Low-latitude storm time ionospheric predictions using support vector machines

The electromagnetic drift plays an important role in low-latitude storm time ionospheric dynamics. In this study we attempt to utilize the electric field data into ionospheric predictions by using support vector machine (SVM), a promising algorithm for small-sample nonlinear regressions. Taking the disturbance electric field data as input, different SVMs have been trained for three seasonal bins at two stations near the north crest of the Equatorial Ionization Anomaly (EIA). Eighteen storm events are used to check out their predicting abilities. The results show fairly good agreement between the predictions and observations. Compared with STORM, a widely used empirical correlation model, the SVM method brings a relative improvement of 23% for these testing events. Based on this study we argue that the SVM method can improve the storm time ionospheric predictions.