V. International Scientific and Vocational Studies Congress Engineering (BILMES EN 2020), İzmir, Turkey, 12 - 15 December 2020, pp.186-194
The ionosphere is an atmosphere layer that causes delay/acceleration effects on electromagnetic signals traveling between space and earth due to the containing number of free electrons. Forecasting the number of free electrons is an important field of study for space weather, communication, and navigation applications. In this study, artificial intelligence (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, were utilized in producing predicted Global Ionosphere Maps (GIMs). The vertical Total Electron Content (VTEC) values of the International GNSS Service (IGS)-GIMs between March 1 and March 31, 2020 were trained at all grid points (71x73 grids) and VTEC values on April 1-2 were forecasted with the proposed models. Also, the forecast VTEC values obtained from the 1 and 2 days predicted GIMs (pGIMs) produced by CODE (c1pg, c2pg) and the forecast VTEC values of proposed methods were compared with the observed VTEC values of the final IGS-GIMs (igsg). The results showed that ARIMA predicted VTEC values with lower accuracy than LSTM and c1pg-c2pg. LSTM method has very close prediction accuracy to c1pg-c2pg, but the results of CODE-pGIMs are more consistent with the final IGS-GIMs. Future work aims to achieve better results than pGIMs of IGS analysis centers (CODE, ESA, UPC, IGS combined) by changing the hyperparameters of the LSTM (e.g. increasing the number of hidden layers or iterations, using long-term training data) or using modified versions of LSTM (e.g. Bi-LSTM). In particular, we foresee that the multivariate approach will further improve the results of the proposed methods.