Advanced technique for producing accurate space weather maps of the polar ionosphere — Combining observations and AI models to reproduce a more realistic space environment

A research group from the Institute of Statistical Mathematics (ISM), the National Institute of Polar Research (NIPR), the Okinawa Institute of Science and Technology Graduate University (OIST), the National Institute of Information and Communications Technology (NICT), and the Graduate University for Advanced Studies (SOKENDAI) has developed a new method for accurately reproducing the electric field distribution in Earth's polar ionosphere. This method makes it possible to reproduce detailed temporal variations that were difficult to capture with conventional numerical models, and to interpolate regions where direct observation is impossible using results calculated according to physical laws. The researchers succeeded in constructing an unprecedentedly accurate "space weather map" of the ionosphere. Their results were published in the American Geophysical Union journal Space Weather.

The region roughly 100 to 1,000 kilometers above Earth is known as the ionosphere, where part of the atmosphere separates to form a plasma state. In the high-latitude ionosphere, the spatial distribution of the electric field changes from moment to moment, and the ionospheric currents driven by this field can affect satellite orbits and ground-based infrastructure. Tracking these electric field variations is therefore essential for the safe use of space and for understanding the space environment.

The electric fields and currents of the ionosphere are generated by physical processes in the magnetosphere, the region above the ionosphere. The research group had previously developed an emulator called "SMRAI2" that uses machine learning to reproduce the output of the magnetospheric MHD model "REPPU (REProduce Plasma Universe)," which numerically simulates physical processes in the magnetosphere. By providing solar wind conditions as input, the researchers were able to create a system to predict the electric field distribution of the polar ionosphere.

However, detailed temporal variations arising from the complex physical processes of the magnetosphere and ionosphere were not well reproduced. Although ground-based observation methods such as radar and magnetic field measurements exist for monitoring electric fields and currents, they only yield data from limited regions. It was difficult to produce a "space weather map" covering electric fields and currents across the entire polar ionosphere.

In this study, the research group made use of data assimilation—a technique also used in weather forecasting—to incorporate ionospheric plasma velocity data obtained by the international radar network "SuperDARN (Super Dual Auroral Radar Network)" into an improved version of SMRAI2 called "SMRAI2.1." This enabled the development of a method that accurately reproduces variations in electric field distribution across the entire polar ionosphere.

Data assimilation is normally used to integrate physics-based model simulations with observational data. Implementing data assimilation requires ten times the computation time needed to run a simulation alone. It is challenging to directly apply data assimilation to magnetospheric models—which involve more complex physical processes than the lower atmosphere that is the focus of weather forecasting.

The research group addressed this computational challenge by using an "emulator"—a model that mimics machine learning-based simulation output—and for the first time succeeded in constructing a "space weather map" of the polar ionosphere that reflects the physical processes of the magnetosphere and ionosphere.

In contrast to conventional methods that assume a fixed functional form, using results calculated according to physical laws eliminates unnatural temporal variations and enables the production of more accurate "space weather maps."

As a result, it became clear that the actual electric field distribution varies more strongly overall than MHD model simulations had suggested. In addition, it is now possible to easily monitor ionospheric variations even in regions where direct observation is not feasible.

This study demonstrates that combining an emulator of the magnetospheric MHD model with real observations makes it possible to reproduce the ionospheric environment more realistically. Going forward, the use of real-time ground-based observational data is expected to contribute to improved space weather forecasting accuracy, operational support for various social systems including satellites, and a wider range of applications for understanding the space environment.

Journal Information
Publication: Space Weather
Title: Data Assimilation Into a Machine Learning-Based Emulator of a Global MHD Simulation for Analyzing the Polar Ionosphere
DOI: 10.1029/2025SW004488