Bayesian Tsunami-Waveform Inversion and Tsunami-Source Uncertainty Estimation for the 2011 Tohoku-Oki Earthquake

Abstract:

This paper develops a Bayesian inversion to infer spatio-temporal parameters of the tsunami source (sea surface) due to megathrust earthquakes. To date, tsunami-source parameter uncertainties are poorly studied. In particular, the effects of parametrization choices (e.g., discretisation, finite rupture velocity, dispersion) on uncertainties have not been quantified. This approach is based on a trans-dimensional self-parametrization of the sea surface, avoids regularization, and provides rigorous uncertainty estimation that accounts for model-selection ambiguity associated with the source discretisation. The sea surface is parametrized using self-adapting irregular grids which match the local resolving power of the data and provide parsimonious solutions for complex source characteristics. Finite and spatially variable rupture velocity fields are addressed by obtaining causal delay times from the Eikonal equation. Data are considered from ocean-bottom pressure and coastal wave gauges. Data predictions are based on Green-function libraries computed from ocean-basin scale tsunami models for cases that include/exclude dispersion effects. Green functions are computed for elementary waves of Gaussian shape and grid spacing which is below the resolution of the data. The inversion is applied to tsunami waveforms from the great Mw=9.0 2011 Tohoku-Oki (Japan) earthquake. Posterior results show a strongly elongated tsunami source along the Japan trench, as obtained in previous studies. However, we find that the tsunami data is fit with a source that is generally simpler than obtained in other studies, with a maximum amplitude less than 5 m. In addition, the data are sensitive to the spatial variability of rupture velocity and require a kinematic source model to obtain satisfactory fits which is consistent with other work employing linear multiple time-window parametrizations.