Distribution and Characteristics of Repeating Earthquakes in Northern California

Repeating earthquakes are playing an increasingly important role in the study of fault processes and behavior, and have the potential to improve hazard assessment, earthquake forecast, and seismic monitoring capabilities. These events rupture the same fault patch repeatedly, generating virtually identical seismograms. In California, repeating earthquakes have been found predominately along the creeping section of the central San Andreas Fault, where they are believed to represent failing asperities on an otherwise creeping fault. Here, we use the northern California double-difference catalog of 450,000 precisely located events (1984-2009) and associated database of 2 billion waveform cross-correlation measurements to systematically search for repeating earthquakes across various tectonic regions. An initial search for pairs of earthquakes with high-correlation coefficients and similar magnitudes resulted in 4,610 clusters including a total of over 26,000 earthquakes. A subsequent double-difference re-analysis of these clusters resulted in 1,879 sequences (8,640 events) where a common rupture area can be resolved to the precision of a few tens of meters or less. These repeating earthquake sequences (RES) include between 3 and 24 events with magnitudes up to ML=4. We compute precise relative magnitudes between events in each sequence from differential amplitude measurements. Differences between these and standard coda-duration magnitudes have a standard deviation of 0.09. The RES occur throughout northern California, but RES with 10 or more events (6%) only occur along the central San Andreas and Calaveras faults. We are establishing baseline characteristics for each sequence, such as recurrence intervals and their coefficient of variation (CV), in order to compare them across tectonic regions. CVs for these clusters range from 0.002 to 2.6, indicating a range of behavior between periodic occurrence (CV~0), random occurrence, and temporal clustering. 10% of the RES show burst-like behavior with mean recurrence times smaller than one month. 5% of the RES have mean recurrence times greater than one year and include more than 10 earthquakes. Earthquakes in the 50 most periodic sequences (CV<0.2) do not appear to be predictable by either time- or slip-predictable models, consistent with previous findings. We demonstrate that changes in recurrence intervals of repeating earthquakes can be routinely monitored. This is especially important for sequences with CV~0, as they may indicate changes in the loading rate. We also present results from retrospective forecast experiments based on near-real time hazard functions.