Forecasting the total electron content TEC of the ionosphere using space weather parameters

Nowadays, an increasing number of studies on the influence of space weather on the ionosphere appeared. The main parameter of the ionosphere is the total electron content TEC, the prediction of which is an important task. One of the approaches to forecasting TEC is the use of the most common and successful statistical methods of the forecast from other areas of science. However, preliminary research has shown that direct application of these methods to TEC time series may not always give positive results. Therefore, in this paper, two modifications of the existing autoregressive moving average (ARMA) and polynomial regression methods are proposed. TEC values obtained for modified methods are compared with values obtained using standard approaches. Additionally, the results for the proposed regression models are compared with two existing Recurrent Neural Network RNN architectures (Gated Recurrent Unit, Long Short-Term Memory) for several parameter sets. The results are illustrated on an example of one of the most disturbed years (2015) according to the data of the European mid-latitude station Juliusruh. It is shown, that: (1) the proposed regression methods modifications essentially improve results in comparison with initial versions of methods, (2) RNN have performed at the comparable level, (3) the best estimate of RMSE is in the range 1–2.5 TECU, (4) accounting for space weather factors did not improve the results compared to NN methods.