Automated ionospheric anomaly detection during a geomagnetic storm using a hybrid wavelet isolation forest approach with explainable artificial intelligence

Abstract

Detection of anomalies in the ionosphere is essential for ensuring reliable timing, navigation, and positioning services of the Global Navigation Satellite System (GNSS). This study proposes a machine learning approach to anomaly detection during severe geomagnetic storms based on the Wavelet Transform with the Isolation Forest algorithm. Explainable Artificial Intelligence (XAI) methods, specifically Shapley Additive exPlanations (SHAP), are employed to interpret feature contributions and improve model transparency. Total Electron Content (TEC) time-series data derived from the ionospheric dataset of the geomagnetic storm event for time-frequency analysis. Then, an Isolation Forest model is applied to identify anomalous patterns based on mean, variance, standard deviation, kurtosis, entropy, and energy to represent the temporal dynamics of the ionosphere. Results showed that the model identified 3% of anomalies in the ionosphere with an overall detection accuracy of 95%. In addition, Root Mean Square Error (RMSE) demonstrated reasonable agreement of 20% between the proposed model and the baseline Z-score. Among the wavelet features, the entropy and energy are found to be the most important for anomaly detection. Overall, the proposed model provides a reliable and interpretable solution for unsupervised ionospheric anomaly detection, with potential applications in GNSS monitoring and space weather research.