Evaluation of ghost cities based on spatial clustering: a case study of Chongqing, China

Abstract

With the acceleration of the urbanization process, construction areas increase rapidly. However, excessive development will lead to high vacancy rates, resulting in the “ghost town phenomenon.” In this research, Tencent LBS (Tencent location-based services) population location heat data and Baidu community POI data in Chongqing are used to divide the residential group areas by the k-means++ algorithm and Graham’s scan algorithm. A new ghost city index (g) that is calculated by relating the population heat data of a cluster to the area of a convex hull cluster is proposed to calculate the ghost city index of each convex hull cluster. The g values of 38 districts in Chongqing are obtained by the weighted average of the population location heat data of convex hulls. The results show that the g in Chongqing is generally high in the west and low in the east. In the “one-hour economic circle and two wings” proposed by the government of Chongqing, the average g of the one-hour economic cycle is 7.64. In a one-hour economic cycle, the g of the main city of Chongqing is 16.41, and the average g of the parts other than the main city is 1.27, which could be a ghost town area. In the northeastern wing and the southeastern wing of Chongqing, the average g values are 0.59 and 0.32, respectively. The research also finds that for clustering Chongqing community POI data, the effect of k-means++ is better than that of DBSCAN algorithm.