A novel hybrid Machine learning model to forecast ionospheric TEC over Low-latitude GNSS stations

Understanding the ionospheric variability caused by solar and geomagnetic space weather can be achieved using Total Electron Content (TEC) parameter. Developing a novel artificial intelligence-based machine learning model that forecast TEC parameter would primarily benefit Global Positioning System (GPS) users. This paper holds one such hybrid empirical exploits nonlinear autoregressive neural network with external input (NARX) for hour in advance forecasting over Hyderabad (17.45°N,78.47°E) Bengaluru (12.95°N, 77.68°E) GPS stations. The 11-year data spans from 2009 to 2019 corresponding 24th cycle obtained these stations operated International GNSS Service (IGS) center. performance of this multiple model, which accepts TEC, Ap index, Solar flux time day, geographical coordinates, others, have been evaluated during both High Activity (HSA) year 2014 Low (LSA) period 2019. A comprehensive evaluation proposed has carried out comparison Reference Ionosphere 2016 (IRI-2016) NeQuick-G Auto Regressive Moving Average (ARMA), Neural Network (NN), combined NN ARMA models. Results reveal ML could well. model's root-mean-square error (RMSE) ranges 0.5–1 TECU LSA 1.63–2.2 HSA period. Maximum Absolute Percentage Error (MAPE) is between 1.6% 4.5%, 0.99 correlation coefficient (R 2 ) value consistently. Thus, provides promising results.