CAIG: AI-Based Reconstruction of Coronal Mass Ejection Structure and Evolution

This project will develop artificial intelligence (AI) methods to reconstruct the three-dimensional shapes of coronal mass ejections (CMEs), large eruptions of magnetized plasma from the Sun that can disturb near-Earth space. These solar storms can affect satellites, navigation, aviation, electric power systems, and astronaut safety, making improved understanding of their evolution important for both science and society. Current methods infer CME shapes from two-dimensional coronagraph images and often rely on simplifying assumptions that introduce poorly quantified uncertainty. The project will combine physics-based simulations, multi-view solar observations, and uncertainty-aware AI to produce more objective reconstructions of CME structure and evolution. The work will advance understanding of how CMEs expand, rotate, and deflect in the solar corona, while also creating open datasets, software, and educational materials that train students and researchers at the intersection of heliophysics and AI. The project aims to develop an uncertainty-aware artificial intelligence (AI) framework for reconstructing the three-dimensional morphology of coronal mass ejections (CMEs) from single- and multi-view coronagraph images. Because observed CMEs lack known three-dimensional ground truth, the project will generate a large synthetic training dataset using magnetohydrodynamic simulations of flux-rope CMEs in data-driven coronal backgrounds. Synthetic white-light coronagraph images, probabilistic silhouette masks, and known three-dimensional CME surfaces will be produced across diverse viewing geometries, eruption parameters, and solar-wind environments. The reconstruction model will extend signed-distance-function scene-representation methods to the heliophysics domain by enforcing consistency between a shared implicit three-dimensional CME surface and its projected silhouettes from one to three viewpoints. The framework will incorporate uncertainty from both silhouette segmentation and three-dimensional shape inference through probabilistic masks, learned signed-distance distributions, and model ensembles. The trained model will be validated against simulated ground truth and against real multi-spacecraft observations by comparing forward-projected reconstructions with independent coronagraph viewpoints. The framework will then be applied to approximately 94 Earth-directed CMEs observed by the Solar and Heliospheric Observatory (SOHO) and the Solar TErrestrial RElations Observatory (STEREO) satellites during 2009–2013 to quantify CME deflection, rotation, expansion, and deformation with event-specific confidence intervals. These measurements will be related to coronal magnetic-field structure, including heliospheric current sheet location and orientation, to test how the structured corona shapes CME evolution. The project will contribute new AI methodology for scientific reconstruction under sparse viewpoints and new insight into CME dynamics. 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.