On the Estimation of the SHARP Parameter MEANALP from AIA Images Using Deep Neural Networks

Abstract

Space-weather HMI Active Region Patches (SHARPs) data from the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO) provides high cadence data from the full-disk photospheric magnetic field. The SHARP’s MEANALP ($\alpha _{m}$) parameter, which characterizes the twist, can provide a measure of nonpotentiality of an active region, which can be a condition for the occurrence of solar flares. The SDO/Atmospheric Imaging Assembly (AIA) captures images at a higher cadence (12 or 24 seconds) than the SDO/HMI. Hence, if the $\alpha _{m}$ can be inferred from the AIA data, we can estimate the magnetic field evolution of an active region at a higher temporal cadence. Shortly before a flare occurs, we observed a change in the $\alpha _{m}$ in some active regions that produced stronger (M- or X-class) flares. Therefore, we study the ability of neural networks to estimate the $\alpha _{m}$ parameter from SDO/AIA images. We propose a classification and regression scheme to train deep neural networks using AIA filtergrams of active regions with the objective to estimate the $\alpha _{m}$ of active regions outside our training set. Our results show a classification accuracy greater than 85% within two classes to identify the range of the $\alpha _{m}$ parameter. We also attempt to understand the nature of the solar images using variational autoencoders. Thus, this study opens a promising new application of neural networks which can be extended to other SHARP parameters in the future.