Determination of land cover change with multi-temporal Sentinel 2 satellite images and machine learning-based algorithms

Detecting and monitoring change on Earth has always been a subject of considerable interest. Over time, human activities have expanded and the impact of these activities on the land cover has been clearly seen. Detecting and monitoring the change in land cover has become a critical issue for decision-makers due to issues such as the increase in industrial activities and the increase in settlement. Several works have been performed on this subject in the field of remote sensing, and methods and tools have continuously improved to determine the change in the earth to achieve the most accurate result. Within the scope of the study, multi-temporal Sentinel 2 satellite images were used in order to determine the land cover change due to urbanization and agricultural activity in Kocaeli province within the framework of dynamic change determination according to LandCover 2.0 standards. Four different data reduction - classification method combinations were applied, which are Built-up Index-Random Forest, Principal Component Analysis-Random Forest, Built-up Index-Regression Tree and Principal Component Analysis-Regression Tree and their performances were evaluated comparatively. The results of the classification analyses performed on the Google Earth Engine platform were turned into thematic maps and an accuracy assessment was carried out. As a result of the study, it has been revealed that Principal Component Analysis-Regression Tree method pair is the approach that provides the highest accuracy, with an accuracy rate of 83.88 percent.