Spatiotemporal dynamics of urban sprawl in China from 2000 to 2020

ABSTRACT

Accurately and timely quantifying the dynamics of urban sprawl is essential for improving land use efficiency and land use planning. However, existing research mainly focused on the sprawl of a single city or urban agglomerations in China, while national scale studies mainly used short-term remote sensing data to quantify urban sprawl. Therefore, we still lack a long term and most up-to-date understanding of urban sprawl in China, especially in different regions, cities of different sizes. Here, we quantified the spatiotemporal dynamics of urban sprawl using urban sprawl index (USI) based on the latest population census, and analyzed its driving force in China during 2000 ~ 2020 using the optimal parameters-based geographical detector. The results showed that in the past two decades, Chinese cities were still experiencing urban sprawl, with an average USI of 3.04, indicating that the average annual growth rate of urban lands was 3.04% higher than that of the urban population. Overall, the sprawl showed a slowdown, with the USI dropping from 3.55 in 2000 ~ 2010 to 2.53 in 2010 ~ 2020. Among regions, urban sprawl was more severe in the western region, where the USI was 5.33 during 2000 ~ 2020 and 89.3% of the cities exceeded the national average sprawl speed. Recently, the driving force of transportation on urban sprawl substantially increased. In the future, the Territorial Spatial Planning should pay attention to confining excessive sprawl of small- and medium-sized cities in the western region of China.

GRAPHICAL ABSTRACT

KEYWORDS:

• urban geography
• urbanization
• urban expansion
• geodetector model
• territorial spatial planning
• urban sustainability

1. Introduction

Urban sprawl refers to a low-density, non-compact urban spatial expansion with discontinuous urban fabric (Downs 1999; Hamidi and Ewing 2014; Steurer and Bayr 2020). A prominent feature of such development is the extensive utilization of urban land caused by the mismatch between the speed of urban land expansion and that of urban population growth (Bhatta, Saraswati, and Bandyopadhyay 2010; Frenkel and Ashkenazi 2008; Gao et al. 2016; Jaeger et al. 2010). Urban sprawl may not only hinder high-efficient development of cities and economies but also creates a series of problems, including the recession of urban centers (Ewing 1997), land resource mismatches (Frenkel 2004), traffic congestion (Newman and Kenworthy 2011), substantial reductions in biodiversity (Liu et al. 2024; McKinney 2002), and declining air quality (Stone 2008), all of which seriously threaten the sustainable development of cities.

In the past century, the global urbanization rate (i.e. the portion of the urban population relative to total population) has increased from 10% in 1900 to 52.6% in 2011 (van de Wal et al. 2011; Chen et al. 2022). With the increasing level of urbanization, urban sprawl has gradually become a global issue that has been widely observed in the urbanization processes of various countries (Fulton et al. 2001; Gao et al. 2016). Between 1985 and 2015, the global urban area increased from 362,700 km² to 653,400 km², an increase of 80%, far higher than the urban population growth rate (52%) (Liu et al. 2020). In the United States, from 2000 to 2010, the compactness index value of 162 major cities decreased from 405.8 to 400, indicating continuous urban sprawl (Hamidi and Ewing 2014). The European Environment Agency found that between 2006 and 2009, cities in most European countries continued to sprawl at a rate of more than 1% per year, with some countries even exceeding 2% (Agency 2016). In China, the issue of urban sprawl also manifested. Between 2004 and 2014, the urban built-up areas in China increased by 78.5%, whereas the urban population growth rate during the same period was only 46% (Bai, Shi, and Liu 2014).

Urban sprawl has received widespread attention from many European and American scholars since the 1970s (Burchell 1997; Ewing 1997; Fernandez Milan and Creutzig 2016; Hamidi and Ewing 2014). Many theoretical and case studies have been conducted on urban sprawl, including on its definition (Ewing 1997; Goetz 2004; Peiser 2001), measurement and evaluation (Fulton et al. 2001; Lopez and Hynes 2003), driving factors (Salvati, Sateriano, and Bajocco 2013; Vicino, Hanlon, and Short 2007), and impact and regulation (Bengston, Fletcher, and Nelson 2004). The study of urban sprawl has also been conducted for portions of Asia. For example, Chettry (2022) combined multi-temporal Landsat images and the urban expansion index to detect the extent of urban sprawl in four rapidly expanding cities in India. Antalyn and Weerasinghe (2020) used Shannon entropy to evaluate urban sprawl and its impact on rural land quality in the Colombo region of Sri Lanka between 1997 and 2018. Fung (1981) first introduced the term “urban sprawl” in China to describe national urban development. Subsequently, along with the rapid urbanization process in China, studies related to urban sprawl dynamics have gradually increased. Empirical studies focusing on China has not only covered Beijing (Frederic and Huang 2004; Jiang et al. 2007), Shanghai (You 2016), Guangzhou (Yu and Ng 2007), Shenzhen (Lv et al. 2011) and other large cities but has also involved the urban sprawl characteristics of small and medium-sized cities (Gao et al. 2016) as well as the metropolitan area and urban agglomeration levels (Feng et al. 2019).

