Developing a mapping procedure for urban forests using online map services and Sentinel-2A images

Abstract

Accurate area statistics at the local government level are important for the scientific management of urban forests. A more consistent and robust mapping of urban forests can enable a comparison of the size and quality management guidelines at major cities in Asia. However, few administrative statistics have employed these scientific techniques. Therefore, a cover map of urban forests focusing on their types is needed for the quantitative and qualitative estimation and evaluation of urban forests. This study presents an approach for mapping urban forest cover for 2021 using Sentinel remote sensing images and aerial photographs from commercial maps. Cover maps of urban forests at a city scale were developed for three pilot study areas in Korea’s capital area. Labeling was performed for deciduous forests, coniferous forests, grasslands, wetlands, bare land, croplands, urban areas, and water bodies. The land cover classification was performed using machine learning models, including random forest (RF), support vector machines (SVM), and gradient tree boost classifier (GBT), and a comparison was made among them. All three methods performed well in the three study areas; the RF model was simple to use, and the SVM was better at detecting forest connections. The proposed mapping method, which focuses on forest types in cities, can be used to compare the area and connection of urban forests in cities.