Satellite monitoring of shrinking cities on the globe and containment solutions.

Shrinking cities are often neglected in the context of global urbanization, the tip of the iceberg that was driven by underlying complex sets of causes. Therefore, it is urgent and crucial to investigate the invisible aspects of global urbanization propelling specific challenges to attain Sustainable Development Goal 11 (SDG 11) related to sustainable cities and communities. Here, we identify shrinking cities in 1992-2000, 2000-2012, and 2013-2018 and predict them in 2018-2050, using nighttime light images and redefined natural city boundaries. The proportion of shrinking cities increased from 9% to 16 and 25%. Looking ahead, there will be 7,166 predicted shrinking cities in 2050, accounting for 37% of all cities. In this context, synergistic efforts like regreening vacant lands and constructing compact cities would help achieve SDG 11 in consideration of the new urban shrinking landscape with multisource data like CO2 emissions and points of interests (POIs).

Introduction

A number of studies focus on the global urban growth in the past (Liu et al., 2020; Sun et al., 2020) and to the future (Chen et al., 2020; Gao and O’Neill, 2020); however, urban shrinkage, as the other side of urban growth, is often neglected. Urban shrinkage is not a new phenomenon; though the driving factors for shrinking cities are complex and variegated; the deceleration of urban growth across the globe is heterogeneous but is becoming a new reality. Shrinking cities are usually characterized by significant population loss (Güneralp et al., 2017), economic decline (Hudson, 2005), and physical disorder, including spatial decay (Lee and Newman, 2017), and have become a common phenomenon throughout the world. The existing SDG 11 was designed in the context of rapid urbanization; however, shrinking is a nonnegligible part of the urban future, which has not received enough attention. To sustain global economic prosperity in the face of plateauing urbanization and global urban shrinking, city authorities are urged to implement coping strategies to address population loss challenges (Lu et al., 2015). For city administrators, this means policy responses that shift the focus from curbing urban sprawl to governing residents within shrunk city boundaries. However, our understanding of how much the spatiotemporal changes of shrinking cities encompass the whole picture is limited. Therefore, it is necessary to identify these shrinking cities around the world accurately enough to provide empirical support for the effective government responses needed to form a healthier urban development mode. This analysis at a global scale would also help urban decision-makers make informed decisions regarding the dialectical view of the impacts of population loss and urban shrinkage.

The current understanding of shrinking cities is largely based on population data and the official administrative boundaries for one or several countries rather than global data, and the spatial characteristics of shrinking cities are not examined. For instance, Martinez et al. used population data to identify shrinking cities in Australia, Japan, Germany, the United Kingdom, France, and the United States and proposed corresponding political strategies for city development (Martinez-Fernandez et al., 2016). Long et al. identified 180 shrinking administrative cities in China alone, each accompanied by rapid urban expansion (Long, 2016). However, previous studies of the spatial distribution of shrinking cities at the global scale have their limitations. The only study on this topic was based on 1950–2000 statistics and did not cover every urbanized area or consider any temporal changes (Oswalt and Rieniets, 2006). The Shrinking Cities Project of Germany’s Federal Cultural Foundation revealed the polarization tendency of cities around the world in 2005–2015 (Project, 2015). However, the statistical data fell short in accurately describing the evolution of every city in the world and objectively evaluating shrinking cities between countries with different statistical standards (Long, 2016). In addition, there is a lack of quantitative research on how shrinking cities developed in the past decades and will evolve in the future.

Compared with demographic data and statistical data, nighttime light remote sensing data are more objective and can more quickly reflect human activities and habitation (Chen and Nordhaus, 2011; Li et al., 2017; Yeh et al., 2020). To investigate the global spatiotemporal evolution of shrinking cities, therefore, we seek to understand the concurrent urbanization era and demographic transformation using nighttime light remote sensing data on a worldwide scale. Instead of administrative boundaries, we utilize natural cities as a benchmark to depict the precise distribution of such spatial cities at a global scale. Natural city is a product of the bottom-up thinking in terms of data collection and geographic units or boundaries proposed by Bin Jiang (Jiang and Jia, 2011; Jiang and Miao, 2015). To accomplish this, we first evaluated historical urban shrinkage to establish that the proportion of identified global shrinking cities increased from 9% to 16 and 25% during 1992–2000, 2000–2012, and 2013–2018, respectively. There were 183,451,413, 547,940,208, and 2,061,743,186 inhabitants in shrinking cities in 2000, 2012, and 2018, according to the distribution of nighttime light pixels. Shrinking cities spread from relatively developed areas to underdeveloped areas and from relatively small-sized cities to medium- and large-sized cities. It is predicted that a total of 7,166 shrinking cities would come into existence, accounting for 37% of all cities, covering 29% of the total natural city area with 2,921,647,019 inhabitants by the year 2050.

