L1G Prediction of Geomagnetically-Induced Currents: Dataset Partitioning and Performance Analysis

Abstract:

During space weather events, fast fluctuations in the surface geomagnetic field can generate large currents in power lines (known as geomagnetically-induced currents, or GICs). These fast magnetic field fluctuations are caused by the impact of a coronal mass ejection onto Earth's magnetosphere. Prediction of these currents is enabled by upstream solar wind measurements (e.g., measurements taken at the L1 Lagrange point). Here, we discuss neural network prediction of the surface magnetic field time derivative (dB/dt) using the historical solar wind data from the Advanced Composition Explorer (ACE). We start by examining statistically-preservant methods for partitioning the historical solar wind data into training, validation, and testing sets. We will then explore the overall predictive performance and the ability to make predictions throughout different geomagnetic storm phases.