Japanese Islands: From Earth’s Structure to Seismic Sources

Studies of the Earth’s internal structure as well as the physics of earthquake source processes have been the cornerstones of seismology. Both source and structure contribute to seismic recordings, thereby posing the challenge to constrain them independently. In this thesis we construct a three-dimensional seismic velocity model with the aim to diminish the effects that unaccounted Earth’s structure has on the source estimates. In the first part of the thesis, we present a full-waveform seismic tomography model for the Japanese archipelago, constructed with 15–80 s period data. The model is radially anisotropic, visco-elastic, and fully heterogeneous. Differences between the calculated and the observed waveforms are quantified using time-frequency misfits. This allows us to explore the temporal evolution of the frequency content of the data and to jointly invert body and surface waves with no need to identify specific seismic phases. We use adjoint methods as an effective means to obtain sensitivity kernels and ultimately gradients, required for iterative gradient-based minimization techniques. In addition to the high-velocity signatures of the well-studied Pacific and Philippine slabs, our model reveals a pronounced low-velocity anomaly beneath the volcanic island of Ulleung in the Sea of Japan. While our tomographic model can be viewed as a stand-alone study, enhancing our knowledge of the subduction dynamics and intraplate volcanism, it also plays a pivotal role in earthquake source inference. In the second part of the thesis, we develop a methodology for seismic source inversion using a three-dimensional Earth model and present a proof-of-concept catalog of seismic source solutions. We study small-to-moderate magnitude (Mw 4.8 – Mw 5.3) offshore earthquakes at the Izu-Bonin trench, which have a significant non-double-couple component. We compute receiver-side Green’s strains numerically and store them in a database for an easy access, which in turn enables a probabilistic inversion approach. The sampling is performed with a variant of the Hamiltonian Monte Carlo algorithm, which allows us to assess the uncertainty of the model parameters and to study the inter-parameter trade-offs. We simultaneously infer a centroid location, centroid time and moment tensor of an earthquake. The inversions are performed across multiple period bands. To assess the effect of the parametrization we run two separate inversions for each case: i) inverting for six independent moment tensor components, and ii) imposing a zero-trace constraint. Our results indicate that a purely deviatoric inversion leads to a higher double-couple share. This suggests a possible overestimation of the double-couple component in the case of reduced parametrization. Multi-period band inversions tend to show a variability in source solutions. We hypothesize that the discrepancy between short- and long-period data inversions might be due to weaker data constraints at longer periods, near-source anisotropy, source complexity, or an implicit bias towards the prior in our implementation. Accounting for a 3D structure of the Earth at short periods (15–80 s, 20–80 s) generally leads to an increase in double-couple component compared to the GCMT solution. This suggests that at least some part of the non-double-couple component in the GCMT catalog might be apparent – resulting from unmodeled Earth’s structure. In summary, a credible seismic velocity model is critical to extract the subtle waveform differences, which hide large variations in source parameters. A three-dimensional model also enables us to use shorter-period data, which have the potential to improve the centroid moment tensor solutions