Deep Learning Analysis of Urban Growth Boundaries: An Evaluation of Effectiveness in Mitigating Urban Sprawl in China

Abstract

Although urban growth boundaries (UGBs) have been employed as a planning tool in certain Western cities for decades, they have recently been adopted in Chinese cities to address urban sprawl. However, the effectiveness of UGBs in different types of Chinese cities, particularly those on flatlands versus mountains, has not been studied. This study aimed to compare the efficacy of UGBs in a city on flatlands (Chengdu, China) and a mountainous city (Chongqing, China). We used a deep learning architecture (U-Net) to project urban expansions in 2035 with the presence/absence of UGBs and landscape metrics to evaluate UGBs’ effectiveness in mitigating urban sprawl. We found significant differences in historical urban expansion between Chengdu and Chongqing from 1992 to 2019. Chengdu experienced spillover sprawl under a monocentric-dominated urban form, while Chongqing witnessed leapfrog and piece-mall sprawl under a polycentric form. Despite the differences in UGBs designation, the simulations demonstrated that UGBs could mitigate urban sprawl in both cities in 2035, with Chengdu exhibiting more pronounced effectiveness. Notably, UGBs were more effective in controlling spillover sprawl in Chengdu, whereas they could effectively reduce leapfrog sprawl in Chongqing. However, UGBs were constrained by strict top-down land quotas, limiting their potential. These findings suggested that implementing UGBs should adopt differentiated goals and strategies for different types of cities.