Iono‐PhyTAU: A Lightweight Global Ionospheric TEC Prediction Model Based on Temporal Attention Unit and Physical Condition Constraints of the Ionosphere

Abstract

Ionospheric Total Electron Content (TEC) is the primary error source causing signal delay in Global Navigation Satellite Systems (GNSS). High‐precision TEC prediction is essential for positioning, navigation, and space weather monitoring. Although deep learning models including ConvLSTM, ConvGRU, and Transformer have advanced TEC forecasting, they still struggle to capture long‐range global spatiotemporal dependencies, suffer from severe error accumulation in multi‐step prediction, and involve high computational cost and large parameters. To address these limitations, this paper proposes a lightweight deep learning model called Iono‐PhyTAU for global ionospheric TEC prediction. The model integrates a Temporal Attention Unit (TAU) and external physical driving constraints, which reduces computational complexity, enhances long‐sequence dependence modeling, and extracts fine‐grained spatiotemporal features. The model significantly improves TEC prediction accuracy during geomagnetic storms. Experiments conducted in the high solar activity year (2015) and low solar activity year (2020) show that Iono‐PhyTAU achieves the best prediction accuracy, with RMSE values of 2.97 TECU and 1.08 TECU, respectively. The model effectively alleviates error accumulation in multi‐step prediction: after the fourth hr of daily forecasting in 2015, its RMSE is reduced by 0.3–0.5 TECU compared with the suboptimal ConvLSTM model and by 0.5–1.5 TECU compared with C1PG. In addition, Iono‐PhyTAU exhibits strong robustness across different latitudes, seasons, and times, with particularly outstanding performance during geomagnetic storms. Experimental results confirm that the model can adapt to various space weather environments and performs excellently in geomagnetic storm events. It has great application potential in improving the reliability of global TEC forecasting and provides strong support for GNSS ionospheric delay correction.