Short‐Term Prediction of the Dst Index and Estimation of Efficient Uncertainty Using a Hybrid Deep Learning Network

Abstract In this study, we developed a novel deep learning model to predict the disturbance storm time (Dst) index 1–4 hr ahead. We also employed the Monte Carlo (MC) dropout technique to estimate the uncertainty and provide the prediction interval by introducing a recalibration factor. The proposed model achieves excellent scalability by extracting representative embeddings from the Dst index time series through an encoder‐decoder framework and integrating these with external solar wind parameters via a prediction network. We utilized magnetic storm data from 1998 to 2018 to evaluate the performance of the prediction model. Experimental results indicate that the proposed model reduces root mean square errors (RMSE) and improves the coefficient of determination (R 2 ) compared to existing methods. Quantitative uncertainty analysis demonstrates that the prediction interval is reliable in most cases. The percent distribution across test storms by comparing the uncertainty‐based prediction of the minimum peak with the mean prediction indicates the probabilistic forecast model is competitive.