Quantifying and measuring urban sprawl is fundamental for urban sustainable development. The index method includes the single-dimension index method and the multi-dimensional index method, which has been widely adopted in previous research (Chettry 2023; Hamidi and Ewing 2014). The single-dimensional index method mainly focus on the relationship between urban population and urban land. For example, Lopez and Hynes (2003) used population density to measure urban sprawl and found the persistence of sprawl in metropolitan areas of United States in the 1990s. The multi-dimensional index method uses multiple indicators to comprehensively measure the degree of urban sprawl. For instance, Galster et al. (2001) proposed eight dimensions of urban sprawl and confirmed the utility of the approach in 13 cities. Jia et al. (2022) quantified urban sprawl in the Beijing-Tianjin-Hebei region of China from six dimensions and found that small cities faced an increasingly trend of urban sprawl. Although the multi-dimensional index method can comprehensively measure the degree of urban sprawl, it requires a large amount of data and was mainly used in small-scale empirical research (Ewing, Pendall, and Chen 2003; Lopez 2014; Zeng et al. 2015). By contrast, the single-dimension index method can effectively measure urban sprawl when national urban data is difficult to obtain (Bhatta, Saraswati, and Bandyopadhyay 2010).

In terms of research data, remote sensing data can observe urban sprawl from a multi-scale and long-term perspective. Garouani et al. (2017) measured urban sprawl using multi-temporal satellite images, and analyzed the dynamic process of urban sprawl in the Morocco Fez area. Ehrlich et al. (2018) used high-resolution remote sensing data to quantify urban sprawl in Europe, and emphasized the importance of the accuracy and timeliness of remote sensing data in measuring urban sprawl. Nighttime light (NTL) data provide a new approach to effectively and accurately obtain urban land area. Gao et al. (2016) used NTL data to extract the urban land and measured urban sprawl during 1990–2010, and found that small cities in western China showed the highest degree of sprawl. Lu et al. (2020) also used NTL images to analyze the spatiotemporal pattern of urban sprawl in the West Taiwan Strait of China and.

However, studies of urban sprawl in China still face some challenges. First, existing research mainly has used short-term remote sensing data to measure urban sprawl in China, and we still lack a long term and most up-to-date understanding of urban sprawl in China, especially in different regions, cities of different scales. Second, the Chinese government proposed the National New Urbanization Plan in 2014, and the coordinated development of urban land expansion and urban population growth became the key to the people-oriented urbanization (Chen et al. 2019). It is urgent to explore the dynamic evolution and driving forces of urban sprawl in China within the past two decades before and after the policy implementation, in order to inform future policies on preventing and controlling urban sprawl.

In this study, we used the dataset of urban built-up areas in China extracted from nighttime stable light (NTL), normalized difference vegetation index (NDVI), land surface temperature (LST) data, and combined it with population census data to analyze and quantify the spatiotemporal pattern dynamics and driving forces of urban sprawl in China from 2000 to 2020. First, we measured China’s urban sprawl status via the urban sprawl index (USI) among 315 cities (prefecture-level and above), and extended the measure to 2020. Then, we compared and analyzed the regional differences and city-size differences in urban sprawl and explored the variations in hotspots of urban sprawl. Finally, we used Geodetector to quantify the driving forces of urban sprawl in China over the past two decades. This study provides a scientific basis for Territorial Spatial Planning for constraining urban sprawl scientifically and effectively.

2. Study area and data

This study included 315 cities (prefecture-level and above) in China. As of 2022, the total number of cities at the prefecture level and above (excluding the Hong Kong, Macao, and Taiwan regions) in China was 337. Since the areas of some urban built-up areas were relatively small and barely changed from 2000 ~ 2010, we excluded the cities with built-up area less than 1 km² during 2000 ~ 2010 as well as several cities whose administrative divisions were adjusted during 2000 ~ 2020 (Figure 1).

Figure 1. The division of prefecture-level and above cities in different regions and city sizes in China. The four regional division (eastern, central, western and northeastern regions) was obtained from the classification scheme of the National Bureau of Statistics, the values represent the number of cities in each region.

