Using repeating earthquakes to correct high-precision earthquake catalogs for time-dependent station delays

Waveform cross-correlation allows one to measure the relative arrival times of similar microearthquakes with errors of less than 1/10 of 1 sample. Location algorithms based on these measurements have greatly improved images of earthquake distribution. For the Northern California Seismic Network catalog, however, the relative location errors implied by the scatter of relocated repeating earthquakes (sometimes 10s of meters) is often considerably larger than the errors estimated from Monte Carlo simulations (meters) using a priori estimates of the cross-correlation errors. I find that most of this discrepancy arises from unmodeled time-varying station delays. By identifying many clusters of repeating earthquakes presumed to rupture the same source, and assuming the measured arrival-time differences between event pairs to be due to the difference in origin time plus the difference in station delay, it is possible to invert for time-dependent station corrections. The method is robust, in that these corrections are consistent over many kilometers of fault and greatly reduce the scatter (and residuals) when relocating earthquake clusters not used in the station correction determination. Many of the largest delay changes, some approaching 2 samples (20 ms), are associated with changing station electronics. At several stations an annual cycle is also clearly seen, with amplitudes up to 0.5 samples. Some M >4.7 earthquakes have produced comparable transient delay changes. After applying the station corrections, the scatter within clusters of repeating earthquakes is reduced to a few meters and the apparent seismogenic thickness of many kilometer-long sections of faults is only 10-20 m.