Harnessing ML Methodologies to Forecast TEC and Classify Ionospheric Scintillations

Ionospheric scintillation, a common occurrence in equatorial, polar, and auroral regions, poses challenges to Global Navigation Satellite System (GNSS) performance. Over the past two decades, GNSS Radio Occultation (RO) plays a vital role in supplying top-tier atmospheric data that seamlessly integrates with Numerical Weather Prediction (NWP) models and contributes to Meteorological research. However, scintillations impact measurements in satellite-to-satellite GNSS-RO geometry, impacted by solar flares, seasons, geomagnetic activity, geographical positions, and local time. These disturbances introduce errors in positioning and degrade GNSS performance, making detection vital for accuracy. Traditional algorithms often lack sensitivity, especially in identifying strong scintillation due to its sporadic nature, resulting in imbalances in datasets. To address this, we employed diverse machine learning algorithms to detect varying degrees of ionospheric scintillation, aiming to improve accuracy by rectifying dataset imbalances. The article explores Total Electron Content (TEC) and Ionospheric Phase Scintillation identification, TEC prediction using regressors like Gated Recurrent Unit (GRU), Decision Tree Regressor, and Random Forest Regressor, and ionospheric phase scintillation classification using classifiers like Support Vector Machines (SVM), Decision Tree, and Random Forest. The comprehensive comparative analysis evaluates regressors and classifiers using standard metrics such as Mean Squared Error, R2 Score, Accuracy, Recall, Precision, F1 score, and confusion matrix.