Predicting Global Ionosphere in Three Dimensions: Integrating Data Assimilation with Convolutional Neural Networks

In this study, we present a novel approach to improve ionosphere prediction by combining the Ionospheric Data Assimilation Four-Dimensional (IDA4D) algorithm with Convolutional Neural Networks (CNNs). The IDA4D algorithm constructs a three-dimensional global electron density by assimilating ground-based GPS slant total electron content (STEC), radio occultation STEC, and radio occultation NmF2 data into the IRI model. The IDA4D outputs are fed into CNNs to learn spatiotemporal patterns. Results are validated with ionosonde and CODE TEC data, demonstrating significant improvements and reducing the root-mean-square error (RMSE) of Nmf2 and vertical TEC by 34% and 51%, respectively, compared to the IRI model. Furthermore, the IDA4D technique successfully reconstructed storm time enhancement of the northern crest equatorial ionization anomaly during late evening hours, resulting from upward and northward plasma transport. The combination of IDA4D and CNNs predicts a three-dimensional electron density more accurately than the IRI model for up to two days.