Modeling the dynamic variability of the outer radiation belt fluxes using machine learning

The outer radiation belt of the Earth consists primarily of energetic electrons ranging in energy from tens of keV to several MeV and is known to be a significant threat to a variety of satellite systems. Thus, developing a useful model to describe the trapped electron flux has long been a challenging task. In this work, we present a trapped electron radiation belt model developed using a neural network with electron flux measurements obtained from NASA's Van Allen Probes. The model is driven by a set of geomagnetic indices (and their time histories), supplemented by solar wind parameters. It can accurately reconstruct the observed electron flux for different energy channels from Van Allen Probes: MagEIS and REPT, anywhere in the outer radiation belt between L~3-6 for any time in the past. Its predictive ability is tested on the out-of-sample time range, which shows excellent agreement. This trapped electron radiation model has wide space weather applications. It can help us understand dropouts and refilling events in the radiation belts as well as provide real-time or hours ahead predictive information about radiation belt dynamics.