A New and Tested Ionospheric TEC Prediction Method Based on SegED-ConvLSTM

Yuanhang Liu, Yingkui Gong, Hao Zhang, Ziyue Hu, Guang Yang, Yuan HongRemote 2025 · DOI

TL;DR

This paper introduces SegED-ConvLSTM, a novel deep learning model for predicting Total Electron Content (TEC) in the Earth's ionosphere. By integrating spatial, temporal, and space weather data, and employing a unique piecewise processing approach, SegED-ConvLSTM significantly outperforms existing methods, especially during challenging space weather events.

What Problem Does This Paper Address?

Accurate prediction of the ionosphere's Total Electron Content (TEC) is crucial for the precision of Global Navigation Satellite Systems (GNSS) and reliable radio communication. The ionosphere, however, is a complex, dynamic system influenced by solar activity, geomagnetic fields, and diurnal cycles. While previous research has focused on spatial variations, they often overlooked the periodic diurnal changes. Existing models struggle to capture the intricate spatiotemporal dynamics and the differential impact of space weather across various latitudes, leading to inaccuracies, particularly during geomagnetic storms and in specific regions like the equatorial anomaly.

Methods

The core of this work is the SegED-ConvLSTM model. It builds upon the ConvLSTM architecture by incorporating an encoder-decoder structure to extract high-dimensional spatiotemporal features. Key innovations include:

  • PieceWise Processing: This technique segments continuous TEC and space weather data into daily time segments, anchoring them to fixed daily timestamps. This helps eliminate day-to-day variability and allows the model to capture persistent ionospheric evolution patterns, effectively acting as a "periodic filtering" mechanism.
  • Space Weather Feature Processing: Multiple space weather elements (solar flares, F10.7, SSN, X-rays, solar wind, Dst, ap) are processed and fused. Notably, Gaussian Process Regression (GPR) is used to interpolate sparse geomagnetic data, providing a more comprehensive representation of space weather's influence across different latitudes.
  • SegED-ConvLSTM Architecture: This combines the piecewise processing with an encoder-decoder ConvLSTM structure. The encoder extracts high-level features, and the decoder reconstructs the TEC map. This hybrid approach aims to capture both fine-grained local variations and large-scale spatial correlations.

Datasets

  • Global Ionospheric Maps (GIMs): Primarily from the Center for Orbit Determination in Europe (CODE), providing TEC data at a 1-hour resolution on a 2.5° latitude x 5° longitude grid.
  • Space Weather Elements:
    • Solar activity indices: Levels of flares, F10.7, sunspot number (SSN), X-rays.
  • Geomagnetic activity indices: Solar wind, Dst, ap.
  • SuperMAG data: High-temporal-resolution (minute intervals) geomagnetic data, interpolated using GPR.
  • Time Period: Data from August 2012 to August 2024 were used for training and testing.

Key Insights and Ideas

  • Diurnal Cycles Matter: Neglecting the periodic diurnal variations of the ionosphere leads to suboptimal prediction accuracy. The PieceWise method effectively addresses this by aligning data to daily timestamps.
  • Multi-Source Space Weather Data is Crucial: Integrating a diverse set of space weather parameters, especially those with different temporal resolutions and spatial impacts (like X-ray and Dst), significantly improves prediction accuracy.
  • Latitude-Specific Features Enhance Performance: Processing space weather features with consideration for different latitude bands (high, mid, low) leads to more accurate predictions, particularly in challenging regions like the equatorial anomaly.
  • Robustness During Storms: The proposed SegED-ConvLSTM model demonstrates superior resilience and accuracy during geomagnetic storm events compared to other models.

Key Results

  • Overall Superiority: SegED-ConvLSTM-6 achieved the lowest RMSE (2.97 TECU) and MAE (1.35 TECU), the highest SSIM (0.93), and the best CC (0.99) among all compared models.
  • Error Reduction: SegED-ConvLSTM-6 showed a significant error reduction (e.g., 23.5% lower RMSE than TS-ED-ConvLSTM-6).
  • Feature Impact: Removing X-ray and Dst features led to the most substantial increase in RMSE and MAE, highlighting their critical role.
  • Latitude Performance: The model demonstrated balanced performance across high, mid, and low latitudes, outperforming models with spatial biases.
  • Storm Event Performance: SegED-ConvLSTM-6 showed significantly higher prediction accuracy and better robustness during typical ionospheric storm events.
  • Batch Size Impact: The model achieved its best performance with a batch size of 24 during training.

Limitations and Future Work

  • Computational Cost: While effective, models incorporating self-attention mechanisms (like SAM-ConvLSTM) can incur higher computational costs. The SegED-ConvLSTM architecture, while efficient, still requires significant resources.
  • Future Work: The authors suggest further enhancing the model by incorporating additional space weather data and other high-dimensional observational data to improve comprehensiveness and precision. Developing real-time prediction systems is identified as a crucial next step for widespread application. The study also notes that while SSN and F10.7 had smaller impacts, further investigation into feature selection optimization could be beneficial.

Summary updated

Abstract

Total electron content (TEC) serves as a key parameter characterizing ionospheric conditions. Accurate prediction of TEC plays a crucial role in improving the precision of Global Navigation Satellite Systems (GNSS). However, existing research have predominantly emphasized spatial variations in the ionosphere, neglecting the periodic changes of the ionosphere with the diurnal cycle. In this paper, we propose a TEC prediction model, which simultaneously considers both spatial and temporal characteristics to extract spatiotemporal features of ionospheric distribution. Additionally, we integrate several space weather element datasets into the prediction model framework, allowing the generation of multiple space weather feature values that represent the influence of space weather on the ionosphere at different latitudes and longitudes. Moreover, we apply Gaussian process regression (GPR) interpolation to geomagnetic data to characterize impact on the ionosphere, thereby enhancing the prediction accuracy. We compared our model with traditional image-based models such as convolutional neural networks (CNNs), convolutional long short-term memory networks (ConvLSTMs), a self-attention mechanism-integrated ConvLSTM (SAM-ConvLSTM) model, and one-day predicted ionospheric products (C1PG) provided by the Center for Orbit Determination in Europe (CODE). We also examined the effect of using different numbers of space weather feature values in these models. Our model outperforms the comparison models in terms of prediction error metrics, including mean absolute error (MAE), root mean square error (RMSE), correlation coefficient (CC), and the structural similarity index (SSIM). Furthermore, we analyzed the influence of different batch sizes on model training accuracy to find the best performance of each model. In addition, we investigated the model performance during geomagnetic quiet periods, where our model provided the most accurate predictions and demonstrates higher prediction accuracy in the equatorial anomaly region. We also analyzed the prediction performance of all models during space weather events. The results indicate that the proposed model is the least affected during geomagnetic storms and demonstrates superior prediction performance compared to other models. This study presents a more stable and high-performance spatiotemporal prediction model for TEC.
BibTeX
@inproceedings{liu2025new,
  author = {Yuanhang Liu and Yingkui Gong and Hao Zhang and Ziyue Hu and Guang Yang and Yuan Hong},
  title = {{A New and Tested Ionospheric TEC Prediction Method Based on SegED-ConvLSTM}},
  booktitle = {Remote},
  year = {2025},
  doi = {10.3390/rs17050885},
}
Figure