Prediction Models of ≥2 MeV Electron Daily Fluences for the Following 3 Days at GEO Orbit Based on Long Short-term Memory (LSTM) Model

The data used in this work include ≥2 MeV electron fluxes, ≥10 MeV proton fluxes and magnetic field data from GOES-08, GOES-10, and GOES-11 satellites, solar wind parameters, geomagnetic disturbance indices, magnetopause subsolar distances from 1995 to 2010. GOES satellites are equipped with the Energetic Particle Sensor (EPS) and Magnetometer (MAG), which can provide ≥2 MeV electron fluxes, ≥10 MeV proton fluxes and vector magnetic fields. The ≥2 MeV electron fluxes and the total magnitude of magnetic field (Ht) are with a 1-min resolution, and ≥10 MeV proton fluxes are with a 5-min resolution. The data of GOES satellites are available from the National Geophysical Data Center (NGDC) website. Solar wind parameters include speed (V), density (N), dynamic pressure (Pd), the total magnitude of interplanetary magnetic field (Bt), Bx, By, Bz components of interplanetary magnetic field (IMF) in the GSM coordinates, electric field (E), temperature (T), and plasma beta (beta). Geomagnetic disturbance indices consist of kp, AE, and Dst. Solar wind parameters and AE index are with a 1-min resolution. Dst index is with a 1-h resolution, and kp index is with a 3-h resolution. The solar wind parameters and geomagnetic disturbance indices are all from OMNI database. The magnetopause subsolar distances, R0, are calculated by Lin et al. (2010)’s model based on the solar wind data from the OMNI database with a 1-min resolution.Missing values are repaired by the linear interpolation method. The daily averaged values of these parameters are used as inputs when modeling.We develop the model for 75°W using ≥2 MeV electron fluxes from GOES-08 satellite between 1995 and 2002. For the model for 135°W, we use data from GOES-10 satellite between 1999 and June 2006 as well as data from GOES-11 satellite between July 2006 and 2010. We will delete all the ≥2 MeV electron fluxes, when the ≥10 MeV proton fluxes at GEO orbit are greater than 3 cm-2·s-1·sr-1 because the data of ≥2 MeV electron fluxes with a 1-min resolution have not been corrected to avoid proton contamination. If there is less than one hour of poor data in one day, including missing data from GOES satellites as well as data with proton contamination, linear interpolation is used to complete the data, otherwise the data for that day is removed. The ≥2 MeV electron daily fluences from different GOES satellites are calculated and converted to log10 (≥2 MeV electron daily fluences), abbreviated as log10 (daily fluences).We obtain the training set and testing set based on this method. The final available data in each year are listed in Table 1 and plotted in Figure 1. Figure 1(a) shows the available ≥2 MeV electron daily fluences from GOES-08 at roughly 75°W between 1995 and 2002 and Figure 1(b) displays the available ≥2 MeV electron daily fluences from GOES-10 and GOES-11 satellites at about 135°W between 1999 and 2010. The distribution of ≥10 MeV proton fluxes from 1995 to 2010 is shown in Figure 1(c). There are several solar proton events with ≥10 MeV proton fluxes greater than 3 cm-2·s-1·sr-1 occurred between 1998 and 2006. The poor data will be kept in the training set but skipped during training to ensure data consistency.