Speaker
Description
Timely and accurate solar flare forecasting is vital for minimizing the adverse effects of space weather on Earth and in space environments. We present a deep learning framework that integrates multi-modal solar observations—line-of-sight (LoS) magnetograms, continuum intensity images, and EUV observations (171Å, 193Å, and 304Å from SDO/AIA)—alongside physical parameters derived from SHARP vector magnetograms, specifically total unsigned magnetic flux and current helicity. The flare sample consists of several hundred events (GOES class C5.0 and above) spanning the period from 2010 to 2019, while the flare-quiet sequences are drawn from the broader 2010–2024 interval. For each event, the model receives a 36-minute sequence of observations. In the case of flaring events, this sequence begins two hours prior to the flare peak, allowing the model to learn pre-eruptive conditions in active regions. A deep convolutional autoencoder is used to extract spatial features from the multi-channel inputs, which are then passed to a recurrent deep learning model (LSTM) to capture temporal dependencies. Results show that combining imaging data with contextual physical parameters substantially enhances predictive performance. These findings demonstrate the potential of this approach for future integration into operational solar flare forecasting frameworks.
| Do you plan to attend in-person or online? | In-person |
|---|
