Deep learning for ionospheric TEC forecasting at mid-latitude stations in Turkey

Abstract

Earth's ionosphere is an important medium for navigation, communication, and radio wave transmission. The inadequate advances in technology do not allow enough realization of ionosphere monitoring systems globally, and most research is still limited to local research in certain parts of the world. However, new methods developed in the field of forecasting and calculation contribute to the solution of such problems. One of the methods developed is artificial neural networks-based deep learning method (DLM), which has become widespread in many areas recently and aimed to forecast ionospheric GPS-TEC variations with DLM. In this study, hourly resolution GPS-TEC values were obtained from five permanent GNSS stations in Turkey. DLM model is created by using the TEC variations and 9 different SWC index values between the years 2016 and 2018. The forecasting process (daily, three-daily, weekly, monthly, quarterly, and semi-annual) was carried out for the prediction of the TEC variations that occurred in the first half-year of 2019. The findings show that the proposed deep learning-based long short-term memory architecture reveals changes in ionospheric TEC estimation under 1–5 TECU. The calculated correlation coefficient and R2 values between the forecasted GPS-TEC values and the test values are higher than 0.94.