Spatio-temporal analysis of TEC during solar activity periods using support vector machine

We propose a new method for spatio-temporal modeling of the ionospheric total electron content (TEC) at severe solar activity periods using a support vector machine (SVM). Using the observations from 37 GPS stations of the Iranian permanent GPS network (IPGN) in 2013–2014, the new model has been evaluated. Observations of 33 stations are used for training. Also, four stations are selected to test and evaluate the new model. In the 2 years under evaluation, the solar activity index is the maximum. The new model results are compared with the adaptive neuro-fuzzy inference system (ANFIS) and the artificial neural network (ANN) as other soft computing algorithms. Training of ANFIS and ANN is done with the error back-propagation (BP) method. The new model has been evaluated on different days, months, and seasons. To validate the results, the root mean square error (RMSE), correlation coefficients, residual histograms, relative errors, and standard deviations are used. Also, all the results are compared with the TEC of the global ionosphere map (GIM) as the traditional ionospheric model. For the SVM, GIM, ANFIS, and ANN models, the average RMSE at test stations are 3.642, 6.723, 4.844, and 5.011 TECU, respectively. Also, the average correlation coefficients of the four models evaluated at the test stations are calculated as 0.910, 0.742, 0.863, and 0.841, respectively. The maximum and minimum standard deviations for the SVM model obtained are 4.064 and 3.075 TECU, respectively. The standard deviations of the GIM, ANFIS, and ANN models are greater than the standard deviation of the SVM. Evaluations show that the SVM has high accuracy in modeling the temporal and spatial variations of the ionospheric TEC during periods of severe solar activity. Also, the analysis performed in precise point positioning (PPP) shows a higher accuracy of the new model compared to the other three models at the test stations. The results show that the SVM model is an accurate and reliable alternative to conventional ionospheric models for modeling the temporal and spatial variations of the TEC in the Iranian region.