Object-Based Temporal Analysis of Ice Mass Change in the Antarctic Region Using Machine Learning Algorithms
Saziye Ozge Atik, Mehmet Arkali, Muhammed Enes Atik

Istanbul Technical University

Abstract

Climate change is causing changes in many areas of the world, and changes in some natural classes also constitute a significant part of this. Analyzing the impact of climate change on glaciers in the southern hemisphere is crucial. Satellite imagery, which is openly shared and can be acquired periodically, is a suitable data source for these analyses. Machine learning methods are among the most promising cutting-edge algorithms in this field. Using Sentinel-2A imagery, we analyzed the ice mass change in a selected region of Horseshoe Island, Antarctica, over 15 years, with five-year intervals. Using low-cloud images taken every five years, both in summer and winter, segments were generated using object-based image analysis (OBIA) in the study area and classified into three different classes. The support vector machine algorithm produced results with higher accuracy than the k-nearest neighbor algorithm. These analyses analyzed the time-dependent rate of ice mass decline over the last 15 years, providing researchers with insights into future predictions. Such studies can be used to provide ideas for measures that can be taken for the future.

Keywords: Machine Learning , Object-Based , Antarctica, SVM, k-NN

Topic: Topic B: Applications of Remote Sensing