4.1. Inversion Methodology
To obtain the spatial and temporal evolution of the slip for the mainshock, we invert near-field strong-motion displacement time-series recorded by ten three-components SSA-2 Kinemetrics digital accelerometers from the ISMN network. The stations are located at distances ranging between 4 km and 60 km from the rupture area (Fig. 1b). The acceleration data is integrated twice to displacements. The mean and trend of the waveforms are corrected and the horizontal components are rotated to an NS/EW coordinate system. The waveforms cut using a time window of 25.6 s after the respective origin time (Fig. 3d). The data were band-pass filtered using a Butterworth one-pass causal filter in the frequency band 0.08-0.7 Hz. We observed some low-frequency noise below 0.08 Hz. Also, the upper limit of the frequency band is chosen based on the resolution of the crustal velocity model and simplifications assumed in the used model.
The processed data has inverted for the rupture evolution using the elliptical sub-fault approximation method (i.e., Ruiz and Madariaga, 2013; Twardzik et al., 2012; Ruiz et al., 2019; Momeni et al., 2019). It approximates the rupture distribution with a few elliptical patches on a planar fault, and, has the advantage of reducing the number of parameters of inversion in comparison to the more commonly used rectangular sub-faults parametrization. Each of the elliptical slip patches is described by just nine parameters: five to define its geometry. The other four parameters to describe the rupture process, which are slip amplitude, slip duration, slip direction, and onset time. While this method is not suited to retrieve fine details of the rupture process, it focuses on the more robust features of the source.
Proper geometry is grid-searched for the mainshock near the two nodal planes obtained in section 2 (Fig 3b). One and two elliptical patch(s) were investigated to estimate the rupture process. During the inversions, for each of the tested geometries, we consider a wide range of source parameters (see Figures S2 to S11). The inversions were carried out using the Neighborhood Algorithm (Sambridge, 1999) to search for the rupture model that fits best the strong-motion displacements. The Green’s Functions were computed using AXITRA (Cotton and Coutant, 1997), a program that is based on a discrete wavenumber method (Bouchon, 2003), and adopting Tatar et al., (2012) velocity model. For each inversion, the hypocenter is allowed to move ±1 km on the fault plane along strike and dip to allow small corrections for errors on the origin time. Up to 500 iterations were applied during inversions, and each iteration had 35 different trial rupture models to ensure convergence (for more details see Figures S2 to S11).