Speaker
Description
Auroras can intuitively reflect the energy coupling between the Sun and the Earth and are an excellent indicator for monitoring and predicting space weather effects. Establishing an auroral oval model driven by the geomagnetic index, predicted up to three days ahead, can effectively assess energy transfer in space. Based on the data spanning from 2005 to 2016 obtained from DMSP/SSUSI, we explore several machine learning algorithms, such as KNN, RF, and XGBoost, to construct an auroral oval prediction model. The input parameters of the models are the magnetic local time, magnetic latitude, and Kp index. The comparison of the three models shows that the XGBoost model performs better at predicting auroral oval locations and dealing with noise than the RF and KNN ones. The equatorward boundaries of the auroral oval predicted by the XGBoost model demonstrated a better performance on the test data set than the Kp-dependent empirical model, especially at geomagnetic disturbed conditions (Kp=5-6). In addition, the XGBoost model predicts that the magnetic latitude of the auroral oval's equatorward boundary decreases linearly with increasing Kp from 1 to 6, with a greater reduction on the duskside. Our comparisons indicate that while relying solely on the Kp index can effectively capture the variations in the nightside auroral oval, it has limited performance in predicting the dayside auroral oval, suggesting the need to incorporate additional parameters in the future. The Kp-driven ultraviolet emission auroral oval model developed in this study significantly contributes to the long-term advanced prediction of auroral oval distribution.
