Artificial intelligence to improve forecast for solar storms

When a solar storm hits the Earth, it can have serious consequences for communication and energy supply systems.

So far, however, it has hardly been possible to forecast such events. Researchers from Graz are developing an early warning programme to better predict the strength of solar storms. They recently published their results in the journal "Space Weather".

Solar storms are usually so weak that the Earth's atmosphere and magnetic field can sufficiently protect the planet, informed the Graz Know-Center. In the case of massive solar storms, however, the effects can be huge, ranging from severe voltages in the power grid, disrupted GPS and aviation radio communications to country-wide internet outages and widespread power outages. The better the prediction, however, the better the timely response and the lower the damage. According to the Graz data science expert, machine learning can provide a better forecast for the strength of solar storms.

Prediction tool important

In Graz, the data experts of the Know-Center are working together with researchers of the Institute of Space Research (IWF) of the Austrian Academy of Sciences to develop an appropriate prediction tool based on artificial intelligence. They are part of the EU project "Europlanet 2024 - Research Infrastructure", which has the goal of more strongly networking and advancing European research in the field of planetary sciences as a whole.

As the Graz authors explain, the Sun constantly sends radiation and charged particles into space - the solar wind. However, so-called coronal mass ejections, also known as solar storms in common parlance, sometimes cause the particle density to rise sharply.

The further south, the greater the danger of a massive geomagnetic storm

The ability of solar storms to cause extreme geomagnetic storms depends in turn largely on the orientation of their magnetic field. In technical language, one speaks of the "Bz magnetic field component". Here, the relative orientation of this component to the Earth's magnetic field determines how much energy is transferred to the Earth's magnetic field: The stronger the Bz component points to the south, the greater the danger of a massive geomagnetic storm. The problem: Until now, the Bz magnetic field component cannot be predicted with sufficient warning time before the solar storm arrives on Earth. Machine learning, however, allows researchers to train algorithms to analyse huge amounts of data and derive predictions and new solutions.

"It only takes a few minutes for data measured directly in the solar wind by probes to be transmitted to Earth. We first looked at whether information about the first hours of a solar storm is at all sufficient to predict its strength," explained Hannah Rüdisser from the Know-Center.

Huge amounts of data already available

"The European research network 'Europlanet 2024' hosts a huge treasure trove of data coming from space missions, simulations and laboratory experiments. Our goal is to bring out knowledge contained in this data and make it usable," says Rüdisser. To gain the prediction tool that can predict the Bz magnetic field component, the researchers have therefore "trained" and tested a programme with data from around 350 different solar storms since 2007.
To test the prediction tool in experimental real-time mode, the team simulates how solar storms are measured by space probes and evaluates how the continuous feed of new information improves the predictions. "Our prediction tool can predict the Bz component quite well. It works particularly well when we use data from the first four hours of the magnetic core of the solar storm," Rüdisser summarised the initial results.
It is hoped that new space missions will provide more data that will further increase the accuracy of the predictions. Finally, the researchers want to use AI methods to automatically detect solar storms in the solar wind. This is necessary to be able to apply the method in real time without a human user having to constantly identify the solar storms.



Machine Learning for Predicting the Bz Magnetic Field Component From Upstream in Situ Observations of Solar Coronal Mass Ejections", Space Weather,