To reveal where people from shrinking cities migrate to, we take China, a country with the most shrinking cities in 2013–2018, as an example for further examination. Baidu Huiyan migration statistics from Nov. 2017 to Nov. 2019 were acquired to analyze the migration among 3,022 natural cities in China. The movements demonstrated that the out-migration from shrinking cities contributes to the emergence of large cities, leading to other small and medium-sized cities’ loss of workplaces and waves of population out-migration. We further predicted that the number of natural cities with a population of less than 10,000 will decrease by approximately 32% in the world by 2050. By using large-scale urban data with geotags, the dual existence of mega and shrinking cities emerges which has coincidently contributed to concentrating environmental impacts spatially, enhancing the public resources’ equity and efficiency (e.g., education and medical). This could help mobilize the achievement of sustainable cities and communities of SDG 11.

In consideration of a new urban shrinking landscape, we therefore proposed two innovative solutions with quantitative simulation experiments toward enriching SDG 11: 1) Regreening vacant lands in shrinking cities. It will benefit in preserving green space, increasing the flexibility of energy system design, rendering stormwater management services, improving the urban thermal environment, expanding terrestrial carbon sinks, adapting to climate change, and finally accomplishing the sustainable development goals of shrinking cities. 2) Constructing compact cities. It has potential to improve the living environment, maintain subject well-being and reduce the expenditure on public services.

Results

Spatial distribution characteristics of global shrinking cities in the past (1992–2018): Global and local scales.

To identify the spatial evolution of global shrinking cities at the city level, we analyzed the first three phases, 1992–2000, 2000–2012, and 2013–2018, based on nighttime light remote sensing data at the global level (Figures S1 and S2). Figure 1A–1C shows there are increasingly more shrinking cities with lower initial nighttime light intensity, which indicates that the cities in underdeveloped areas have been shrinking. In the three phases, 18%, 32%, and 62% of the total cities with an initial nighttime light intensity of less than ten were shrinking. Figure 1D–1F indicate that natural cities with larger areas have been experiencing shrinking. For the cities smaller than 10 km2, the mean shrinking ratios are 56%, 51, and 43%; in contrast, for the cities larger than 50 km2, the ratios are 5%, 7, and 12% of all the cities globally. In addition, we did not find a causal relationship between urbanization rates and proportions of shrinking cities in different countries. Shrinking cities emerge in countries with different levels of urbanization as indicated in Figure 2. Belgium and Japan have the highest urbanization rates, while the proportions of shrinking cities in these two countries do not exceed 25%. Countries with a high urbanization rate like Sweden, Denmark, and the United Kingdom and countries with a low urbanization rate like Egypt, Uzbekistan, and Syria suffer from a high proportion of shrinking cities.

Figure 1

Relationship between the shrinking ratio and the initial nighttime light intensity

(A), 2000 (B), and 2013 (C) or the natural city area in 1992–2000 (D), 2000–2012 (E), and 2013–2018 (F). The initial nighttime light intensity represents the development level of the natural city.

Figure 2

Relationship between urbanization and urban shrinkage

We plotted the urbanization rate in 2015 and the proportion of shrinkage from 2013 to 2018 of 102 countries in the world with more than 10 natural cities and at least one shrinking city. The urbanization rates in different countries were achieved from the United Nations. Thirty countries with the largest number of shrinking cities are labeled. The world average urbanization rate and proportion of shrinking cities are 54 and 25%.

Figure 3 shows the overall changes in the shrinking cities of different countries in the first three phases. In general, shrinking cities spread from Eastern Europe and Central Asia to the developed countries in Europe and the United States and then worldwide. Moreover, we present cartograms to reveal the spatial distribution of global natural cities and their proportions of shrinking cities in the three stages From 1992 to 2000, shrinking cities were relatively uncommon at only 1,295 (9%) of all cities worldwide — and they were mainly concentrated in Eastern Europe and Central Asia. Ukraine was the country with the highest proportion with nearly 60% of its natural cities shrinking. Others were sparsely distributed in Western Europe, the United States, the Middle East, and North Africa. From 2000 to 2012, we identified 2,821 shrinking cities (accounting for 16% of all cities) that were increasing and stretched globally, especially in developed countries. In the United States, which ranked first among the nations, there were 612 cities extending from the northeast to the east and west coasts. Many cities were shrinking in Western Europe, including cities in Germany (305), France (224), and the United Kingdom (185). From 2013 to 2018, the number further accelerated at a global scale, with nearly a quarter of all cities shrinking and around 27% of people living in shrinking cities in the world (Figure 4). A substantial number of shrinking cities appeared in Africa, Latin America, and East Asia. They were no longer restricted to developed countries but spread to developing countries such as BRICS (Brazil, Russia, India, China, and South Africa). For instance, the number of cities in China reached 754, which was almost seven times higher than in the previous phase, outnumbering that of the United States, and China became the country with the most shrinking cities. There were 167 in India, which was eight times that in 2000–2012.

