Proxy-based Prediction of Solar Extreme Ultraviolet Emission Using Deep Learning

Abstract High-energy radiation from the Sun governs behavior of Earth’s upper atmosphere and such any planet-hosting star can drive long-term evolution a planetary atmosphere. However, much this is unobservable because absorption by interstellar medium. This motivates identification proxy that be readily observed ground. Here, we evaluate in near-infrared 1083 nm triplet line neutral orthohelium as for extreme ultraviolet (EUV) emission 30.4 He ii 17.1 Fe ix Sun. We apply deep learning to model nonlinear relationships, training validating on historical, contemporaneous images solar disk acquired i ground-based SOLIS observatory EUV NASA Solar Dynamics Observatory. The fully convolutional neural network incorporates spatial information accounts projection spherical 2d images. Using normalized target values, results indicate median pixelwise relative error 20% mean disk-integrated flux 7% held-out test set. Qualitatively, learns complex correlations between has predictive ability superior pixel-by-pixel model; it also distinguish active regions high-absorption filaments do not result emission.