Detection and Modeling of Timing Errors at Seismic Arrays

Abstract Accurate timing is both an essential requirement and a perennial problem for seismology. With the adoption of satellite-based global positioning systems for the timestamping of data in the 1990s, this problem appeared to have been solved. There is, nonetheless, still a need for techniques for in situ verification of timing, and keen investigators continue to turn up issues. I have applied multi-channel cross-correlation (VanDecar and Crosson, 1990) to recordings of teleseisms across seismic arrays, followed by empirical moveout correction, to estimate the relative timing error between array elements. At the Yellowknife Seismological Array, timing errors are observed at some array elements, up to a maximum of 0.4 s, whereas others exhibit no apparent timing errors. The relative timing errors between 2014 and 2020 were quantized, with a base value of 0.125 s, exactly five samples at 40 samples per second. The timing errors have random onset and only increase until the datalogger is reset. This quantization is inconsistent with what is expected for a free-running clock, and no timing errors have been observed since a digitizer firmware upgrade on 8 April 2020. Because of the quantization, and because the error only increases between clock resynchronizations, it is possible to make a table of time spans and timing errors for each element—a model—that can be used to correct the timing of recorded waveforms. It is recommended that operators of other seismic arrays periodically use similar methods to check for timing errors.