Urban shrinkage has emerged as a global challenge, but its drivers in metropolitan regions remain poorly understood. In Japan, demographic decline and economic stagnation have extended shrinkage beyond rural areas into suburban and metropolitan zones, making the Tokyo Metropolitan Area an important case study. This research integrates Suomi National Polar-orbiting Partnership–Visible Infrared Imager Radiometer Suite nighttime light data with an advanced machine learning framework to analyze shrinkage and growth in the Tokyo Metropolitan Area between 2012 and 2022. A systematic framework of environmental factors was developed, covering both built and natural dimensions. The Random Forest algorithm was used for variable screening and nonlinear interpretation, while the Geographically Weighted Random Forest and SHapley Additive exPlanations methods captured spatial heterogeneity and local effects. The results show that shrinkage and growth coexist, forming a three-tier ring-like pattern, with 19.84% of grid cells identified as shrinking. Ten environmental factors significantly shaped these dynamics, with built environmental factors exerting a greater influence than natural factors. Their effects were nonlinear and spatially heterogeneous, and could be classified into positive, negative, and composite types. Four dominant drivers (residential intensity, floor area ratio, land and sea gradient, and industrial intensity) structured shrinkage into sequential west–east zonal belts. This study advances urban shrinkage research by systematically incorporating environmental drivers and developing an interpretable spatial machine learning framework. The findings provide spatially tailored insights for urban governance in Tokyo and offer broader lessons for other metropolitan regions confronting demographic decline. • High-resolution identification of urban shrinkage in the Tokyo Metropolitan Area. • Ten drivers of urban shrinkage identified from a 24-factor environmental framework. • Nonlinear and spatially heterogeneous effects revealed by machine learning methods. • West–east zonal belt pattern of shrinkage dominated by four environmental factors.
Zheng et al. (Mon,) studied this question.