Machine learning used to predict hazardous solar flares

Machine learning used to predict hazardous solar flares

Support vector machine algorithm used to forecast M- and X-class solar flares

Physicists at Stanford University have applied a machine learning algorithm to better predict solar flares that could expose high-flying airline passengers to radiation and disrupt power grids and communication satellites.

Physicists Monica Bobra and Sebastien Couvidat thought about applying machine learning to the huge amount of data they had from NASA’s Solar Dynamics Observatory (SDO) satellite. The Stanford Solar Observatories Group processes and stores 1.5 terabytes of that data per day.

The SDO's Helioseismic Magnetic Imager was also used to collect vector magnetic field observations.

After taking an online machine learning course at Stanford, Bobra and Couvidat applied what they learnt to their study on solar flares to see if the support vector machine algorithm could provide early warning of the most hazardous types of solar flares: M-class and X-class.

M-class solar flares are medium-large flares that cause minor radiation storms that could endanger astronauts and cause short radio blackouts at Earth's poles. X-class are the largest flares and can cause a lot more damage.

Bobra and Couvidat characterised 25 features – such as energy, current and field gradient – of flaring and non-flaring regions, and identified which features are a predictor for solar flares.

It was found that the topology of the magnetic field and the energy stored in the magnetic field are very relevant to predicting solar flares.

Seventy per cent of the data was used to train the machine learning model and identify relevant features, with the remaining 30 per cent used to test its accuracy in its predictions.

The model was able to nail down active and non-active solar flare regions.

In addition to data from the solar surface, Bobra and Couvidat plan to incorporate data from the sun's atmosphere into the model.

It is not the first time researchers have machine learning algorithms to predict solar flares. However, according to Stanford, nobody else has done it with such a large set of data and with vector magnetic field observations.

“Machine learning provides a significant improvement because automated analysis is faster and could provide earlier warnings of solar flares,” a statement from the university said.

Bobra and Couvidat’s work comes after NASA’s SDO captured an X-class solar flare that caused radio blackouts on Earth in October 2014.

Bobra and Couvidat’s work is published in <i>The Astrophysical Journal</i>.

Follow Rebecca Merrett on Twitter: @Rebecca_Merrett

Follow CIO Australia on Twitter and Like us on Facebook… Twitter: @CIO_Australia, Facebook: CIO Australia, or take part in the CIO conversation on LinkedIn: CIO Australia

Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.

Join the newsletter!


Sign up to gain exclusive access to email subscriptions, event invitations, competitions, giveaways, and much more.

Membership is free, and your security and privacy remain protected. View our privacy policy before signing up.

Error: Please check your email address.

Tags physicsmachine learningsupport vector machineSolar Dynamics Observatorysolar flaresHelioseismic and Magnetic Imager

More about FacebookNASAStanford University

Show Comments