Detection of Pre-Seismic TEC Anomalies Using Isolation Forest: Virtual Station Analysis for Elazig and Maras Earthquakes

This study examines the relationship between Total Electron Content (TEC) anomalies in the ionosphere and seismic activity, with a particular focus on the Elazig (2020) and Maras (2023) earthquakes. An unsupervised machine learning approach, namely the Isolation Forest (iForest) algorithm, was employed to analyze TEC data from Global Navigation Satellite Systems (GNSS) stations situated in the regions affected by the seismic events. The analysis has been conducted over a two-month period, encompassing one month prior to and one month following each earthquake. A novel methodology is introduced, whereby TEC data from multiple stations is aggregated into virtual stations, with each station weighted according to its proximity to the earthquake epicenter. The findings reveal a distinctive pattern of TEC anomalies occurring especially 10–15 days before both earthquakes, followed by a reduction in fluctuations as the seismic event approaches. It is noteworthy that significant anomaly synchronization is observed across all virtual stations, which lends support to the hypothesis that TEC variations may serve as a pre-seismic indicator.