Forecasting the Peak Intensities of Energetic Storm Particle (ESP) Events using Machine Learning

Energetic Storm Particle (ESP) events are enhancements of energetic particles in association with the passage of an interplanetary (IP) coronal mass ejection (ICMEs). ESPs can produce significant increases in the near-Earth particulate radiation and pose severe hazards to astronauts and hardware in space. The primary candidate for producing these enhancements is diffusive shock acceleration (DSA). Physical parameters thought to affect ESP production include IP shock properties (e.g., speed, strength, obliquity) and upstream conditions ahead of the propagating shock (e.g., turbulence, seed populations, solar wind and IP magnetic field conditions). While several observational studies and theories have attempted to link ESP production to these drivers, reliable prediction of ESP properties (e.g. intensities, spectra, abundances), including their event-to-event variability, has so far proven elusive, indicating an incomplete understanding of how ICME-driven IP shocks accelerate ESPs. Goal and Science Questions: The goal of this project is to identify the dominant upstream and shock parameters that influence ESP intensities and their event-to-event variability, and determine how these parameters can be used to predict ESP intensities at 1 AU. This would result in advancing our current understanding of ICME-driven shock particle acceleration, provide constraints to shock particle acceleration and propagation models, and improve our capabilities of forecasting the physical properties of shock-accelerated particles. We will achieve this goal by answering the following two science questions: SQ1. How do upstream conditions and IP shock properties affect ESP intensities? SQ2. Can upstream conditions and IP shock properties be used to predict ESP peak intensities at 1 AU? Methodology: We use energetic H-Fe ion, plasma, magnetic field, radio emissions, and X-ray measurements from ACE, Wind, STEREO-A/B, and GOES during solar cycles 23 and 24. Using comprehensive selection criteria, we will identify all shocks and ESP events measured at 1 AU. For each ESP and when available, we will derive a matrix of parameters that characterizes the upstream conditions, IP shock, and ESP properties. Statistical and correlation studies will follow to identify the dominant drivers and related combinations that influence ESP properties (SQ1). Once the Upstream-Shock-ESP linkage is determined, we will utilize Machine Learning algorithms to determine if and how upstream and shock parameters can be used to predict ESP peak intensities (SQ2). To perform these tasks, we have a tightly integrated team that combines expertise in data analyses, modeling, statistical analysis, and machine learning. The relationships uncovered in these analyses are expected to lead to a more complete understanding of ICME-driven particle acceleration and forecasting at 1 AU.