Construction of LSTM model for total electron content (TEC) prediction in Thailand

Abstract

Total electron content (TEC) is an important parameter often used to explain the ionosphere characteristics and disturbances. Severe local disturbance often originates in the equatorial region then expand to low- and mid-latitude regions. Vertical TEC (VTEC as well as slant TEC (STEC) modeling's and predictions attract attention from researchers worldwide since they are essential for characterization and warning to users. Therefore, in this work, we design a local VTEC prediction model based on the Long-Short Term Memory (LSTM) Neural Network by using the GPS data from 12 stations in Thailand. The results show that the root mean square error (RMSE) of LSTM loopback 24 together with the 120 hidden layers from all stations in 2008-2016 is the best model. The RMSE of the proposed model from the actual VTEC reach about 3.26 TECu, less than that from the IRI 2016 model at 6.5 TECu. In addition, the R-square values of the proposed model and the IRI 2016 model reach 78.33% and 63.7892%, respectively, during storm and quiet periods in 2020. The designed LSTM model is a promising method to predict VTEC in this region.