Short-Term Prediction of Ionospheric 3D Electron Density in Low Latitude Region Based on LSTM Neural Network

Abstract

The monitoring and prediction of ionospheric electron density is of great significance for ionospheric physics research and navigation positioning technology. Based on the observed slant total electron content (STEC) along different satellite-receiver rays, computerized ionospheric tomography (CIT) is used to reconstruct the 3-D electron density. In this paper, the four dimensional imaging and short-term prediction of electron density of ionosphere are studied. A data-driven tomography and prediction method has been achieved, which combines the long short-term memory (LSTM) neural network to predict the ionospheric electron density. Firstly, the data-driven sparse ionospheric tomography method will be used to achieve 4D electron density with high spatial and temporal resolution. Secondly, the prediction of electron density at hour level are constructed by LSTM neural network based on the tomography data set. Thirdly, the dSTEC error of the independent reference station is adopted to compare the prediction accuracy based on the LSTM model. The proposed tomography and prediction method will be verified based on real measurement of GNSS data in Hainan, China collected from Qianxun ground-based augmentation system on October 26, 2020. The temporal and spatial variation of ionospheric electron density with 5 minutes temporal resolution can be constructed by the data-driven compressed sensing method, and can be predicted by the LSTM neural network. There is a corresponding relationship between ionospheric electron density by the tomography method and the S4 parameters at Sanya station. The RMSE of dSTEC of the prediction model is about 0.6 TECU. The relative error of 50-minute short-term prediction is about 3.5%. The ionospheric electron density prediction method proposed in this paper is expected to be used to realize real-time and accurate ionospheric monitoring and prediction in China.