Probabilistic Machine Learning Techniques for Field Aligned Current Predictions Using AMPERE NEXT and Connections to a Data-Driven Magnetosphere Model.

Field-Alligned Currents (FACs) play a critical role in the coupled Magnetosphere - Ionosphere - Thermosphere (MIT) system, facilitating the transfer of energy and momentum from the solar wind into near-Earth space. Since 2019, the next generation of Active Magnetosphere and Planetary Electrodynamics Response Experiment (AMPERE Next) satellites have produced global maps of ten-minute-averaged FACs in both the Northern and Southern Hemispheres. This volume of new data provides an opportunity to learn more about the influence of the solar wind on the coupled MIT system.

This study presents a probabilistic Machine Learning (ML) approach for forecasting Northern Hemisphere FAC distributions using upstream solar wind and interplanetary magnetic field measurements. The model is designed to capture the structure of FACs on a 1-degree Magnetic Latitude and 1-hour Magnetic Local Time resolution grid, while explicitly representing predictive uncertainty and identifying solar wind drivers. We describe the model architecture and training method, and present preliminary validation results, including performance during geomagnetic storm events selected from the ML-based Geospace Environment Modeling (MLGEM) resource group at the Geospace Environment Modeling (GEM) workshop. Finally, we outline plans for integration of this work into the Artificial Intelligence Modeling Framework for Advancing Heliophysics Research (AIMFAHR) project.