Deep Learning Based Solar Flare Forecasting Model. II. Influence of Image Resolution

Abstract Due to the accumulation of solar observational data and development data-driven algorithms, deep learning methods are widely applied build a flare forecasting model. Most works focus on how design or select proper networks for task. Nevertheless, influence image resolution based model has not been analyzed discussed. In this Paper, we investigate magnetograms accuracy forecasting. We study active regions by Solar Dynamics Observatory/Helioseismic Magnetic Imager (SDO/HMI) from 2010 2019. Then, downsample them get database containing with several resolutions. Afterwards, three neural (i) AlexNet, (ii) ResNet-18, (iii) SqueezeNet implemented evaluate performance compared different resolutions magnetogram. experiments, first did comparative experiments our own simulated HMI Then conducted two selected actual overlapping databases, Hinode–HMI Michelson Doppler Imager–HMI, reconfirm conclusions. The experiment results show that all insensitive certain extent. visualized interest network an interpretable perspective found pays more attention global features extracted sensitive local information in magnetograms.