A Model for Predicting Ionospheric Scintillation Intensity in High Latitude Region Based on VMD‐CNN‐BiLSTM‐Attention

Abstract Ionospheric scintillation degrades the reliability of satellite navigation, making accurate forecasting essential for high‐precision Global Navigation Satellite System (GNSS) services. In the high‐latitude region, the ionosphere exhibits complex behavior, and scintillation signals show inherent multi‐frequency mixing; therefore, existing methods often fall short in effective long‐term forecasting. To address these challenges, we propose a hybrid forecasting model integrating Variational Mode Decomposition (VMD), an improved Gray Wolf Optimizer (GWO), a Convolutional Neural Network (CNN), a Bidirectional Long Short‐Term Memory network (BiLSTM), and an attention mechanism. First, VMD decomposes the raw phase scintillation index (σ ϕ ) into multi‐scale components, including high‐frequency non‐stationary terms, low‐frequency trend terms, and residual terms; the improved GWO adaptively selects the VMD mode number and penalty factor. Building on this, a CNN extracts spatial features from the σ ϕ signal, such as local high‐frequency disturbances and instantaneous bursts, capturing the abrupt changes and localized fluctuation patterns of ionospheric scintillation; the BiLSTM captures bidirectional temporal dependencies, and the attention module emphasizes historical time points most informative for the current prediction. The model is validated using 2023 observations from the CHUC station of the Canadian High Arctic Ionospheric Network (CHAIN). For forecasting 15 time steps ahead, compared with existing ionospheric scintillation forecasting methods, the root mean square error is reduced by 63.01%, and the success rate for scintillation epochs reaches 83.62%. Moreover, the model demonstrates favorable transferability and robustness across multiple Arctic stations, supporting disturbance analysis and reliable high‐precision GNSS positioning under disturbed ionospheric conditions.