Fault geometry and finite fault inversion
We use the geodetic observations and the relocated aftershocks [W Wang et al. , 2021] to define the geometry of rectangular fault segments that will be used in the finite fault model (FFM) inversion. Ten fault segments are needed to mimic the first-order strike variations due to bending and bifurcation, as shown in the geodetic observation (Fig.1a and Fig.2a). These ten segments are sub-vertical faults, as aftershocks are distributed quite close to the surface rupture trace (Fig.2b), and the E-W deformations across the fault are nearly symmetric (Fig.1b). Aftershocks are located mostly to the north of surface rupture at fault segments 1-6 while aftershocks are distributed primarily to the south of surface rupture at fault segments 7-10. Based on this relative location between seismicity and surface rupture trace, we set the dipping direction of fault segments 1-6 and 7-10 towards north and south, respectively (Fig.2). Note, that the seismic stations used in earthquake relocation study are distributed quite uniformly in the source region [W Wang et al. , 2021]. Therefore, we can rule out the systematic location bias between the seismicity to the west and east of the epicenter. Geodetic-only inversions can then be conducted to determine the dip angle of each fault segment (Table S2).
Based on this fault geometry, we jointly invert geodetic data, nearby high-rate Global Positional System (GPS) waveforms, regional broadband waveforms and teleseismic body waves using a finite fault inversion method [Ji et al. , 2002] to recover the kinematic rupture history and slip distribution on the fault segments (Supplement text-2 ). We divide the rectangular fault segments into 3 km × 2 km patches, and invert for the slip amplitude, slip direction, rupture time and rise time on each patch. A Laplacian smoothing algorithm is applied for the slip distribution during the inversion. High-quality nearfield geodetic and high-rate GPS observations greatly suppress the trade-offs between the parameters in the inversion.
Our preferred FFM is presented in Fig.2b in a map view and in Fig.3 as a depth view along the strike. This model produces excellent fits to both seismic waveforms (Fig.S3-5) and static surficial deformations (Fig.S6-8), suggesting a reliable model resolution. Most of the slips are distributed at the depth shallower than 8 km except for S5 and S6, where slips are as deep as 15 km. Interestingly, the distribution of the aftershocks also shows a gap at S5 and S6 (Fig.2b and Fig.3d), indicating the rupture released most of the stress accumulated in the entire seismogenic zone, which has a depth of ~15 km. On the other segments, where the seismogenic zone is only partially ruptured, aftershocks are much denser, and distributed in the depth range of 5-15 km, showing a clear complementary feature with the coseismic slip distribution (Fig.3d), similar to that observed for other strike-slip earthquakes (e.g., [Wei et al. , 2011]). The aftershocks were most likely triggered by down-dip post seismic slip and Coulomb stress change imposed by coseismic slip. Therefore, slip deficit happens both at shallow [Jin and Fialko , 2021] and at greater depths, that is, at the upper and lower bounds of the seismogenic zone. The equivalent moment tensor of the FFM shows relatively small non-double-couple component (Fig.3b), which is highly similar to that from multiple point source solution (see next section). Our equivalent moment tensors are more similar to the GCMT solution, including the Mw, but much more different from the W-phase solution, in which the east-west oriented fault plane has a shallower dip angle (67°) and quite strong normal faulting component in rake (-40°). It appears that a reliable FFM of Maduo earthquake provides an independent verification to the global moment tensor solutions.
There are substantial slip distributions on the two branches of the bifurcated fault, where the moment magnitude of the northern branch (S7-8, Mw6.8) is slightly larger than that on the southern branch (S9-10, Mw6.7). The sizable and comparable moment distribution on the two branches is a key feature that allows robust resolution on their ruptures. The waveform observations on high-rate GPS station HSHX exhibit a clear frequency-dependent feature that the rupture from the bifurcated branches (S7-10) produced more high-frequency seismic waves than that from earlier ruptures (S1-6) (Fig.4). HSHX station is located at almost the same distances away from S4-10 (Fig.2b), therefore the geometric spreading induced attenuations at HSHX are similar for ruptures on these fault segments. If we assume pure strike-slip motion uniformly distributed on these fault segments, we would expect stronger seismic waves from S4-6 as the azimuth of HSHX is closer to the strongest SH-wave radiation direction of S4-6. Indeed, in the displacement waveforms, which are dominated by low-frequency energy, seismic waves excited from S1-S6 show larger amplitudes than those from S7-10, as shown in the synthetic waveform decompositions (Fig.4). However, in velocity waveforms, in which higher-frequency features are presented, the waveform amplitudes from S7-S10 are larger than those from S1-S6. This suggests that S7-10 radiated more high-frequency waves than those on the other fault segments. Note that the displacement waveforms are dominated by the periods of 10 - 20 s, while the velocity waveforms show stronger energy at 5 - 10 s. We did not fit the N-S velocity waveform as good as the E-W component for the rupture from S7-10 (highlighted by circles in Fig.4). This portion of the waveform shows more high frequency energy in the N-S component than the E-W component. This larger misfit to the higher frequency waveform is likely because our FFM inversion is dominated by relatively low frequency energy in the observations.