Using AI to Predict the Danger of Solar Storms for Earth 

Using AI to Predict the Danger of Solar Storms for Earth  

Hannah Rüdisser of the Know-Center (Austria) and Ute Amerstorfer of the Space Research Institute (Austria) show how machine learning and artificial intelligence can help protect us from damage caused by solar storms.   

Read article in the fully formatted PDF of the Europlanet Magazine.

The Sun is a source of light and energy for life on our planet but, nonetheless, our nearest star can present some hazards to Earth’s inhabitants. When magnetic field lines at the visible surface of the Sun become tangled and break, large quantities of electrically charged particles can be released into space. If directed at Earth, these ‘solar storms’ can interact with Earth’s magnetic field and have serious effects on power grids, radio networks and satellites, that are key to our daily life. 

Solar activity fluctuates every eleven years between quiet and active phases, and we are currently in an active phase that is expected to reach its peak in 2025. In most cases, Earth’s magnetic field and atmosphere can protect us from solar storms, with effects limited to perhaps spectacular auroral light displays at the poles. However, a strong solar storm could cause around ten percent of all satellites to fail, and this could cause problems in areas where precise positioning is required, such as shipping and air traffic. Induced currents during geomagnetic storms can cause surges in transformer voltages and damage to undersea cables, leading to widespread electricity and Internet outages. A severe example in 1989 caused blackouts for nine hours across Quebec. In our digitally dependent world, a reliable space weather forecast is an increasingly vital resource. 

Several spacecraft have been launched to study the Sun and measure the solar wind, and these can transmit data back to Earth within minutes to give us advanced warning of incoming space weather. However, even if we know a solar storm is heading our way, to date, it has been very difficult to estimate the potential impact of the storm when it hits Earth. 

Machine Learning for Better Forecasts 

The use of Artificial Intelligence (AI) to analyse and classify planetary datasets is still relatively new, but is becoming increasingly important. Machine learning enables algorithms to be trained to analyse huge amounts of data and derive predictions and new solutions from them. Over the past decade, potential applications of machine learning in planetary science have exploded, but tools tailored to this area of research are still rare. 

The Europlanet 2024 Research Infrastructure (RI) Machine Learning activity aims to extract knowledge from the vast treasure-trove of data from space missions, simulations and laboratory experiments. By developing a series of machine learning tools that support planetary scientists in their work, Europlanet aims to promote a broader use of machine learning technologies in data-driven space research. At the Know-Center and the Space Research Institute in Graz, Austria, we have been working to deploy machine learning to help improve our predictions of the potential risk from solar storms to infrastructure. 

Artificial intelligence can improve weather forecasting in space. ©NASA/Johns Hopkins APL/Ben Smith
Artist’s impression of a solar storm. Credit: NASA/JHU-APL/B Smith 

Predicting the Strength of Solar Storms 

In extreme cases, solar storms can reach Earth in less than a day. The amount of energy transferred from the solar storm to Earth’s magnetic field depends largely on the relative orientation of the storm’s magnetic field, in particular its north-south ‘Bz magnetic field component’. The larger the southward Bz component, the greater the risk of a massive geomagnetic storm that could cause damage. Using machine learning, we have developed a tool to predict the Bz magnetic field component within the first few hours of an in-situ observation of a solar storm, to help give advance warning before it reaches Earth. The tool has been trained and tested with data collected by the Wind, STEREO-A and STEREO-B spacecraft for 348 different solar storms since 2007. To evaluate how well the prediction tool works in real time, we have simulated a continuous feed of measurements from spacecraft and monitored how the accuracy of the forecast improves as it receives new information. 

In results recently published in the journal, Space Weather, we have shown that our forecasting tool can predict the orientation of the magnetic field well, particularly when using data from the four hours that follow the first signs of the solar storm in the in-situ spacecraft data.

At present, we still need human eyes to manually spot when a solar storm is occurring, so the next step of our research project will be to use additional AI methods to help automatically detect solar storms in real-time. New space missions will provide even more data in the coming years, further increasing the accuracy of the predictions. However, our prediction tool already shows the great potential of machine learning to improve space weather forecasting and, in the event of a massive solar storm, to provide early warning to affected areas and prevent major damage. 

Machine Learning for Predicting the Bz Magnetic Field Component From Upstream in Situ Observations of Solar Coronal Mass Ejections‘, M. A. Reiss, C. Möstl, R. L. Bailey, H. T. Rüdisser, U. V. Amerstorfer, T. Amerstorfer, A. J. Weiss, J. Hinterreiter, A. Windisch. Space Weather, Volume 19, Issue 32.

Issue 3 of Europlanet Magazine