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

[MGI26-P05] Assimilation of ionospheric non-Gaussian data into an emulator of a magnetosphere-ionosphere model

*Shin ya Nakano^1,3,5, Sachin Reddy^2,6, Ryuho Kataoka^2,5, Aoi Nakamizo⁴, Shigeru Fujita^3,1 (1.The Institute of Statistical Mathematics, 2.National Institute of Polar Research, 3.Joint Support Center for Data Science Research, 4.National Institute of Information and Communications Technology, 5.Graduate Institute for Advanced Studies, SOKENDAI, 6.Now at NASA Jet Propulsion Laboratory, California Institute of Technology)

Keywords:data assimilation, MHD simulation, emulator, reservoir computing

The dynamics of the polar ionosphere is strongly controlled by physical processes in the magnetosphere. In order to rapidly predict the electric field and current in the polar ionosphere, we have developed a machine-learning-based emulator of the magneto-hydrodynamic (MHD) models of the magnetosphere. This emulator provides an empirical prediction of the ionospheric response to the solar wind variation as a result of the magnetospheric processes. We also conduct data assimilation into this emulator to enhance the realism of the prediction. The line-of-sight velocity data by the SuperDARN radars are assimilated into the emulator to obtain the global map of the electric potential distribution. Since the line-of-sight velocity data contain many outliers and have a heavy-tailed distribution, the observations are modelled with a multivariate Student distribution. The data assimilation is achieved by an algorithm which combines the ensemble transform Kalman filter with the importance sampling method to efficiently handle non-Gaussian features.