Self-organized maps in scientific data analysis

Abstract

The Thermal Ion Dynamics Experiment (TIDE) investigates low energy (0 - 500 eV) plasma in the Earth's magnetosphere, especially in the polar regions. It is part of NASA's larger POLAR mission. Shortly after the POLAR spacecraft was in orbit, TIDE's mass spectrometer functionality failed. However, data from other instruments can be used to create energy vs. time spectrograms. The number of peaks in these spectrograms relates to the composition of the plasma, so determining the number of peaks in the spectrogram effectively regains the mass spectrometer functionality. Kohonen self-organized maps (SOMs,) a type of neural network, are particularly suited to this problem due to the amount of data that needs to be analyzed and the algorithm's ability to find patterns within data. The algorithm leads to clustering of similar data points on the map. Ultimately, the location of the input data point on the map allows for determination of how many peaks the data point contains, and thus the composition of the plasma at that time. The SOM correctly classified 99% of the input data, making it a viable solution to the problem. Further research is planned, namely the possibility of extending this concept to investigate energetic neural atom (ENA) images in order to determine the source of these atoms.