Prediction of Extreme Ultraviolet Dynamic Spectrum during Large Flare using Convolutional Neural Network

To forecast the influences of large flares on the earth, it is important to predict the time series of Extreme Ultraviolet (EUV) spectrum during flares. This is because an enhancement of EUVs can increase the density of the upper atmosphere and it can crash satellites in the low earth orbit by atmospheric drag forces. Therefore, the development of methods to nowcast/forecast EUV dynamic spectra is required. We have developed two methods. The first one is to nowcast the EUV spectrum by using 1D hydrodynamic simulation and CHIANTI atomic database. We calculate the light curve of GOES/XRS-B when a flare occurs in the simulation. We convert it to observed one by stacking it with matching these peaks. The EUV dynamic spectrum can be calculated by converting it as same as XRS-B. The another one is to forecast the EUV spectrum by using a Convolutional Neural Network (CNN) and 1D hydrodynamic simulation. We train a CNN to predict GOES/XRS-B light curves during flares as a response to the input of the past EUV maps and magnetograms observed by SDO/AIA. After that, we adjust the flare parameters in the simulation such as heating rate and duration to reconstruct predicted GOES/XRS-B light curves well. We applied these methods to M- and X-class flares occurred between 8:00 - 14:00 UT on 6 September 2017. We evaluate the accuracy of the prediction by comparing observed and predicted GOES/XRS-A light curves and total irradiance of SDO/AIA EUV maps.