Browse Teach Curate

Deep-Learning Ensemble Prediction of Geoeffective Coronal Mass Ejections from SOHO Images

A transfer-learning ensemble (GeoCME) that predicts whether an Earth-directed coronal mass ejection will cause a geomagnetic storm, using SOHO LASCO C2, EIT, and MDI images. Students learn how ImageNet-pretrained convolutional backbones are combined per instrument and averaged into a deterministic or probabilistic geoeffectiveness forecast, and how weighted loss and skill scores handle imbalanced space-weather data.

B9 Text

Solar Physics DOI: 10.1007/•••••-•••-•••-••••-• Prediction of Geoeffective CMEs Using SOHO Images

GeoCME is a deep-learning framework that predicts whether a coronal mass ejection (CME) arriving at Earth will trigger a geomagnetic storm, defined as a minimum Dst index below -50 nT. It is trained …