Comparison of Global TEC Prediction Performance with two Deep Learning Frameworks

Kunlin Yang, Yang Liu

Abstract: The ionosphere is a crucial component of Earth's atmosphere and plays a significant role in radio communication, broadcasting, and radar positioning. Assessing the ionosphere's ionization strength is commonly done by measuring the total electron content (TEC), which serves as an indicator of ionospheric changes. Accurate TEC prediction is essential for anticipating the effects of space weather on trans-ionospheric radio propagation and related applications. In this study, we examine the temporal characteristics of global TEC and propose deep learning frameworks based on long short-term memory (LSTM) and Transformer to forecast short-term TEC on a global scale. We compare the prediction performance of the two models using evaluation metrics such as root mean square error (RMSE) and ?? 2 . Our numerical experiments demonstrate that the LSTM model outperforms the Transformer model in terms of TEC prediction for both the test set years. Specifically, for the high, middle, and low latitudes, the LSTM model reduces the RMSE by 25.9%, 31.1%, and 21.0% in 2014, and by 15.4%, 17.1%, and 11.2% in 2017, respectively. Furthermore, we evaluate the TEC prediction during magnetic storm periods. The results indicate that both the LSTM and Transformer models exhibit improved prediction stability during these periods. However, the LSTM model consistently outperforms the Transformer model. For the high, middle, and low latitudes, the RMSE decreases by 24.7%, 26.4%, and 23.1%, respectively, during the first magnetic storm, and by 15.4%, 21.3%, and 9.70% during the second magnetic storm. Overall, our proposed TEC prediction framework proves to be effective, enhancing our understanding of ionospheric mechanisms, and facilitating trans-ionospheric radio applications under unstable space weather conditions. Keywords: TEC prediction, LSTM, Transformer, time series.
Published in: Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023)
September 11 - 15, 2023
Hyatt Regency Denver
Denver, Colorado
Pages: 3982 - 4001
Cite this article: Yang, Kunlin, Liu, Yang, "Comparison of Global TEC Prediction Performance with two Deep Learning Frameworks," Proceedings of the 36th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2023), Denver, Colorado, September 2023, pp. 3982-4001. https://doi.org/10.33012/2023.19467
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In