5:15 PM - 6:30 PM
[SCG39-P16] A Bayesian inversion for slip distribution of slow slip events beneath the Bungo Channel based on ensemble modeling of the uncertainty of underground structure
Keywords:Slow slip events , Bayesian inference, Uncertainty of the underground structure, Geodetic data inversion
Because long-term slow slip events (L-SSE) are associated with relatively long duration and large magnitude, we can estimate slip distribution of L-SSE using geodetic displacement data. For instance, the L-SSE beneath the Bungo Channel is known to regularly occur in every six or seven years with the help of continuous GNSS observation in the land area. Most of geodetic slip inversions performed for this event were based on regularization using spatially uniform smoothing constraints. On the other hand, Nakata et al. (2017) introduced a sparse-promoting regularization for slip estimation for the past events. Their result of slip distribution with discontinuous boundaries obtained by introducing sparse-promoting constraint agreed better with the spatial distribution of deep nonvolcanic tremors and temperature structure at the plate boundary of Nankai Trough area. It also follows that the choice of empirical constraints for regularization to reduce the instability of estimation due to several error factors often largely affects the estimation results.
Geodetic slip inversion requires a prior assumption of the underground structure. Provided that we only have information on the underground structure with uncertainty, the assumption of “one model” of the underground structure may lead to a significant amount of model prediction errors, which essentially results in significant bias in slip estimation results. Several studies have addressed this problem by considering the contribution of the model prediction error due to the uncertainty of Green’s functions to the data covariance (e.g., Yagi & Fukahata 2008, 2011; Duputel et al. 2014). Agata et al. (2021) recently developed a new full Bayesian inference method for estimating fault slips, which considers the model prediction errors more generally without assumption of Gaussian distribution by incorporating the uncertainty of the underground structure using an ensemble consisting of many models of underground structure: It reduces the error factor associated with the assumption of “one model” by considering an ensemble of multiple underground structure models together. In this study, we apply the new method to estimation of slip distribution in the L-SSE in 2010 that occurred beneath the Bungo Channel, because this method may allow for estimating slip distribution without introducing such empirical constraint, relaxing the instability of estimation.
We use the homogeneous elastic half-space for the underground structure model, whose property is characterized by the plate boundary model to locate the fault slips and Poisson’s ratio of the media. We consider an ensemble of weight averages of three models of the geometry, namely, those provided in the models of Hayes et al. (2018), Iwasaki et al. (2015) and Koketsu et al. (2009, 2012). We assume that the stochastic property of the weights is characterized by the Dirichlet distribution with α=0.5. We also consider an ensemble of Poisson’s ratio calculated based on random samples of density from a uniform distribution between 2000 and 3400 kg/m3 and the empirical relationship between density and other elastic parameters proposed by Brocher (2005). In such a manner, we compose 1,000 samples of underground structure models and incorporate them as an ensemble in the estimation method of Agata et al. (2021). We use the digital data for the observed vertical and horizontal displacements provided by Yoshioka et al. (2015).
As a result, we obtained posterior PDFs for the slip distribution, in which the mean model presents a sharp slip distribution which is similar to that estimated in Nakata et al. (2017) rather than to those estimated based on spatially uniform smoothing constraints. In the conference, we would also like to show results incorporating uncertainty of one-dimensional heterogeneous elastic structure to consider a wider model space for the underground structure, which is expected to further reduce the bias in the estimation.
Geodetic slip inversion requires a prior assumption of the underground structure. Provided that we only have information on the underground structure with uncertainty, the assumption of “one model” of the underground structure may lead to a significant amount of model prediction errors, which essentially results in significant bias in slip estimation results. Several studies have addressed this problem by considering the contribution of the model prediction error due to the uncertainty of Green’s functions to the data covariance (e.g., Yagi & Fukahata 2008, 2011; Duputel et al. 2014). Agata et al. (2021) recently developed a new full Bayesian inference method for estimating fault slips, which considers the model prediction errors more generally without assumption of Gaussian distribution by incorporating the uncertainty of the underground structure using an ensemble consisting of many models of underground structure: It reduces the error factor associated with the assumption of “one model” by considering an ensemble of multiple underground structure models together. In this study, we apply the new method to estimation of slip distribution in the L-SSE in 2010 that occurred beneath the Bungo Channel, because this method may allow for estimating slip distribution without introducing such empirical constraint, relaxing the instability of estimation.
We use the homogeneous elastic half-space for the underground structure model, whose property is characterized by the plate boundary model to locate the fault slips and Poisson’s ratio of the media. We consider an ensemble of weight averages of three models of the geometry, namely, those provided in the models of Hayes et al. (2018), Iwasaki et al. (2015) and Koketsu et al. (2009, 2012). We assume that the stochastic property of the weights is characterized by the Dirichlet distribution with α=0.5. We also consider an ensemble of Poisson’s ratio calculated based on random samples of density from a uniform distribution between 2000 and 3400 kg/m3 and the empirical relationship between density and other elastic parameters proposed by Brocher (2005). In such a manner, we compose 1,000 samples of underground structure models and incorporate them as an ensemble in the estimation method of Agata et al. (2021). We use the digital data for the observed vertical and horizontal displacements provided by Yoshioka et al. (2015).
As a result, we obtained posterior PDFs for the slip distribution, in which the mean model presents a sharp slip distribution which is similar to that estimated in Nakata et al. (2017) rather than to those estimated based on spatially uniform smoothing constraints. In the conference, we would also like to show results incorporating uncertainty of one-dimensional heterogeneous elastic structure to consider a wider model space for the underground structure, which is expected to further reduce the bias in the estimation.