Urban shrinkage and its drivers: time-series clustering and a panel model of Chinese cities

Understanding and characterizing urban shrinkage is essential for informing sustainable development policies. However, accurately tracking its dynamics over time through concurrent analysis of population and economic metrics remains challenging. To address this, we apply time-series machine learning—specifically Dynamic Time Warping (DTW) clustering—to 285 Chinese cities (2004–2022), revealing non-linear, continuous yearly trajectories of both population and economic indicators. This approach groups cities by similar shrinkage/growth patterns. We then employ panel regression to identify drivers of these trajectories. Four distinct groups emerge: (1) 40 cities experiencing simultaneous declines in both population and economy, primarily driven by resource depletion and manufacturing recession; (2) 40 cities with economic decline but population growth, where shifts in industrial structures were the main influencing factors; (3) 42 Cities with economic growth but declining populations, predominantly affected by population migration; and (4) 163 cities exhibiting continuous growth in both population and economy. By mapping unique shrinkage trajectories to group-specific drivers, our framework enhances the precision of urban decline diagnosis. These insights enable policymakers to design regionally tailored strategies that mitigate shrinkage and foster sustainable urban futures.