Prediction of Radiation Belts Electron Fluxes at a Low Earth Orbit Using Neural Networks With PROBA-V/EPT Data

We introduce for the first time the PROBA-V/EPT electron flux data to train a deep learning data-driven model with the purpose of investigating the Earth's radiation belts dynamics. The Long-Short Term Memory Neural Network is employed to predict the electron fluxes between 1 and 8 Earth Radius (R-E) along a Low Earth Orbit. Different combinations of time series inputs involving Solar Wind and geomagnetic data are tested, based on previous knowledge of their impact onto the high energy radiation fluxes. Two Energetic Particle Telescope energy channels feed the learning procedure for nonrelativistic (0.5-0.6 MeV) and relativistic (1.0-2.4 MeV) electron fluxes. A good performance of the model employing different time resolutions from hours to days is demonstrated with a correlation of more than 0.9 between the predicted and out-of-sample fluxes, and a prediction efficiency that can attain between 0.6 and 0.9 depending on the L range. The analysis of different input parameters and time resolutions allows to construct the best data set structure and improve the model to identify relevant effects such as dropouts, flux increase and recovery features.Plain Language Summary The Van Allen belts are regions of intense radiation trapped by the Earth's magnetic field. A continuous stream of charged particles, the solar wind, is ejected from the Sun's outer atmosphere carrying the interplanetary magnetic field. When the solar wind interacts with the Earth's magnetic field, it can have significant effects on the space environment surrounding our planet. Predicting the state of the Van Allen Belts is crucial for satellites orbiting near-Earth, and to aviation and astronaut operations. Disturbances in the Van Allen Belts can represent a significant hazard to human activities, and can also drive out-of-control induced electric currents that can damage equipment on satellites. A Machine Learning model is developed here using electron radiation fluxes measurements from an instrument called Energetic Particle Telescope, onboard the PROBA-V satellite orbiting at an altitude of 820 km. A good performance of the model is demonstrated with a prediction efficiency between 0.6 and 0.9. The results of the model support the prediction of the space-time Van Allen belts evolution to prevent serious damage in case of strong external perturbations from particles and electromagnetic fields.