Identifying and quantizing the non-linear correlates of city shrinkage in Japan

Abstract

City shrinkage is a growing global issue and has raised great attention. Although shrinkage in Japan was mostly contributed to the demographic factors of the aging population and low fertility, a comprehensive understanding of the influence mechanisms of shrinking cities in Japan is still challenging. In this study, we used a random forest (RF) approach to explore the complex features and quantize the non-linear correlates of city shrinkage. Compared to conventional linear regression, RF is more effective in revealing complex relationships. City shrinkage in Japan is manifested in population aging and low birth rate, while other indicators also have a significant impact on the shrinkage. For example, when a city has a total population and a population density below 50,000 people and 1600 people/km2, shrinkage is significantly correlated to the city size. The results revealed the non-linear correlations and the various marginal effects of characteristics on the city shrinkage correlates and indicate the influence of urban scale and local autonomy on shrinkage. The study contributes to improving the understanding of city shrinkage in Japan and further provides insights for urban planners to implement city shrinkage management.