Application of Singular Spectrum Analysis Using Artificial Neural Networks in TEC Predictions for Ionospheric Space Weather

The prediction of the ionospheric state has become increasingly important as more and more terrestrial and space-based radiocommunication systems rely on the ionospheric space weather. This article presents a new ionospheric prediction model, named the singular spectrum analysis-artificial neural network (SSA-ANN) model. SSA is used as a preprocessing tool for the ionosphere total electron content (TEC) prediction based on the ANN approach. The hourly global positioning system (GPS)-TEC observations from the period of 2009-2017 at the Bangalore (13.02° N and 77.57° E) station are taken into account for analysis. The quick convergence of the SSA decomposition makes it possible to use the first four SSA modes to represent 99.57% of the total variance of the GPS-TEC dataset. The root mean square error between the observations and the SSA-ANN model TEC value is 1.40 TECU for the period of 2009-2017, and the correlation coefficient is 0.99. The performance of SSA-ANN is evaluated using autoregressive moving average (ARMA) and International Reference Ionosphere-16 (IRI-16) models in three different cases of solar phase periods, namely the ascending solar phase (ASP), high solar phase (HSP), and descending solar phase (DSP), and in different seasons. The SSA-ANN model predicts the absence of winter anomaly during the ASP period (2012), but the ARMA model and IRI models failed to predict the absence of the anomaly. The proposed SSA-NN model resulted in higher prediction accuracy and reduced training and testing times, which would be useful for the development of an ionospheric predicting system.