Background on the HRSN
This background information on the HRSN was originally put together by Professor T. V. McEvilly in October, 1998. For current HRSN network information as well as individual station pages, please see the Berkeley Seismological Lab's HRSN pages.
The HRSN was initially established in 1986. Over the years, the goals of the network have evolved and the operation of the HRSN can be divided into three stages:
- 1987-1998: Original HRSN data acquisition
- 1998-2000: PASSCAL RefTek recorders
- 2001-present: Q730 dataloggers and expansion to 13 stations
The latest stage was initiated in the fall of 2000 with the installation of Quanterra dataloggers and the addition of 3 new borehole stations. Please see the HRSN overview for more information.
This document describes the operation of the HRSN in the initial period of operation from 1987-1998.
The Parkfield network, called HRSN in publications (High-Resolution Seismic Network) is described in Karageorgi et al., BSSA, 82, 1388-1415, 1992. The Parkfield earthquake data set has been acquired as a secondary effort in our 'Controlled Source Monitoring' (Vibroseis) project, begun in 1986 as a part of the Parkfield Prediction Experiment. Due to the design of the system for the original Vibroseis project, the earthquake data set from the network has some unusual characteristics.
Data Acquisition Format:
The data have been recorded as 32-trace 'shot gathers' each 12 seconds long on 9-track tape in SEG-Y format as 16-bit integers with 2 msec sampling (500 samples per second, 6000 samples per trace) on the VSP acquisition system used in the Vibroseis work through a innovative system modification allowing triggered recording of earthquakes when the system is not in use with the vibrator. The VSP aquisition system, developed around 1980, uses a DEC LSI-11 processor with a 5 MByte removable Bernoulli system disk.
Since the primary purpose of the system was the controlled-source monitoring program, there was initially no need for absolute timing, and because the earthquakes were analyzed only with the network data, there was also no need for absolute time. However, it very quickly became obvious that we were collecting important data, and a portable time-code generator was borrowed and installed at Gold Hill station early in 1987 in place of a failed horizontal component. We later recorded the IRIG-E time code available at the Haliburton house central receiving site.
All seismic data channels recorded by the VSP system have a fixed pre-event delay associated with their recording. The re-event delay was initially set to 2 seconds, but was changed to 3 seconds early on. The time code recorded in the Gold Hill H1 channel experienced the same pre-event processing delays as the seismic data channels, and there is no time offset between the data channels and the time code channel.
However, the IRIG-E time code channel at Haliburton house was was recorded on an auxilliary channel on the VPS acquisition system (e.g. channel used for the pilot signals in the Vibroseis mode) that does not experience the pre-event delay in the triggered event mode so the time code is shifted with respect to the data. The data sample time is exactly 3-second EARLIER that the time indicated by the IRIG-E time code channel. This time offset must be applied to the seismic data channels in order to determine the precise seismic data sample time.
The timing uncertainty is +/- 1 msec (one half sample interval) between stations because the digitization at each site is asynchronous and the synchronization between stations is applied in the multiplexing process at Haliburton where samples are 'grabbed' in 2 msec bins. At each field site the digitization of the 3 chanels is sequential (using a sample & hold approach) with one-third sample interval skew between the each of the 3 data channels.
The records have been 12 seconds long (6000 sampled per trace) since early in the experiment, in order to capture a full frame of the time code. Greater lengths compromised the effectiveness of the triggering, since the system is 'blind' for approximately 5 seconds while writing a captured event to the 9-track tape.
Bad time for events:
There are many events with 'bogus' times - either times that appear plausible but are incorrect, or obviously impossible times. This is due to loss of time code and generally those events float in time. However, as a help in this situation, we have assigned a sequence number to every event based on the order in which they were written to tape, so that those incorrectly timed events can at least be placed in proper time order. Locations in the location catalog are not affected by the lack of absolute time except of course for origin time. If a user has serious need for absolute time on one of these events, there is still the option of finding the same event on the Middle Mountain or Gold Hill NCSN station (if it was large enough to be seen) and correcting the HRSN data using that time with the uphole time between borehole and surface sensors.
The hypocenters have been estimate using a 3-D velocity model and location program (Michelini and McEvilly, BSSA, 81, 524-552, 1991).
Parkfield HRSN data have been processed from the beginning at the Center for Computational Seismology (CCS) at Lawrence Berkeley National Laboratory (LBNL), since the commercial reflection processing software needed for the Vibroseis data processing, and optimal for the earthquake data as well, was in place at CCS (the VSP data acquisition system belongs to LBNL and was 'borrowed' for our project). The full archive of field data (noise triggers, distant locals, regionals and teleseisms) is more than an order of magnitude larger than the event set selected as Parkfield earthquakes for processing in our research efforts, and this 'full archive' resides at CCS.
From the beginning there was an effort to operate a more conventional Calnet-type acquisition system in parallel with our unorthodox one. A Microvax with Metrum tape mass store was installed by USGS (Bob Dollar) as the primary archive to generate CUSP-compatible data directly from the 10-station data stream. The status of those data is unknown, and the system no longer is operational, but the tapes are somewhere in Menlo Park.
Duke University Archive:
From the onset of HRSN operations, Peter Malin (then UCSB, later Duke Univ.) was funded to study Parkfield seismicity and to maintain an archive of events. The field tapes were sent routinely from Berkeley to UCSB or Duke until July 1996, where an archive in UW format was maintained.
The NCEDC data set contains the locations and waveforms of events selected for inclusion in the 'Parkfield set' as the full data set was edited and the data set for processsing was built. The location for all of these events should appear in the location catalog.
There are some events in this archive that are not Parkfield events, but 'interesting' earthquakes that were selected for other reasons to be accessible in the working archive for subsequent study (e.g., Loma Prieta, major teleseisms, etc.). In addition, there are some 'near-Parkfield' locals that were not picked and located, or events that were 'unlocatable' due to insufficient data. For these events, there will be waveform files but there will not be a location for the event in the location archive. There are fewer than 100 such waveform sets.