Clustering-based framework for large urban studies: A case study in Taipei City

The acceleration of urbanization driven by immigration trends has led to half of the world’s population now residing in cities, highlighting the urgent need for sustainable urban planning and design. Consequently, the study of urban typology is essential for comprehending the complex nature of cities and fostering sustainable and livable urban environments. Recently, thanks to advanced simulation tools, numerous researchers have successfully developed techniques for evaluating urban typology performance across various dimensions, including energy, transportation, thermal comfort, and air quality.

However, urban studies seem diverse and scattered, complicating the understanding of its context. This study proposes a new framework using clustering algorithms to better understand the characteristics of various urban blocks. A section of Taipei City was chosen as a case study. Open source data including 1,459 urban blocks and 44,996 buildings were input into a GIS platform. Urban design parameters, such as building density, plot ratio, average building height, building height variation, window-to-wall ratio, block size, and block shape factor, were extracted from GIS for each urban block to serve as inputs for the clustering task.

The study employed both partitioning and density-based clustering algorithms, including K-Means, Spectral, DBSCAN, and Mean-Shift clustering. The performance of these algorithms was compared and validated using the Davies-Bouldin Index and Silhouette Score. The results indicate that only K-Means and Spectral clustering performed well, while DBSCAN and Mean-Shift proved unsuitable for urban studies.

The K-Means model with six clusters and the Spectral model with five clusters demonstrated the best performance. In the K-Means model, Cluster 2, characterized by the highest density and the lowest average height and height variation, reflects a horizontally urbanized, dense pattern. Cluster 6, with the highest average height, plot ratio, and window-to-wall ratio, strongly represents high-rise buildings. Cluster 5, showing the lowest plot ratio and density, exhibits a scattered pattern.

The Spectral model identified similar characteristics, with Cluster 2 mirroring Cluster 2 of K-Means, and Cluster 5 resembling Cluster 6 of K-Means. A notable distinction is that the Spectral model’s Cluster 1 has the lowest height variation and the second-highest density, representing the most typical pattern in the study area.

This proposed clustering framework is crucial for understanding the current urban context and is particularly useful for large-scale urban studies. Moreover, this approach can significantly reduce the time and resources required for simulation tasks. Additionally, analyzing cluster characteristics can assist urban planners in making informed decisions for sustainable urban development.