Operational SYM‐H Forecasting With Confidence Intervals Using Deep Neural Networks

Armando Collado‐Villaverde,Pablo Muñoz,Consuelo Cid
DOI: https://doi.org/10.1029/2024sw004039
2024-10-30
Space Weather
Abstract:In this study, we develop a robust real‐time forecast system for the SYM‐H index using Deep Neural Networks and real‐time Solar Wind measurements along with Interplanetary Magnetic Field parameters. This system provides not only one‐off forecasts but also quantile‐based confidence intervals, offering a range within which the observed SYM‐H values are expected to fall, enhancing forecast reliability and usability. The model is tested both on historical level 2 science‐ready data and on preliminary observations, which are closer to the operational environment that the model is expected to work on, demonstrating the model's robustness and practical utility in real‐time scenarios. The integration of quantile forecasts into SYM‐H prediction models represents a significant advancement, providing decision‐makers with more accurate and trustworthy information to manage the hazard of geomagnetic storms.
meteorology & atmospheric sciences,geochemistry & geophysics,astronomy & astrophysics
What problem does this paper attempt to address?