Machine Learning-Based Supervised Classification of Sentinel-2 MSI and Landsat-8 OLI Imagery in Marguerite Bay of Antarctic Peninsula

Abstract. Especially in the last decade, many innovative advantages of machine learning algorithms have been known, and their use in places where the effects of climate change are closely monitored, such as the polar regions, has introduced revolutionary scientific breakthroughs. In this study, machine learning methods were used to classify Sentinel-2A and Landsat-8 OLI satellite images of Marguerite Bay of Antarctic Peninsula. Four supervised classification algorithms were applied for pixel-based and object-based classification. Random Forest (RF), Decision Tree (DT), Support Vector Machines (SVM), k-nearest neighbor (kNN) are the algorithms selected for object-based image analysis (OBIA). SVM, RF, Light Gradient Boosting Machine (LightGBM) and Extreme Gradient Boosting (XGBoost) were used for pixel-based classification. Each image is labelled into three classes: glacier, water and soil. The classification methods were analysed comparatively for each data set. In both Sentinel-2 and Landsat-8 images, 97.31% and 96.28% overall accuracy were achieved with SVM, respectively.