Deep Learning-Based Ionospheric TEC Prediction Approach

Abstract

The ionosphere layer is an environment that causes a time delay depending on the frequencies of the radio waves of the Global Positioning System (GPS) satellites. Most ionospheric studies are performed using total electron content (TEC) changes obtained from GPS signals. Today, studies on the physical structure of the ionosphere continue in many areas such as the prediction of space weather conditions, positioning, navigation, and communication. This study aims to create a deep learning-based model for the prediction of ionospheric TEC. Artificial Neural Networks (ANN) method was used to create this model. The artificial neural network and related properties designed for this method have been prepared in the MATLAB® environment using the Deep Learning Toolbox. In this study, HRUH permanent station which is located in Harran University Campus was registered Turkey Continuously Operating Reference Station’ (CORS-TR) RINEX observations are used. TEC variations were obtained from GPS observations between the years 2016 and 2019 with two hours of temporal resolution. In this study, the determination of the optimum parameters was investigated which aims to forecast ionospheric TEC variations for the first six months of 2019. In the created model, the number of iterations is selected as constant (i = 100). The minimum RMSE value is ±0.28704 TECU with parameters where the number of hidden layers is selected as 50. The RMSE value of the forecasting model which is calculated in 1 hidden layer is ± 0.47298 TECU.