Analysis of ionospheric TEC anomalies and prediction using XGBoost, GRU and RF algorithms associated with 2024 Japan Earthquakes (Mw > 7.0)

The ionosphere is a portion of Earth’s upper atmosphere, which is ionised by solar and cosmic radiation. It includes parts of the mesosphere, thermosphere and exosphere and is characterised by the presence of free electrons and ions. The Total Electron Content (TEC) denotes the total number of electrons along a signal path through the ionosphere. Variation in TEC has been widely studied in relation to Space-weather effects and Lithosphere-Atmosphere-Ionosphere Coupling (LAIC), particularly in seismically active regions. This research focuses on analysing Machine Learning (ML) models, including a Gated Recurrent Unit (GRU) trained with Root Mean Square Propagation (RMSProp), Random Forest (RF) model optimised using Particle Swarm Optimiser (PSO), XGBoost model optimised using Genetic Algorithm (GA), for predicting TEC values during earthquakes, and the results were compared with the NeQuick model. This research investigates the ionospheric response to earthquakes in Japan, a region prone to earthquakes and volcanic eruptions. The three ML models are designed to predict TEC variations during seismic activity using solar and geomagnetic activity data like Solar wind (Sw), Solar flux index (F10.7), Disturbance Storm Time Index (Dst) and Ap index (Ap), along with TEC data collected from the USUD, AIRA & ISHI stations situated in Japan. The earthquakes considered in this research are the 2024 Suzu earthquake, the 2024 Miyazaki earthquake and the 2024 Ishigaki earthquake. Statistical metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Symmetric Mean Absolute Percentage Error (SMAPE) were used to evaluate the predictive capabilities of the four models. Cross-validation was performed on the GIM data for the two best ML models, and the results suggest that XGBoost has better accuracy than the other models. The comparative results indicate that all models can capture TEC variations associated with seismic events. Across all three earthquakes, XGBoost consistently outperformed the other machine learning models by achieving lower values of RMSE and MAE. Among the remaining approaches, the RF model outperformed the GRU model. Overall, these results suggest that XGBoost with a GA predicts TEC anomalies during earthquakes more effectively than the other models considered.