Regional application of generalized regression neural network in ionosphere spatio-temporal modeling and forecasting

We propose using the generalized regression neural network (GRNN) method for spatio-temporal modeling of ionosphere total electron content (TEC). The GRNN model uses radial basis functions in the pattern layer. Therefore, the accuracy and convergence speed to the optimal solution of this model are higher compared to the other machine learning models. The efficiency of the new model has been evaluated using observations of 30 global navigation satellite system (GNSS) stations in central Europe at 2015. It should be noted that the training of the GRNN model is done using the latitude and longitude of GNSS station, day of year, hours, AP, KP and DST geomagnetic indices and solar activity index (F10.7). Also, the vertical TEC corresponding to these input parameters is desirable output. The results of the new model have been compared with the results of the artificial neural network, adaptive neuro-fuzzy inference system, support vector regression, ordinary Kriging, global ionosphere map and the international reference ionosphere 2016 (IRI2016) empirical model as well as precise point positioning (PPP) method. The obtained results show that in both high and low geomagnetic and solar activities, the GRNN model has a higher accuracy with respect to the other models. The analysis of the PPP method shows an improvement of 37 mm in the coordinate components using GRNN model. The results show that the GRNN model can be considered as an alternative to global and empirical ionosphere models. The GRNN model is a high-precision regional ionosphere model.