Abstract The total electron content (TEC) in the ionosphere is strongly affected by solar activity and geomagnetic disturbances in mid‐ and low‐latitude regions, making it a major source of error in GNSS navigation and communication systems. To improve the prediction accuracy of ionospheric TEC, this study proposes a deep learning model—Beluga Whale Optimization (BWO)‐convolutional neural networks (CNN)‐extended long short‐term memory (xLSTM)—that integrates ground‐based GNSS observations with space‐based COSMIC‐2 radio occultation data and incorporates the BWO algorithm. The model combines CNN and xLSTM structures to extract spatial features and temporal dependencies, respectively. The BWO algorithm is employed to optimize the model's structural parameters, thereby enhancing its predictive performance. Comprehensive evaluations under both quiet and geomagnetically disturbed conditions demonstrate that the BWO‐CNN‐xLSTM model achieves superior prediction accuracy and stability across various geomagnetic environments, significantly outperforming comparative models such as CODE‐GIM, IRI, CNN‐BiLSTM, and BiLSTM. During quiet periods, the model achieves an root mean square error (RMSE) of 0.924 TECU and an mean absolute error (MAE) of 0.675 TECU, while during geomagnetic storm periods, the RMSE and MAE increase to 1.173 TECU and 0.883 TECU, respectively. These results confirm the effectiveness and applicability of the proposed model for mid‐ and low‐latitude ionospheric modeling and space weather monitoring.
Li et al. (Fri,) studied this question.