Artificial intelligence predicts solar activity nearly nine hours before it appears

12:40

By: Azzat Manal

Artificial intelligence predicts solar activity nearly nine hours before it appears

Researchers at the New Jersey Institute of Technology have developed a new artificial intelligence model capable of detecting the early signs of emerging active regions on the Sun, potentially providing a warning several hours before these regions become visible.

Known as EarlyDetect, the system was developed by a research team working with several U.S. universities and scientific institutions. The study, published in the Journal of Geophysical Research: Machine Learning and Computation, used observations collected by NASA’s Solar Dynamics Observatory to train the model to recognize subtle changes associated with developing solar activity.

Active regions are areas of intense magnetic activity on the Sun where sunspots can form and powerful solar flares may originate. Detecting these regions at an early stage could improve scientists’ ability to anticipate periods of increased solar activity and assess potential space-weather risks.

Rather than relying solely on visible changes on the solar surface, EarlyDetect analyzes acoustic waves traveling through the Sun. Researchers found that emerging magnetic fields can leave extremely subtle signatures in these waves, allowing artificial intelligence to identify patterns that are difficult to detect through conventional observation alone.

The model uses a Transformer-based architecture, a type of artificial intelligence technology that has also become widely known through large language models. In this case, however, the system was trained to interpret solar observations rather than human language.

During development, researchers encountered an unexpected problem involving a noise-reduction technique designed to remove irrelevant signals from the data. Instead of improving the system, the filtering process sometimes eliminated the very weak signals that contained valuable information about future solar activity.

After modifying the approach, the researchers trained EarlyDetect on real solar observations and tested it against previously unseen active regions. The model was able to predict the emergence of active regions approximately 9.24 hours in advance on average, outperforming earlier approaches.

The researchers believe the technology could eventually contribute to more effective space-weather forecasting. Strong solar activity can affect satellites, radio communications, navigation systems and electrical infrastructure on Earth, making early warnings potentially valuable for operators of critical systems.

However, EarlyDetect is not yet a fully operational forecasting system. The model can still generate false alarms or delayed predictions, and identifying an emerging active region does not necessarily mean that a solar flare will occur.

The research team is therefore continuing to refine the technology and improve its reliability. To support further scientific work, the researchers have also made the SolARED dataset available and developed an interactive platform known as the SAR Portal, allowing researchers in artificial intelligence and solar physics to explore the observations and test new forecasting methods.

The development of EarlyDetect illustrates how artificial intelligence is increasingly being combined with space science to identify patterns that may be difficult to detect through traditional methods. If its accuracy can be further improved, such systems could become an important component of future efforts to monitor and forecast solar activity.