Figure 3

Overall changes of global shrinking cities in four phases

The numbers and total areas of shrinking cities and natural cities are presented. Defense Meteorological Satellite Program-Operational Linescan System (DMSP-OLS) images with a 1 × 1 km2 spatial resolution are used for the first two phases, and Suomi National Polar-Orbiting Partnership-Visible Infrared Imaging Radiometer Suite (NPP-VIIRS) images with a 500 × 500 m2 spatial resolution are used for the last two phases. Natural city boundaries were converted from Liu et al.’s global land use distribution data from the years 1990, 2000, 2010, and 2015 (Liu et al., 2018). Note: SC refers to a shrinking city. The division of global countries is based on World Urbanization Prospects (2014 revision, UN Population Division). A natural city is recognized as a shrinking city if its average nighttime light intensity decreases by more than 10% in 10 years.

Figure 4

Satellite map of the distribution of global natural cities and the shrinking ratio at the city level by NPP-VIIRS (2013–2018)

We redefined 21,058 natural cities of over 5 km2 for the year 2010 to estimate and compare their areas by using the NPP-VIIRS nighttime light data for 2013 and 2018. The total area was 715,638 km2. A total of 5,237 of the 21,058 natural cities had shrunk during 2013–2018, among which there were 531 natural cities with the shrinking ratios exceeding 50%. The total shrinkage area was 114,158 km2, which covered more than 16% of the total area of natural cities with 2,061,743,186 inhabitants. The shrinking cities were mainly located in Eastern Asia, Western Europe, and the eastern United States.

Four primary driving factors of shrinking cities

By summarizing the reasons affecting urban shrinkage, we drew a schematic diagram as indicated in Figure 5 through reviewing 100 studies (Table S1). The driving factors can be categorized into five types. First, cities that are faced with population decline because of such economic factors as globalization, industrial transformation, high housing costs, and resource depletion. Second, cities that are suffering from a demographic decline, which is the result of low fertility rates, high life expectancies, and aging. There are many factors that may affect the global urban shrinkage which include the wide gap between the number of births and deaths and the declining labor force. Third, several social factors lead to urban shrinkage in many small and medium-sized cities; for example, the social services and urban infrastructure are lacking, which has caused the population’s out-migration. Moreover, the development of transportation and formation of suburbanization accelerate this trend. Fourth, institutional factors such as the transformation of the social system, adjustment of national policy, and even wars will have a lasting impact upon the development of cities. The main functions, industries, and labor forces of some cities are affected or damaged by the aforementioned transformation, which has led to the emergence of shrinking cities. Finally, some shrinking cities are exacerbated by environmental factors such as natural disasters, low temperature, climate change, and air pollution.

Figure 5

A literature review of the driving factors for shrinking cities

Each digit in the diagram represents the number of shrinking city cases resulting from a specific reason in one country. Driving factors contributing to shrinking cities in the literature involve five groups: economic factors, demographic factors, social factors, institutional factors, and environmental factors. The number of cases mentioned for the five groups are 124, 100, 48, 42, and 15. The top four studied countries in the world are the United States, Germany, China, and Japan.

To illustrate the driving factors and mechanism of shrinking involved, we selected a few typical shrinking cities from four representative countries in the aforementioned literature review as shown in Figure 6. First, such cities as Detroit, Cleveland, and Chicago in the Rust Belt of the United States are an example of shrinking cities caused by unequal regional development. The recession of the manufacturing industry has led to a reduction in labor demand, whereas healthcare, education, working environment, and other supporting services are not comparable to those in other cities. Consequently, the population in the Rust Belt has declined dramatically. Second, Dresden and Leipzig in Eastern Germany suffered from great social and economic transformations after the reunification of East and West Germany in 1990. Many problems have appeared since then, including plant closures and bankruptcies, population out-migration, and increasing unemployment rate. Third, Hegang, Anshan, and other cities in Northeast China have gradually lost their vitality, which has led to population loss, economic decline, and spatial decay. Fourth, in Japan, small and medium-sized cities in Hokkaido (Sapporo) experienced shrinking because of severe population aging. The declining demographics are also partly exacerbated by the insufficient supply of labor in these single-industry cities, which further lead to their economic decline.

Figure 6

Four typical shrinking cities (Hegang in China, Detroit in the United States, Dresden in Germany, and Sapporo in Japan) are specifically examined at the regional, city, and block levels

The locations of the four cities are given in the regional level column. The spatial distributions of the intensity of the shrinking in the four cities are shown in the city level column. Two street view pictures in each city are given in the block level columns. Hegang and Detroit are described for the 2013–2018 phase, and Dresden and Sapporo are described for the 2000–2012 phase. The remote sensing images and street view photos are from Google Maps.

Out-migration from shrinking cities mainly move to large cities

Because there are so many shrinking cities on the globe, where do their populations mainly flow? Few studies have investigated the out-migration from shrinking cities because of the missing big data as well as quantification difficulties. As the country with the most shrinking cities in 2013–2018, China is further investigated as an example to quantify the out-migration from shrinking cities employing the Baidu Huiyan migration statistics. To the best of knowledge, it is the first research to investigate where the residents are from shrinking cities moving to, with high-resolution user location information. As a typical representative, China’s population flows reflect the general law of population flow out of shrinking cities. The results indicate that 65% of them migrate to cities at the top 12% in size (Figure 7). Approximately 78% of the distances between destinations and shrinking cities from which residents migrate out are less than 1,000 km; 19% of them are in 1,000–2,000 km; only 3% of the distances are over 2,000 km.

