9:50 AM - 10:10 AM
[MGI26-04] Numerical experiments for estimating frictional properties on an SSE fault using the adjoint method
★Invited Papers
Keywords:Data assimilation, SSE, Friction
Numerical simulations of the fault slip have been performed to understand the slip phenomena and predict the spatiotemporal evolution of the slip. The frictional properties of the fault are the key to determining the slip behavior, but they are generally unknown. So, when performing the simulations, they are usually determined by trial and error to qualitatively reproduce the characteristics of the observed slip. Recently, data assimilation (DA) studies have attempted to estimate the frictional properties directly from observational data. DA incorporates the observed data into the physics-based model to construct a more plausible model. When DA works well, we can obtain the physics-based fault model, including the frictional properties, that can quantitatively explain the observed data. The constructed physics-based model can be used to simulate the slip evolution beyond the data period, i.e., it is expected to be a slip prediction.
In seismology, DA studies have progressed in recent years, but the success has been limited to the problems of wave propagation, and there are still many problems in applying DA to the fault slip system. The fault takes tens to hundreds of years to accumulate the elastic strain and releases it as an earthquake in seconds to minutes. This is a “stiff” system and difficult to apply DA. Therfore, currently, DA is also applied to slower fault slip events, the postseismic slip or the slow slip (SSE), to gain knowledge for predicting the slip event. In this talk, we present the numerical experiments applying the DA method to SSE.
Along the Nankai Trough in southwestern Japan, SSEs occur in the deep extensions of the seismogenic zone, and the Bungo SSE is repeatedly observed in the Bungo Channel. Hirahara and Nishikiori (2019) conducted numerical experiments to simultaneously estimate the slip evolution and the frictional parameters of the Bungo SSE using EnKF, a sequential DA method and showed that EnKF works effectively works. However, the method of selecting the initial ensemble members has not been established, and it is also difficult to determine whether the estimated values are close to the true values. So, instead of EnKF, we apply the adjoint method, a variational DA method, to Bungo SSE, and perform the numerical experiments in the same setting with H&N (2019).
The adjoint method optimizes the time-invariant model parameters and the initial values of the model variables to minimize the time series of the misfit between the model and the observed data. In this study, the former are the frictional parameters, and the latter are the distributions of slip rate and frictional strength. However, when the observed data period is limited, the initial distribution of the frictional strength is in a trade-off with the frictional parameters, and they cannot be resolved. To avoid this problem, we add a new constraint of the periodicity of the SSE to the ordinary adjoint method and propose a two-step iterative method. We performed numerical twin experiments to validate the method and showed that the resulting estimated values were obtained around the true values, starting from the various values for the initial guess.
In seismology, DA studies have progressed in recent years, but the success has been limited to the problems of wave propagation, and there are still many problems in applying DA to the fault slip system. The fault takes tens to hundreds of years to accumulate the elastic strain and releases it as an earthquake in seconds to minutes. This is a “stiff” system and difficult to apply DA. Therfore, currently, DA is also applied to slower fault slip events, the postseismic slip or the slow slip (SSE), to gain knowledge for predicting the slip event. In this talk, we present the numerical experiments applying the DA method to SSE.
Along the Nankai Trough in southwestern Japan, SSEs occur in the deep extensions of the seismogenic zone, and the Bungo SSE is repeatedly observed in the Bungo Channel. Hirahara and Nishikiori (2019) conducted numerical experiments to simultaneously estimate the slip evolution and the frictional parameters of the Bungo SSE using EnKF, a sequential DA method and showed that EnKF works effectively works. However, the method of selecting the initial ensemble members has not been established, and it is also difficult to determine whether the estimated values are close to the true values. So, instead of EnKF, we apply the adjoint method, a variational DA method, to Bungo SSE, and perform the numerical experiments in the same setting with H&N (2019).
The adjoint method optimizes the time-invariant model parameters and the initial values of the model variables to minimize the time series of the misfit between the model and the observed data. In this study, the former are the frictional parameters, and the latter are the distributions of slip rate and frictional strength. However, when the observed data period is limited, the initial distribution of the frictional strength is in a trade-off with the frictional parameters, and they cannot be resolved. To avoid this problem, we add a new constraint of the periodicity of the SSE to the ordinary adjoint method and propose a two-step iterative method. We performed numerical twin experiments to validate the method and showed that the resulting estimated values were obtained around the true values, starting from the various values for the initial guess.