Illustration of the solar wind interacting with Earth's magnetic field.
Credit: NASA's Goddard Space Flight Center
Credit: NASA's Goddard Space Flight Center
The Solar Heliospheric Observatory (SOHO) took this coronagraphic image of a coronal mass ejection on 20 April 1998.
Credit: SOHO (ESA & NASA)
Credit: SOHO (ESA & NASA)
What happens when pileups of solar wind plasma collide with Earth’s protective magnetosphere? New work uses machine learning to examine how strongly these events affect our planet’s magnetic field.
Plasma Pileups
Geomagnetic storms driven by solar activity paint night skies with glowing aurorae, but they also threaten spacecraft electronics with showers of high-energy particles. While immense eruptions of solar plasma and magnetic fields called coronal mass ejections are the most infamous example of solar activity, a team led by Yudong Ye (Sun Yat-Sen University) recently focused on another, less destructive form of activity: stream interaction regions.
Stream interaction regions arise when slow-moving solar wind is struck from behind by faster-moving solar wind emitted later. The collision of the two solar wind streams creates a tangle of compressed plasma and strong magnetic fields capable of peeling back Earth’s protective magnetosphere and dumping in high-energy charged particles, with beautiful yet harmful results.
Stream interaction regions arise when slow-moving solar wind is struck from behind by faster-moving solar wind emitted later. The collision of the two solar wind streams creates a tangle of compressed plasma and strong magnetic fields capable of peeling back Earth’s protective magnetosphere and dumping in high-energy charged particles, with beautiful yet harmful results.
Illustration of the authors’ support vector machine framework. The optimal hyperplane is the boundary that best divides the data by maximizing the distance between the boundary and the data points nearest to it; these points are called support vectors. Square and triangle symbols represent two classes of data. Click to enlarge. Credit: Ye et al. 2025
Machine Learning Method
Though stream interaction regions are less disruptive than coronal mass ejections, they’re far more common; they frequently needle Earth’s magnetosphere, especially during the calmer years of the Sun’s activity cycle. Predicting how strongly a stream interaction region will influence Earth’s magnetosphere — in other words, how geoeffective it is — is challenging, however. When two streams of solar wind collide, their properties combine in complex and nonlinear ways that traditional statistical investigations have struggled to pin down.
Now, Ye and collaborators have used machine learning to study the properties and impact of stream interaction regions in a physically meaningful way. They performed their study on a sample of 879 stream interaction events for which there is abundant information, such as temperature, magnetic field strength and direction, and solar wind conditions before and after the event.
Ye’s team based their framework on a support vector machine classifier: a classical machine learning algorithm that draws a mathematical boundary between groups of data while maximizing the distance between the boundary and the data points nearest to the dividing line. The support vector machine algorithm is well-suited to the task of modeling the geoeffectiveness of stream interaction regions because it doesn’t require a particularly vast dataset, can tolerate misclassified events, and allows for a physical interpretation of the results.
Now, Ye and collaborators have used machine learning to study the properties and impact of stream interaction regions in a physically meaningful way. They performed their study on a sample of 879 stream interaction events for which there is abundant information, such as temperature, magnetic field strength and direction, and solar wind conditions before and after the event.
Ye’s team based their framework on a support vector machine classifier: a classical machine learning algorithm that draws a mathematical boundary between groups of data while maximizing the distance between the boundary and the data points nearest to the dividing line. The support vector machine algorithm is well-suited to the task of modeling the geoeffectiveness of stream interaction regions because it doesn’t require a particularly vast dataset, can tolerate misclassified events, and allows for a physical interpretation of the results.
Illustration of how the interplanetary magnetic field (IMF) interacts with Earth’s magnetosphere. When the IMF points southward, as it does in this diagram, the impact on Earth’s magnetosphere is increased, with magnetic reconnection occurring in the red areas. Credit: NASA
A Physical Interpretation
The team first reined in the model’s complexity by identifying the most important features in the dataset. They then determined which features or combination of features had the largest contribution to the output — in other words, which physical parameters most strongly determined the geoeffectiveness of the event.
Ye and collaborators found that the strongest determinants of an event’s geoeffectiveness were how long the solar wind was directed southward, the strength of the solar wind electric field, and the average and minimum strengths of the southward-pointing solar wind magnetic field. These results align with the current understanding of how energy is transferred from the solar wind to Earth’s magnetosphere through magnetic reconnection, a release of magnetic energy driven by rearrangement of magnetic fields. This shows how classical machine learning methods can enhance our ability to predict the outcome of oncoming space weather while simultaneously examining the physical drivers of the event.
Ye and collaborators found that the strongest determinants of an event’s geoeffectiveness were how long the solar wind was directed southward, the strength of the solar wind electric field, and the average and minimum strengths of the southward-pointing solar wind magnetic field. These results align with the current understanding of how energy is transferred from the solar wind to Earth’s magnetosphere through magnetic reconnection, a release of magnetic energy driven by rearrangement of magnetic fields. This shows how classical machine learning methods can enhance our ability to predict the outcome of oncoming space weather while simultaneously examining the physical drivers of the event.
Citation
“Assessing the Geoeffectiveness of Stream Interaction Regions Through Physically Interpretable Machine Learning,” Yudong Ye et al 2025 ApJ 993 10. doi:10.3847/1538-4357/ae0454
