A lightweight prediction model for global ionospheric total electron content based on attention-BiLSTM

The ionospheric total electron content (TEC) is a critical parameter for space weather and Global Navigation Satellite System (GNSS) applications. A Bidirectional Long Short-Term Memory neural network model with an added attention mechanism (BiLSTM-Attention) was developed in this study to predict 256 spherical harmonic coefficients (SHC). Input data for the forecasting model includes F10.7, Dst, and other feature parameters, along with a lightweight historical time series of SHC from the past day. The model output is the next day's SHC. Then, we compare the results of SHC with those of the 1-day Center for Orbit Determination in Europe (CODE) prediction model. The correlation coefficient form model TEC with respect to the CODE TEC are 0.93 and 0.96 in 2018 and 2022, respectively, while the correlation coefficient of the 1-day CODE prediction model are 0.91 and 0.94. The results illustrate established model both in high and low solar activity years, and exhibiting enhanced robustness during geomagnetic storms. Furthermore, typical ionospheric structures such as Equatorial Ionization Anomaly (EIA) is well reproduced in the TEC prediction maps. Compared to C1PG, the proposed model offers a lighter computational load while maintaining competitive performance in global TEC prediction. (c) 2024 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.