Figure 7

Migration flow in natural cities and shrinking cities

(A and B) There are 64%, 18%, and 18% of migration from large cities flowing to large, medium, and small cities in China (A). As for migration from medium shrinking cities, 70%, 14, and 16% of them migrate to large, medium, and small cities. What is more, as for migration from small shrinking cities, 72%, 13, and 15% of them migrate to large, medium, and small cities. Large, medium and small in this figure refer to 12%, 15, and 73% of natural cities in China according to city size, referring to the five-level urban hierarchy classification standards (Figure S3). The inflow-outflow migration statistics of three typical shrinking cities Lvliang, Jilin, and Liaoyang are presented in inset (B). The outflow outnumbers the inflow of shrinking cities and a majority of outflow move to large cities.

This could be explained by Myrdal’s theory (Myrdal and Sitohang, 1957) of circular and cumulative causation. In Myrdal’s framework, large cities attract population from less developed neighboring areas, leading to the ‘backwash effects’ on the neighbors, whereas the urban agglomeration may increase demand for agricultural products and raw materials and thus create employment opportunities in the peripheral areas, thereby exerting ‘spreading effects’. This model holds that in no circumstances do spread effects exceed backwash effects. In this context, the cities with increasing out-migration were to face a ‘vicious circle’ of cumulative population decline. Urban shrinkage was thus conceptualized as a self-reinforcing process propelled by labor-oriented out-migration.

The negative effects of shrinking cities

A persistent loss of population usually has a negative impact on shrinking cities (Table S2), including economic hardship (García-Ayllón, 2016; Pagano and Bowman, 2004), social disruption (Frazier et al., 2013; Hudson, 2005; Rybczynski and Linneman, 1999), abandonment of built environments (Blanco et al., 2009; Greenberg and Schneider, 1996; Lee and Newman, 2017), and urban planning problems (Pallagst, 2009; Ryan and Gao, 2019). According to the inflow-outflow migration statistics from Baidu Huiyan in China, 43% of labor productivity in urban areas decreased 22% because of urban shrinkage (Yang et al., 2020). Further, 15% of labor productivity in the world decreased with the same scale from 2013 to 2018. Furthermore, the adverse effects of urban shrinkage may lead to causal relationships and form a cyclical chain (Kim, 2019). For instance, a declining population leads to a shortage of labor resources that may cause factories and companies to cease operations, economic crises, unemployment, and more population immigrating into other prospering regions (Haase, 2013). This, as a kind of a vicious cycle, results in further demographic decline (Nuissl and Rink, 2005), which occurs in many such cities (Couch et al., 2005).

The coincidental positive shrinking effects on carbon emission mitigation, healthcare, and education enhancement

Our quantitative experiments in China proved that accompanying the increasing number of shrinking cities, there is an out-migration trend from nonurban areas or small cities to large cities, which is also verified by traditional census data (Mu et al., 2021) and emerging Internet data (Egidi et al., 2020; Keuschnigg et al., 2019; Zhang et al., 2020). This is echoed with our predictions that the number of natural cities with a population of less than 10,000 will decrease by approximately 32% (Figures S4 and S5).

Based on data provided by a global dataset of CO2 emissions, we examined the relationships between annual CO2 emissions per capita (t/year) and city area/population size in 343 administrative cities in 2015. The results indicate that with an increase in city size, the emissions of CO2 in cities gradually decreased (Figure S6). The annual amount of CO2 emissions per capita in administrative cities with an area over 2,000 km2 is 35% of that with an area less than 50 km2. The annual CO2 emissions per capita of administrative cities with a population of over 20 million are 40% of those with a population of less than 500,000.

Then, we explored the healthcare-related and education-related points of interest (POIs) from OpenStreetMap (OSM) to evaluate the densities of the POIs in natural cities with different areas (Figures S7 and S8). Cities with a higher density of healthcare-related or education-related POIs are regarded as cities with better healthcare facilities or educational resources. The result shows that the density of healthcare resources in natural cities with an area over 2,000 km2 is twice that of cities with an area less than 50 km2. Similar studies show that the density of educational resources in natural cities with an area over 2,000 km2 is 2.5 times that in natural cities with an area less than 50 km2 (Figures S7 and S8). To sum up, as some megacities grow, they pull in ever more people, leading to other small-sized and medium-sized cities’ proportionate urban shrinkage, it is coincidentally beneficial to reducing environmental problems, enhancing public resources’ efficiency, establishing good health and well-being, and thus mobilizing the achievement of sustainable cities and communities of SDG 11 (Table S3).

National urbanization strategies across the world: Does policy matter?

