quake. By 7 months after the quake, velocities in the Parkfield area appear to have returned to their pre-earthquake levels.

Fig. 3.

Seismic velocity changes, surface displacements from GPS, and tremor activity near Parkfield. The red curve represents the postseismic fault-parallel displacements along the San Andreas fault as measured by GPS at station pomm (Fig. 1) (29). The tremor rates are averaged over a centered 30-day-length moving time window.

Kinematic and dynamic rupture inversions as well as GPS and INSAR (Interferometric Synthetic Aperture Radar) measurements showed that the Parkfield mainshock released a maximum stress of 10 Mpa and that the average slip was about 0.5 m (19). The Parkfield mainshock was also followed by postseismic afterslip that is still ongoing and broadly distributed between the surface and a depth of 12 km (20, 21). Immediately after the Parkfield earthquake, velocities decreased by 0.08%, and postseismic velocities remained low for almost 3 years (Fig. 3). The long-term decay of the relative velocity perturbation was very similar to the relaxation curve associated with the along-fault displacement deduced from GPS measurements (21, 22). Therefore, our hypothesis is that the evolution of the observed seismic velocity changes after the Parkfield earthquake was governed by the postseismic stress relaxation within deeper parts of the fault zone and the surrounding region.

Observation of nonvolcanic tremors in the vicinity of the Parkfield area supports this hypothesis (Fig. 3). We considered the 30-day averaged rate of tremor activity in the Cholame-Parkfield region computed by using continuous records from the HRSN for the period 2002 through 2007. These tremors are estimated to have occurred between 20- and 40-km depths (23), similarly to the episodic tremor and slip phenomena on subduction zones (24, 25). There is clear evidence of triggering of tremor activity by both San-Simeon and Parkfield earthquakes. After the Parkfield earthquake, tremor activity remained elevated and has yet to return to its pre-event level similarly to the seismic velocity changes. This observation supports our hypothesis that both seismic velocity changes and tremor activity after the Parkfield earthquake are related to postseismic stress relaxation and corresponding slow slip. We also propose that the increased nonvolcanic tremor activity after the San Simeon earthquake may be related to slow slip at depth in response to small stress variations induced by the passing of seismic waves from the Mw = 6.5 event (26).

Differences in the evolution of seismic velocities after the San Simeon and the Parkfield earthquakes indicate that two different physical mechanisms may be responsible for the changes in crustal properties: (i) damage of shallow layers and fault zone caused by the strong ground shaking and (ii) co-seismic stress change followed by the postseismic relaxation. These results demonstrate that measuring small velocity perturbations from correlations of seismic noise can be a useful tool for studying the continuous time evolution of the stress regime in the vicinity of seismogenic faults.

Supporting Online Material

Materials and Methods

Fig. S1

References and Notes

View Abstract

Navigate This Article