Imputation of Geomagnetic Disturbance Fields with Nonlinear Regression based on Synthetic Data

Geomagnetic variation, such as realized during magnetic storms, induces geoelectric fields in the Earth’s electrically conducting interior. These fields can, in turn, drive direct currents in high-voltage electric power systems, interfering with their operations. The sparse geospatial distribution of operational real time ground magnetometers presently leads to an under-constrained continuous distribution of geomagnetic variation needed for modern high-resolution geoelectric field models when using traditional interpolation techniques, or even more sophisticated inverse models (for example, a spherical elementary current system, or SECS) alone. To address this challenge, we generate a multivariate statistical model of SECS basis functions on a regular grid using state-of-the-art global magneto-hydrodynamics (MHD) simulations. These synthetic data resemble real observations in a statistical sense, although they do not generally reproduce the detailed time evolution of observations due to poorly constrained MHD boundary conditions. However, the statistical model can be used to regress on sparse observations in order to generate a statistically optimal SECS that is used to fill in, or “impute”, the unobserved points on a regular grid over North America in near real time.