Most of the existing studies on shrinking cities have focused on the negative effects of urban shrinkage and called on the government to carry out planning interventions. Urbanization strategies in different countries across the world have been adjusted to manage shrinking cities. We thus attempted to investigate whether these policies are effective as containment policies (Table S4). We divided national urbanization strategies into three types according to the purpose of the urbanization strategies formulated by the governments: proactive, negative, and balanced (Mingione, 1982; Renaud, 1979; Richardson, 1981). A proactive strategy promotes the development of metropolises, a negative strategy slows down the urbanization process, and a balanced strategy promotes the development of both urban and rural areas (Pfeffermann, 1982). Owing to the difficulty of data acquisition, we only obtained twelve relevant policy indicators (six positive policy indicators and six negative policy indicators) published by the UN in 2015 to identify different countries’ urbanization strategies in 2013–2018. We classify a country’s urbanization strategy by comparing the numbers of proactive and negative indicators. If the proportions of the two types of policy indicators are equal, the urbanization strategy of the country is balanced. The results show that 89, 48, and 55 countries adopted proactive, negative, and balanced urbanization strategies in 2013–2018, respectively (Figure 8).

Figure 8

Correlation between national urbanization policy and proportion of shrinking cities in 2013–2018

Note: NC and SC refer to a natural city and a shrinking city.

To reveal whether the national urbanization strategy has an impact on urban shrinkage, we applied two hypothesis testing approaches (Tables S5 and S6), including the t-test (two-sample equal variance) (Díaz–Pachón et al., 2020; Schnuerch et al., 2020) and the Friedman test (two-way ANOVA by rank) (Beasley and Zumbo, 2003; DeJuan and Seater, 2007). We set the null hypothesis that national urbanization strategies have no effect on shrinking cities and analyzed the significance of the shrinkage rate at the national level under the three types of urbanization strategies in 2013–2018 to test the authenticity of the null hypothesis. The results show that the null hypothesis is true; that is, each of the nations’ urbanization strategies has no significant effect upon the shrinking ratios across these countries. It could be inferred that the national urbanization strategy has little impact on the migration and spatial distribution of population (Banerjee and Schenk, 1984). This result agrees well with the outcome of the urbanization strategies implemented in historical periods in some countries.

Global urban shrinkage: Prospecting urban futures

A quantitative prediction of shrinking cities can reveal the development tendency of shrinking cities in the world; however, such projections are limited at present. We used night light data to analyze the distribution of global shrinking cities over a long-time span from 2000 to 2018; however, grasping the distribution of shrinking cities in the future is essential for effectively guiding the healthy development of cities. We made future predictions of urban shrinkage from 2018 to 2050 based on the past developmental trend identified by historical nighttime light images, which were also constrained by global country-level population prediction data. The technological details for the future predictions can be found in the STAR Methods section. According to our forecast, the decline of cities will continue from 2018 to 2050 worldwide. There are expected to be 7,166 shrinking cities in 2018–2050, accounting for 37% of all cities and covering 29% of the total natural city area. Shrinking cities in 2050 involve 2,921,647,019 inhabitants. The increase of inhabitant quantity in shrinking cities decelerates. Their quantity and intensity will keep increasing in both developed and developing countries. Approximately 32% of cities in the world will shrink to more than 50% of their original area. For cities with an area smaller than 10 km2, the mean shrinking ratio will be approximately 39%. In contrast, the ratio is 13% of all the cities larger than 50 km2 globally. During this phase, significant increases in shrinking cities will appear in regions such as Australasia and Northeast Asia. Figure 9 displays their predicted distribution and proportion of shrinking cities at the country level from 2018 to 2050. The United States, China, Russia, Brazil, Germany, and India are the top six countries with the highest numbers worldwide. We compared the urban shrinking forecast developed by the United Nations, Department of Economic and Social Affairs, Population Division with the forecasts developed by the five scenarios of shared socioeconomic pathways (SSPs); the six results are consistent (Table S7).

Figure 9

Cartogram of the distribution of global number of shrinking cities and the proportion of shrinking cities at the country level by NPP-VIIRS (2018–2050)

Of the 30 countries with the largest numbers of shrinking cities, the proportions of shrinking cities in France, Canada, Uzbekistan, Belgium, Egypt, Denmark, and Syria exceed 50%, whereas those in China, India, Ukraine, Italy, Australia, Kazakhstan, and Argentina are below 30%. At the city level, the numbers of shrinking cities and the shrinking ratios will accelerate around the world.

Shrinking cities have existed for decades and will continue to spread in the future. At present, it seems that the policies in different countries have limited influence and that this trend is irreversible. Their presence may not be conducive to the development of specific cities. Nevertheless, our experimental results from Baidu Huiyan migration statistics demonstrated that shrinking cities have also contributed to the emergence of additional large cities, which can be beneficial in various respects. We tested the trend of shrinking cities around the world with different population situations, as indicated in Figure 10. In addition to the number of shrinking cities calculated based on the population distribution under the UN normal scenario, we calculated the situations under high fertility, low fertility based on estimates from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017 (Murray et al., 2018; Vollset et al., 2020) and the situation under zero migration based on the UN forecast data. Next, we calculate the number of cities in the world in 2050 under the SSP1 scenario with global aging population (Samir and Lutz, 2017). Moreover, we assume that China’s relaxation of the Hukou and the migration from small cities to large cities will lead to a 10% increase in the population of the top seven natural cities. Finally, it is assumed that cities will shrink when affected by an epidemic situation, such as because of COVID-19 deaths, in an unmitigated scenario in different regions. The results indicate that the relaxation of the Hukou in China, a low fertility policy, a zero migration, and the influences of pandemic diseases (e.g., COVID-19) will accelerate the shrinking of cities around the world, whereas a high fertility will have a mitigating effect.

