Machine learning approach for ionospheric scintillation prediction on ROTI parameter over the African Region during solar cycle 24

Abstract

Scintillation is a dynamic phenomenon of the earth's ionosphere that adversely affects satellite-based communication and navigation systems. It is characterized by rapid fluctuations in phase and amplitude of trans-ionospheric radio waves, posing immense risks to systems that operate at radio frequencies, such as the Global Positioning System (GPS). In this regard, harnessing modern technology and state-of-the-art datasets to predict their occurrence is crucial for mitigating their effects. This paper presents a neural network approach to predict ionospheric scintillation using datasets obtained from distributed geodetic receivers in the African region. The motivation for this work is to develop a model backed by an extensive database for scintillation prediction over the region. Eleven years of data from the stations were obtained for magnetically quiet days and the Rate of TEC index (ROTI) was computed as a proxy for scintillation. The model development was backed by data from the ascending to the maximum phases of solar cycle 24 while the years in the descending phase were used for model validation and prediction. Using the solar flux (F10.7 cm), elevation and critical frequency (FoF2) as physical model parameters, the model achieved a prediction accuracy of about 70 %. A control experiment using the wavelet features increased the model's accuracy to about 91 % during the testing phase with an 86 % prediction accuracy. The model was extensively evaluated using metrics such as the Root Mean Square Error (RMSE), statistics from the Residuals and the Wavelet Coherence Analysis (WCA) technique. The standard deviation was used alongside the RMSE to gauge dispersion and ascertain model stability. The developed model demonstrated the ability to reconstruct the ROTI with low errors.