Quantitative uncertainty estimation of the coseismic fault model based on the real-time GNSS time series

Rapid estimation of the coseismic fault model will be extremely important to predict the accurate tsunami forecasting. After the 2011 Tohoku-Oki earthquake, Tohoku University and Geospatial Authority of Japan are jointly developing the real-time crustal deformation monitoring system (REGARD). The REGARD adopts the maximum likelihood approach for the coseimsic fault model estimation. One of the problems in REGARD is the difficulty of the quantitative uncertainty estimation of the estimated coseismic fault model. Based on these backgrounds, we investigated the uncertainty estimation approach of the coseismic fault model, which contain two types of the models (single rectangular fault and slip distribution) by real-time GNSS time series. We adopt the full Bayesian inversion approach. We constructed a discrete representation of the posterior probability density function (PDF) by sampling with a MCMC method. In addition, considering utilization in real time, we adopted the Parallel tempering approach to obtain the reasonable result more efficiently.

In the single rectangular fault model estimation, we developed the new sampling flow to seek the reasonable search setting for the real-time purpose. We applied the developed method to the 2011 Tohoku-Oki and 2016 Fukushima-Oki earthquake based on the actual displacement field data. The obtained results clearly suggest that the onshore GNSS data cannot constrain the trade-off between fault area and slip amount of offshore fault without the a priori information, which are important information for precise near-field tsunami forecasting.

In the slip distribution model estimation, we focus on Nankai area in Japan where megathrust earthquakes have repeatedly occurred. We utilized the simulated Hoei type earthquake by Todoriki et al. (2013). We divided into 2951 rectangular subfaults along the subducting plate interface and designed Quadtree-like sampling flow by 4-step sequential segmentation. As the result, we successfully obtain the expected slip distribution and its uncertainty as the 95% C.I of posterior PDF. We will discuss how this diversity impacts on the tsunami inundation.