Ionospheric TEC prediction using SVM during fifteen X-class solar flares occurred in the 25th solar cycle and comparison with LSTM and IRI-Plas 2020

Solar Flares (SFs) are powerful radiation bursts from the Sun’s atmosphere, primarily occurring in active regions, caused by the acceleration of charged particles and magnetic field reconfigurations in plasma. SFs produce sudden ionospheric Total Electron Content (TEC) fluctuations, directly affecting radio wave propagation between satellites and Earth, usually degrading communication and navigation satellite-based system performances. In this research, we applied Support Vector Machine (SVM) machine learning algorithm to predict TEC changes during SF events. SVM model is highly regarded for its strong handling of both linear and non-linear patterns of data and is thus generally appropriate for modeling complex ionospheric responses. The research is based on high-resolution TEC data retrieved from the IISC and BAKO stations, as well as co-relevant SF parameters, for a period of thirteen months from October 2023 to October 2024. Here, the addition of the BAKO station for analyzing three SF shows the universality of the SVM model in predicting the changes in TEC due to SF. The dataset was pre-processed to remove outliers and down-sampled to match the temporal resolution of the International Reference Ionosphere plasmasphere (IRI-Plas) 2020 model. This cleaned dataset was used for training the SVM model with linear and non-linear kernels, choosing the best method with GridSearchCV. The model was tested and cross-validated against Long Short-Term Memory (LSTM) neural networks as well as the IRI-Plas 2020 empirical model. Moreover, this research examines the TEC behavior during the occurrence of SF under both non-geomagnetic storm times and geomagnetic storm (especially March and May 2024) times. The performance of the SVM model was measured in terms of Root Mean Square Error (RMSE), Normalized RMSE (NRMSE), Mean Bias Deviation (MBD), and Relative Length Error (RLE). For the considered fifteen X-class SF events of 2024 at IISC Bangalore Station, the SVM model reported average RMSE of 5.742 TECU, NRMSE of 11.173%, MBD of 4.910 TECU, and RLE of 13.12%, outperforming the IRI-Plas 2020 model. This research shows that the SVM model is efficient and also reliable at forecasting TEC variations at times of intense SF activity. This work adds to the improvement of space weather forecasting, with the practical implications on enhancing positional accuracy and with operational stability in satellite communication and navigation systems during space weather events.