Deep Learning—Based Airglow Fusion and Forecasting for Equatorial Plasma Bubble Monitoring and GNSS Scintillation Prediction

Abstract

Equatorial plasma bubbles (EPBs) are major low‐latitude ionospheric irregularities that can cause scintillation and degrade Global Navigation Satellite System (GNSS) performance. Existing amplitude‐scintillation prediction methods primarily use historical GNSS observations and ionospheric parameters, but rarely incorporate the evolving spatial morphology of EPBs. This study develops an airglow‐enhanced framework for regional EPB monitoring, short‐term airglow forecasting, and minute‐resolution scintillation prediction. Images from four 630.0 nm all‐sky imagers were projected onto a common geographic grid and fused using coordinate matching, Gaussian weighting, and a stacked autoencoder (SAE). A residual U‐Net then forecast the short‐term evolution of the fused airglow field. Airglow depletion, gradient, texture, and contrast features around each GNSS ionospheric pierce point were combined with historical S4, total electron content, temporal variables, and satellite geometry in a long short‐term memory model. The fusion improved the spatial continuity of multi‐station observations. At a lead time of 15 min, the airglow forecast achieved a correlation coefficient of 0.755 and an MS‐SSIM of 0.746, while the main EPB depletion structures remained identifiable. Compared with the airglow‐free model, the airglow‐enhanced model better reproduced the timing and intensity variations of scintillation events, particularly during rapid local changes. For two relatively smooth events, the predicted S4 values achieved determination coefficients of 0.916 and 0.931. These results show that airglow‐derived EPB structural information provides complementary spatial constraints for minute‐resolution S4 prediction when preceding S4 observations are available. Extending the approach to locations without reliable S4 measurements is a potential future application that requires further validation.