Prediction and analysis of potential locations of Antarctic meteorites based on machine learning

Meteorites, as rock samples from outside the Earth, play an important role in studying the evolutionary history of the solar system. This paper combines multi-source remote sensing data and uses the Kernel Density Estimation-based Bayesian Method (KDE-Bayes) semi-supervised classification algorithm to predict potential stranding zones of Antarctic meteorites. It compares with various other semi-supervised learning methods and finds that the KDE-Bayes semi-supervised classification algorithm has the best performance with the prediction accuracy reaches 77.5%. The potential stranding areas of Antarctic meteorites are mainly concentrated on the east side of the Hengduan Mountains and the Queen Maud area, and less in the Antarctic Peninsula. Further analysis of environmental factors in typical meteorite-rich areas found that surface temperature, ice sheet flow velocity and surface characteristics are key factors affecting meteorite distribution. In addition, this paper also compares the extraction results of the model in different regions. The differences are mainly due to the low resolution of remote sensing data and the deviation from the actual recovery time of meteorites.