Estimating source parameters by dynamic source inversion of apparent source time functions or spectra

In dynamic source inversions, observed waveforms are used to infer frictional parameters on a fault, obtaining a rupture model constrained by both data and physics. Assuming a slip-weakening friction law, Bayesian dynamic rupture inversions of regional waveforms have already been performed for several large events (e.g., Gallovič et al., 2019;  Gallovič, et al., 2020; Kostka, et al., 2022). On the other hand, apparent source time functions (ASTF) or apparent source spectra are commonly used to estimate source parameters (corner frequency, stress drop, etc.) under the assumption of simplified source models (e.g., Brune source). The ASTFs can be inferred from observed waveforms by, e.g., empirical Green’s function deconvolution (Plicka et al., 2022), and characterize a station-specific source radiation free of path and site effects. Employing the ASTFs or apparent source spectra at each receiver instead of the full (low-frequency) waveforms in Bayesian dynamic source inversion is a promising way of surpassing the simplifying assumptions on the rupture process and radiation in earthquake source analyses.

 

Here we inspect the performance of the Bayesian dynamic source inversion applied to ASTFs and apparent source spectra on a series of synthetic tests. As a target model, we assume a Mw 6.2 dynamic source model on a 30x14km fault embedded in a 1D layered medium. The model parameters of the slip-weakening friction law are heterogeneous on the fault. The inversion is performed using the fd3D_tsn_pt code (Premus et al., 2020) with the same parametrization as the target dynamic model. The posterior dynamic model samples are used to assess the reliability of various kinematic or dynamic parameters, such as rupture size, duration, corner frequency, or static stress drop, including their uncertainty, when inverting ASTFs, apparent source spectra, or full waveforms. The synthetic tests reveal which of the source parameters can be estimated reliably using each dataset and which are biased due to the settings of the McMC sampling process. Finally, a real-data application of the ASTF dynamic source inversion for the 2010 Mw 6.9 deep earthquake in East China is shown.