Predicting the Ionospheric Total Electron Content using Adaptive Regression Model

The Ionospheric Total Electron Content (TEC) is a major factor affecting the positional accuracy of satellite-based navigation systems. The Ionosphere causes a group delay in the signals received from the satellite resulting into a range error and hence reduced accuracy. The performance and accuracy of these systems can be improved by advanced estimation of TEC so that the delays are available along with the real measurements and suitable corrections can be incorporated to reduce the range error. As the ionosphere is not homogeneous and constantly changing, it also requires a dynamic correction in real-time. This paper proposes the use of regression method to predict the TEC using online machine learning model with new training data at regular intervals. The model is tested for location IISC, Bangalore, India near the magnetic equator for both quiet and disturbed days. The model is adaptive and chooses appropriate regression method to reduce the computation time and make it suitable for short term predictions in real time. Apriori estimation of TEC by twenty minutes is done using polynomials in linear regression. The accuracy of the predicted output is verified using Mean Squared Error, Mean Absolute error and the coefficient of determination. The reconstructed forecast data for the entire day is compared with the measured TEC from dual frequency receiver using the Kullback Leibler divergence method. The comparative evaluations suggest that the adaptive apriori estimation model is reliably accurate for forecasting the TEC values which will help in reducing the GPS error.