High‐Precision Prediction of Ionospheric TEC in the China Region Based on CMONOC High‐Resolution Data and an Auxiliary Attention Temporal Convolutional Network

• Published in: Journal of geophysical research. Machine learning and computation Vol. 2; no. 2
• Main Authors: Chen, Jianghe, Xiong, Pan, Wu, Haochen, Zhang, Xiaoran, Zhang, Xuemin, Chai, Rongzi, Zhang, Ting, Wang, Kaixin, Wang, Chaoyu
• Format: Journal Article
• Language: English
• Published: Wiley 01.06.2025
• Subjects: AuxATTCN; CMONOC; deep learning; forecasting; ionospheric; TEC
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2025JH000608

Accurate prediction of Total Electron Content (TEC) in the ionosphere is crucial for navigation, communication, and space weather forecasting. However, the Global Ionosphere Maps provided by the International GNSS Service have limitations in resolution and adaptability in the China region, making high‐precision predictions difficult. This study constructs a high‐precision regional TEC map with 1° × 1° spatial resolution, 2‐hr temporal resolution, and coverage from 2019 to 2023, based on data from 270 Crustal Movement Observation Network of China (CMONOC) GNSS stations. At the data level, a non‐integrated spherical harmonic model and Differential Code Bias correction method are employed to significantly reduce interpolation errors and improve model accuracy. At the algorithmic level, an Auxiliary Attention Temporal Convolutional Network (AuxATTCN) model is proposed, integrating an auxiliary attention mechanism with a Temporal Convolutional Network (TCN) to effectively capture long‐term dependencies and dynamically incorporate external driving factors such as geomagnetic activity and solar radiation. Comparative analysis with multiple experiments under varying geomagnetic and solar conditions shows that the AuxATTCN model significantly outperforms traditional time‐series methods (such as ARIMA, Prophet), mainstream deep learning models (including ConvLSTM, CONGRU, and TCN), and international ionospheric models (IRI2020, NeQuick2) in terms of overall error, seasonal and diurnal variations, and prediction accuracy during geomagnetic storms and solar activity peaks. The results demonstrate that the synergistic optimization of high‐quality CMONOC data sets and innovative algorithms achieves exceptional spatiotemporal accuracy and robustness in TEC prediction for the China region, providing new insights and technical support for fields such as navigation, communication, and space weather forecasting. Plain Language Summary This study presents a new model to predict the ionospheric Total Electron Content (TEC) in China with high precision, using data from 270 GNSS stations between 2019 and 2023. The model has a high spatial resolution of 1° × 1° and a temporal resolution of 2 hr, allowing it to capture local ionospheric changes better than global models. The model, called Auxiliary Attention Temporal Convolutional Network (AuxATTCN), combines advanced deep learning techniques such as Temporal Convolutional Networks and an auxiliary attention mechanism that factors in solar radiation and geomagnetic activity. This makes the model especially effective during geomagnetic storms or solar flares. Compared with other models, AuxATTCN shows significantly better accuracy, particularly during times of high solar or geomagnetic activity, and handles daily and seasonal TEC variations well. It also performs well in areas with dense GNSS stations, though it is less accurate in remote regions. This research provides a powerful tool for more accurate space weather forecasting and other applications relying on TEC predictions. Key Points Developed a high‐precision ionospheric Total Electron Content (TEC) prediction model for China with 1° resolution, improving local variation capture The model combines causal dilated convolution and attention, improving accuracy in modeling TEC's temporal, seasonal, and diurnal variations The Auxiliary Attention Temporal Convolutional Network model outperforms traditional methods in predicting TEC, especially during geomagnetic storms and solar activity peaks