ML‐EPM: A Machine Learning Model of Auroral Electron Precipitation Spectra Using Long‐Term DMSP Observations

Abstract

Auroral precipitation plays a key role in coupling the Magnetosphere–Ionosphere–Thermosphere system by modifying ionospheric conductance, currents, atmospheric heating, and upper‐atmospheric dynamics. Leveraging 17 years of electron spectrum data from the Defense Meteorological Satellite Program (DMSP), we developed a Machine Learning‐based Electron Precipitation Model (ML‐EPM) that nowcasts differential energy fluxes of precipitating electrons in 19 energy channels ranging from 30 eV to 30 keV, using 6‐hr histories of solar wind (SW), interplanetary magnetic field (IMF) and geomagnetic indices as inputs. The model performance was evaluated using a 2‐year DMSP data set (2013–2014) and a geomagnetic storm event on 27–28 February 2014, both of which were not included in training. ML‐EPM achieved moderate‐to‐strong correlation coefficients (0.5–0.72) for 16 out of 19 channels while maintaining root‐mean‐square and mean‐absolute errors within the same order of magnitude. Although the model underestimated total energy flux compared to OVATION and DMSP observations, it reproduces key auroral features, including (a) auroral oval broadening, equatorward shifting of auroral boundaries, and enhanced precipitation during disturbed periods, (b) low‐energy precipitation (<1 keV) such as soft electrons in the cusp, polar rain in the polar cap, and secondary electrons generated by primary auroral electrons, and (c) various auroral spectral shapes, including diffuse, monoenergetic, and broadband auroras. Rather than relying on total energy flux and mean energy to estimate an idealized Maxwellian shape, ML‐EPM directly nowcasts electron fluxes across 19 distinct energy channels, allowing global circulation models to more accurately resolve altitude‐dependent ionization rates and predict ionosphere‐thermosphere dynamics.