Description
Predicting solar eruptions from physics-based simulations is a challenging uncertainty-quantification problem: the governing magnetohydrodynamic (MHD) evolution is nonlinear, the coronal magnetic field is poorly constrained by observations, and the various simulation codes available use distinct numerical, physical, and boundary-condition assumptions. In this talk I will present statistical methods, borrowed from the terrestrial weather forecasting literature, for constructing and interpreting ensembles of deterministic solar-eruption simulations. I will discuss ensemble generation, ensemble calibration, and ensemble-based prediction. The first part of the presentation will describe the ensemble model statistics and post-processing algorithms used to convert a finite set of deterministic simulation outputs into probabilistic forecasts. Particular attention will be given to the choice of forecast variables, the treatment of time-dependent diagnostic quantities, and the importance of retaining trajectory-level information rather than reducing each simulation to a scalar statistic. The second part of the presentation will demonstrate ensemble construction for our motivating example of prediction of solar eruptions. We will demonstrate how initial-condition uncertainty, boundary-driving uncertainty, and numerical-method uncertainty are represented, and how these choices control the interpretation of ensemble spread. Finally, we will explore how the same underlying ensemble can produce materially different probabilistic predictions depending on the statistical algorithm used to post-process it. The broader aim of this work is to establish statistically defensible ensemble modelling principles for data-driven solar-eruption prediction, analogous in spirit to terrestrial weather forecasting but adapted to the constraints of computational solar physics.
| I am the presenting author | Yes |
|---|