SOLACE: A Contrastive Learning Framework for Predicting Solar Proton Events and Geomagnetic Storms from Flare–Coronal Mass Ejection Events
- 1Chinese Academy of Sciences
- 2University of Chinese Academy of Sciences
Abstract
Intense solar eruptions can be followed by solar proton events (SPEs) and geomagnetic storms (GMSs), posing potential risks to spacecraft operations, communication and navigation systems, and power grids. Existing studies mainly forecast a single disturbance type, with limited attention to subsequent disturbances associated with an individual solar eruption. In this study, we construct a space weather event database in which each solar eruption is treated as the basic sample unit. The database records observational features of flares and coronal mass ejections (CMEs) and labels whether SPEs and GMS subsequently occurred. Based on this database, we propose Solar Contrastive Embedding (SOLACE) with Attentional Factorization Machines, a representation learning framework that integrates contrastive learning and feature interaction modeling to support historical-event retrieval and probabilistic prediction. Experimental results show that SOLACE performs well in probabilistic prediction of SPEs and GMS. For SPE prediction, the mean F1 score, false-alarm ratio (FAR), and area under the receiver operating characteristic curve (AUC) are 0.734 ± 0.065, 0.287 ± 0.077, and 0.951 ± 0.021, respectively. For GMS prediction, the corresponding values are 0.733 ± 0.036, 0.311 ± 0.062, and 0.924 ± 0.014. Interpretation analyses show that SPE prediction mainly relies on CME speed, CME angular width, heliolongitude, and flare soft X-ray fluence, whereas GMS prediction emphasizes CME angular width and source-region location, with propagation characteristics and interactions involving angular width also contributing. By learning physically meaningful representations of solar eruptive events, SOLACE provides a unified framework for the early prediction of subsequent space weather disturbances.
Related
- Understanding the Complex Morphology of a Coronal Mass Ejection. III: How to Create a Severe Geomagnetic Storm
- Contrastive Astromer: Two-Stage Contrastive Representation Learning for Astronomical Light Curve Classification
- Lyman Break Galaxy selection and redshift measurement with supervised contrastive learning
- Multi-Scale Contrastive Attention for Light-Curve Representation Learning
- Impact of geomagnetic storm intensity on thermospheric density and orbital decay of LEO satellites: a G1–G5 per-event analysis
BibTeX
@article{yan2026solace,
title = {{SOLACE: A Contrastive Learning Framework for Predicting Solar Proton Events and Geomagnetic Storms from Flare–Coronal Mass Ejection Events}},
author = {Shuainan Yan and Yenan Zou and Liqin Shi and Yanmei Cui and Bingxian Luo and Enxv Dong and Yanxia Cai},
journal = {The Astrophysical Journal Supplement Series},
year = {2026},
month = nov,
doi = {10.3847/1538-4365/aeb2db},
url = {https://iopscience.iop.org/article/10.3847/1538-4365/aeb2db},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - SOLACE: A Contrastive Learning Framework for Predicting Solar Proton Events and Geomagnetic Storms from Flare–Coronal Mass Ejection Events
AU - Shuainan Yan
AU - Yenan Zou
AU - Liqin Shi
AU - Yanmei Cui
AU - Bingxian Luo
AU - Enxv Dong
AU - Yanxia Cai
PY - 2026
DA - 2026/11/06
JO - The Astrophysical Journal Supplement Series
DO - 10.3847/1538-4365/aeb2db
UR - https://iopscience.iop.org/article/10.3847/1538-4365/aeb2db
AB - Intense solar eruptions can be followed by solar proton events (SPEs) and geomagnetic storms (GMSs), posing potential risks to spacecraft operations, communication and navigation systems, and power grids. Existing studies mainly forecast a single disturbance type, with limited attention to subsequent disturbances associated with an individual solar eruption. In this study, we construct a space weather event database in which each solar eruption is treated as the basic sample unit. The database records observational features of flares and coronal mass ejections (CMEs) and labels whether SPEs and GMS subsequently occurred. Based on this database, we propose Solar Contrastive Embedding (SOLACE) with Attentional Factorization Machines, a representation learning framework that integrates contrastive learning and feature interaction modeling to support historical-event retrieval and probabilistic prediction. Experimental results show that SOLACE performs well in probabilistic prediction of SPEs and GMS. For SPE prediction, the mean F1 score, false-alarm ratio (FAR), and area under the receiver operating characteristic curve (AUC) are 0.734 ± 0.065, 0.287 ± 0.077, and 0.951 ± 0.021, respectively. For GMS prediction, the corresponding values are 0.733 ± 0.036, 0.311 ± 0.062, and 0.924 ± 0.014. Interpretation analyses show that SPE prediction mainly relies on CME speed, CME angular width, heliolongitude, and flare soft X-ray fluence, whereas GMS prediction emphasizes CME angular width and source-region location, with propagation characteristics and interactions involving angular width also contributing. By learning physically meaningful representations of solar eruptive events, SOLACE provides a unified framework for the early prediction of subsequent space weather disturbances.
SN - 0067-0049
SN - 1538-4365
ER -