SpatioTECformer: Global ionospheric VTEC Forecasting via spatio-temporal modeling and adaptive driver fusion

The ionosphere plays a vital role in energy exchange between Earth and outer space, and its dynamic variations significantly impact satellite navigation and radio communication. Vertical Total Electron Content (VTEC), a key parameter that quantifies ionospheric electron density, is critical for space weather monitoring and Global Navigation Satellite System (GNSS) error correction. However, current forecasting models struggle to accurately represent spatial heterogeneity, capture long-range temporal dependencies, and remain robust under external disturbances such as geomagnetic storms. We propose SpatioTECformer, a novel model that integrates local and global spatiotemporal features to accurately forecast VTEC dynamics. It employs a Transformer encoder for long-sequence temporal modeling, an enhanced convolutional neural network (CNN) module for spatial feature extraction, and an adaptive feature fusion module to integrate solar wind and geomagnetic indices. Validation on Global Ionospheric Map (GIM) data from the Center for Orbit Determination in Europe (CODE) shows that SpatioTECformer achieves state-of-the-art performance, with RMSE and MAE of 1.80 and 1.23 TECU in 2014 (high solar activity), and improved values of 0.78 and 0.58 TECU in 2017 (low solar activity). The model exhibits superior robustness and predictive accuracy across global regions, particularly within the Equatorial Ionization Anomaly (EIA) zone and under geomagnetic disturbance conditions. The source code is publicly available at: https://github.com/jiawenchen1011/SpatioTECformer.