Ionospheric TEC Prediction Performance of ARIMA and LSTM Methods in Different Space Weather Conditions

Abstract

The ionosphere has some temporal regular changes under the dominant control of the Sun. The stationary structure of the ionospheric time series (e.g. TEC, foF2) allows it to be modeled on a specific time. In this study, we tested the performance of the artificial intelligent (AI) techniques e.g. a machine learning-based method, autoregressive integrated moving average (ARIMA), and a deep learning-based method, long short-term memory (LSTM) network to the prediction of Total Electron Content (TEC) values. The TEC data of six different locations in low, middle, and high latitudes were selected from the Center for Orbit Determination in Europe – Global Ionosphere Maps (CODE-GIMs). To show the performance of the proposed methods during quiet space weather and a severe geomagnetic storm, we trained the 60 days TEC data (24 data points in one day) and forecasted the TEC data of the subsequent five days by fitted models with optimal hyperparameters. The forecasted TEC values were compared with observed TEC through some statistical metrics (RMS, MAE). The results indicated that the LSTM is more successful in TEC prediction than ARIMA. This study brings new insights into the AI techniques in the ionospheric TEC prediction.