NJIT, Israeli Researchers Use AI and Computer Vision to Predict Space Weather
It's not your typical five-day forecast, as researchers with the Institute for Future Technologies, a joint venture of NJIT and Israel's Ben-Gurion University of the Negev, are improving solar weather prediction by adding data visualization techniques to their existing artificial intelligence methods.
The new method will allow for more accurate forecasts of the Sun's activities, such as sunspots, solar flares and coronal mass ejections, explained Ying Wu College of Computing Professor Jason T.L. Wang, who is collaborating with physics department Distinguished Professor Haimin Wang, along with BGU faculty Ohad Ben-Shahar and Jihad El-Sana.
Through their collaboration, the four hope to advance the field of space weather forecasting and analytics with a different, but complementary, approach that leverages the latest in artificial intelligence and computer vision technology.
Space weather refers to extreme solar events that occur on the surface of the Sun and their consequences in the near-Earth environment. These events can travel toward Earth and affect satellites, aviation and other Earth-based technology systems. The NJIT professors have long studied this subject through the NSF-funded EarthCube project. Their work focuses on using AI to analyze solar magnetic field parameters, in turn accurately predicting if an extreme space weather event will occur within 24 hours to a few days.
All extreme space weather events in our solar system begin on the Sun, they explained. Multiple agencies and even other NJIT researchers conduct extensive studies of the Sun and activities on its surface to better understand those events, with solar images serving as an essential component of those studies. According to NJIT's Jason Wang, a new set of magnetic measurements of the entire Sun — a high-resolution solar image — is taken approximately every twelve minutes from NASA’s Solar Dynamics Observatory.
So many solar images generate a wealth of visual information that Ben-Shahar and El-Sana, experts in computer vision and AI image processing, can analyze the visual information using pattern recognition techniques. By understanding how visual patterns shift and move over time, they can then develop a computer vision model to better understand, and predict, the Sun’s behavior. One pattern may warn of an upcoming solar flare, while another pattern predicts a coronal mass ejection.
The hope is that this new approach to forecasting space weather can be used in conjunction with the existing machine learning systems that use magnetic field parameters to make similar predictions.
“AI in general is a very powerful technique,” added Haimin Wang. “We are applying AI to understand and forecast the onset of solar flares in our EarthCube project.”
A geomagnetic storm that hit Earth on May 12, 2021, highlights the significance of this research. The storm was the result of a coronal mass ejection that left the Sun’s surface on May 9. When a storm of its size reaches Earth, power systems can experience voltage alarms and transformer damage while, above the Earth, low-Earth orbit satellites are at risk of component charging. Charging increases satellite drag, the atmospheric friction that acts against the motion of an object in space, resulting in orbital decay. Left unchecked, a satellite in orbital decay will ultimately fall back to Earth. GPS navigation and some low-frequency radio communications also experience storm-related interference. While these outcomes can go largely unnoticed by people on Earth, what did draw attention were the rare sightings of aurora borealis, better known as the Northern lights, that could be seen in Alaska, Minnesota, Canada and even across parts of Europe, and Southern lights (aurora australis) seen in New Zealand that the May 12 storm created.
“It remains to be explored how a computer vision model can work with a magnetic field parameter model,” said Jason Wang. “It is clear to us, however, that the two approaches complement each other and, when used together, could help researchers derive more meaningful insights about these solar events that will lead to more accurate and complete space weather predictions.”
