The moment tensor (MT) determination in the real-time earthquake notification
system by BSL currently relies on the earthquake locations and origin times
determined by the USGS using travel-time picks from the dense short period
Northern California Seismic Network (NCSN). If only for redundancy purposes,
it is desirable, however, to be
able to determine a centroid-location and MT (and Mw) of significant earthquakes
(M
4.5) using only waveform data from the sparse (25 stations) broadband
Berkeley Digital Seismic Network (BDSN) without relying on the dense short
period data. With the quality of broadband waveform data it is
possible to develop such a system. This kind of a stand alone system
development is especially valuable for the following reasons: (1) in the occurrence
of a catastrophic earthquake (M5.5) the link to obtain location and
origin time information from the USGS may be destroyed or unavailable;
(2) such a system would also find applications in the areas of less frequent,
but potentially large earthquakes, where the operation of more than
a few broadband stations may never be affordable.
As shown by Kawakatsu (1998) the seismic wavefield below 
0.1 Hz may be
continuously monitored to simultaneously solve for potential point sources
and their moment tensors over a spatial grid covering the region of interest.
Critical issues in this procedure are to assess whether the system can
detect an event over a certain threshold magnitude and determine the location
and MT accurately. Recently we have configured such a system using the waveform
MT inversion method of Dreger and Romanowicz (1994) with Green's
functions tabulated for simple 1-D models of the crust that generally
give good results in our standard automatic moment tensor inversion procedure
(e.g. Pasyanos et al., 1996).
We carried out a feasibility study of the system in northern California
using a sample dataset of 66 events (ML
4.2) which were
recorded by the BDSN network since 1993. The large region of northern
California was divided into several subregions, i.e., Mendocino, South of
the Bay, Mammoth, Northern California, and Geysers, each of which is covered
by a mesh with an interval of 0.5 deg, and waveforms were filtered in the
frequency band of 20 (or 10) to 50 sec in which MT source inversion is
reasonably stable from our REDI experience. We have tested the accuracy and
stability of the analysis in the grid search scheme focusing on the
uniqueness of the solution in space and time as well as its frequency
dependency of waveforms. The results were compared with the ground truth
parameters determined by the REDI system, the Berkeley/USGS joint
notification system. We also evaluated results to find adequate threshold
values for event detection, and stable determination of centroid-locations
and MTs.
This feasibility study showed promising results at a magnitude level
above ML
4.5 using data from only 3 stations. Out of a total of
28 such events, 15 were located within 40 km (the grid spacing coverage for
this experiment) from the ground truth locations, two events in northern
California were mislocated by more than 80 km, and the others fell in between.
Including data from additional stations improved the performance. The test
results also indicate that a variance reduction (VR) of 75% for the MT
inversion seems to be an adequate measure to detect an event occurrence
and to obtain a stable solution for it. In the band-pass
filtered waveforms, relatively long-period structural noise comparable
to source signals can also be emphasized for smaller events. Refinements
of the method on a denser mesh (with an interval of 0.1 deg)
also require more accurate path calibration, determining optimal station
configuration and number, as well as ways to minimize computing time.
The results of this feasibility study will be presented at the Fall 1999
AGU meeting (Tajima et al., 1999)
There are some events for which MT inversions were not very stable in the northern California, Mendocino and south Bay regions. At present Green's function are chosen from a group of Greens' functions that were calculated for every 5 km in the distance range between 100 and 400 km, and every 3 km in the depth range from 5 to 39 km using either the 1-D southern California model or Gil7 model, not accounting for the 2-D or 3-D structure along specific paths between grid points (virtual source locations) and station locations. To obtain stable solutions using the new system, the structures along paths between grids and stations should be modeled properly to compute Green's functions. The improvement of the structural model is critical for the Mendocino/Gorda plate region and northern California where no optimal models have been established to our knowledge. This work is in progress.
When installation of the new stand alone system is completed, we will
be able to immediately determine the earthquake source parameters (location,
size and mechanism) using real-time data from the sparse BDSN, and/or the backup
subsystem at Sacramento, should
a significant event take place.
 |
Dreger, D. S., and B. Romanowicz, Source characteristics of events in the San Francisco Bay region, U. S. Geol. Surv. Open-File Rept. 94-176, 301-309, 1994.
Kawakatsu, H., On the realtime monitoring of the long-period seismic wavefield, Bull. Earthquake Res. Inst., 73, 267-274, 1998.
Pasyanos, M. E., D. S. Dreger, and B. A. Romanowicz, Toward real-time estimation of regional moment tensors, Bull. Seism. Soc. Am., 86, 1255-1269, 1996. Tajima, F., C. Megnin, D. Dreger, B. Romanowicz, and R. Uhrhammer, Simultaneous determination of earthquake location and moment-tensor using waveform data from a sparse regional network, EOS Trans. Am. Geophys. Union, 80, 1999.
is very close to the
true location (provided by the USGS) denoted with a star (upper panel)
and the VR is maximum at the origin time (lower panel)