Figure 10

Numbers of shrinking cities under different situations

Under normal developmental situations (black line), there will be 7,166 shrinking cities in the world in 2050 and the increasing rate will reach 37%. Low fertility (orange line) and high fertility (purple line) lead to slow and fast increases of the population, respectively, resulting in increasing rates of 47 and 26%, respectively. In the zero-migration situation (yellow line), there are more shrinking cities given comparatively smaller populations in West Europe and North America. Population aging (blue line) will be aggravated in SSP1, and there will be an additional 42% shrinking cities in 2050 in this trajectory. If China relaxes the Hukou restriction (red line), migration to large natural cities will lead to the depletion of small cities. Infectious diseases (green line) such as COVID-19 have caused a population decline over the world, which will contribute to an increase in the number of shrinking cities.

Synergistic efforts toward SDG 11 in consideration of the new urban shrinking landscape

The Sustainable Development Goals (SDGs) were adopted by the United Nations in 2015, providing a shared blueprint for peace and prosperity for people and the planet now and into the future. The previous SDG 11 was designed and defined in the context of rapid urban growth, although shrinking is a nonnegligible part of the urban future. There is an urgent need for a planning paradigm shift toward ‘planning for shrinking’ besides ‘planning for growth’. In this research, we propose two solutions to shrinking cities for the enrichment of SDG 11.

Regreening vacant lands in shrinking cities

The outflow of population and capital from a shrinking city result in massive vacant lands. Shrinking spaces are converted from nighttime light image pixels whose intensities decrease over 10% within shrinking city boundaries (Figure S9). Existing studies suggest that vacant land constitutes over 15% of the total land area in cities throughout the United States (Bowman, 2004; Nickerson et al., 2011). According to the previous studies, vacant land can be converted into green infrastructures (GI ) such as temporary or permanent community gardens, urban farmland, and community forests (Hummel, 2015). The regreening of shrinking cities will benefit in preserving green space (Choi et al., 2021), increasing the flexibility of energy system design (Jing et al., 2020), rendering stormwater management services (Kelleher et al., 2020; Lovell and Taylor, 2013), improving the urban thermal environment (Bosch et al., 2021), increasing terrestrial carbon sinks (Anderson and Gough, 2020), adapting to climate change (Gaffin et al., 2012), and finally achieving the sustainable development goals of shrinking cities.

For instance, green infrastructure has been introduced across many urban areas as a decentralized, economical approach to reducing sewer system overflows and to supply urban ecosystem services (Lovell and Taylor, 2013). If all vacant land in shrinking cities were converted into GI, we applied a simple model to estimate the rainfall detention capacity (Kelleher et al., 2020) with the average precipitation in depth in different countries. In this model, we take average parcel infiltration as the maximum rate that water can be infiltrated over the percentage of previous area, and we assume unvegetated surfaces (impervious surface, bare ground) cannot contribute to rainfall detention capacity and will contribute to the volume of overland flow (Kelleher et al., 2020). The results suggested that rainfall of 1.26 × 108 m3 for a representative 1 h storm event will be absorbed and thereby diminish stormwater runoff (Table S8). The elimination is approximately two times of the average discharge of the Mississippi river per hour.

In addition, the newly built green infrastructures in shrinking cities form ‘cooling islands’ to improve the urban thermal environment. Recent work demonstrated a linear correlation between cooling intensity and green infrastructure patch area and another linear correlation between cooling range and green infrastructure patch area (Tan et al., 2021). We referred to the quantitative relationships to estimate the effects generated from the newly built green infrastructures in shrinking cities. The results revealed that if all vacant lands of shrinking space in the world are converted into green infrastructures, the total area of green infrastructures over 50 ha reaches 7,334 km2, leading to an air temperature reduction of 1.83–2.17°C with a 256–291 m range. The total area of green infrastructures between 10 and 50 ha reaches 3,621 km2, reducing air temperature by 1.37–1.59°C with a 202–226 m range. As for the green infrastructures smaller than 10 ha, the total area reaches 2,832 km2, and the cooling intensity is 1.05–1.29°C with a 162–178 m range.

Moreover, the newly built green infrastructures function as terrestrial carbon sinks. We referred to the relationship from an intensive field campaign involving 14,371 field plots (Tang et al., 2018) and applied a linear estimation on the increase of the total carbon pool in the world. It is estimated that 1.51 ± 0.08, 0.83 ± 0.04, 0.83 ± 0.05, or 0.88 ± 0.02 Pg C will be supplemented to the terrestrial carbon pool if the vacant lands are converted into forests, shrublands, grasslands, or croplands, respectively (Table S9).

