Evaluating the Role of AIA 193 Å Solar Images Beyond Historical Dst and Near‐Earth Solar Wind and Interplanetary Magnetic Field Information for 1–6 Hour Dst Forecasting

Abstract

Most short‐term disturbance storm time (Dst) index forecasting studies use historical Dst and near‐Earth solar wind and magnetic field information, while solar remote‐sensing imagery can provide upstream information related to geomagnetic‐storm drivers; however, its additional predictive value has not been systematically evaluated. We developed a multimodal deep‐learning framework integrating Atmospheric Imaging Assembly (AIA) 193 Å images with historical Dst and near‐Earth solar wind and magnetic field information, and systematically evaluated the role of solar imagery in 1–6 hr Dst forecasting. Building on this framework, our analysis provides the first systematic assessment of the role of AIA 193 Å images in multimodal short‐term Dst forecasting and demonstrates the feasibility of incorporating solar remote‐sensing information into such forecasts. The results show that historical Dst and near‐Earth solar wind and magnetic field information form the primary information basis for short‐term Dst forecasting, while increment prediction anchored to the current Dst can more effectively utilize recent Dst state information. In contrast, AIA images provide only limited auxiliary information about the solar coronal background. Furthermore, the Integrated Gradients attribution analysis shows that, compared with the full test set, the attribution to historical Dst and near‐Earth solar wind and magnetic field inputs is greater for disturbed samples and is more strongly concentrated at recent time steps. Meanwhile, the attribution to AIA images decreases and shows no clear preference among historical times. Overall, this study clarifies the role and limitations of solar imagery in multimodal short‐term Dst forecasting and provides a basis for future research on solar imagery.