Work

Development and Application of Full-Waveform Inversion for Regional-Scale Crust and Upper Mantle Imaging

In recent years, geophysical research has increasingly shifted from traditional traveltime tomography to full-waveform inversion (FWI), a technique that utilizes complete seismic waveforms to achieve high-resolution subsurface imaging. By capturing both phase and amplitude information, FWI offers significantly improved sensitivity to complex geological structures.

This research theme, initiated in 2022, evolved through a logical sequence: establishing a theoretical framework, validating the method in a well-studied subduction zone, and applying it to the structurally complex region of Taiwan. It represents a coherent and integrated development in both methodology and geophysical application.

1. Theoretical Foundation (Kan et al. 2022, GJI)

Published in Geophysical Journal International, the 2022 study introduced a consistent Bayesian multiparameter inversion framework for isotropic elastic media. The novelty lies in incorporating prior correlations between model parameters—density, P-wave velocity (VP), and S-wave velocity (VS)—through a non-diagonal model covariance matrix.

Synthetic tests demonstrate that this approach significantly reduces parameter trade-offs and enhances the recovery of density and shear wave anomalies at depth, particularly in teleseismic settings where sensitivity to deep structures is limited.

2. Application to the Cascadia Subduction Zone (Kan et al. 2023, JGR: Solid Earth)

The second study applied this inversion scheme to the Cascadia subduction zone in central Oregon, using teleseismic P and SH waveforms from the CASC93 array. The resulting models revealed:

• A thin (<10 km) east-dipping low-velocity layer interpreted as the fluid-saturated subducting Juan de Fuca oceanic crust.
• A progressive increase in velocity and density below 40 km, marking the onset of eclogitization.
• Upward migration of silica-rich fluids into the forearc mantle, triggering serpentinization and reducing seismic velocities.
• Extremely low VP/VS ratios in the forearc crust, indicating extensive quartz mineralization.
• A low-velocity, high VP/VS anomaly beneath the volcanic arc at ~75 km depth, consistent with partial melting triggered by slab dehydration.

These results provide direct seismic evidence for deep fluid transport, metamorphic reactions, and melting processes, linking them with structural and compositional changes in the lithosphere.

3. Imaging Complex Structures Beneath Taiwan (Kan et al., 2026, Tectonophysics, DOI: 10.1016/j.tecto.2026.231301)

The most recent work applied the same methodology to Taiwan, a region shaped by arc–continent collision and subduction polarity reversal. By inverting complete P and SH teleseismic waveforms recorded across the island, the study produced high-resolution 3-D models of density, VP, VS, and VP/VS. Key findings include:

• Imaging of the aseismic and detached segment of the Eurasian slab beneath central Taiwan, with a slab gap forming at 130–200 km depth north of 23.5°N, consistent with slab breakoff or tearing.
• A thickened crustal root (>50 km) beneath the Central Range, with elevated VP/VS values in the lower crust, possibly reflecting fluid infiltration from slab dehydration.
• Detailed velocity anomalies beneath the Tatun Volcano Group and Turtle Island, corresponding to shallow magma reservoirs.

Resolution tests (checkerboard models) confirmed that the method can reliably resolve structures at 20–60 km scale throughout the lithosphere, providing sharper images than conventional traveltime tomography.

Summary and Outlook

This series of studies demonstrates a complete research arc—from theoretical development, through field validation, to practical application in tectonically complex regions.

The results highlight the potential of multiparameter FWI to recover key geophysical properties and illuminate deep Earth processes such as fluid release, metamorphism, and crustal thickening.

Future applications may extend to other subduction zones, orogenic belts, and volcanic regions, contributing to a more accurate understanding of lithospheric dynamics.

Source Finite Fault Inversion

Finite fault inversion is a powerful method for characterizing the spatiotemporal rupture processes of moderate-to-large earthquakes. It plays a critical role in advancing the understanding of earthquake source physics and improving ground motion predictions.

This research theme focused on two case studies: the 2016 Mw 7.0 Kumamoto earthquake in Japan and the 2014 Mw 6.1 Ludian earthquake in China, with key results summarized below:

For the Kumamoto earthquake, waveform inversion using strong-motion (K-NET and KiK-NET) and teleseismic (IRIS) data revealed that the main rupture propagated northeastward at shallow depths starting ~4 seconds after origin time and continued until ~14 seconds. The largest slip (~2.5 meters) occurred ~17 km northeast of the hypocenter. A simulated annealing algorithm was used to explore fault parameters on a 2.5×2.5 km subfault grid with a 1D velocity model.

