One of the most fundamental pursuits in the field of space science is to understand the dynamical connections between the sun’s driving forces and the response of Earth’s space environment. Classical reasoning holds that the sun’s plasma outflow and magnetic field characteristics (known as the Solar Wind) propagate all the way to the earth and interact with its
magnetic field, serving to directly shape and control the geospace environment, known as the magnetosphere. Conditions within Earth’s magnetosphere directly affect many aspects of human life, such as harmful radiation conditions in space, radio communication blackouts, satellite
operational failures, and disruption of electrical power grids, to name a few. Understanding and establishing predictability in the solar and magnetospheric conditions is therefore paramount to furthering our knowledge of space, as well as protecting civilization from the perils of space weather.
Research in space physics is typically conducted by first collecting data using space- and ground-based assets such as satellite missions and observatories, followed by analysis using a plethora of mathematical and statistical techniques. These data are eventually used to derive models of space phenomena such as solar storms and aurorae. However, because of the vastness of space, these data are fundamentally limited in that it is never possible to simultaneously measure the conditions across all regions that participate in phenomena of interest, in both space and time. As a result, scientists are always seeking novel analysis techniques to more fully
extract the content hidden within the available sparse data.
Machine learning and information theory present such possibilities, and are only now beginning to see use in the field of space science. In particular, techniques such as Transfer Entropy and Mutual Information present exciting possibilities to extract nonlinear correlation and causal dynamical connections between state variables in the coupled Solar
Wind-Magnetosphere system. We present an initial survey of recent results using these techniques, along with a proposed approach to robustly apply Transfer Entropy and Mutual Information to analysis of a number of observational data, with the aim of validating known results as a proof of concept, and subsequently for model discovery of phenomena which could
not have been captured using traditional analysis techniques employed in the field such as linear time lag correlation.

Author: Colin Wilkins