Towards an Antarctic meteorite hotspot map

• Main Author: Tollenaar, Veronica (author)
• Other Authors: Zekollari, Harry (mentor), Lhermitte, Stef (mentor), Tax, David (graduation committee), Lindenbergh, Roderik (graduation committee), Delft University of Technology (degree granting institution)
• Format: Master Thesis
• Language: English
• Published: 2020
• Subjects: Antarctica; meteorites; Remote sensing; Big Data Analysis; Machine Learning; Antarctic; Antarc*; Ice Sheet
• Online Access: http://resolver.tudelft.nl/uuid:a1b5b764-0ba0-42d1-8918-3e681a24a4a1

Meteorites contain information on the formation and evolution of the Solar System. Antarctica is the most productive region for collecting meteorites, as the visually contrasting meteorites are easily detectable and tend to concentrate at specific areas exposing blue ice. Blue ice areas act as meteorite stranding surfaces (MSSs) if the flow of the ice sheet and specific geographical and climatological settings combine favorably. Previously, possible meteorite stranding surfaces were identified by chance or through visual examination of remote sensing data, which have limitations in discovering new locations for future meteorite searching campaigns. In this study, datasets are combined in a novel machine learning approach in order to estimate the likeliness of a blue ice area to be a meteorite stranding surface. Input data consists of positive and unlabeled observations. The ca. 2,500 positive observations are defined as the centers of regularly spaced grid cells containing one or more meteorite finds. The ca. 2,000,000 unlabeled observations, for which the presence of meteorites is unknown, are defined as the centers of regularly spaced grid cells overlaying blue ice areas. The size of a grid cell is 450 by 450 meter. Features of the observations, such as the surface velocity, the surface temperature, and the ice thickness, are extracted from geospatial datasets. Individual features and correlations between features indicate that positive observations differ from unlabeled observations. The unlabeled observations are classified as MSS or non-MSS by training a classifier with the nontraditional training set consisting of positive and unlabeled data. The obtained classification is validated and evaluated quantitatively with positive and negative observations, where the latter are defined after investigating fieldwork reports. With an estimated accuracy of 80%, the classification shows promising results. The influence of the different features on the classification does confirm the current, qualitative, ...