Machine learning in space physics
Advisor: František Němec (DSPS FMF CUNI)
Funding: Fully funded
The number of spacecraft and the amount of available data relevant for space physics studies increase tremendously in recent years. Automated ways of extraction of useful scientific information from big data volumes thus become a question of the utmost importance, which is the reason why machine learning algorithms are becoming more and more popular in the space physics field.
The aim of the study is to use machine learning techniques (neural networks, in particular) to analyze selected spacecraft data sets. A special focus will be on modeling significant plasma boundaries in space (bow shock, magnetopause, plasmapause). These are readily identifiable in the data as sudden specific changes of measured plasma parameters (density, flow velocity, magnetic field), and they are typically described by empirical models following prescribed shape/distance formulas. Machine learning techniques should allow not only to automatically identify the respective plasma regions and boundary crossings, but also to model the boundary locations themselves without the need for additional a priori assumptions. The evaluation of parameters controlling the shape and location of the derived model boundaries will then allow to improve our understanding of factors controlling the boundaries.
This is a project focused on a computer analysis of large existing satellite data sets. It involves programming and usage of dedicated machine learning packages, as well as a scientific interpretation of the dependences obtained.
 M. G. Kivelson, C. T. Russell: Introduction to Space Physics. University
Press, Cambridge, 1995.
 E. Camporeale, J. Johnson, S. Wing: Machine Learning Techniques for Space Weather. Elsevier, Radarweg, 2018.
 Q. Zong, P. Escoubet, D. Sibeck, G. Le, H. Zhang: Dayside Magnetosphere Interactions. AGU, Geophysical Monograph Series, 2020.