Abstract Details


Name: Priyansh Jaswal
Affiliation: Center of Excellence in Space Sciences India (CESSI), IISER Kolkata
Conference ID: TVS202510140
Title: Implementing Neural Networks to predict whether an Active Region will spawn a Coronal Mass Ejection in the near future or not
Authors and Co-Authors: Dibyendu Nandi
Abstract Type: Contributory Presentation
Abstract: Coronal mass ejections (CMEs) cause only a third of the geomagnetic storms at Earth. However, this subset contains more than 70% of the most severe ones. Therefore, early predictions of CMEs are necessary for mitigating the impacts of extreme space weather on our space- and ground-based technological infrastructure. Physical models have not yet been able to systematically predict the occurrence of CMEs. Bobra and Ilonodis 2016 is one of the very few studies that leverage the flare-CME connection to indirectly predict the occurrence of CMEs after a flaring event. Nevertheless, direct predictions of CME occurrence are yet unexplored. In this work, we implement convolutional neural networks over the direct (line-of-sight and vector) magnetogram images and predict whether an active region will give rise to a CME in the near future, or not. Our paradigm is first of its kind, achieving a reasonably high test accuracy for a timing window of 6 hours prior to the eruption. This significantly extends the forecasting window of a CME to instances before its birth. We anticipate our study to be the starting point for the development of more sophisticated models that would eventually harness the untapped potential of Machine learning techniques in predicting the occurrence of CMEs.