Authors: Aniket Jivani (University of Michigan), Xun Huan (University of Michigan), Yang Chen (University of Michigan), Bart van der Holst (University of Michigan), Shasha Zou (University of Michigan), Zhenguang Huang (University of Michigan), Nishtha Sachdeva (University of Michigan), Daniel Iong (University of Michigan), Ward Manchester (University of Michigan), Gabor Toth (University of Michigan)

The Space Weather Modelling Framework (SWMF) offers efficient and flexible sun-to-earth simulations based on coupled first principles and/or empirical models. This encompasses computing the quiet solar wind, generating a coronal mass ejection (CME), propagating the CME through the heliosphere, and calculating the magnetospheric impact via geospace models. The predictions from these different steps are affected by uncertainty and variation of many model inputs and parameters, such as the Poynting flux emanating from the photosphere and driving and heating the solar wind. In this presentation, as part of the NextGen SWMF project funded by NSF, we perform uncertainty quantification (UQ) for the quiet solar wind simulations produced by the Alfven Wave Solar atmosphere Model (AWSoM). We first catalogue the various sources of uncertainty and their distributions, and then propagate the uncertainty to key predictive quantities of interest, the in-situ solar wind and magnetic field at 1 au, through space-filling designs of high-fidelity simulations. Using this dataset, we then build polynomial chaos surrogate models that offer a convenient route to global sensitivity analysis (GSA), which quantifies the contribution of each input parameter’s uncertainty towards the variability of the QoIs. The Sobol’ sensitivity indices calculated allow us to rank and retain only the most impactful parameters going forward, thereby achieving dimension-reduction of the stochastic space. We show these results for events corresponding to solar maximum and solar minimum conditions, as well as their variability with sample size. We will summarize these findings as well as their potential use in downstream models of the simulation, including CME propagation.