Deep Learning Based UNet for Global Ionospheric Forecasting

Vertical Total Electron Content (VTEC) in the ionosphere is a crucial parameter for describing dynamic changes. Due to limitations in observational techniques, real-time global distribution of ionospheric VTEC is challenging to obtain. Therefore, short-term forecasting of VTEC holds significant theoretical and practical importance. In this study, we utilize global ionospheric VTEC data and employ aU-Net network to capture the spatiotemporal characteristics of global VTEC, predicting the distribution for the next 24 hours. We analyze and evaluate the U-Net model and three other models in 2015 year and 2020 year. Under varying ionospheric activity conditions, the forecasting accuracy of the U-Net network is notably superior to the other three models. In 2015, its root mean square error (RMSE) is 4.045 TECU, and in 2020, it is 1.519 TECU.