CPAESS Discovery Seminar: Guided by Physics, Powered by AI: Physics-Informed Machine Learning in Solar Physics

Author:

alexmeyer

May 6, 2025

Please join CPAESS for a virtual seminar talk with Dr. Robert Jarolim, NASA Jack Eddy Postdoctoral Fellow

headshot of Robert Jarolim

Title: Guided by Physics, Powered by AI: Physics-Informed Machine Learning in Solar Physics

When: Wednesday, May 21, 2025
11:00 AM MT (Virtual)

Add this event to your google calendar

How To Watch

Please join us on UCAR's Live Broadcast site. Beneath the broadcast screen is a Slido link so that you can ask questions of the speaker.

View the May seminar flyer

About Dr. Robert Jarolim

Robert Jarolim is a NASA Jack Eddy Postdoctoral Fellow at the High Altitude Observatory in Boulder, Colorado. He completed his PhD in 2023 at the University of Graz in Austria and received the ESPD–Patricia Edwin Award, the PhD Prize of the International Astronomical Union, and the Josef Krainer Promotional Award for his thesis. With a background in physics and computer science, Robert has contributed to the field of AI in solar physics, particularly in the areas of automatic feature detection, solar image enhancement, and data-driven simulations. His research topics include coronal holes, solar flares, coronal mass ejections, and ground-based solar observations. His recent publications focus on the use of physics-informed neural networks for magnetic field simulations and tomographic reconstructions of the solar atmosphere.

Description

The limited ability to directly observe the 3D distribution of plasma and magnetic fields in the solar atmosphere remains a central challenge in solar and heliophysics. However, understanding the structure and evolution of the solar atmosphere is crucial for advancing our knowledge of heliospheric processes and their effects on Earth. Although the Sun is continuously observed, imaging data remain inherently difficult to interpret and cover only a fraction of the heliosphere.

In this talk, I will introduce Physics-Informed Neural Networks (PINNs) and discuss their transformative potential for data-driven simulations in solar physics. I will highlight two recent applications of physics-informed machine learning that contribute to a more complete picture of the solar atmosphere:

We apply PINNs for coronal magnetic field extrapolations of solar active regions, which are essential to understand the genesis and initiation of solar eruptions and to predict the occurrence of high-energy events from our Sun. With this approach, we achieve the first method that can perform realistic coronal magnetic field estimates in quasi real-time, enabling advanced space weather monitoring.
We present a method for 3D tomographic reconstruction of the plasma distribution in the solar corona based on multi-viewpoint imaging data. This approach provides unprecedented views of the solar poles and enables detailed reconstructions of coronal structures, filaments, coronal holes, and coronal mass ejections.

Finally, I will discuss how combining PINN-based methods can further enhance our modeling capabilities and maximize the use of current solar observations.