Constructing compact cities

There are also other alternative solutions to remake a shrinking city into a compact city, such as solutions to improve the living environment (Lee and Erickson, 2017), maintain the subject well-being (Mouratidis, 2019), and reduce the expenditure on public services (van Vliet, 2019). Compact development has to be implemented to go against other dimensions of sustainable urbanization to preserve the livability of cities for their inhabitants. Owing to the outflow of population and capital, it is costly and redundant to keep the original urban size and infrastructure intensity (Schilling and Logan, 2008). If the residence, working place and public services are shrinking to concentration, the efficiency will be promoted in a shrinking city for different aspects. Several shrinking cities such as Detroit in the United States, Toyama in Japan have attempted to downsize their shrinking space by means of consolidating schools as well as their public transport routes.

We conducted several simulation experiments to evaluate the potential cut in expenditures in different aspects of public infrastructures for local governments. Education and healthcare-related POIs and road networks on the globe from OSM were acquired, representing the spatial information of three typical public infrastructures. We assumed that the city size and the number of public infrastructures in each shrinking city should be compressed proportionally to the reduction of nighttime light intensity. The results indicate that the urban land of 91,926 km2 in shrinking cities can be converted into nonurban areas to adapt to the outflow of population. In addition, the financial burden on 20% of both educational infrastructures and healthcare infrastructures in shrinking cities can be eliminated. Detailed effects from constructing compact cities within the 30 countries that have the largest number of shrinking cities in 2013–2018 are presented (Table S10). The eliminated proportions on educational infrastructures and healthcare infrastructures are above average in France (24%, 24%), China (23%, 21%), and Russia (23%, 24%). Moreover, the financial burden on 19% of road networks with a total length of 207,245 km in shrinking cities can be reduced, among which the roads and traffic occupy 49.3 and 50.7% (Table S11). In particular, the reductions of the financial burden on pathways in Africa, Asia, Australia, North America, South America, and Europe are 23%, 26%, 17%, 14%, 22%, and 20%, respectively.

Limitations of the study

Few limitations remain. First, the division in four phases is phenomenological, which is restricted by the continuity of remote sensing images.

Although the division of four phases can be utilized to investigate the trend of shrinking cities in the world, the time lags for different countries still remain. In this regard, global shrinking cities evolve differently and face variegated challenges.

Second, it is perhaps not effective to use a linear regression for nighttime light data prediction as the nighttime light image may serve as a type of manifestation of human distribution only and the linear regression model may fail to include important information in time series analysis. Hence, the driving factors contributing to shrinking cities should be further analyzed in future studies with a more detailed analytical framework.

Third, the lagging effect of policies should be considered when analyzing the relationship between urbanization strategies and shrinking cities.

Finally, the list of shrinking cities is narrowed in this study, which is because of the complexity of the formation of shrinking cities. Each shrinking city may have specific evolutionary trajectories, which were affected by different policies; therefore, it is difficult to draw a generalized law/rule from a global perspective. In fact, we do not attempt to explain the causal effect of national policy on the formation of shrinking cities in a country, though it is still important to do so in the future studies.

STAR★Methods

Key resources table

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Ying Long (ylong@tsinghua.edu.cn).

Materials availability

This study did not generate new unique materials.

Data and code availability

The data used in this study are all available from public data resources that have been appropriately cited in the manuscript. Code and any other additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.

Method details

Division of the four phases

We investigated the global shrinkage of cities in four phases by using night-time light remote sensing images. Compared with the ordinary demographic and statistical data, night-time light remote sensing data are considered more objective which could quickly reflect human activities and nature habitation (Shi et al., 2014; Wardrop et al., 2018; Wei et al., 2014) and widely applied to urban sciences in recent times.

DMSP-OLS nighttime light data with a 1 × 1 km2 spatial resolution are available from 1992 to 2012, and they were used to study the shrinking cities from 1992 to 2012. The year 2000, as the turn of the century, was used as the division between the first two phases. NPP-VIIRS nighttime image data with a 500 × 500 m2 spatial resolution are available since 2013, and they were applied to evaluate the global shrinking cities in the past few years (phase 3) and forecast the future ones (phase 4).

Natural city construction from nighttime images

This study proposed an alternative approach to identify shrinking cities based on the transformation of urban nighttime light data at the global level in three phases and further predict the future forecasts of shrinking cities. Every pixel in the nighttime light image for 2050 was predicted by a linear regression model for each corresponding pixel in 2013–2018. In particular, we identify urban shrinkage by utilizing natural cities as a benchmark to make it comparable for shrinking cities in different countries. The phenomenon of shrinking cities has been ignored by urban planners and policymakers, especially in developing countries. According to our prediction, more than one-third of the world’s cities will shrink to varying degrees by 2050.

