5 - Nationwide Urban Vacant Land Identification in China Using Segment Anything Model and Remote Sensing Imagery

Urban vacant land (UVL) is both a resource and a challenge in urban development, critical for the sustainable development of future cities. With China reaching its population peak and advancing urbanization, China’s cities are transitioning from rapid expansion to high-quality development. This shift implies that numerous cities will experience urban shrinkage, manifested by the presence of UVL in the urban space. Existing research on the identification of UVL in China mostly relies on field survey data, leading to issues of inconsistent definitions and significant variations, making comparisons difficult. With the advancement in computer vision research, the utilization of image semantic segmentation technology to analyze remote sensing imagery for obtaining UVL data has been a recent breakthrough in related studies. In this study, we employed the well-known "Segment Anything Model (SAM)” by Meta AI to identify urban vacant land for all Natural Cities (NCs) in China (a total of 3666 cities). This method showed improved accuracy compared to existing approaches and demonstrated efficiency, standardization, robustness, and transferability. The approach provides a solid foundational dataset for the study of urban vacant land in China’s cities, thereby propelling the advancement of relevant research.