LAUSR

Abstract Rapid urbanization has become an increasingly serious issue worldwide. While most previous studies focused on two-dimensional urban development, the spatial characteristics of building heights are rarely explored. Such information could provide valuable implications for smart urban planning and management. However, previous attempts did not systematically investigate the spatial factors that influence building heights and their associations with urban development. Therefore, this study developed a machine learning-based method to compare the distributions of building heights in Guangzhou and Shenzhen, two cities with different development patterns. First, we collected detailed information on the buildings, such as the location and land values. Second, the height of each building was simulated based on the above information using the well-known random forests, k-nearest neighbor algorithm, and artificial neural network. The random forests algorithm outperformed the other two in both cities. We also found that the commercial land value is the most important factor associated with building heights. Moreover, the building heights in Guangzhou are more sensitive to the distances to administrative centers, while the distances to transportation networks exert stronger influences on the building heights in Shenzhen. Overall, these findings could support urban planning and management. More importantly, the proposed method can be used to predict the heights of new buildings and investigate the distributions of building heights in other regions.