Total electron content forecasting with neural networks during intense geomagnetic storms of the solar maximum and moderate years of solar cycle 24 in the low latitude Indian region

Abstract

A geomagnetic storm is a brief disruption in the magnetosphere of the Earth that depresses the magnetic field. This activity causes variations in the ionospheric Total Electron Content (TEC) and other physical properties because it is connected to solar coronal mass ejections, coronal holes, or solar flares. A modeling method called a Neural Network (NN) displays nonlinear characteristics which include physical variables. The present research aims to develop a feed-forward back-propagation neural network-based forecasting approach to ionospheric TEC throughout a geomagnetic storm. We have contrasted the TEC gained through neural network estimation (NN TEC), GPS TEC, and NeQuick TEC via the IRI model during four intense geomagnetic storms on June 23, 2015 (Kp index of 8.33 and Dst index −198 nT), December 20, 2015 (Kp index of 6.67 and Dst index −166 nT), May 28, 2017 (Kp index of 7.0 and Dst index −125 nT) and August 26, 2018 (Kp index of 7.67 and Dst index −175 nT) across two Indian stations Bangalore (Geog. latitude 12∘, 58’N, longitude 77∘, 35’E), and Lucknow (Geog. latitude 26∘, 50’N, longitude 80∘, 55’E) and analysis have been made. It is observed that modeled NN TEC values, and GPS TEC values, match well during the entire duration of all the geomagnetic storms at both stations. Whereas, except for the geomagnetic storm beginning phase, NeQuick TEC has been underestimated during the entire event across Bangalore and Lucknow during the selected intense geomagnetic storms. It is noted that Lucknow’s maximum deviation of NN and GPS TEC is higher than Bangalore’s. The greatest correlation coefficient 0.99, was found and shows that the modeled NN TEC values, and GPS TEC values, are in excellent agreement. In order to assess the preciseness of the model’s output, Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) values were used to compare the projected TEC relative to the measured GPS-TEC and NeQuick TEC model output. The RMSE value of the NN model is shown to be lower throughout the storm’s initial and final phases, but greater throughout its main phase. Additionally, it is found that throughout the intense geomagnetic storms, Lucknow’s RMSE and MAPE values were higher than Bangalore’s. As a result, when compared to the NeQuick approach, the NN method demonstrated greater accuracy and TEC estimations. The present research concludes that neither the NN nor the NeQuick models can accurately forecast ionospheric TEC for an equatorial station (Bangalore) and an equatorial ionization anomaly (EIA) station in Lucknow throughout intense geomagnetic storms. However, during the recovery period (post-storm), the precision of the forecast increased. © 2023, The Author(s), under exclusive licence to Springer Nature B.V.