Analysis of data from multiple geodetic networks for a spatio-temporal fault slip history at Parkfield, CA during the 1990s

Transitions from creeping to locked portions of the San Andreas fault have been seen to experience transient deformation. One such transition zone occurs at Parkfield, CA, a locale well-known for its history of M6 earthquakes. Despite a prediction that another M6 earthquake would take place in 1988 +/- 5 years, the most recent such event at Parkfield was in 1966. In the early 1990s, however, rate changes were observed in data from the permanent two-color geodimeter network and tensor strain meters. There was a general increase in seismic moment release, and three M ∼4.5 earthquakes nucleated near the inferred hypocenter of previous Parkfield M6 earthquakes. More recently a concurrent magnetic field anomaly was identified. These observations have been interpreted to result from a transient increase in fault slip-rate on the San Andreas. However, to date analyses of the geodetic data have been limited to forward models and inversions for instantaneous rate changes. Here we present results of a time-dependent inversion using data from the permanent two-color network, two portable two-color networks, and GPS surveys in the Parkfield area. Our analysis uses an efficient nonlinear Kalman filtering technique well-suited to inferring slip-rate changes from data contaminated by random walk and white noise. This method exploits the fact that, unlike local benchmark noise, a deformation signal is spatially coherent. We have also explored means for simultaneously estimating the seasonal noise that is common in geodetic time-series. This type of analysis enables estimation of fault slip at every epoch for which there are data, resulting in a time-history of slip rather than a single rate change. Results to date show an increase in slip-rate on the upper ∼6 km of the fault near Middle Mountain between 1993 and 1995 followed by a decrease to the background rate by 1998. The Parkfield two-color network, established in 1984, provides a long baseline of temporally dense measurements. Thus, the analysis methods we discuss are applicable to modeling data from continuous GPS networks like PBO. While large deformation events are easily observed; here we demonstrate the potential for highlighting subtle slip-rate transients with continuous geodetic monitoring.