Prediction of GNSS-Based Regional Ionospheric TEC Using a Multichannel ConvLSTM With Attention Mechanism

Monitoring and predicting ionospheric space weather is important for global navigation satellite system (GNSS) navigation, positioning, and communication. Ionospheric total electron content (TEC) is a vital indicator to measure ionospheric space weather. This study utilizes a multichannel convolutional long short-term memory (ConvLSTM) with attention mechanism to predict ionospheric TEC maps considering the relevance of physical observations for TEC variations. The multichannel ConvLSTM with attention mechanism (MConvLSTM-Attention) method is trained and tested on regional ionospheric maps (RIMs) for three years (2015–2017) by using GNSS observations from the Crustal Movement Observation Network of China (CMONOC). The experimental results show that the root-mean-square error (RMSE) values of MConvLSTM-Attention model during quiet, moderate, and geomagnetic storm periods are 2.58, 2.60, and 3.21 TECU, respectively. Also, the MConvLSTM-Attention model performs better than MConvLSTM, ConvLSTM, and international reference ionosphere (IRI) 2016 models in predicting RIMs. In addition, the MConvLSTM-Attention prediction model shows good generalization performance, relatively good stability, and high precision during both geomagnetic quiet and storm time.