Suvadip Sinha, "A Comparative Analysis of Machine-learning Models to Identify Active Regions Associated with Intense Solar Flares"

Wednesday 26 Oct 2022, 13:30 → 14:30 Europe/Stockholm

Hybrid talk: Mega (6228, Hus 3, Albano) + https://stockholmuniversity.zoom.us/j/530682073

Solar flares release a large amount of energy over a short period and can impact us by interfering with radio communication, damaging sensitive electronic equipment and sensors on satellites, and risking astronaut’s health with high doses of electromagnetic radiation. Hence predicting solar flares is a demanding problem of space weather forecasting, but the challenges are accentuated by the lack of any unique flare trigger or a single physical pathway. Studies indicate that multiple physical properties contribute to active region flare potential, compounding the challenge. Recent developments in machine learning (ML) have enabled the analysis of higher-dimensional data leading to increasingly better flare forecasting techniques. In this study, we use four popular machine learning algorithms (k nearest neighbors, logistic regression, random forest classifier, and support vector machine) to identify the active regions capable of producing intense flares by looking at the magnetic parameters of associated AR obtained from the Helioseismic and Magnetic Imager on board the Solar Dynamics Observatory. Comparing the performance of different algorithms, we have demonstrated that both the Logistic regression (LR) and Support vector machine perform very well, with a high True Skill Score of 0.967 ± 0.018 obtained with LR. Our analysis also constrains magnetic parameters that are crucial in deciding the flare potential of an active region. We have shown that current helicity, current density, total unsigned flux, flux near the polarity inversion line, and total absolute twist are the top-ranked magnetic parameters contributing significantly to identifying flaring active regions.

Sinha_2022-10-26.mp4