International Journal of Science, Engineering and TechnologyAuthors: Rajesh Kumar Mishra, Divyansh Mishra, Rekha Agarwal
Abstract: The coupling between solar activity, interplanetary magnetic field (IMF) dynamics, and Earth’s magnetosphere-ionosphere system gives rise to geomagnetic storms and substorms that represent a primary space weather hazard for technological infrastructure, including power grids, navigation systems, and low-Earth orbit (LEO) satellites. Traditional physics-based models rooted in magnetohydrodynamics (MHD) and empirical index formulations—such as the Dst and Kp indices—are physically interpretable but computationally expensive and insufficient for real-time forecasting at the 1–6 hour horizons required for operational applications. In this work, we develop and rigorously evaluate a hybrid Artificial Intelligence (AI) framework that fuses deep learning with physics-informed constraints to deliver accurate, uncertainty-quantified predictions of geomagnetic activity across multiple temporal horizons. Our ensemble approach achieves a root-mean-square error (RMSE) of 9.4 nT at 1-hour lead time and 18.7 nT at 6-hour lead time, representing improvements of 34% and 41%, respectively, over the best-performing operational empirical model (Temerin-Li). The full AI pipeline produces a complete geomagnetic forecast cycle in under 2 seconds on commodity GPU hardware, meeting real-time operational constraints.
