Development of an Automatic Detection System for Satellite Surface Charging Using DMSP Observation Data and Investigation of the Surrounding Plasma Environment

Electrostatic discharge and potential differences from surface charging during satellite operations can cause serious damage and control malfunctions. Detecting and preventing such charging events is therefore critical for safe and reliable spacecraft servicing.This study developed an automatic, high-accuracy detection method for surface charging of Defense Meteorological Satellite Program (DMSP) using a Convolutional Neural Network (CNN), which identifies the “ion-line structure” in ion energy-time diagrams observed by the plasma instrument during charging. Unlike the previous conventional threshold-based approach using fixed criteria for auroral electron flux and plasma density, our CNN employs Grad-CAM visualization to extract ion-line regions continuously, enabling event detection independent of thresholds.Using ion-line events detected between 2009 and 2019 as training data, the CNN achieved 99.6% accuracy in cross-validation. Applying the model to DMSP F16 ion flux data between 2005 and 2023 automatically detected 2,955, among which 1323 exhibited charging potentials exceeding 100 V.Detected events were concentrated between 17–03 magnetic local time (MLT), with strong (>100 V) negative charging frequently occurring between 18–24 MLT. About 86% corresponded to monoenergetic electron fluxes precipitations. As the energy peak of these fluxes shifted toward higher energies, the corresponding current density increased from 4.71 to 14.4 μA/m². However, when plasma density exceeded 10⁴ cm⁻³, the ion current became dominant, preventing surface charging—confirming the validity of conventional thresholds from a current balance perspective.Moreover, the mean energy of electrons contributing to charging varied with plasma density, indicating that conditions previously discussed independently can be integratively explained by considering plasma density effects.