Forecasting Solar Irradiance During Solar Flares via Deep Learning Trained on FISM2 Dataset

In this study, we present a deep learning-based approach to predict solar irradiance variations induced by solar flares—an essential factor in understanding ionospheric disturbances that impact Earth's upper atmosphere and space-based communication systems. Solar flares significantly enhance solar radiance, intensifying photo-ionization in the ionospheric D and E layers. However, current space weather monitoring systems are limited in their ability to provide timely forecasts of these irradiance changes. To address this gap, we developed a predictive model utilizing deep learning techniques to estimate solar irradiance in the X-ray to EUV wavelength range, focusing on a three-hour window following solar flare onset. The model is trained on a flare-event dataset reconstructed using the FISM2 framework, covering 964 M-class and stronger events from 2003 to the present. Input features consist of 90 one-minute resolution data points collected from 1.5 hours before the flare peak in the 0.1–0.8 nm X-ray range. The model predicts 180 one-minute data points covering the post-peak recovery phase, across four spectral bins including both X-ray and EUV bands. We adopted a simple multi-layer perceptron (MLP) model and split the dataset into training, validation, and testing sets at an 8:1:1 ratio. This work is part of the ongoing SpaceAI initiative led by the Korea Astronomy and Space Science Institute (KASI). In this presentation, we share our results, highlight key challenges encountered, and discuss future directions for enhancing the model’s predictive capability.