A Neural Network Model of Equatorial Electric Field Structures in the Inner Magnetosphere

Abstract Enabled by state‐of‐the‐art electric field measurements from the Van Allen Probes and careful calibration of the high‐quality data, we developed the first machine‐learning based inner‐magnetosphere electric field model, which covers L = 2.5–6.0 within 20 around the magnetic equator. The model output is the DC electric field perpendicular to the background magnetic field, including the poloidal and toroidal components. The most informative drivers, including the SYM‐H and AE indices and solar wind speed, are automatically identified during the model training process. The model input consists of these parameters along with time and spatial coordinates. The model successfully reproduces the storm‐time evolution of meso‐scale electric field structures, potentially related to subauroral polarization streams and dawnside auroral polarization streams. Given the growing recognition that meso‐scale electric field structures modulate the transport of high‐energy electrons, our model can incorporate these structures into studies of ring current and radiation belt dynamics.