72-hour Forecasting of Electron Fluxes at Geostationary Orbit by Deep Learning

In this study, we forecast hourly >2 MeV electron fluxes at geostationary orbit for the next 72 hours using a deep learning model. The input data of the model are solar wind parameters (temperature, density and speed), IMF (|B| and Bz), geomagnetic indices (Kp and Dst), and electron fluxes themselves. All input data are hourly averaged ones for the preceding 72 consecutive hours. We use electron flux data from GOES-15 and -16, and perform cross-calibration to match the two data. Total period of the data is from 2011 January to 2021 March (GOES-15 data for 2011~2017 and GOES-16 data for 2018~2021). We divide the data into training set (January~August), validation set (September), and test set (October~December) to consider the solar cycle effect. Our main results are as follows. First, our model successfully predicts hourly electron fluxes for the next 72 hours. Second, root mean square error (RMSE) of our model is from 0.18 (for 1h prediction) to 0.68 (for 72h prediction), and prediction efficiency (PE) is from 0.97 to 0.53. Third, PE of daily averaged prediction shows higher score than previous studies: 0.92 for 1-day prediction, 0.74 for 2-day, and 0.61 for 3-day. Our study implies that the deep learning model can be applied to forecasting long-term sequential space weather events. This work was supported by Institute for Information & communications Technology Promotion (IITP) grant funded by the Korea government (MSIP) (2018-0-01422, Study on analysis and prediction technique of solar flares).