Observations of time-dependent errors in long-period instrument gain at Global Seismic Stations

We present evidence for significant deviations of the true instrument gain from that reported for several modern broadband permanent seismograph stations. Our result derives from a systematic comparison of observed and synthetic long-period seismograms for approximately 600 large earthquakes. Seismograms were collected from globally distributed stations and analyzed using the centroid-moment-tensor (CMT) algorithm for estimating earthquake parameters. Following the source inversion, synthetic seismograms corresponding to the final earthquake parameters were compared with the observed seismograms and an optimal amplitude-scaling coefficient for each seismogram was determined. Scaling coefficients for earthquakes occurring in a given calendar year were then averaged to investigate the temporal stability of instrument gain. Data for up to 15 years (1990-2004) for more than 200 stations were processed. Most stations show good agreement (+/- 10%) between the observed and reported gains. A small number of stations display a larger constant offset in the gain, probably caused by errors in the reported absolute gain or, potentially, by unmodeled systematic effects resulting from the Earth's lateral heterogeneity. The existence of errors in the reported long-period gain is confirmed through a station-by-station comparison with results from an independent analysis of station-gain bias at similar periods. More than 15 stations display significant reductions in the true long-period gain that occur gradually over several months to a few years. At similar to 250-s period, these changes are as large as 50% of the reported gain. The changes are smaller at shorter periods, suggesting a frequency-dependent modification of the instrument response. All of the affected stations showing a time-dependent deterioration are equipped with the Streckeisen STS-1 seismometer, suggesting a common cause for the observed behavior.