A dataset of manually annotated filaments from H-alpha observations

Area of Science:

• Solar Physics
• Astronomy
• Machine Learning

Background:

• Solar filaments are key structures in the Sun's atmosphere, crucial for understanding magnetic activity.
• Previous datasets lacked the scale necessary for comprehensive analysis and advanced modeling.
• Accurate annotation of filament properties, including magnetic field chirality, is essential.

Purpose of the Study:

• To introduce the Manually Annotated GONG Filaments in H-alpha Observations (MAGFiLO v1.0) dataset.
• To provide a large-scale, high-quality dataset for solar filament research.
• To enable advancements in deep learning models for solar physics.

Main Methods:

• Collected 10,244 filament annotations from 1,593 Global Oscillation Network Group (GONG) observations (2011-2022).
• Detailed annotations include segmentation, bounding box, spine, and magnetic field chirality.
• Ensured data quality through over 1,000 person-hours of annotation and a double-blind review process, achieving a Kappa score of 0.66.

Main Results:

• MAGFiLO v1.0 is the largest dataset of its kind, facilitating unprecedented precision in filament feature identification.
• Verified alignment of hemispheric filament preference with previous smaller-scale studies.
• Demonstrated the dataset's utility as a testbed for large-scale solar filament analysis.

Conclusions:

• The MAGFiLO dataset significantly advances the potential for deep learning applications in solar physics.
• Provides a robust resource for solar physicists studying large-scale filament behavior and magnetic field properties.
• Establishes a new benchmark for the scale and quality of annotated solar filament data.