Bayesian exploration of coseismic seafloor deformation process during the 2011 Tohoku-Oki earthquake using near-field tsunami records

The diverse range of data for the 2011 M_w 9.0 Tohoku-oki earthquake are unprecedented. However, several models using joint data sets still disagree on the estimated slip distribution due to (1) choice of fault geometry and elastic structure; (2) choice of regularization scheme in optimization approaches; (3) lack or under-estimation of model prediction error due to imperfect forward modeling. Some important questions therefore remain unresolved, e.g., what is the profile of fault slip as one approaches the trench, and how much of the co-seismic signals is recorded in seafloor geodetic measurements.

To obtain robust and detailed features of the co-seismic process, we present an approach to invert for the seafloor deformation field using only tsunami waveforms recorded by near-field stations, including GPS buoys, ocean bottom pressure gauges, cable pressure gauges and open ocean DART buoys. In addition to observational error, we incorporate model prediction error by considering the uncertainty in dispersion characteristics of tsunami propagation simulations. We adopt an analytical Bayesian approach to derive the posterior distributions for the coseismic seafloor deformation with minimal a priori assumptions. The analytical approach provides fast and robust characterization of coseismic seafloor deformation using the first arrivals of tsunami waveforms with potential for real-time applications.

Our models show that large seafloor uplift is required at the trench, with maximum seafloor uplift occurring about 50 km from the trench. The actual fault slip depends on the assumed elastic structure and fault geometry; in the case of a homogeneous half-space and simplified fault geometry, slip decreases towards the trench. This method also provides direct comparisons with seafloor geodetic measurements and a quantitative estimation of the respective contributions from co- and post-seismic processes.