Affiliation:
1. Predictive Science Inc. San Diego CA USA
2. Community Coordinated Modeling Center Code 674, NASA GSFC Greenbelt MD USA
3. Austrian Space Weather Office Zentralanstalt für Meteorologie und Geodynamik Graz Austria
Abstract
AbstractAccurately predicting the z‐component of the interplanetary magnetic field, particularly during the passage of an interplanetary coronal mass ejection (ICME), is a crucial objective for space weather predictions. Currently, only a handful of techniques have been proposed and they remain limited in scope and accuracy. Recently, a robust machine learning technique was developed for predicting the minimum value of Bz within ICMEs based on a set of 42 “features,” that is, variables calculated from measured quantities upstream of the ICME and within its sheath region. In this study, we investigate these so‐called explanatory variables in more detail, focusing on those that were (a) statistically significant and (b) most important. We find that number density and magnetic field strength accounted for a large proportion of the variability. These features capture the degree to which the ICME compresses the ambient solar wind ahead. Intuitively, this makes sense: Energy made available to coronal mass ejections (CMEs) as they erupt is partitioned into magnetic and kinetic energy. Thus, more powerful CMEs are launched with larger flux‐rope fields (larger Bz), at greater speeds, resulting in more sheath compression (increased number density and total field strength).
Funder
National Aeronautics and Space Administration
Publisher
American Geophysical Union (AGU)
Reference37 articles.
1. Bailey R. L. Reiss M. Arge C. N. Möstl C. Owens M. J. Amerstorfer U. et al. (2020).Improving ambient solar wind model predictions with machine learning. arXiv preprint arXiv:2006.12835.
2. A Criterion for Stepwise Regression
3. Dalex: Explainers for complex predictive models in R;Biecek P.;Journal of Machine Learning Research,2018
4. Bobra M. G. &Couvidat S.(2014).Solar flare prediction using SDO/HMI vector magnetic field data with a machine‐learning algorithm. arXiv preprint arXiv:1411.1405.
5. Solar wind turbulence and shear: A superposed-epoch analysis of corotating interaction regions at 1 AU