Large-scale analysis of filaments' magnetic evolution from curated ground-based observations

The National Science Foundation's Division of Astronomical Sciences awarded $558,316 to the University of Missouri System (award date: September 1, 2025; completion date: August 31, 2028) under the Mathematical and Physical Sciences (CFDA 47.049) program to develop machine learning tools for large-scale analysis of solar filament magnetic evolution. The primary deliverables include: (1) robust algorithms using convolutional neural networks and variational autoencoders to identify corrupted observations within over one million hydrogen-alpha and magnetogram observations; (2) a reliable filament tracking system capable of processing millions of filament instances from curated ground-based data; and (3) open-source tracking software for use by the solar research community. These tools will enable researchers to recover subtle filament dynamics and significantly enhance catalogs of filament physical and magnetic properties, thereby advancing understanding of space weather events driven by filament eruptions and coronal mass ejections.

The project includes a planned sub-award to the New Jersey Institute of Technology (awarded March 15, 2026), where Dr. Qin Li will lead processing of Big Bear Solar Observatory data, oversee analysis and validation of project results, and direct investigation of filament physical and magnetic properties derived from curated observations from the Global Oscillation Network Group and Global High-resolution H-alpha Network facilities. This collaborative effort positions the award to deliver comprehensive solar observation datasets and analytical capabilities to the broader scientific community for advancing space weather research and prediction capabilities.

Abstract

Filaments are prominent structures in the sun?S chromosphere, consisting of cool, dense plasma suspended in the hot solar corona by magnetic fields. Understanding their formation, progression, and disappearance is essential, as their eruptions often lead to coronal mass ejections, which are key drivers of space weather. Currently, there are no operational filament-detection or filament-tracking algorithms that provide the community with large and up-to-date datasets. Using machine learning, this proposal will produce a large-scale system capable of accurately and reliably tracking filaments and will provide the solar community with potentially millions of instances of tracked filaments. The proposal will benefit society by increasing our understanding of space-weather events driven by filament eruptions. This project aims to build robust machine learning tools for identifying corrupt observations and for tracking solar filaments. The team has developed a model that they will modify and expand to detect corrupt incidences in over a million H-alpha and magnetogram observations. Convolution neural networks and variational autoencoders will be used for anomaly detection and the resulting algorithm will be used to train a reliable filament tracking system. Through the efforts of data cleaning and filament tracking, the team will aim to recover subtle filament dynamics from curated data and significantly enhance the catalog of filament physical and magnetic properties. The project will deliver an open-source tracking software for general use by the solar community. 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.