[9-4]

Simulation-/Data-driven Data Assimilation Project toward "Terra Forecasting"

H. Nagao (Earthquake Research Institute, The University of Tokyo), M. Kano (Earthquake Research Institute, The University of Tokyo), S. Mizusako (Earthquake Research Institute, The University of Tokyo), A. Suzuki (Graduate School of Information Science and Technology, The University of Tokyo)

The current framework of data assimilation essentially relies on a priori given simulation model rather than observation data, so that phenomena which the simulation model does not assume are neither extractable nor predictable. Since it is impossible to provide a perfect simulation model that reproduces all complex phenomena, data assimilation that implements a cutting-edge data-driven approach would open a new paradigm in geophysical modeling.

Our group aims to establish methodology of such data assimilation driven by both simulation models and observation data, toward "terra forecasting" in the solid Earth science like "weather forecasting" in meteorology. This paper overviews our activities related to terra forecasting, i.e., estimation and prediction of states in the Earth's crust. Kano et al. reproduces the spatiotemporal distribution of afterslip triggered by the 2003 Tokachi-oki Earthquake by applying an adjoint data assimilation to GNSS data, simultaneously estimating the frictional parameters at the oceanic plate boundary. Suzuki et al. develops a region dividing method using the k-means clustering to clarify large-scale structure of frictional properties in an afterslip region. Mizusako et al. proposes a sparse modeling procedure based on "lasso" to image seismic wave field in the Tokyo metropolitan area from a dense seismic array MeSO-net.