New Approaches to Solar Flare Forecasts

This poster presented at the European Space Weather Week 2023, Toulouse, discuss :

• a deep learning model able to forecasts flare at regional levels while only using full-disk labels during training, thus bypassing the misslables issues encountered for flare forecasting at Active Region levels
• general limitations of flare forecasting models, mainly their inability to foreccast changes of ativity and underperformances compared to persistent models, along with tailored evalation approaches to highlight such limitations
• new labels based on SXR-fluence to better characterize flare activity in large time - windows and provide valuable information on radiation exposure

Related papers:

Limits of solar flare forecasting models and new deep learning approach