Global Prediction of Sub-relativistic and Relativistic Electron Fluxes in the Geosynchronous Orbit Using Artificial Neural Networks

High-energy particles in geosynchronous orbit (GEO) present significant hazards to astronauts and artificial satellites, particularly during extreme geomagnetic activity conditions. In the present study, based on observations onboard the GOES-15 (Geostationary Operational Environmental Satellites) spanning from 2011 to 2019 as well as the historical values of solar wind and geomagnetic activity indices, an artificial neural network (ANN) model was established to predict the temporal evolution of the GEO sub-relativistic and relativistic (>0.8 MeV and >2 MeV) electron fluxes one day in advance. By adding the last-orbital observations of electron flux in each of all 24 different MLTs (magnetic local times) and its two MLT-adjacent values into inputs, the current model can provide accurate predictions with an MLT-resolution of one hour for the first time. Moreover, it achieves the best performance in comparison with previous methods, with overall root-mean-square-errors (RMSEs) of 0.276 and 0.311, prediction-efficiencies (PEs) of 0.863 and 0.844, and Pearson-correlation-coefficients (CCs) of 0.930 and 0.921 for >0.8 MeV and >2 MeV electrons, respectively. More than 99% of the samples exhibit an observation-prediction difference of less than one order of magnitude, while over 90% demonstrate a difference of less than 0.5 order. Further analysis revealed that it can precisely track the global variations of the electron flux during both quiet times and active conditions. The present model would be an important supplement for examining the temporospatial variations of inner magnetospheric particles and helping to establish a warning mechanism for space weather disaster events.