Characterizing Performance Limitations on the L1-to-Ground Prediction of Geomagnetically-Induced Currents

Prediction of geomagnetically-induced currents (GICs) flowing in power lines during space weather is a challenging problem that encompasses multiple branches of physics and engineering. The genesis of these currents, closely related to fast changes in the magnetic field on the Earth's surface, ultimately traces back to the Sun. Collection of high-rate ground magnetic field measurements, coupled with the recent deployment of large-scale magnetotelluric surveys and the development of sophisticated power system models, has solidified our understanding of the relationship between Earth's surface magnetic field and GICs. However, the complex relationship between powerful solar ejecta (coronal mass ejections) and Earth's surface magnetic field remains the most challenging link in the GIC prediction chain. Here, we discuss the empirical "L1-to-ground" prediction problem in the context of learning theory, and present data-derived upper bounds on approximation error induced by coarse temporal sampling. Our results suggest that to keep peak GIC estimation errors below ~25%, sample-to-sample predictions of the surface magnetic field must characterize geomagnetic spectral content up to at least 25-30 mHz. Additionally, we briefly discuss how these ideas translate to prediction of signal properties, and show some preliminary empirical GIC predictions trained on historical solar wind data from the Advanced Composition Explorer (ACE) satellite and magnetic field data from USGS geomagnetism observatories in the US.