Predicting radiation belt relativistic electron flux from sub-relativistic electron fluxes using machine learning

Relativistic electrons (>1 MeV) confined within the Earth's radiation belt, which presents significant hazards to astronauts and spacecraft, potentially arise from sub-relativistic electrons (a few hundred keV) through radial diffusion and wave–particle interactions. This study presents a novel artificial neural network (ANN) prediction model for radiation belt relativistic electron flux at a typical energy of 1.8 MeV, utilizing concurrent measurements of sub-relativistic electron fluxes at Ek = 78–871 keV from Van Allen Probe observations between 2012 and 2019. The historical values of the solar wind and geomagnetic indices are also adopted as inputs of the model. The ANN model has remarkable performance in L = 2.5–6.5, with overall root mean square errors (RMSEs) of 0.2287, prediction efficiencies (PEs) of 0.9241, and Pearson correlation coefficients (CCs) of 0.9478 between observations and predictions in the test set (from April 2018 to March 2019). Statistical analysis indicated that 98.56% of the samples exhibit an observation–prediction difference of less than one order of magnitude, while 85.94% demonstrate a difference of less than 0.5 order. Moreover, predictions can precisely replicate the dynamics of relativistic electrons within the outer radiation belt and the slot region under realistic solar wind conditions and geomagnetic activitiesies. The present model offers novel insights into radiation belt electron predictions and facilitates the validation of results across several instruments onboard spacecraft.