Estimation of solar flare loop length by machine learning

Abstract:

X-rays and extreme ultraviolet (EUV) emissions from solar flares rapidly change the physical composition of the Earth's thermosphere and ionosphere, thereby causing space weather phenomena such as communication failures. To predict the effects of flare emissions on the Earth's upper atmosphere and the occurrence of communication failures, numerous empirical and physical models have been developed which derive the flare spectral changes.

We have established a method to reproduce the flare loop emission using one-dimensional hydrodynamic calculation and the CHIANTI atomic database (Kawai et al., 2020). This method has successfully reproduced the time-integrated irradiance and time evolution of flare EUV lines. An important input parameter when using this method is the flare loop length. If the flare loop length can be estimated from the information of the Sun before flare, flare emissions can be estimated before the occurrence of flare by this method.

In this study, we attempted to estimate the flare ribbon distance related to the flare loop length from the observed images before flare using some machine learning techniques such as Convolutional Neural Network (CNN). We used active region images by Helioseismic and Magnetic Imager (HMI) onboard Solar Dynamics Observatory (SDO) and loop images before flare by Atmospheric Imaging Assembly (AIA) onboard SDO as input to machine learning method. We obtained the ribbon distance from the observations of SDO/AIA 1600 Å as teacher data.

we tried to construct a CNN model using all flare events greater than M class observed by SDO. The CNN model, which uses the active area image as input, can estimate the ribbon distance with error within 45% for majority events. This result indicates that it is possible to estimate the ribbon distance from the active region image with a certain level accuracy.

In this presentation, we report the accuracy and latest results of the models for estimating the flare loop length from solar multi-wavelength real-time observations.