Using synthetic kinematic source inversions with dynamic rupture models to evaluate the effect of seismic network density and geometry in near-field source inversions
Abstract
Detailed source imaging of the spatial and temporal slip distribution of earthquakes is a main research goal for seismology. In this study we investigate how the number and geometrical distribution of seismic stations affect finite kinematic source inversion results by inverting ground motions derived from a known synthetic dynamic earthquake rupture model, which is governed by the slip weakening friction law with heterogeneous stress distribution. Our target dynamic rupture model is a buried strike-slip event (Mw 6.5) in a layered half space (Dalguer & Mai, 2011) with broadband synthetic ground motions created at 168 near-field stations. In the inversion, we modeled low frequency (under 1Hz) waveforms using a genetic algorithm in a Bayesian framework (Moneli et al. 2008) to retrieve peak slip velocity, rupture time, and rise time of the source. The dynamic consistent regularized Yoffe function (Tinti et al. 2005) was applied as a single window slip velocity function. Tikhonov regularization was used to smooth final slip. We tested three station network geometry cases: (a) single station, in which we inverted 3 component waveforms from a single station varying azimuth and epicentral distance; (b) multi-station configurations with similar numbers of stations all at similar distances from, but regularly spaced around the fault; (c) irregular multi-station configurations using different numbers of stations. For analysis, waveform misfits are calculated using all 168 stations. Our results show: 1) single station tests suggest that it may be possible to obtain a relatively good source model even using one station, with a waveform misfit comparable to that obtained with the best source model. The best single station performance occurs with stations in which amplitude ratios between the three components are not large, indicating that P & S waves are all present. We infer that both body wave radiation pattern and distance play an important role in selection of optimal station. 2) Multi-station tests indicate irregular distribution of stations with different azimuths and distances around the fault provides the best source models. The minimum waveform misfit is obtained using the all-168 stations, but source model is not significant improved by using denser network. It suggests the best source model is not necessarily derived from the model with minimum waveform misfit. 3) Number of stations affects the estimated source image, but a surprisingly small number of well-spaced stations appear sufficient to obtain acceptable solutions in our study. This study is done under unrealistic conditions, e.g. no noise on ground motions, fault geometry and velocity structure are perfectly known. However, we argue that it provides basic guidelines for seismic / GNSS network geometry for the study source models from real earthquakes. Also, this study suggests a-priori physical constraints for the earthquake source is required to exclude unrealistic models. A pseudo-dynamic source inversion, in which the correlation structures between source parameters inferred from dynamic rupture models (Song et al 2013, in review), will support such constraints, and is currently work in progress.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.T43A2641Z
- Keywords:
-
- 7200 SEISMOLOGY;
- 7294 SEISMOLOGY Seismic instruments and networks;
- 7290 SEISMOLOGY Computational seismology