A Novel Reconstruction of the 3D Morphology of Solar Coronal Mass Ejections by Combining Multi-viewpoint Observations and Machine Learning
Abstract
Coronal Mass Ejections (CMEs) are colossal plasma bubbles filled with magnetic fields that are expelled from the Sun in all directions. They are capable of unleashing geomagnetic storms at Earth, which have recorded detrimental effects on various technological systems that are critical to both civilian and military activities. They can affect satellites, cause radio signal degradation, disrupt GPS signals, increase radiation exposure for flight crews, and degrade radar systems detection capabilities, among many other consequences. In addition, CMEs effects are fundamental for the planning of manned missions to the Moon and Mars. While our understanding of CMEs has improved in the last two decades, due to the success of dedicated ground- and space-based observatories, predicting their occurrence and impact remains a challenge. A key area of focus is the study of CME plasma morphology, as it closely relates to their magnetic field configuration and ability to interact with Earth#s magnetosphere. An obstacle in this endeavor is that the images available from specialized instruments (coronagraphs) in space have limited viewpoints and thus are unable to provide a 3D view ofCMEs. To overcome this, the Graduated Cylindrical Shell (GCS) model is widely used to reconstruct the CME outer shell morphology, by manually adjusting an ad hoc 3D hollow geometrical shape to two or three nearly simultaneous views of a CME recorded by coronagraphs from different vantage points. However, the adjusted GCS parameters are typically unreliable, because the procedure is highly dependent on the operator, spacecraft spatial configuration and the CME characteristics. In this three-year investigation, we propose tackling these limitations by using state-of-the-art Machine Learning (ML) to develop a method to automatically adjust the GCS model to the observed CME outer shell. This approach is motivated by our recent successful application of Deep Neural Networks (DNNs) to CME identification and segmentation using GCS-based synthetic images for supervised training. We propose developing, training and validating a DNN model formed by a deep convolutional backbone followed by a fully connected linear stack, that can infer the GCS parameters that best adjust the shape of the CME given in the input images. We will develop a dataset for supervised training based on synthetic CME brightness images, obtained using known GCS shapes, raytracing, and real coronagraphic background images, among other elements. Furthermore, we will employ our method to characterize the 3D evolution of a large number of events. We focus on three mainaspects: the statistical 3D description of the outer shell morphology of many individual CMEs, the morphological evolution of the outer shell in relation to the coronal magnetic environment for selected events presenting large deflections and rotations, and the 3D organization of the main internal and external constituents of selected, well-structured CMEs.The latter involves modifying and retraining the DNN to also infer the parameters of basic 3D shapes modeling the CME inner bright core and front shock. This novel and challenging project will provide valuable statistical insights into the 3D evolution of CMEs that can enhance our understanding of the physical processes involved, constraint numerical models and improve the prediction of their impact on technologies, Earth and other bodies.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 10, 2025
- Source ID
- N629092512017
Entities
People
- Eduardo F. Luna
Organizations
- Office of Naval Research
- United States Navy
- University of Mendoza