(1) Inversion of source rupture process
I have investigated the source rupture process of several large earthquakes occurred in Taiwan. A parallel non-negative least square (NNLS) inversion technique is developed and applied on these studies to obtain precise source rupture information (Lee et al., 2008). By using the parallel NNLS technique in the source inversion, not only the computing time can be substantially decreased (20 times faster compared to the ordinary least square inversion) but also the resolution of spatial and temporal slip distribution on the fault plane can be highly improved. With the ability of high resolution inversion, several specific rupture behaviors are found, such as rupture directivity effect, supershear rupture and pre-slip during the rupture (Lee et al., 2008; 20012; 2013). Another issue is the repetition of slip during rupture process of earthquake which is a debate issue that had never been confirmed clearly in the past big events due to the lack of dense near-field observations and limited resolution in time of source model. By using the full time-space inversion approach based on parallel NNLS technique, the repeating slip rupture behavior had been found during the giant 2011 Tohoku-Oki earthquake (Lee et al., 2011). These important inversion results can help seismologist review the source rupture from kinetic and dynamic points of view as well as its physics processes during the earthquake.
(2) Forward ground motion simulation
In order to simulate the 3-D seismic wave propagation with high resolution, we build a spectral-element mesh which incorporates high resolution surface topography and 3-D basin model in the study area. A critical issue for the successful application of the spectral-element method (SEM) lies in the design of a good mesh. I introduced a new mesh implementation using a so-called “control layers” to improve mesh quality and numerical stability of the explicit time integration scheme (Lee et al., 2009a). Based upon this implementation, realistic topography and complex subsurface structures can be efficiently incorporated in a SEM mesh. With detailed source and structure models, I carefully simulated the specific characteristics of the ground motion due to the hazardous earthquakes. A serial of papers about these study results had been published in SCI journals. We found the location of the basin amplification, reverberation, topography scattering and several wave propagation effects; all of these have been discussed in published papers (Lee et al., 2009a; 2009b).
(3) Real-time computational seismology
Real-time earthquake analysis based on computational seismology is currently possible to be achieved which needs a high connection between seismic network and high performance computing. We have developed a real-time moment tensor monitoring system (RMT, Lee et al., 2013) by using continuous BATS broadband records and moment tensor inversion technique. RMT provides all the point source parameters including the event origin time, hypocentral location, moment magnitude and focal mechanism within 2 minutes after the occurrence of an earthquake. Then, all of the source parameters are automatically forwarded to the Real-time Online earthquake Simulation system (ROS, Lee et al., 2014) to perform an island-wide earthquake simulation based on spectral-element method. Combine RMT with ROS, an earthquake report based on computational seismology can be provided within 5 minutes after an earthquake occurred (RMT obtains point source information < 120 sec; ROS completes a 3D simulation < 3 minutes). All of these computational results are posted on the internet in real-time.
