Earthquake source characterization algorithms usually consist of a two-step process requiring the determination of hypocenter location before a moment tensor can be derived. Source location is obtained by triangulation through an automatic body-wave picking scheme based on the travel-time information for P wave arrivals. To minimize location error, a large number of seismic stations is required. In places exposed to significant seismic risk and where a dense network of seismic stations is presently unaffordable, the creation of a real-time source characterization system based on a small number of seismic stations is desirable.

We present an automated system to simultaneously determine centroid location, origin time and moment tensor for regional earthquakes. This system monitors real-time seismic waveforms continuously from a sparse 3-component broadband instrument network of between two and four stations out of a pool of six. Location and moment tensor are determined by a hybrid grid-search/direct inversion scheme. We overlay a
grid over the region of interest and invert a continuous stream of broadband waveforms for moment tensor at each point on the grid (*Kawakatsu*, 1998). In the presence of a seismic event, the best fit to the data should be obtained at the grid point nearest to the source location.

To test the feasibility of this system we use a data set of waveforms recorded for 62 events with *M*_{L} 4.2 which occurred in northern and central California between 1993 and 1999. These events have well calibrated solutions obtained using an established procedure that relies on travel time picks from a dense short-period network. The solutions determined by the new system are compared with those available from the established system for appraisal.

In Figure 18.1, we show an example of spatial (top) objective functions for the 08/12/1998 *M*_{L}=5.4 San Juan Bautista event. The red star and the blue `beach ball' correspond to the calibrated solution. The red beach ball points to the location of the best solution (with a variance reduction of 89.7%). To investigate whether a unique solution exists in time, we inverted the data allowing the origin time to vary (Figure 18.1, bottom). In all cases, a unique origin time was determined.

Results show that onshore events with *M*_{w} 4.5 are detected and adequately characterized in terms of moment tensor and centroid location.
In contrast, events that occurred off the Mendocino coast aren't as reliably constrained, reflecting the complexity of the crustal structure in the transition region from oceanic to continental inadequately represented by the present 1D model, and the gap in azimuthal coverage.

Although computationally intensive, our results show that the implementation of this system in real-time is within the reach of a medium-size PC cluster. The code is currently being parallelized and current research includes investigating the computational feasibility of the inclusion of path-dependence effects.

Kawakatsu, H., 1998, On the realtime monitoring of the long-period seismic
wavefield, *Bull. Earthq. Res. Inst.*, ** 73**, 267-274.

The Berkeley Seismological Laboratory, 202 McCone Hall, UC Berkeley, Berkeley CA 94720

Questions and comments to www@seismo.berkeley.edu

Copyright 2000, The Regents of the University of California.