A local 3-D ionospheric model based on COSMIC-2 radio occultation GNSS data and deep neural networks

Abstract The growing availability of ground and space-based Global Navigation Satellite System (GNSS) observations has significantly enhanced our ability to monitor ionospheric variability with improved spatial and temporal resolution. The FORMOSAT-7/COSMIC-2 (F7/C2) mission provides a comprehensive database of high vertical resolution radio occultation (RO) profiles, enabling detailed long-term ionospheric studies. This work presents a three-dimensional ionospheric electron density model over Egypt, developed using Deep Neural Networks (DNN) trained on F7/C2 RO data from 2020 to 2024, designated as 3D-DNN Ion-EG. The optimized DNN architecture captures complex nonlinear relationships between electron density and nine geophysical and temporal inputs: latitude, longitude, altitude, universal time, day of year, year, solar flux (F10.7), and geomagnetic indices (ap and Dst). The 3D-DNN Ion-EG model demonstrates high predictive accuracy for ionospheric electron density, achieving an R 2 of 0.93 and RMSE of 121,000 el/cm 3 during independent testing on 36,856 F7/C2 observations. Comprehensive validation shows that the model outperforms the International Reference Ionosphere (IRI-2020 and IRI-Plas) across multiple parameters including electron density, total electron content (TEC), F2-layer peak density (NmF2), and complete vertical profile reconstructions. The model successfully captures ionospheric variability under both quiet and disturbed geomagnetic conditions, demonstrating superior performance in Egypt’s complex low-latitude environment near the Equatorial Ionization Anomaly (EIA). These results highlight the model’s potential for high-resolution ionospheric modeling over Egypt, with promising applications in GNSS error correction, space weather forecasting, and upper atmospheric research.