Astronomy and Astrophysics Seminars

Prediction of Solar Flares Using Machine Learning

by Mr. Dattaraj Dhuri (DAA - TIFR)

Monday, March 27, 2017 from to (Asia/Kolkata)
at DAA SEMINAR ROOM ( A269 )
TIFR
Description
Sun, our closest star, displays a range of dynamic and occasionally violent magnetic phenomena. Elucidating the mechanism that governs generation of solar magnetism and emergence of magnetically active regions on the surface of Sun is an outstanding problem in solar physics. Solar flares are eruptions on the surface of Sun caused by the rapid restructuring of magnetic field lines in active regions. The radiation and charged particles released in the process pose a threat to space and ground based communication instruments. Prediction of these flares is therefore an important problem in the field. Helioseismic and Magnetic Imager (onboard NASA's Solar Dynamic Observatory) makes available high resolution solar vector-magnetic-field data with 12 minutes cadence. Advanced data analysis techniques like machine learning are required to efficiently utilise this data. We use active region parameters derived from solar vector-magnetic-field data to train a machine learning algorithm (Support Vector Machines) to forecast solar flares in the following 24 hrs with accuracy greater than 85%. We also study variation of accuracy for predicting flares in the following 6 hrs to 72 hrs and shed light on underlying physics responsible for triggering solar flares.