Ensemble regression for GNSS TEC prediction using SF model fusion

Abstract Ionospheric Total Electron Content (TEC) forecasting during geomagnetic storms is crucial for reliable Global Navigation Satellite System (GNSS) operations. Traditional single-frequency ionospheric broadcast correction models (Klobuchar and NeQuick-G models) are likely to have large ionospheric prediction errors under disturbed conditions, particularly at various latitudinal regions. In this study, a two-stage machine learning framework that integrates Light Gradient Boosting Machine (LightGBM) and an ensemble regression approach to improve storm-time TEC prediction. First stage, LightGBM is trained on the GNSS TEC data using the geomagnetic, solar, and temporal indices as inputs. In the second stage, the predictions from Klobuchar and NeQuick-G ionospheric model are integrated using traditional, optimized, and Regression based ensemble techniques. The Model performance was tested during the geomagnetic storm of October 2024 (DOY 283–287) at seven IGS stations spanning high, mid, and low-latitude regions. Results show that the ionospheric broadcast models, NeQuick-G and Klobuchar both either overestimated or underestimated storm-time TEC, while the proposed Regression-based ensemble consistently achieved better accuracy, with RMSE values reduced by more than 50 % compared to standalone ionospheric prediction TEC models. These findings demonstrate that combination of data-driven methods with physics-based models as a powerful and scalable method for real-time ionospheric modelling and GNSS position error mitigation during quiet and disturbed space weather conditions.