Double pincer movement: Encircling rupture splitting during the 2015 Mw 8.3 Illapel earthquake

The 2015 moment magnitude (Mw) 8.3 Illapel earthquake, that ruptured the central section of the Chilean subduction zone, is among the largest megathrust events in recent years. The coseismic rupture processes of the Illapel earthquake are imaged by the back-projection (BP) method in previous studies. But these models differ significantly in the extent of high-frequency radiations in the along-dip direction. Here, we conduct a refined High-resolution MUSIC BP imaging analysis of the Illapel earthquake based on teleseismic recordings in continental US. In contrast to conventional BP imaging, we add a slowness (ray parameter) error term calculated based on aftershock locations to effectively mitigate the spatial biases of BP. This correction accounts for the P-wave travel time errors at each receiver as a result of approximating the 3D Earth structure with 1D models. The calibrated BP images of aftershocks indicate that the root-mean-square location error was reduced from 24.17 km to 8.11 km. Our refined BP of the mainshock reveals geometrical rupture complexity with unprecedented details, involving stages of diverse rupture speeds and simultaneous up-dip and down-dip high-frequency bursts. The earthquake starts with a slow initiation phase propagating northward at a speed of 1 km/s in the first 13 s. Between 14 s to 34 s, the rupture diverges into two simultaneous fronts seemingly unzipping the rim of a circular patch of large slip at a speed of 3.5 km/s. The two fronts reemerge as a single front between 35 s and 45 s. The rupture splitting repeats in a second episode from 46 s to 60 s. The two episodes of encircling rupture involve intermittent high-frequency radiations both up-dip and down-dip, which reconcile the discrepancy of the extent of along-dip ruptures reported in previous BP studies. Key features of the rupture process correlate with the prominent pulses recorded by local strong-motion network. In one possible scenario, the rupture initially encounters and splits around a barrier of higher strength or an asperity of higher prestress but eventually breaks into the asperity/barrier and produces large slip in the center. Another scenario is the cascade-up growth model in which the nucleation process initiates inside a small weak patch and tends to grow into large-scale rupture surrounding the rim of a larger and stronger patch. Such degree of complexity is previously only reproduced in dynamic simulations but is directly observed with sufficient level of details for the first time. This case study demonstrates the capability of the BP method, enhanced by aftershock calibrations, to observe and probe the geometrical complexity of dynamic ruptures.