Artificial Neural Network–Based Modeling and Prediction of GNSS Ionospheric Errors in Egypt

Abstract

One of the major error sources in Global Navigation Satellite Systems (GNSS) positioning is the signal propagation in the ionospheric layer. For low-accuracy real-time applications, the basic ionospheric models can be used to estimate the error budget of this effect. On the other hand, high-accuracy real-time applications necessitate the availability of a dual-frequency receiver to employ the ionosphere-free linear combination and eliminate most of the ionospheric effect. However, achieving a high positioning accuracy using a single-frequency receiver remains a challenge. This study explores the potential of using artificial neural networks (ANNs) for predicting the ionospheric errors influenced by the vertical total electron content (VTEC) across Egypt. The research utilizes data from 11 permanent GNSS stations covering Egypt over 10 years, in addition to information about the sunspot numbers, the planetary K-indices, and the electron densities at the F2 peak. Several ANNs are developed to identify the optimal number of hidden layers and number of neurons. The results indicate that an ANN with three hidden layers and 50 neurons can deliver the best performance. The prediction capability of the developed model is assessed using GNSS observations from an independent station. This shows promising VTEC predictions with a mean error value of 2.11TECU and a root mean square error (RMSE) value of 2.67TECU. Moreover, two field experiments (static and kinematic) are conducted to emphasize the significance of employing ANN ionospheric models in improving positional solutions. The epoch-wise accuracy improvement reaches 2.32 m (95.5%) and 6.86 m (46.8%) in the static and kinematic tests, respectively.