Auroral Image Analysis: From Raw Data to Modelling and Understanding

Modern magnetospheric and ionospheric research utilises images from various sources such as ground-based all-sky imagers and more global satellite images. Although these data convey information relating to intensity, boundaries and auroral type, only the intensity and boundaries are dealt with quantitatively in typical studies. Since auroral type (e.g., patchy, arcs, omega bands, etc.) is closely related to magnetosphere-ionosphere coupling, the lack of ability to deal with type is a significant weakness. In other words, features that auroral researchers identify with ease by eye cannot easily be quantitatively dealt with. Such quantitative treatment would, for example, allow for more realistic truthing of global simulations. We are developing computer vision techniques to address this shortcoming. In recent work, we have developed shape classifiers that have been successful in automatically classifying large image data sets. In this paper, we review this recent work, and discuss how these new data mining techniques could be utilized to improve the usefulness of global physics-based models.