Japan is facing a demographic crisis, an aging population, a shrinking workforce, and declining birth rates, which pose significant challenges for its socio-economic stability and urban development. This study investigates these demographic trends up to the year 2100 at an unprecedented 500-meter grid level, employing advanced deep learning models with more than 3 million parameters. We create a paradigm by combining deep learning and dynamically updated datasets to forecast long-term high-resolution data accurately. Specifically, our average forecast accuracy is consistently above 99%. The forecasts reveal that Japan’s population will fall below 100 million by around 2055, decline to 90.5 million by 2070, and stabilize at the end of the century with a modest rebound. At the spatial level, more than half of Japan’s grids will become uninhabited by 2100, while Tokyo remains the dominant growth pole, attracting migrants from other cities. Household sizes decrease nationally from 2.26 in 2020 to 1.39 by 2100, signaling rising risks of social isolation. Meanwhile, the proportion of foreign residents rises above 6% of the national population by 2100, with over 3000 grids becoming immigrant-majority communities. The results imply that governments must prepare flexible, spatially targeted policies, balancing the consolidation of urban centers with tailored support for aging and immigrant communities, to ensure social and economic resilience. Our findings contribute to both the academic understanding of population dynamics and offer practical tools for policymakers to navigate and respond to Japan’s evolving demographic landscape for a long-term sustainable society.
Li et al. (2026) studied this question.