Innovative Pro-Boost Multi-Fidelity Machine Learning Method on Extreme Space Weather Events

We introduce an innovative ensemble machine learning (ML) method
designed for forecasting extreme space weather events, called proboost
(probabilistic boosting). The ensemble technique is a powerful tool
that enhances prediction accuracy and has been widely applied to
physical model predictions. However, implementing ensemble tech-
niques on ML methods faces challenges as ML models typically do not
provide well-calibrated uncertainty estimates for predictions. To ad-
dress this issue, we have developed an uncertainty quantification (UQ)
based neural networks method, which enables simultaneous forecast-
ing of uncertainty for model predictions.
In this study, we apply the proposed method to three distinct space
weather applications:
• A one-to-six-hour lead-time model predicting the value of Dis-
turbance storm time (Dst) using solar wind data;
• A geoelectric field model with multi-hour lead time, incorporat-
ing solar wind and SuperMag data;
• Ambient solar wind velocity forecast, up to 5 hours ahead.
It is important to emphasize that the proboost technology is model-
agnostic and thus can be applied to other forecasting applications.