Artificial Intelligence for the Characterization of the 2024 May Superstorm: Active Region Classification, Flare Forecasting, and Geomagnetic Storm Prediction

Abstract Space weather, driven by solar flares and coronal mass ejections, poses significant risks to technological systems. Accurately forecasting these events and their impact on Earth’s magnetosphere remains a challenge because of the complexity of solar–terrestrial interactions. This study focuses on the solar and geomagnetic extreme events associated with the 2024 May superstorm and shows that artificial intelligence (AI) tools are able to characterize this storm at three different levels. First, using magnetogram cutouts, a vision transformer was able to classify the morphological evolution of NOAA Active Region 13644 (primarily involved in storm generation), and then a video-based deep learning method predicted the occurrence of the associated solar flares, and a data-driven method exploited in situ measurements to raise 1-hour in advance alerts of the geomagnetic storm during the entire event. These AI models outperformed traditional methods in predicting solar flare occurrences, onset, and recovery phases of the geomagnetic storm. These findings highlight the impressive potential of AI for space weather forecasting as a tool to mitigate the impact of extreme solar events on critical infrastructure.