Jeongheon-Kim/2026SW_Solar Irradiance Prediction Model:Dataset

Paper #2026SW (Journal of Space Weather)

Title: "Real-Time Prediction of Solar Irradiance during the Post-Peak Decay Phase of Solar Flares Using Deep Learning and FISM2 Data"

Authors: Jeong-Heon Kim, Junmu Youn, Sung-Hong Park, Seungwoo Ahn, Young-Sil Kwak, Yong-Jae Moon, P.C. Chamberlin

Abstract: This study presents a deep learning model that predicts solar irradiance changes induced by solar flares. The model targets the soft X-ray band (0.1–0.8 nm) and three extreme ultraviolet (EUV) wavelength bands (9.0–9.9 nm, 13.0–13.9 nm, and 30.0–30.9 nm), and produces forecasts at one-minute cadence for three hours after the flare peak. The model was trained on 964 M-class or larger solar flare events from January 2003 to March 2023, using Geostationary Operational Environmental Satellite (GOES) X-ray flux as input and Flare Irradiance Spectral Model Version 2 (FISM2) reconstructed irradiance as the target. Each event contributes 60 minutes of pre-peak X-ray data as input and three hours of post-peak irradiance as output, and the dataset is divided into training, validation, and test sets in a 8:1:1 ratio. A multi-layer perceptron (MLP) architecture is adopted for its interpretability and computational efficiency. The model achieves a correlation coefficient of 0.92 in the X-ray band and 0.62, 0.82, and 0.47 in the three EUV bands, with mean relative error (MRE) values of −28.7%, −21.4%, −22.1%, and −18.1%, respectively, indicating an overall tendency toward underprediction. These results show that the proposed framework can provide a real-time forecast of post-peak solar irradiance across multiple wavelength bands and serve as a practical input source for ionospheric models, supporting space weather operations such as the prediction of flare-driven communication disruptions.

Data: 

This dataset provides the numerical data and figure files used to validate the deep-learning-based solar flare irradiance prediction model presented in this paper. The dataset includes the FISM2 flare-mode irradiance values used as the observational reference and the corresponding deep-learning prediction results for 94 solar flare events. The data are organized for four wavelength bands, labeled w0–w3, representing the X-ray band and three selected EUV wavelength ranges used in the study. For each event, the reference and predicted irradiance time series are provided for each wavelength band.

The dataset also includes CSV files containing event-by-event statistical evaluation results. The model performance was evaluated using the correlation coefficient, Root Mean Square Error (RMSE), and mean relative errors (MRE), calculated according to the equations described in the paper. In addition, the figure files generated from the reference and prediction data are included for each event and wavelength band. These files correspond to the validation plots and statistical analyses presented in the manuscript.