A Deep Learning Approach to the Forecasting of Ground Magnetic Field Perturbations at High and Mid-Latitudes

Abstract:

Ground magnetic field fluctuations (dB/dt) play a key role in the occurrence of geomagnetically induced currents. Since ground magnetometers measurements are readily available, sometimes continuously for decades, prediction of dB/dt has been a long-term goal of the scientific community as a proxy to assess the risk of GIC occurrence and intensity. Empirical, data-driven modelling, in particular machine learning algorithms can provide low computational cost and near instantaneous forecasting capabilities while obtaining similar performance to well-proven physical based models. In this work, we have trained a series of time-dependent neural networks over different ground magnetometer stations located at high and mid latitudes, and covering different MLTs to forecast the horizontal component of the ground magnetic field fluctuations (dB_H/dt) using only parameters from the solar wind that can be obtained in near-real time. The choice of stations has been initially determined following the guidelines of the 2008-2009 “ground magnetic field perturbations” GEM challenge, as it allows us to compare our results with already tested empirical and physics-based models using a set of established metrics.