Machine learning based method for dynamic forecasting of total electron content in the equatorial ionosphere

The total electron content in the ionosphere is a vital parameter for the users of the Global Navigation Satellite System (GNSS) as it causes a delay in the satellite signal propagating through it, which in turn degrades the positional accuracy of the receiver. Thus, improving the GNSS positioning requires precise ionospheric Total Electron Content (TEC) prediction, especially in the equatorial region where complex electrodynamics and erratic space weather events introduce substantial short-term variability. Although deep learning techniques have shown promise, they frequently rely heavily on large historical datasets, are computationally demanding, and are not interpretable. In this work, we propose a Multiclass Classifier Short-Term Dynamic Prediction Model (MSTDM) that uses reliable and interpretable machine learning techniques to forecast Vertical TEC (VTEC) 30 min ahead of time. The model optimizes the training set using a threshold-based learning algorithm to identify nonrepetitive and relevant VTEC patterns from recent data. The model uses polynomial interpolation to impute missing values, and a sliding window method is used to extract temporal features, which are then further refined through statistical feature selection. The continuous VTEC values are first discretized into distinct classes, after which Support Vector Machines (SVM) and Random Forests (RF) were employed for supervised classification. Different feature selection techniques were used. Prediction accuracy for SVM and RF with Recursive Feature Elimination (RFE) demonstrated the best results during geomagnetic storm days. The results varied from 85 to 91 % and 87–92 %, respectively. The RF-RFE model outperformed other configurations with 99 % training accuracy and 96 % test accuracy. Thus, this method provided a high-performing, interpretable, and computationally efficient solution for short-term VTEC forecasting.