Real-time regional ionosphere modeling with RFR-Net over China

Abstract

The mid-to-low-latitude ionosphere, influenced by phenomena such as the Equatorial Ionization Anomaly, responds more sensitively to changes in solar activity, which negatively affect the transmission of various electromagnetic signals. Moreover, next-generation technologies, particularly Precise Point Positioning-Real-Time Kinematic (PPP-RTK), require more instant and detailed information on near-earth space environments. However, current ionospheric Total Electron Content (TEC) maps are often post-processed and designed for global applications. Under this challenge, we develop a real-time, high-precision regional ionospheric TEC map service using a deep learning inpainting Recurrent Feature Reasoning (RFR) method. Given the limited ionospheric observation resources, our approach significantly reduces the scale of observational data by utilizing only 2.5% of the total TEC data. This is achieved through the RFR and the Knowledge Consistent Attention (KCA) module embedded in the RFR-TEC model, where the RFR module leverages pixel correlations for robust estimation, and the KCA mechanism enforces patch consistency. Results indicate that the real-time RFR-TEC achieves TEC accuracy comparable to the post-processed CODE-TEC and surpasses the real-time UPC-TEC by 47.8% in long-term validation. Additionally, the RFR-TEC map demonstrates superior stability compared to the real-time UPC-TEC, while its performance varies with the seasons.