Ghost Cities, Housing Vacancy and Idle Urban Space

This study estimates block-level housing vacancy by comparing housing stock with resident household counts. It combines 3D building attributes, natural-block boundaries, gridded population, household size and dwelling-area information across 2,813 spatial cities in mainland China. The model estimates reflect a 2020–2021 study snapshot rather than a household census or live occupancy. Results are compared with Wi-Fi probes, ground-truth video and other independent observations. The paper reports that data are available on request.

Due to rapid urbanization over the past 20 years, many newly developed areas have lagged in socio-economic maturity, creating an imbalance with older cities and leading to the rise of ghost cities. However, the complexity of socio-economic factors has hindered global studies from measuring this phenomenon. To address this gap, a unified framework based on urban vitality theory and multi-source data is proposed to measure the Ghost City Index (GCI), which has been validated using various data sources. The study encompasses 8,841 natural cities worldwide with areas exceeding 5 km2, categorizing each into new urban areas developed after 2005 and old urban areas developed before 2005. Urban vitality was gauged using the density of road networks, points of interest and population density with 1 km resolution across morphological, functional and social dimensions. By comparing urban vitality in new and old urban areas, the study quantifies the GCI globally for the first time. The results reveal that the vitality of new urban areas is 7.69% that of old ones. The top 5%, or 442 cities, were designated as ghost cities, a finding mirrored by news media and other research. This study sheds light on strategies for sustainable global urbanization and the United Nations Sustainable Development Goals.

In high-density urban areas, accurately identifying the vacancy status of multi-family residential buildings is crucial for forecasting urban energy consumption, mitigating vacant-housing proliferation, preventing crime and enhancing residents' experiences. Existing vacancy research has mainly focused on single-family housing. To address this gap, the study examines 16 multi-family residential buildings containing 1,020 residential units in Inner Mongolia, China. It develops a vacancy-identification framework involving data collection, housing identification and coding, multi-method vacancy identification and validation. Residential units are coded using daytime point-cloud data, while vacancy is analyzed using nighttime point clouds, front-door images and facade images with machine learning, multimodal large language models and spatial analysis. Electricity-consumption data are used for validation. Nighttime point-cloud data offer the highest accuracy and practicality, front-door images offer moderate accuracy and facade images perform less well. The proposed and validated methods provide a scalable approach for identifying unit-level short-term vacancy in multi-family housing.

High-rise residential buildings have become a predominant housing form, yet a substantial share exhibits high vacancy rates and associated resource inefficiencies. Accurately estimating housing vacancy rates at the residential-community scale is therefore essential for improving resource allocation. This study compares methods based on location-based services data, mobile signaling data and three types of nighttime-light data across 5,737 residential communities within Beijing's Fifth Ring Road. Field surveys in 103 representative communities use Wi-Fi probes to estimate household counts and validate the methods. The results show that location-based services data achieve the highest accuracy, with a correlation coefficient of 0.76, demonstrating their effectiveness and potential for large-scale housing-vacancy assessment. The global accessibility of such data also suggests that the approach can be transferred to other high-rise urban contexts.

Housing vacancy in China has become increasingly prominent, while the absence of official statistics makes accurate assessment and policy response difficult. This dissertation develops a multi-source, multi-scale measurement framework covering housing units, residential communities and residential land parcels. At the unit scale, nighttime point clouds, facade and front-door images, machine learning and multimodal models are validated with electricity-consumption data across 1,020 units in Inner Mongolia; nighttime point clouds achieve an F1-score of 0.85. At the community scale, five population and nighttime-light datasets are tested for 5,737 communities within Beijing's Fifth Ring Road and validated with Wi-Fi probes; Baidu location-based services data perform best, and the estimated vacancy rate is 9.37%. At the national scale, high-resolution imagery, deep learning, building data and open population data are integrated for 2,859 cities. The estimated national urban housing vacancy rate is 23.46%, corresponding to approximately 77.14 million vacant units and 10,580 km2 of vacant floor area. Six typical spatial patterns are identified, led by suburban vacancy and multi-center localized vacancy. Accessibility and population structure emerge as stronger drivers than economic level or built-environment quality. The dissertation connects these measurements to scale-specific planning and governance strategies for more efficient housing-resource allocation.

Rapid urbanization in China has produced extensive new urban development, but high vacancy in some newly developed areas has also created the phenomenon of ghost cities. In the absence of a clear evaluation criterion, this study uses urban vitality as the conceptual basis for identifying and evaluating ghost cities. It profiles 535,523 project-level residential developments built from 2002 to 2013 and measures their morphological, functional and social vitality using nationwide road junctions, points of interest and location-based-service records from 2014 and 2015. Project-level results are aggregated to cities, and thirty ghost cities are identified by comparing residential vitality in old urban areas developed by 2000 with that in newer areas. The average vitality of residential projects in new urban areas is only 8.8% of that in old urban areas. The results are benchmarked against existing rankings, search-engine attention and nighttime-light imagery, providing a systematic national assessment and policy evidence for urban development.

Rapid urbanization and large-scale new-town development have made ghost cities a prominent public concern in China. This article interprets ghost cities as newly developed urban areas with persistently low vitality and summarizes a data-driven evaluation based on residential-development records, road junctions, points of interest and location-based-service data. Morphological, functional and social vitality are compared between old and new urban areas, leading to the identification of thirty candidate ghost cities. The article argues that policy responses should move beyond generalized housing de-stocking. Urban expansion should be coordinated with demographic and economic capacity, excessive new development should be constrained, and employment, services and everyday activity should be strengthened in low-vitality districts. The work translates the underlying national assessment into direct recommendations for differentiated planning and urban governance.