th Seismic Research Review – Nuclear Explosion Monitoring : Innovation and Integration
Clifford H. Thurber,Haijiang Zhang,Charlotte A. Rowe,William J. Lutter
2002-01-01
Abstract:Our research program consists of four components, each involving some aspect of multiple-event analysis: (1) highprecision waveform cross-correlation (WCC) for arrival time estimation, (2) robust event clustering, (3) waveform decomposition and source wavelet deconvolution reshaping, (4) double-difference (DD) multiple-event location and tomography. Our research is focusing initially on the development and testing of these seismic analysis methods using "ground-truth" datasets at different scales (local and regional), and will be followed by the application of these methods to "test-bed" seismic datasets, including tests on real-time data streams. We are constructing two regional ground-truth datasets, one for New Mexico and one for California. The former dataset is from the New Mexico Tech short-period network and is dominated by small explosions, whereas the latter is from the UC-Berkeley broadband network and contains relatively large earthquakes (about 260 events of magnitude 4.3 and above). In each case, the ground-truth dataset will be used initially for algorithm assessment purposes and later as a "reference event" dataset for clustering and locating new events. The New Mexico "training" dataset is undergoing final grooming. For the California dataset, we have affirmed the quality of the ground-truth location information by carrying out a DD location analysis using catalog time-difference data. We also present initial results of hierarchical cross-correlation analysis applied to the California dataset. We have completed the modifications to the integer portion of the cross-correlation software, adding the multiple eigentaper approach to provide improved integer correlation statistics. The modifications to the program logic provide the initial step for subsequent adaptation towards multiple frequency-scale correlations. In addition, modifications have been completed to interface the WCC software with our new DD tomography code (discussed below). We are preparing the California dataset for further DD location analysis and for processing with the regional-scale tomography code under development. At the same time, we will experiment with clustering and waveform decomposition and reshaping techniques to expand the ability to use WCC on large datasets. Efforts have also begun on an Earthworm-MatSeis real-time interface, including the generation of CSS database files from Earthworm. We have developed a new DD tomography method that uses high-precision WCC relative arrival time picks as well as differential and absolute catalog arrival-time picks. "Double-difference" refers to the residual between the observed differential arrival-time picks and the predicted differential arrival-times. If two sources have almost identical source mechanisms and the differential signal scattering due to velocity heterogeneity along the ray path is minor, they will produce similar waveforms at the same station, allowing for the use of WCC to determine precise differential times. Compared to absolute travel-time picks, relative arrival-time picks obtained using WCC methods have a precision to the sub-sample level. When two sources are far away from each other so that their waveforms are not similar, we just use the relative catalog arrival-time pick differences. We also use the absolute travel-time picks to constrain the absolute event locations. Our new DD tomography method has been applied to the Hayward fault, California, as an initial test to compare to previously published location results. In comparison to the standard seismic tomography method, the doubledifference tomography method results in a sharper-featured velocity structure that is consistent with the geologic setting. At the same time, it produces absolute and relative event locations accurately, similar to the results obtained by the DD location method developed by Waldhauser and Ellsworth (2000). Our next steps are to apply this algorithm to another test dataset, from Parkfield, California, and then to develop the regional-scale version and apply it to our regional California ground-truth dataset. 438 24th Seismic Research Review – Nuclear Explosion Monitoring: Innovation and Integration