Homogenizing SOHO/EIT and SDO/AIA 171 Å Images: A Deep-learning Approach

Abstract Extreme-ultraviolet (EUV) images of the Sun are becoming an integral part space weather prediction tasks. However, having different surveys requires development instrument-specific algorithms. As alternative, it is possible to combine multiple create a homogeneous data set. In this study, we utilize temporal overlap Solar and Heliospheric Observatory Extreme ultraviolet Imaging Telescope Dynamics Atmospheric Assembly 171 Å train ensemble deep-learning models for creating single survey EUV two solar cycles. Prior applications deep learning have focused on validating homogeneity output while overlooking systematic estimation uncertainty. We use approach called “approximate Bayesian ensembling” generate whose uncertainty mimics that fully neural network at fraction cost. find goes down as training set size increases. Additionally, show model adds immense value by showing higher in test not well represented data.