Authors: Gergely Koban (University of Michigan), Lulu Zhao (University of Michigan) Fan Guo (Los Alamos National Laboratory)

Solar Energetic Particle (SEP) events occur when the Sun emits bursts of high-energy protons, electrons, and heavy ions, often accelerated to near-relativistic speeds. These events pose serious risks to spacecraft, astronauts, and technology, making it essential to understand the mechanisms behind SEP acceleration for improved space weather forecasting. Understanding why some shocks efficiently accelerate particles while others do not will offer new insights into the broader processes governing SEP events and space weather phenomena.

The acceleration of particles by shocks is an important mechanism in SEP production, as evidenced by the presence of Energetic Storm Particles (ESPs) near interplanetary (IP) shocks. However, not all shocks produce significant particle enhancements, and those that do exhibit diverse temporal profiles—ranging from short, spike-like increases to extended, gradual rises or irregular fluctuations. Previous studies have proposed different methods to classify ESP events based on various criteria, aiming to understand the relationship between shock-driven particle acceleration and the resulting flux profiles. However, despite these numerous attempts, the factors governing the variations in particle flux profiles remain a fundamental open question in heliophysics.

Given the diverse nature of ESP events and the complex interplay of acceleration mechanisms, we developed a comprehensive classification method to uncover patterns in particle flux profiles. We employ an unsupervised Machine Learning (ML) approach to classify ESP events based on their time intensity profiles. Specifically, we apply Dynamic Time Warping (DTW) to quantify the similarity between time series, which enables clustering through the k-means algorithm. Finally, a detailed statistical analysis is performed to associate shock properties like speed, shock angle and compression ratio with particle flux characteristics.

By integrating ML-based classification with statistical analysis, this study  uncovers the underlying physical mechanisms driving ESP variability.