GCBi-ScintNet: Predicting GNSS Positioning Errors under Ionospheric Scintillation with a GA-CNN-BiLSTM Hybrid Model

Ionospheric scintillation during periods of high solar activity significantly degrades the accuracy and reliability of Global Navigation Satellite System (GNSS) positioning. While prior research has largely focused on predicting long-term positional variations under relatively quiet conditions or scintillation indices separately, the direct forecasting of positioning errors under strong ionospheric disturbances remains scarcely explored. In this study, we develop a hybrid machine learning model integrating Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Genetic Algorithm (GA) to predict scintillation-induced 3D positioning errors. Using the Rate of Total Electron Content Index (ROTI) values—computed from 30-second GNSS data—as well as satellite numbers, Geometric Dilution of Precision (GDOP), and timestamps, the model was trained on a high-latitude station (LERI, 2023) and tested across multiple stations in 2023 and 2024. Results show strong agreement between predicted and actual errors, with Pearson Correlation Coefficients (PCC) reaching 0.816 at BOAV in 2024, respectively. The model reduced Root Mean Squre Error (RMSE) by a maximum of 50.7% at NYA1 in 2024. Performance varied by receiver type and latitude: stations sharing the same receiver type as the training station showed higher prediction accuracy, and low-latitude sites showed more days with very high correlation (PCC ≥ 0.9). These findings highlight the model’s ability to capture error trends across diverse ionospheric conditions, while also underscoring the impact of data source and geographic location on prediction accuracy.