Fast Inversion of the Earthquake Rupture Processes with Complicated Velocity Structure: An Application to the Earthquake of 2017 Mw 6.5 Jiuzhaigou, China

Abstract

Due to the limitation of seismic station coverage or the network transport interrupted when the earthquake occurred, an accurate seismic shakemap may not be released to the public quickly. When the near-source observed waveforms for the intensity prediction technology used are incomplete, we synthesize the seismic waveform into observation waveforms. An accurate seismic rupture process is necessary to synthesize virtual station observations. So, we should release the rupture process as soon as possible after a large earthquake. Most large earthquakes occur at the junction of two or three tectonic terranes. With violent tectonic movements, fault basins and uplift zones are distributed on the edge of the plateau. With complex structural conditions, the 1D layered half-space velocity structure model could not meet the requirement of earthquake rupture process inversion. It takes much time to calculate 3D Green’s function with a 3D velocity model for the complete waveform inversion of the earthquake rupture process. To rapidly invert the rupture process as accurately as possible, according to the geological conditions of the station, we calculated several Green’s function libraries in advance. We extracted Green’s functions from these libraries for each site based on the sites’ coordinates once an earthquake occurs. The time we spend in extracting Green’s functions from several Green libraries equals that we spend in extracting Green’s functions from one single library. The applicability of this method was tested in the 2017 Jiuzhaigou M6.5 earthquake with complex structural conditions in the mountain uplift zone. With our model, the time we spent in calculating the rupture process was almost the same as that we spent with the 1D velocity structure model, which was far less than that we could have spent in calculating 3D Green’s function. The degree of fitting between the synthetic data and the observation data of our model was much higher than the fitting of the 1D velocity model, which means that the earthquake rupture process we determined was more reliable.