AI-Powered 3D CME Reconstruction Method Developed

Coronal Mass Ejections (CMEs) are large-scale plasma structures ejected from the Sun into interplanetary space, representing the largest energy release events in the solar system and serving as the primary driver of hazardous space weather. When CMEs reach Earth, they can disrupt ground-based and space-based infrastructure, leading to significant economic losses. Therefore, understanding the propagation of CMEs through the solar corona and interplanetary space, and analyzing their dynamic characteristics, is a crucial research topic in space physics and space weather forecasting. Recently, a research team led by Professor Shen Fang at the National Space Science Center of the Chinese Academy of Sciences has developed a novel method for automatic three-dimensional (3D) reconstruction of CMEs. The approach integrates dual-view coronagraph observations with machine learning techniques. By employing convolutional neural networks for feature extraction, principal component analysis, and the Otsu method for thresholding, the team accurately identifies CME signatures in two-dimensional coronagraph images from different perspectives. A target function is then constructed to quantify the similarity between the 2D projected images and the reconstructed 3D structure. The 3D reconstruction problem is reformulated as an optimization task, solved using a differential evolution algorithm to determine the optimal CME parameters. The method was applied to fit nearly 97 CME events, enabling the construction of a comprehensive CME dataset. Statistical analysis of both 2D and 3D parameters revealed that traditional 2D observations may introduce errors due to projection effects, whereas the new method significantly reduces such uncertainties. A key advantage of this approach is its automation—unlike conventional methods that require manual feature matching and parameter tuning, this technique performs full 3D reconstruction without human intervention. The reconstructed 3D structures closely match the observed coronagraph images, demonstrating high accuracy in capturing the true morphology of CMEs. Moreover, the method provides a fast and reliable way to derive initial conditions for interplanetary CME propagation simulations and improve predictions of CME arrival times at Earth. It also holds promise for future applications using data from solar orbiters and satellites positioned at Lagrange points such as L5, enabling multi-angle observations and enhancing overall forecasting accuracy. The findings have been published in The Astrophysical Journal Supplement Series. The research was supported by the National Natural Science Foundation of China and the National Key Research and Development Program.