Speaker
Description
Geomagnetic storms are large disruptions of the magnetosphere. These events can interfere with satellites, communication systems, and power grids, causing significant technological and economic damage. Current forecasting models utilise L1 satellite data, constraining lead time to a few hours', often insufficient for effective mitigation. Accurate long-lead forecasts would help protect infrastructure and ensure operational continuity.
We investigate how to extend the lead times of geomagnetic storm forecasts by using solar data. Associated spatial and propagation uncertainties of solar data are captured with a solar-wind ensemble of the computationally efficient one-dimensional HUXt numerical model, rather than a 3D-MHD based model. HUXt allows us to simulate significantly more solar wind profiles than 3D-MHD models. This work builds on a previous study on binary classification of geomagnetic storms, to now providing regression based forecasts targeting the open-ended Hp30 global geomagnetic index, which offers higher temporal resolution (30 minutes) compared to the more commonly used Kp index (3 hours), enabling finer-scale forecast evaluation.
The HUXt solar-wind ensemble is processed through a series of regression based models trained on individual solar wind profiles, giving us an array of Hp30 forecasts. The ensemble forecasts are aggregated to produce a final prediction, with uncertainty estimated from the ensemble spread and the historical correlation between observed solar wind (OMNI) and the simulated profiles. Using 30 years of historical data, model performance is analysed over a variety of lead-times from 1 to 36 hours, and performance over a variety of storm intensities is assessed.
In this work, we evaluate multiple machine learning architectures and input variables for regression based forecasting, and focus particularly on reducing the Mean Absolute Error (MAE), and R-squared of our forecasts. Overall, we show the predictive capabilities of coupling computationally fast numerical modelling of the solar wind with machine learning algorithms to increase the lead time of Hp30 forecasting.
| Do you plan to attend in-person or online? | In-person |
|---|
