Speaker
Description
Solar flares are primary drivers of space weather and play a crucial role in the Sun-Earth connection. The physical mechanisms underlying solar flare initiation remain a topic of intense research. It is widely accepted that flares result from the rapid release of magnetic energy stored in the stressed configurations of ARs. Several competing, and possibly concurrent, mechanisms have been proposed to explain this energy release, including magnetic reconnection, flux emergence, and shear-driven instabilities. Each of these processes contributes in a distinct way to the destabilization of the magnetic configuration, ultimately triggering the flare.
A central challenge in solar flare forecasting is the identification of reliable precursors. Over the past decades, feature-based machine learning approaches have been explored to tackle this task, relying on physically meaningful parameters extracted from solar magnetograms, particularly from instruments like the Helioseismic and Magnetic Imager onboard Solar Dynamics Observatory.
In this work, we propose a physics-informed extension to traditional feature-based machine learning. Our method constructs non-linear combinations of features guided by physical laws and dimensional analysis. This not only improves the machine learning model interpretability but also allows us to discern how many and which mechanisms govern flare initiation, through sparsity-enhancing and feature ranking methods.
We highlight the significance of the product of magnetic flux and electric current, a quantity related to magnetic helicity, as a powerful physics-informed feature, which is a well-investigated candidate for representing a significant portion of the energy budget stored in the active regions. Our findings suggest that this physics-informed strategy holds promise for uncovering novel descriptors of energy distribution in ARs, potentially improving the identification of flare precursors.
| Do you plan to attend in-person or online? | In-person |
|---|
