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

In recent years, prediction models of radiation belt electron fluxes or phase space density have been established and optimized by numerical simulations and machine learning based on measurements near the geomagnetic equator. In the present work, using observations from low Earth orbit satellites, the Meteorological Operational Satellite Program of Europe-A (MetOp-A), we constructed a novel artificial neural network (ANN) model to predict the electron fluxes in low equatorial pitch angles at 40 and 130 keV. The historical solar wind and geomagnetic indices are adopted as model inputs. The ANN model achieves excellent performance in the main region of the outer radiation belt (L = 4–6), with overall root mean square errors of 0.3468 (0.3567), prediction efficiencies of 0.9381 (0.9343), and Pearson correlation coefficients of 0.8893 (0.8628) for electrons at 40 keV (130 keV). Moreover, 51.76% of samples for electrons at 40 keV exhibit an observation–prediction discrepancy of fewer than 0.2 orders of magnitude, 87.21% demonstrate a difference of less than 0.5 orders, and 98.58% show a difference of less than one order. For electrons at 130 keV, the three critical values are 51.29% for 0.2 order, 86.33% for 0.5 order, and 98.43% for one order. Moreover, the model can precisely monitor variations in radiation belt electron fluxes during a realistic geomagnetic storm event, with no substantial errors. By adopting observations in low Earth orbit, the present model concentrates on the electron fluxes in low equatorial pitch angles, which broadens the scope of the radiation belt electron forecast.