Multivariate Supervised Classification for Instabilities at Saturn: Comparison of Methods for Automated Event Detection in Magnetospheres
Shortened Abstract: In 2004 the Cassini mission arrived at Saturn. For the next 13 years the mission collected large amounts of data, resulting in a highly sampled magnetosphere of Saturn. Saturn is now the second most observed magnetosphere after that of Earth now allowing opportune applications of large scale statistical methods.
In this work we will first present a previous effort to both identify and characterize interchange injections from high-energy ion intensities using the methods commonly employed in supervised classification tasks merged with required physical assumptions of the Saturnian environment. This work created a unique and reproducible list of events by combining predictive data analytics with background plasma environment characterization, uniquely allowing for subsequent statistical analysis on these events. This represents the first automated event detection algorithm implemented to detect such events. We then discuss issues in automated data analysis within dynamic planetary system including non-equal sampling, extreme temporal and spatial variability, and missing or invalid values. We focus on solutions to allow for the applications of classification tasks and automated event detection methods to benefit from the new surge of planetary space physics data now available to characterize the outer planets.
Azari, A. R., Lockhart, J. W., Liemohn, M. W., and Jia, X. “Multivariate Supervised Classification for Instabilities at Saturn: Comparison of Methods for Automated Event Detection in Magnetospheres”, Machine Learning in Heliophysics. September, 2019. Amsterdam, The Netherlands.
I would like to thank the conference organizers and the Thomas Metcalf Award Committee for their support in my attendance to this meeting.
Full text at ml-helio.io.