For the Ludian earthquake, comparisons were made between inversions using 1D and 3D velocity models. The results showed that 3D structure significantly affects the accuracy of waveform fits and rupture models. The inverted slip distribution was further used in ground motion simulations via a finite-difference method, successfully reproducing the peak ground velocity (PGV) pattern, including areas of concentrated shaking.

This line of research integrates source inversion theory, waveform processing, structural modeling, and ground-motion simulation, demonstrating the utility of finite fault models for both seismic hazard assessment and fundamental earthquake science.

Finite-Frequency Body Wave Sensitivity Kernels for Inner Core PKP Phases

My master's research focused on the finite-frequency effects of body waves used to probe the Earth's inner core structure, particularly the sensitivity of PKP phases to velocity anomalies. Traditional studies often rely on differential traveltime measurements between PKP branches (e.g., PKPab, PKPbc, PKIKP), modeled using ray theory. However, ray theory assumes infinitely high-frequency waves and restricts sensitivity to the geometric ray path, failing to capture the volumetric nature of real seismic wave propagation.

To overcome this limitation, I adopted the axisymmetric spectral-element method (AxiSEM) to simulate wave propagation in a spherically symmetric Earth model. This approach reduces the 3-D spherical problem to a 2-D semi-disk domain under axisymmetric conditions, significantly reducing computational cost and allowing high-frequency simulations (up to 1 Hz) at global scale. By computing both forward and adjoint wavefields and performing their convolution using the Monte Carlo Kernel method, I obtained finite-frequency delay-time sensitivity kernels for various PKP phases.

The study analyzed the sensitivity patterns of key inner-core phases including PKPab, PKPbc, PKIKP, and PKiKP, and examined their differential traveltime kernels (e.g., PKPbc–PKIKP, PKPab–PKIKP, and PKiKP–PcP) to assess the influence of lower mantle heterogeneities on inner-core phase measurements. The results highlight the significance of the banana-doughnut–shaped sensitivity kernels, showing that observed traveltime anomalies do not solely arise from anomalies along the ray path but from broader volumes defined by the first Fresnel zone. This challenges the conventional interpretation of differential times as being sensitive only to turning depths or path intersections.

This work improves our understanding of how seismic waves sample deep Earth structures and demonstrates the feasibility of incorporating high-frequency finite-frequency kernels into global-scale inverse problems. It provides a computationally efficient framework for future studies aiming to resolve deep Earth heterogeneity with greater accuracy and realism, particularly in applications such as full-waveform tomography or high-resolution mantle and core imaging.

Fault Scenario Simulation

This study evaluates the seismic hazard and stress perturbation at the Tieh-Chen-Shan site, a potential location for geological carbon storage in western Taiwan. We conducted scenario-based numerical simulations of ground motion and static stress/strain changes associated with four active faults near the site (Tunzihjiao, Sanyi, Dajia, and Tieh-Chen-Shan faults), as well as the 1999 Chi-Chi earthquake (Mw 7.6), to assess the impact of strong earthquakes on site stability and storage integrity.

We employed the Spectral Element Method (SEM) to model seismic wave propagation, incorporating a 3D velocity model (Kuo-Chen et al., 2012), surface topography (ETOPO1), and a 1D attenuation model (Wang et al., 2010). Fault geometries and source mechanisms were defined based on published fault catalogs and empirical scaling relationships. The Sanyi Fault was further subdivided into EW- and NS-trending segments to capture the observed strike change. For each fault, three rupture scenarios with varying hypocenter locations were simulated using point-source approximations.

For each scenario, we calculated the Peak Ground Velocity (PGV) and mapped the corresponding intensity at the injection site. Static stress and strain changes were computed using the Okada (1992) elastic dislocation model via the Coulomb 3.3 code, assuming a dipping reservoir layer at ~1.4 km depth.

The simulations show that:

• Most earthquake scenarios result in intensities no higher than Intensity 3 (Central Weather Bureau scale) at the site.
• In a few scenarios (e.g., Sanyi NS rupture), the PGV may reach ~25 cm/s, corresponding to Intensity 5−, but the resulting co-seismic normal stress changes remain small, with a maximum increase of only +0.168 bar.
• Even under the Chi-Chi earthquake scenario (which generated Intensity 5 at the site), the modeled stress increase is only +0.28 bar, far below the reservoir's estimated in-situ stress (~30 MPa).

These results demonstrate that, even under worst-case local and regional seismic conditions, the Tieh-Chen-Shan site exhibits high resilience against dynamic and static stress perturbations. The estimated stress variations are several orders of magnitude lower than the background pore pressure, implying negligible risk of structural compromise or induced leakage from CO₂ injection.

Publications