The Tohoku Megathrust Earthquake Cycle: Unravelling Key Processes Using Data Assimilation

Abstract

Earthquakes at subduction zones are among the most powerful natural events on Earth. During the interseismic, coseismic, and postseismic phases, interseismic coupling, co- and postseismic slip along the megathrust, and viscoelastic relaxation in the mantle produce the observed surface displacements. Different combinations of these processes can reproduce similar displacement patterns, resulting in a non-unique inverse problem. This thesis aims to quantify these processes using models that consistently describe stress accumulation and release throughout the earthquake cycle. The 2011 Tohoku-Oki earthquake serves as the central case study because of the extensive geodetic observations available. Data assimilation is used to estimate model parameters and uncertainties and to investigate whether the parameters controlling aseismic slip and viscoelastic relaxation can be estimated independently. First, a particle method is investigated using a simplified elastic flexural model of the overriding plate. Synthetic experiments show that onshore observations may constrain short-wavelength deformation, whereas seafloor observations near the trench are essential for resolving long-wavelength deformation. Assimilation of interseismic observations yields unrealistically low effective elastic thicknesses of 5–7 km. This indicates limitations of the simplified model, missing physical processes, and insufficient near-trench observations rather than shortcomings of the data assimilation method. Next, a two-dimensional earthquake cycle model is developed with realistic slab geometry and a temperature- and stress-dependent viscosity structure. The ensemble smoother with multiple data assimilation estimates rheological parameters describing the mantle and shear zone. Nonlinear dislocation creep in both domains is required to reproduce the observed interseismic and postseismic deformation. Aseismic slip occurs predominantly at depths of 40–80 km, whereas viscoelastic relaxation in the shallow continental mantle at 30–60 km depth is required to reproduce sufficient flexure. Nonlinear rheology reduces parameter trade-offs and allows the parameters controlling both processes to be estimated independently. Landward motion near the trench is explained by relaxation beneath the subducting plate and recovery of interseismic shortening, without requiring an explicit low-viscosity layer. Finally, the model is extended to three dimensions to account for along-strike variations in slab geometry, coupling, rheology, and aseismic slip. The interseismic and ten-year postseismic displacement time series contain sufficient information to independently estimate the parameters controlling aseismic slip and viscoelastic relaxation. Aseismic slip is concentrated at 40–80 km depth, reaches approximately 5 m after ten years, and occurs predominantly during the first five years following the earthquake. The inferred viscosity structure requires a composite rheology. Dislocation creep dominates deformation within the continental mantle at depths of 40–60 km, with effective viscosities of ~10¹⁹ Pa·s, whereas a viscosity of ~10²⁰ Pa·s is required at depths of 60–100 km. Beneath the subducting plate, deformation is dominated by dislocation creep, with effective viscosities of approximately 5 × 10¹⁸ Pa·s. The model successfully reproduces the magnitude, direction, and persistence of the postseismic landward motion observed in Hokkaido. This motion results from in-plane bending of the overriding plate in response to postseismic stress changes. Overall, the results demonstrate that aseismic slip and viscoelastic relaxation leave distinct signatures in the observed surface displacements and can therefore be quantified independently using realistic earthquake cycle models.