LASP Magnetosphere Seminars

Abstract

Geomagnetic storms, driven by solar activity, pose a significant threat to modern technological infrastructure. Accurate prediction of these events hinges on robust physical models of the Sun-Earth system. This work introduces a novel framework that leverages Physics-Informed Neural Networks (PINNs) to systematically evaluate and refine such models. Our approach is founded on the traditional model for the ring current evolution developed by Burton et al. (1975). By employing PINNs, we solve the inverse problem: determining the optimal parameters for several candidate solar wind-magnetosphere coupling functions while enforcing physical consistency. To enhance model fidelity and quantify uncertainties, we utilize an ensemble PINN approach and incorporate Random Fourier Features to mitigate the network’s spectral bias, enabling the capture of high-frequency storm dynamics. As an initial application, we validate the framework using the SMR SuperMAG index and solar wind observations from the May 2024 storm. Our results quantify significant performance differences among the coupling functions, providing new insights into the physical mechanisms of solar wind-magnetosphere interaction.