Machine learning for Heliophysics

Our sun is incredibly active, and some of that activity can have tremendous consequences here on Earth! Solar flares are explosive outbursts of electromagnetic energy which occur on the sun, which can have damaging consequences on Earth, including harming satellites and resulting breaking the power grid. Extreme flares can release 10^15 joules of energy – this is the equivalent of 2 billion megatons of TNT! The precise causative mechanism underlying solar flares is not well understood.

Thus predicting when solar flares will occur and whether the flares will have an impact on earth – whether they will be geoeffectve – is important both for scientific and terrestrial applications.

Here we use techniques from machine learning, signal processing, and computer vision to predict the occurrence of solar flares based upon data generated by NASA's Solar Dynamics Observatory (SDO) . Working with Monica Bobra and her group at the Hansen Experimental Physics Laboratory at Stanford, we developed feature extraction methods and machine learning techniques to predict these flares from a hybrid of image and historical data.

Movies

Here We show hyperspectral data across AIA three channels and magnetogram data from HMI on SDO over the course of an active region moving across the sun.

Avatar
Eric Jonas
Assistant Professor of Computer Science

My research focuses on the application of machine learning to inverse problems and scientific measurement.