Shrinking city identification

We collected monthly NPP-VIIRS nighttime remote sensing images captured in Jan.–Mar. and Oct.–Dec. from 2013 to 2018 spanning from 75 °N to 65 °S latitude to avoid the effects of missing data over the high-latitude regions of the Northern Hemisphere in summer months and computed the average intensity of each year pixel by pixel. To make the nighttime light data in different years comparable, we used the brightest areas and the darkest areas as pseudo-invariant features to correct the images (Wei et al., 2014). The identification of the shrinking cities in four phases was based on natural cities, and this was achieved by delineating the global urban land distribution images in 1990, 2000, 2010, and 2015, reserving only those with an area greater than 5 km2, following previous studies (Long et al., 2018). Next, we computed the ratio of change of the mean pixel values within each natural city in the four phases. The shrinking ratio at the city level is defined by the decreased percentage of nighttime light intensity from the start year to the end year in one phase. Following previous studies (Table S12), a natural city with a shrinking ratio lower than −10% over ten years is defined as a shrinking city. The proportion of shrinking cities at the country level is defined by the ratio of the number of shrinking cities among all natural cities within one country. In the actual experiments, although the time spans in each phase are not exactly ten years, they do not affect the robustness of their results according to our additional experiments (Table S13).

Literature review of the driving factors of shrinking cities

We conducted a literature review of the driving factors of shrinking cities on the Web of Science, and the searching query is (AK=(shrinking cities) OR AK=(urban shrinkage)) AND (TS=(reasons) OR TS=(causes) OR TS=(factors) OR TS=(resources) OR TS=(population) OR TS=(regional development) OR TS=(transformation) OR TS=(policy) OR TS=(war) OR TS=(demographics decline) OR TS=(low fertility rates) OR TS=(lack of labor)). A total of 284 research papers were retrieved, of which 100 were empirical papers related to shrinking cities after manual review and excluding papers unrelated.

Outmigration from shrinking cities in China

We acquired 4,538,201 migration trajectories in China in Nov. 2017–Nov. 2019 from Baidu Huiyan. First, approximately 170 million personal residence information were gathered from software and services of Baidu on different terminals with privacy information removed. Second, a movement of personal residence over 6 months from one city to another is regarded as a migration. Based on the natural cities in 2018 (Song et al., 2018) with a minimum size of 2 km2, we finally identified the migration of residents in China among 5,950 natural cities.

Global nighttime light forecast

We acquired NPP-VIIRS global nighttime light images from 2013 to 2018. A linear regression was applied to each corresponding pixel of the six global annual nighttime images to obtain an estimated global annual nighttime image in 2050, which had a size identical to those of the images from 2013 to 2018. The brightest areas and the darkest areas were selected as pseudo-invariant features to correct the nighttime light images in 2050. We scaled the values of the nighttime light data in 2050 constrained by the population data of each country from the United Nations and five SSPs so that the distribution of the sum of the nighttime light intensity was equal to the population distribution in each country. In this way, a global nighttime light map for 2050 with a 500 × 500 m2 resolution was achieved. A sensitivity analysis of natural cities and the constraint selection strategy for the forecasted shrinking cities was conducted for the validation (Table S14).

Quantification and statistical analysis

In this research, we analyzed the data using MATLAB R2018a and ArcGIS 10.5. The pixel-based average calculation for the DMSP-OLS nighttime light data was processed by the MATLAB R2018a mapping toolbox. The global natural city areas were generated with the assistance of the “raster to polygon” tool from the ArcGIS Toolboxes. The mean nighttime light pixel value of each natural city area was computed with the “zonal statistics as table” tool, which is also generated from the ArcGIS 10.5 system toolboxes. The cartogram figures in this study were created with the “cartogram geoprocessing tool” (http://arcscripts.esri.com/details.asp?dbid=15638).

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
DMSP-OLS nighttime light data (1992, 2000 and 2012)	NOAA	https://www.ngdc.noaa.gov/eog/dmsp/downloadV4composites.html
NPP-VIIRS nighttime light data (2013 and 2018)	NOAA	https://www.ngdc.noaa.gov/eog/viirs/download_dnb_composites.html
High-resolution global urban land (1990, 2000, 2010 and 2015)	GlobalUrbanLand	http://www.geosimulation.cn/GlobalUrbanLand.html
The world population prospects (2020–2100)	Department of Economic and Social Affairs (US)	https://population.un.org/wpp/Download/Standard/Population
The urbanization rate in 2015	UN	https://population.un.org/wup/Download/
SSP database	Climate change research community	https://tntcat.iiasa.ac.at/SspDb/
World POIs data	Open Street Map	http://download.geofabrik.de/index.html
China’s POIs (2018)	AMap	https://lbs.amap.com/
The GDP per capita, total population data (1992–2018) and average precipitation (2014)	Worldbank	https://data.worldbank.org/indicator
The global dataset of CO2 emissions and ancillary data (for 343 administrative cities)	GlobalCarbonCities	https://katirg.github.io/GlobalCarbonCities/
Migration statistics in China (2017–2019)	BaiduMap	https://huiyan.baidu.com/
Software and algorithms
ArcGIS 10.5	ESRI	https://www.arcgis.com/
Origin 12	OriginLab	https://www.originlab.com/
SPSS 22	IBM	https://www.ibm.com/
Matlab 2020	Mathworks	https://www.mathworks.com/