Changing environmental conditions in near-Earth space driven by solar activity can potentially disrupt our lives on Earth. Although the most visible solar events - auroras like the Northern Lights - are beautiful and harmless, not all the Sun’s activities are so benign.

One of the most concerning risks is the creation of geomagnetically induced currents (GICs) in power grids, which can cause blackouts and damage to electrical infrastructure. Solar storms, including coronal mass ejections (CMEs), send bursts of charged particles and a stronger, erratic magnetic field toward Earth which interact with Earth’s magnetic field and potentially create powerful electric currents in the ionosphere. These currents, in turn, induce GICs in the ground.

Large GICs could disrupt power grids across the UK, although northern areas like Scotland are more at risk as the Earth’s magnetic field is more easily disturbed in the north. Such events can cause significant economic and practical effects; the 1989 Quebec blackout left millions without electricity and similar scale events today have been estimated to have the potential to cause a $2-3 trillion global impact.

Given such potential impacts, improving space weather forecasting is crucial. Dr Maria Walach from Lancaster University is involved in research to determine whether a machine learning model can more accurately predict space weather.

Given the significant risks posed by space weather, improving the accuracy of forecasting capabilities is essential. The model, developed by researchers at Northumbria University, seeks to predict when rapid changes in the magnetic field go beyond the threshold that may trigger GICs by analysing solar wind parameters, including interplanetary magnetic field strength, solar wind speed, and particle density.

The model was shown to perform best during periods of intense magnetic disturbance, such as during geomagnetic storms driven by strong solar wind conditions. This was especially true when the interplanetary magnetic field (IMF) pointed southward, an alignment that allows for a more effective coupling with Earth’s magnetic field. During such active periods, the model successfully predicted when the magnetic field variability would exceed dangerous thresholds.

However, its accuracy was less when short, impulsive events occurred, such as sudden magnetic field changes caused by interplanetary shocks. It was also less effective during quiet times when the solar wind’s magnetic field orientation fluctuated close to zero. These conditions are harder for the model to interpret, leading to false alarms or missed predictions.

Dr Walach said: “It is crucial we understand what happens during geomagnetic storms. Whilst no two storms are ever truly the same, by consistently identifying and defining what a geomagnetic storm is, we can find common features in the data, and this has helped our understanding tremendously. What is really interesting about this study is that the ‘quiet’ intervals, for which we have the most training data, are some of the trickier ones to model.”

Machine learning models can be a powerful tool, but they depend upon the level and extent of data they receive during training. Future models could improve accuracy by including more detailed magnetospheric data, such as the intensity of the ring current, and the level of energy stored in the system.

Other performance improvements could be achieved by focusing the training phase on stormy periods, rather than gathering data constantly. Although consistent monitoring can help the model generalise more scenarios, it reduces the accuracy during rare but significant events whereas training only during stormy periods could improve performance during those critical moments when the power grid is at risk.

Dr Walach’s research into developing a better understanding of geomagnetic storms and predictive tools through machine learning is important in protecting UK infrastructure. As society becomes more reliant on interconnected power systems, early warning of potentially damaging space weather will allow power companies to reduce the system load or adjust transformer configurations to lessen the impact of GICs.

The full paper, published in Space Weather, can be found here