Beacon2Science: Enhancing STEREO/HI beacon data1 with machine learning for efficient CME tracking
To address the limited real-time coronal mass ejection (CME) warning accuracy caused by poor quality, low spatial resolution, and coarse temporal sampling of STEREO/HI beacon data, this paper proposes the first end-to-end deep learning enhancement framework. Methodologically, it jointly integrates image super-resolution, temporal interpolation, and inter-frame consistency modeling via a dedicated neural architecture and physics-informed loss functions—simultaneously improving signal-to-noise ratio, spatial resolution, and temporal resolution to approach science-grade data fidelity. Experiments demonstrate that the enhanced beacon data significantly improves CME structural visibility; trajectory tracking error decreases from 1.0° to 0.5°, achieving science-data-level accuracy. This work achieves, for the first time, high-fidelity reconstruction and temporally consistent dynamic representation of CMEs from low-quality beacon data, thereby providing robust observational support for operational space weather forecasting.