[M-GI26] Data assimilation: A fundamental approach in geosciences

[MGI26-P14] Data assimilation into a machine learning-based emulator of global MHD simulation

*中野 慎也^1,2,4、片岡 龍峰^3,4、藤田 茂^1,2 (1.情報・システム研究機構 統計数理研究所、2.情報・システム研究機構 データサイエンス共同利用基盤施設、3.情報・システム研究機構 国立極地研究所、4.総合研究大学院大学)

キーワード：データ同化、磁気圏電離圏系、機械学習

The ionospheric electric field is one of fundamental factors controlling the dynamics of the polar ionosphere. The electric potential distribution in the polar ionosphere is imposed as a result of physical processes in the magnetosphere. Hence, the modeling of the magnetospheric process is essential to reproduce the ionospheric dynamics. Recent magneto-hydrodynamic (MHD) models of the magnetosphere enable to successfully simulate the ionospheric potential pattern. However, it is computationally demanding to run the MHD models. In addition, assimilation of the ionospheric observations is desired for obtaining reliable results.

To overcome the problem of the computational cost, we have developed a machine learning-based emulator of the MHD model of the magnetosphere. Our emulator, which is based on an echo state network model, allows us to efficiently predict the ionospheric potential distribution under a given solar wind condition. We then assimilate the SuperDARN (Super Dual Auroral Radar Network) data, which provides the information on ionospheric plasma flow, into the emulator. This approach is promising for real-time monitoring of the ionospheric state.