| 概要 |
Data assimilation (DA) is a computational technique that integrates numerical simulation models and observational data based on Bayesian statistics.
The four-dimensional variational method (4DVar), or adjoint method, is widely used in DA for large-scale models, such as weather forecasting, to optimize model parameters and/or initial conditions and to evaluate their uncertainties.
We review the foundation of the 4DVar, including our algorithm that is capable of uncertainty quantification [1].
Then we show recent application examples of the 4DVar to the models in solid Earth science, such as the slow slip event model to quantify uncertainties in the estimated frictional parameters at a plate boundary [2], and the mantle convection model to estimate its status more than several tens of years ago [3, 4].
References
[1] Ito, S., H. Nagao, A. Yamanaka, Y. Tsukada, T. Koyama, M. Kano, and J.
Inoue, Dataassimilation for massive autonomous systems based on a second-order adjoint
method, Phys. Rev. E 94 (2016) 043307.
[2] Ito, S., M. Kano, and H. Nagao, Adjoint-based uncertainty
quantification for inhomo-geneous friction on a slow-slipping fault, Geophys. J. Int. 232(1) (2023) pp. 671–683.
[3] Nakao, A., T. Kuwatani, S. Ito, and H. Nagao, Adjoint-based data assimilation for reconstruction of thermal convection in a highly viscous fluid from surface velocity and temperature snapshots, Geophys. J. Int. 236(1) (2024) pp. 379–394.
[4] Nakao, A., T. Kuwatani, S. Ito, and H. Nagao, Adjoint-based marker-in-cell data assimilation for constraining thermal and flow processes from Lagrangian particle records, J. Geophys. Res. Machine Learning and Computation in revision.
|