Student researchers from Jesmond Park Academy, along with Orbyts Fellows Paloma Jol and Shivdev Turkay from Northumbria University, embarked on a project to predict solar flares, a crucial step in safeguarding our technology from the unpredictable nature of space weather. Using a machine learning approach with a random forest classifier, the teams analysed data from the Helioseismic and Magnetic Imager (HMI) on the Solar Dynamics Observatory (SDO), uncovering compelling patterns that link specific magnetic features of sunspots to the likelihood of a flare.

The research conducted on high-activity years (2013-2016) produced promising results, with the model achieving an impressive 89.81% accuracy for the year 2013. This success was largely driven by a clear correlation between flare occurrences and key magnetic attributes such as total helicity, which measures the entanglement of magnetic field lines, and total unsigned flux. The team’s analysis also confirmed that active regions tend to cluster in specific equatorial bands, suggesting a strong latitudinal dependency on the sun’s magnetic activity.

In a separate study focused on low-activity years (2017-2019), the students developed a model that achieved a notable overall accuracy of 94.8%. However, a more detailed examination revealed that the model had a bias towards predicting the majority class of no flare. This finding highlights a critical limitation that the model’s performance is hindered by the relative scarcity of flare events during these years. The combined data from low activity years did, however, reveal a clear correlation between the sum of currents in the magnetic field and the total unsigned flux near inversion lines.

Collectively, this research confirms that machine learning can be a powerful tool for forecasting solar flares. By identifying and correlating key magnetic features with flare production, the students have taken a step toward developing predictive models that can help mitigate the disruptive effects of space weather on Earth!