Predicting radiation belt electrons in the low Earth orbit using machine learning methods.

The article focuses on the development of an artificial neural network (ANN) model to predict radiation belt electron fluxes at low equatorial pitch angles and lower energies (40 and 130 keV) using low Earth orbit (LEO) satellite data from the Meteorological Operational Satellite Program of Europe-A (MetOp-A). By incorporating historical solar wind parameters and geomagnetic indices as inputs, the model achieves strong predictive performance in the outer radiation belt region (L = 4–6), with root mean square errors below 0.35, prediction efficiencies above 0.93, and Pearson correlation coefficients exceeding 0.86. The model reliably captures electron flux variations during geomagnetic storms, with over 98% of predictions deviating by less than one order of magnitude from observations. This work extends radiation belt forecasting capabilities by addressing low equatorial pitch angles, a parameter less explored in prior models focused on equatorial measurements, and demonstrates potential for improving space weather prediction relevant to satellite operations.