Applications of machine learning for inner magnetospheric space weather prediction

Abstract:

The volume of space physics data has been rising exponentially over the past several years, and promises to accelerate its growth in the near future, far outpacing even Moore’s law. The projected data volumes of several upcoming solar and heliospheric projects are so large that the traditional method of analysis (i.e., downloading the data to a personal computer or small workstation and performing various calculations, extracting statistics or plotting) will very quickly become impractical.

At the same time, it is not clear that our physical understanding of the system has kept pace with the rapid growth of the data, or that our current methods of analysis exploit the capabilities of the data to its fullest potential. What future innovations are needed to move the field forward, deepening our understanding of the system and enhancing our ability to predict it?

In this talk we focus on a number of novel applications of machine learning in space weather specification and prediction. We discuss methods to reconstruct several spatiotemporal quantities in the near-Earth space environment, quantify uncertainties, track information flow in the system, and dramatically speed up physics-based simulations.