Journal Article

Storm‐Time Dst Forecast: An Innovative Approach

Zhang, Yongliang, Paxton, Larry J

2026

Overview

One of the most persistent challenges in the space weather field is predicting the storm‐time response of the geospace without knowing the predicted drivers in the solar wind. Here, a new pattern recognition algorithm is developed to predict storm‐time Dst index from 1 hr to ∼4.5 days into the future. Storm‐time Dst patterns (or reference Dst) are obtained by a superposed epoch analysis of normalized Dst where storm‐time Dst is scaled or normalized so the minimum of the scaled Dst is fixed at −100 nT. About 148 isolated storms (minimum Dst ≤ −35 nT) between 2000 and 2014 were used. Two types of storms are identified with a fast (∼12 hr, Type‐1) and slow (∼30 hr, Type‐2) growth phase. Both Types show nearly identical recovery phases (∼5 days) and are likely driven by coronal mass ejection and corotating interaction region, respectively. This establishes the reference Dst profile. During the initial growth phase or any other phase of an isolated storm, Root Mean Squared Errors (RMSE) are calculated between the observed Dst (e.g., 5 data points) and a reference Dst. The Dst profile with the minimum RMSE serves as a forecast of the storm over the next 1 hr to ∼4.5 days. The algorithm has been tested for a few isolated storms and shows a good agreement between predicted and observed Dst with correlation coefficients up to ∼0.9 and RMSE as low as ∼5–10 nT. Caveats and a few future approvements are also discussed.

• Publisher: John Wiley & Sons, Inc
• Subject: Algorithms / Coronal mass ejection / Corotating Interaction Regions (CIR) / Correlation coefficient / Correlation coefficients / Data points / DST Index / Machine learning / Magnetic fields / Neural networks / Pattern recognition / Solar wind / Space weather / Storm forecasting / Storms / Time response