Utilizing machine learning techniques along with GPS ionospheric TEC maps for potentially predicting earthquake events

The scientific use of ground and space-based remote sensing technology is inherently vital for studying different lithospheric-tropospheric-ionospheric coupling mechanisms, which are imperative for understanding geodynamic processes. Current remote sensing technologies operating at a wide range of frequencies, using either sound or electromagnetic emitted waves, have become a valuable tool for detecting and measuring signatures presumably associated with earthquake events. Over the past two decades, numerous studies have been presenting promising results related to natural hazards mitigation, especially for earthquake precursors, while other studies have been refuting them. While highly impacting for geodynamic processes the controversy around precursors that may precede earthquakes yet remains significant. Thus, predicting where and when natural hazard event such as earthquake is likely to occur in a specific region of interest still remains a key challenging task in geo-sciences related research. Recently, it has been discovered that natural hazard signatures associated with strong earthquakes appear not only in the lithosphere, but also in the troposphere and ionosphere. Both ground and space-based remote sensing techniques can be used to detect early warning signals from places where stresses build up deep in the Earth’s crust and may lead to a catastrophic earthquake. Here, we propose to implement a machine learning Support Vector Machine (SVM) technique, applied with GPS ionospheric Total Electron Content (TEC) pre-processed time series estimations, extracted from global ionospheric TEC maps, to evaluate any potential precursory caused by the earthquake and is manifested as ionospheric TEC anomaly. Each TEC time series data was geographically extracted around the earthquake epicenter and calculated by weighted average of the four closest points to evaluate any potential influence caused by the earthquake. After filtering and screening our data from any solar or geomagnetic influence at different time scales, our results indicate that with large earthquakes (> 6 [Mw]), there is a potentially high probability of gaining true negative prediction with accuracy of 85.7% as well as true positive prediction accuracy of 80%. Our suggested method has been also tested with different skill scores such as Accuracy (0.8285), precision (0.85), recall (0.8), Heidke Skill Score (0.657) and Tue Skill Statistics (0.657).