SHINE: Multimodal Machine Learning Approaches for Solar Energetic Particles Events Prediction and Posthoc Analysis

Solar Energetic Particle (SEP) events are bursts of high-energy particles from the Sun that can disrupt satellites, navigation systems, and human spaceflight. Predicting these events remains challenging because they are rare and driven by complex solar activity. This project will apply machine-learning methods to space-based observations spanning two solar cycles to improve identification of patterns that precede SEP events. By strengthening space weather forecasting, the research will enhance protection of systems that support national security, economic activity, and space exploration. Findings will also be incorporated into data science coursework to train students in advanced analytical approaches for space weather applications. The project will develop a comprehensive multimodal dataset by integrating data from space-based observatories, including the Solar Dynamics Observatory (SDO), Solar and Heliospheric Observatory (SOHO), Wind, Geostationary Operational Environmental Satellites (GOES), and Advanced Composition Explorer (ACE) (Thrust 1). This dataset will consolidate magnetograms, extreme ultraviolet images, radio observations, in situ particle flux measurements, and solar wind data. The structured dataset will facilitate detailed investigations into the solar conditions that precede SEP events and improve the identification of precursor conditions and event timing. Building on this dataset, the project will develop an advanced ensemble model that combines global and local learning strategies for SEP prediction (Thrust 2). The global model will learn from the comprehensive dataset gathered in Thrust 1, capturing complex interactions and complementarities among various data sources. In parallel, local models will analyze specific data modalities to extract unique patterns, providing more granular insights into SEP events. This dual strategy ensures that the ensemble framework leverages both the strength of integrated data and the specialized knowledge offered by individual observational sources. Finally, the project will apply unsupervised learning techniques to analyze SEP events post-occurrence to better understand their acceleration mechanisms and propagation characteristics (Thrust 3). Clustering algorithms will group SEP events based on charge states, decay times, and interplanetary conditions, helping to distinguish between different acceleration mechanisms (e.g., solar flares vs. coronal mass ejections) and identify patterns in SEP transport through the heliosphere. The final operational model will be deployed at the Community Coordinated Modeling Center (CCMC), supported by NSF and NASA, and made publicly available through open-access repositories. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.