WQC Noise Analysis Identifies Flaw in GSN Datalogger
The main task of the Harvard Waveform Quality Center (WQC) is to identify and report on problems with data from GSN stations. We often discover such problems when data from a particular station are unusable in the routine analysis of global seismicity (the Harvard CMT Project). The types of problems that we can identify this way (through comparison with synthetic waveforms) are typically related to malfunctioning seismometers or erroneous system response information, such as incorrectly specified gain or seismometer orientation.
During 2001 the WQC has also been developing methods to identify less obvious data problems, such as deteriorations over time in the quality of data from a given station. We were particularly interested in investigating the long-period noise levels at GSN stations over the last 10 years, since some of us at the Harvard WQC had the impression that background noise levels had increased, especially at some of the historically quiet GSN sites. An increase in background noise at previously quiet stations directly affects our ability to successfully analyze smaller earthquakes, which are typically well recorded only at a small number of stations.
We are currently at the tail end of a systematic analysis of all long-period data collected by the GSN since 1990. We process all LH- and VH-channel seismograms to determine monthly estimates of background seismic noise, here defined as the rms signal level not exceeded 10% of the time. The resulting monthly estimates at some particular period are then plotted as a function of time to identify significant changes in the background noise levels; any sudden change is likely to indicate an instrument problem. The details of the analysis are described on our web site.
For many stations, the background noise levels are remarkably constant; one example is the Chinese Network station WMQ. A seasonal component is clearly visible for many stations that resolve ground noise, reflecting the variations in meteorological or surf conditions, as at KIP. At some stations we discovered large and abrupt temporal changes in the background noise, in particular at some traditionally quiet IRIS/IDA sites such as ERM, ESK, NNA, and SUR. In the mid-1990s, the background noise level at 100 seconds period increases abruptly by more than 15 db for these stations. At shorter periods (around 30 seconds), the increase in background noise is even greater.
After further analysis, and a confirmation of our results by Bob Uhrhammer at Berkeley, we also discovered that the abrupt deterioration of long-period data, which is clearly seen for several more IRIS/IDA stations, correlates with the time of upgrade to the IDA Mark 7 datalogger. We notified the IDA group, who, following their own investigation, confirmed that the increase in long-period noise is caused by a design error in the digital signal processing (DSP) unit of the Mark 7 datalogger. A report that describes the problem and possible remedies was subsequently prepared. Some of the long-period signal lost in the DSP can be recovered by a reprocessing of the broad-band (BH) data stream. An effort to reprocess all of the affected IRIS/IDA data is underway, as is a program to repair the Mark 7 dataloggers in the field.
The WQC has prepared a web site with the long-period background-noise histories for all GSN stations, and we are planning to develop this site further. Though the Mark 7 problem was the first “signal” that caught our attention in this analysis, it is clear that there are several other temporal changes in noise that await an explanation.
by Göran Ekström (Harvard University)