Forecasting One-day Global Ionospheric TEC Maps based on a Modified 3D Convolution U-Net Incorporating Fused Index Features

Ionospheric Total Electron Content (TEC) serves as a fundamental parameter for characterizing ionospheric morphology. Ionospheric TEC exhibits irregular disturbances driven by solar and geomagnetic activities. TEC forecasting products enhance GNSS positioning precision through error correction and space weather assessment, consequently improving satellite navigation system reliability and space weather warning systems. This paper proposes a modified 3D convolutional U-Net architecture incorporating fused index features for forecasting 1-day global ionospheric TEC maps. The input includes the previous day’s TEC maps and corresponding solar and geomagnetic indices, such as F10.7, SSN, Vsw, IMF Bz, Dst, and Kp indices. Our model comprises an MLP-based fused feature generation module for solar and geomagnetic indices and a 3D convolutional spatiotemporal forecasting module. The primary contribution of this work involves the dimensional expansion of 1D solar and geomagnetic indices to achieve spatiotemporal alignment with 2D TEC grid maps. We benchmark our model against the C1PG model, evaluating prediction performance under different geomagnetic storm intensities. The results demonstrate that our fused index generation module significantly enhances 1-day TEC prediction accuracy during storm periods, particularly in low-latitude regions where our model better captures large-scale TEC anomalies. Compared to the C1PG model, our model reduces forecast errors by 35%-40% at low latitudes. The feature fusion approach provides new insights into the spatiotemporal TEC modeling field.