PreMevE: A Machine-Learning Based Predictive Model for MeV Electrons inside Earth’s Outer Radiation Belt

The presence of megaelectron-volt (MeV) electrons in the Earth’s outer radiation belt poses a hazardous radiation environment for spaceborne electronics through the total ionization dose effect and deep dielectric charge/discharge phenomenon. Thus, developing a reliable forecasting model for MeV electron events has long been a critical but challenging task for space community. Here we update our recent progresses on the PREdictive model for MEV Electrons (PreMevE). This model exploits the power of machine learning algorithms, takes advantage of the coherence caused by local wave‐electron resonance, and uses electron observations from NOAA POES satellites in low‐Earth orbits as inputs—along with the upstream solar wind speeds and densities and GEO measurements—to provide high‐fidelity 1- and 2-day predictions of 1 MeV, 2 MeV and > 2 MeV electron flux distributions across the whole outer radiation belt. Using near-equatorial long-term electron data from the NASA Van Allen Probes mission, we trained, validated and demonstrated that the PreMevE model has L-shell averaged performance efficiencies of ~0.6 for out-of-sample 1-day forecasts and ~0.5 for 2-day forecasts. This study adds new science significance to an existing LEO and GEO space infrastructure, provides reliable and powerful tools to the whole space community, and also suggests for the development of more future tailored space weather models driven by similar methodologies.