Cross-gradient Constrained Seismic Velocity and Attenuation Tomography of the Aftershock Zone of the 2013 Mw 6.6 Lushan Earthquake, China
Abstract:For seismic velocity tomography, Vp, Vs and Vp/Vs models can be inverted by first arrival P, S times and their differences, respectively (Zhang et al., 2009). These models are only indirectly linked by common earthquake locations. Because of different data quality and quantity, these models may have different resolutions and uncertainties, thus leading to some artifacts in model structures. As a result, it causes some difficulties in interpreting anomalies in these models because anomaly boundaries could be shifted among Vp, Vs, and Vp/Vs models. In this research, we enforce structure similarity among different models using the cross-gradient approach proposed by Gallarado and Meju (2003), where the cross product of the spatial gradients of different models is forced to be zero.
For seismic attenuation tomography, the velocity model is fixed and the attenuation model is inverted using the t* values. In generally, the attenuation model has a lower resolution compared to the velocity model because of lower amount of t* data. For this reason, the structures in the attenuation model may also be different from the velocity model although their real structures could be same. For mitigating this issue, we enforce the attenuation model to be structurally similar to the velocity model by the cross gradient constraint. For both seismic velocity and attenuation tomography, the structure constraint and data fitting are balanced by a weighting parameter found by the L-curve analysis.
We test the new methods on the 2013 Lushan earthquake dataset. On April 20, 2013, an Mw6.6 earthquake occurred in the Lushan, Sichuan area, about 80 km southwest of the epicenter of the catastrophic 2008 Mw 8.0 Wenchuan earthquake. To monitor aftershocks, we installed a temporary seismic network consisting of 29 short period seismic stations in the aftershock zone. Within 3 weeks of operation, ~4000 events were detected and located. Previous velocity models clearly showed a velocity contrast at a depth of ~8 km, below which the velocity is much higher than the above zone. As expected, although Vp, Vs and Vp/Vs models share similar features, they show different model structures to some extent. By incorporating the structure constraint, the velocity models have more similar structures and at the same time they can also fit the separate data at a similar level.