Seismic moment tensor analysis can help in two ways. It not only gives information about the size and mechanism of a source in terms of its seismic moment and the moment tensor components. It provides, in addition, an estimate of the source's depth, which cannot always be reliably determined using normal location techniques. Thus, a large non double-couple component () may be an indication for a nuclear explosion, as compared to the typically more than 70-80% double couple for an earthquake (Dreger and Woods, 1999). The source depth determined from moment tensor analysis may also help to weed out tectonic events from among the more than 100000 events of magnitude 4 and greater that occur annually. Only events at shallow depths need be scrutinized by the monitoring process of the Comprehensive Test Ban Treaty (CTBT).
The goal of this project is to implement the process for the automatic determination of seismic moment tensors in real-time routinely used at the University of California at Berkeley (UCB, Romanowicz et al., 1993; Dreger and Romanowicz, 1994; Pasyanos et al., 1996) at the PIDC. Although the moment tensor process will not be working in real-time at the PIDC, in its final implementation it will be running automatically, triggered off of the Reviewed Event Bulletin (REB). Thus, it will add an additional, potentially powerful event screening procedure (Pechmann et al., 1995; Dreger and Woods, 1999) to the tools used at the PIDC by providing estimates of: (1) the moment magnitude, , a more accurate measure of event size; (2) the source depth, which will help distinguish natural events, typically much deeper than 1 km depth, from nuclear explosions and (3) radiation characteristics, such as deviations from the typical double-couple radiation of earthquakes.
The automated procedure as implemented both at UCB and at the PIDC uses two methods for determining the moment tensor. The first is a time domain, waveform fitting procedure that utilizes the complete, long-period recordings (CW, Dreger and Romanowicz, 1994; Pasyanos et al., 1996; Fukuyama et al., 1998; Fukuyama and Dreger, 2000). The second moment tensor method fits the surface waves in the frequency domain (SW). It is adapted from the two-step method of Romanowicz (1982).
In the current testing phase, the automated procedure at the PIDC extracts both event information and waveforms from the PIDC database. The data are preprocessed using the programs SAC (CW) and sapling (SW, Pasyanos, 1996). The two inversions produce independent solutions, which may be used to assess solution quality. We are tuning the procedure at the PIDC using events from the 90 day interval between days 200 - 290, 1999 (July 19 to October 16). We have divided this dataset into two subsets consisting of the events with and those with . The map in Figure 23.1 shows the primary stations of the IMS used for the inversions, as well as the two sets of events selected for this interval.
As initially implemented at the PIDC, the moment tensor processing uses parameters determined from our experience in California, as well as general global assumptions. As the testing proceeds, we are attempting to improve the bulk processing performance and the quality of the inversion results by changing the runtime parameters. Due to other problems with the SW procedure, we have thus far only investigated changes to parameters for the CW inversion.
In addition to testing the inversion using three different frequency passbands (0.005 - 0.02 Hz, 0.00714 - 0.03 Hz, and 0.02 - 0.05 Hz), we have compared the results produced when an automatic crosscorrelation selects the time shifts between Greens functions and recordings with the moment tensor results generated using a grid-search of time shifts. We have also studied the difference between inverting the body wave section of the seismogram, the complete waveform and differentially weighted combinations. Figure 23.2 compares waveform inversion results for the 5.9 Kamchatka event of September 16, 1999. The solution shown in Figure 23.2B is comparable to the Harvard solution in terms of the relatively deep source depth (50 km vs. 68 km), the non-double-couple nature of the radiation pattern, and in terms of the scalar seismic moment (8.7x vs. 1.2x dyne-cm). This particular solution is derived using the complete three-component displacement waveforms in the 0.02 to 0.05 Hz passband from three stations located between 28 to 35 from the epicenter. The P and S body wave portions of the seismograms are weighted more heavily to increase their importance in the inversion. Figures 23.2A and 23.2C show results for the complete waveform inversion without differential weighting, and for an inversion using only body waves, respectively. In all three cases there is reasonably good agreement with solutions obtained by other researchers. Complete waveforms in the 0.00714 - 0.03 Hz passband yielded results comparable to those in Figure 23.2A.
These results indicate that complete three-component waveforms from a few distant stations are sufficient for the recovery of the seismic moment tensor and also source depth. At these long-periods however, we expect that the resolution for the depths of shallow sources will be poor due to the vanishing tractions at the free-surface. In these cases regional distance stations, and shorter periods may be used. These results also show that it will be advantageous to use both body and surface waves in the inversions.
The preliminary results we have obtained indicate that it is possible to recover reasonable estimates of the seismic moment tensor using only a few stations and a globally averaged 1D seismic velocity model. We are adjusting the code for the SW procedure, so that we will be able to determine moment tensors using both methods. By comparing two independent results we can improve our confidence in their implications. In addition, we will be better able to adapt the moment tensor inversion process to different situations determined by such things as event-station distances or configurations, or signal-to-noise issues. Future work will include the bulk processing of the 330 test events, investigating the results for different global 1D-velocity models, the development of an automated method of assessing solution, and the focussed monitoring of a region. The latter task will involve the use of ground truth and calibration information available at the Center for Monitoring Research. Possible regions for this focused analysis include the Lop Nor and Novaya Zemlya test sites, former Soviet Union, and the Middle East.
This project is funded under the Defense Threat Reduction Agency contract DTRA01-00-C-0038.
Dreger, D. and B. Romanowicz, Source characteristics of events in the San Francisco Bay Region, USGS Open-file report, 94-176, 301-309, 1994.
Dreger, D. and B. Woods, Regional moment tensors as discriminants?, Seism. Res. Lett., 70, 212, 1999.
Fukuyama, E., M. Ishida, D. Dreger, and H. Kawai, Automated seismic moment tensor determination by using on-line broadband seismic waveforms, Jishin, 51, 149-156, 1998.
Fukuyama, E. and D. Dreger, Performance test of an automated moment tensor determination system, Earth Planets Space, 52, 383-392, 2000.
Pasyanos, M., D. Dreger and B. Romanowicz, Toward real-time estimation of regional moment tensors, Bull. Seism. Soc. Am., 86, 1255-1269, 1996.
Pechmann, James C., et al. The February 3, 1995, ML 5.1 seismic event in the Trona mining district of southwestern Wyoming, Seis. Res. Lett., 66, 25-34, ????.
Romanowicz, B., Moment tensor inversion of long period Rayleigh waves: a new approach, J. Geophys. Res., 87, 5395-5407, 1982.
Romanowicz, B., M. Pasyanos, D. Dreger and R. Uhrhammer, Monitoring of strain release in central and northern California using broadband data, Geophys. Res. Lett., 20, 1643-1646, 1993.