Forecast of Ionospheric TEC Maps Using ConvGRU Deep Learning Over China

Abstract

In this article, we propose a convolutional gated recurrent unit (ConvGRU) deep learning method to forecast ionospheric total electron content (TEC) over China based on the regional ionospheric maps (RIMs) from 2015 to 2018. First, we use Global Navigation Satellite System observations from the Crustal Movement Observation Network of China to generate the RIMs of China (CRIMs). Second, we use the CRIMs of 2015–2017 as the training set to predict the ionospheric TEC over China in 2018. Finally, comparative experiments are carried out with ConvLSTM, International Reference Ionosphere (IRI), and Center for Orbit Determination in Europe's (CODE's) 1-day predicted Global Ionospheric Map (C1PG) released by CODE. In addition, we add geomagnetic indices (ap, Kp, and Dst) and solar activity index (F10.7) as the training set to analyze the prediction accuracy of the model (using -A if there are no indices, and -B if there are indices). The results illustrate that the prediction accuracy of ConvLSTM-B and ConvGRU-B models are improved on both geomagnetic storm and quiet days, and the improvement is more obvious on geomagnetic storm days. Furthermore, the root mean square error (RMSE) of the ConvGRU-B model decreases by 28%, 22.4%, and 5.9% compared to that of the ConvGRU-A, IRI-2016, and ConvLSTM-B models during geomagnetic storm days, respectively. For the prediction accuracy of a certain grid point, the RMSE of the ConvGRU-B model decreases by 23%, 32.6%, and 19.3% during geomagnetic quiet days and 24.4%, 30.6%, and 15.7% during geomagnetic storm days compared to that of the ConvGRU-A, IRI-2016, and ConvLSTM-B models, respectively. For the forecast accuracy of TEC in different seasons, the performance of the ConvGRU-B model is also better than that of the ConvLSTM-B model in 2018. These results show that the ConvGRU-B model has competitive performance in RIMs prediction over China during the geomagnetic quiet and storm days.

Keywords

• Convolutional gated recurrent unit (ConvGRU)
• deep learning
• forecast
• ionosphere
• total electron content (TEC)