Abstract The plasmapause position is crucial for understanding magnetospheric dynamics and space weather forecasting. This study pioneers the integration of lunar phase (LP) into plasmapause modeling using two neural network architectures (BP and fully connected neural network) and a large database of 37,869 crossing events from 1977 to 2015. Our LP‐coupled models achieved a 15% reduction in root mean square error compared to prior artificial neural network models and outperforms empirical benchmarks (e.g., new solar wind‐driven global dynamic plasmapause model), with optimal performance in the dusk sector (18–24 magnetic local time) where lunar tidal effects peak. This work establishes LP as a critical modulator of plasmapause dynamics, challenging the conventional solar wind‐driven paradigm. The neural network framework combining LP modulation with solar wind/geomagnetic parameters yields significant improvements in global plasmapause prediction accuracy, providing a foundation for more precise space weather forecasting. Future research could further refine predictions by incorporating real‐time tilt angle data and coupling with first‐principles simulations of neutral atmosphere tides.

Read original article

Read original article