Four types of data are mainly used in the study: urban area data, population census data, socio-economic data, and geographic auxiliary data (Table 1). The urban area data were derived from the Chinese urban built-up area dataset (1992 ~ 2020) (He et al. 2022), with a spatial resolution of 1 km. Specifically, built-up area is defined as the area with more than 50% in coverage by non-vegetated, human-constructed elements, and this definition excludes rural settlements. He et al. (2014) and Xu et al. (2016) constructed this dataset by combining nighttime light (NTL), the normalized difference vegetation index (NDVI), and land surface temperature (LST) data through the training sample selection, support vector machine classification, and interannual series correction processes. This dataset integrates multi-source remote sensing data, which can effectively extract residential, commercial, and public facility land within the city, and it can also identify industrial land, warehouse, transportation, and road on the edge of the city. Compared with other commonly used urban built-up land data (i.e. the GHS built-up data, the GHS SMOD data, and the ESACCI data), this dataset has a higher accuracy (an overall accuracy of 91% and a Kappa index of 0.50) than those of other datasets (ranging from 84%-90% and 0.15–0.20), respectively. This dataset has been widely used to extract urban land information at national and regional scales (Huang et al. 2020; Huang et al. 2021; Liu et al. 2024; Luan and Li 2021).

Table 1. Data sources and descriptions.

Type	Name	Description	Source	Resolution	Year
Urban area data	Chinese urban built-up area dataset	Based on NTL; NDVI; LST, classified by support vector machine	National Tibetan Plateau Data Center (https://data.tpdc.ac.cn/)	1 km	2000–2020
Remote sensing image data	Nighttime stable light data	Based on DMSP/OLS and VIIRS, produced composite images of NTL	The US National Geophysical Data Center (https://ngdc.noaa.gov/eog/)	1 km	2000–2020
	Land surface temperature data	Composed of average LST values of MODIS 8-day composite imagery	The National Aeronautics and Space Administration (https://ladsweb.modaps.eosdis.nasa.gov/)	1 km	2000–2020
	Normalized difference vegetation index data	Composite of maximum NDVI values over 16 days	The National Aeronautics and Space Administration (https://ladsweb.modaps.eosdis.nasa.gov/)	1 km	2000–2020
Urban population data	Population census data	Accurate statistics of the permanent population living in the city	National Bureau of Statistics of China (https://www.stats.gov.cn/sj/pcsj/)	/	2000/2010/2020
Socio-economic data	China City Statistical Yearbook	Exploring the driving force of urban sprawl	China Data Insights (https://data.cnki.net/yearBook/)	/	2000–2020
Geographic auxiliary data	The administrative boundaries	The Vector Boundary of Chinese Cities	The National Geomatics Center of China (http://bzdt.ch.mnr.gov.cn/)	Vector	2019

The urban permanent population data came from the decadal population census data set released by the National Bureau of Statistics in 2000, 2010, and 2020. Compared with the annual population data in the Statistical Yearbook, decadal population census data counted the number of registered nonagricultural people, migrating peasant workers, and even foreigners living in cities, which can more accurately reflect the number of the urban population (Gao et al. 2016; Wang and Zhou 1999). After considering both continuity and accuracy, we have ultimately decided to use the decadal census data prepared by Liu et al. (2022).

Urban area data was extracted based on the administrative boundary of Chinese cities. The population census data included all urban permanent population within the same city administrative boundary (prefecture level), which includes multiple municipal districts, county-level cities, and counties administrated by the city. In terms of the time consistency of the data, the urban area data was annually continuous from 2000 to 2020. They were extracted from the annual NTL, LST and NDVI (see Table 1 for detailed description and image sources). All of the NTL, NDVI and LST data were resampled to a spatial resolution of 1 km. For the urban population data, population census data was surveyed and estimated for each decade. To guarantee the time consistency, we used the decadal population and urban area data for 2000, 2010 and 2020, respectively.

The socio-economic data included 10 indicators under the five dimensions of national policies, economic development, society, transportation, and education level in 2000, 2010, and 2020, respectively. The data came from the China City Statistical Yearbook and the China Provincial Statistical Yearbook released by the National Bureau of Statistics of China and local governments.

The geographic auxiliary data mainly included 1:100,000 administrative boundary data of China; these data were released by the National Geographic Information Center of China. Due to the absence of urban demographic data for Taiwan, Hong Kong and Macao, this study mainly analyzed the urban sprawl dynamics in mainland China and used ArcGIS 10.7 software to process the geographical data.

3. Methods

This study basically included three parts (Figure 2). First, we used the USI to measure the extent of sprawl in 315 Chinese cities from 2000 to 2020. After that, we analyzed the general dynamic characteristics of urban sprawl in China over the past two decades. Then, after classifying the city size and dividing the regions, we analyzed the spatiotemporal differences and dynamic characteristics of urban sprawl among different regions and city sizes, followed by applying the hotspot analysis method to identify the spatial aggregation characteristics of urban sprawl and the dynamic change processes in these hotspots. Finally, we used geodetector to analyze the driving force of socioeconomic factors on urban sprawl in China.

Figure 2. The methodological workflow of the study.

3.1. Measuring the degree of urban sprawl

The index method has generally been adopted to measure urban sprawl. These indices can be classified as the single-dimension index method and multidimensional index method (Hamidi and Ewing 2014; Jaeger et al. 2010). The multidimensional index method can comprehensively gauge the degree of urban sprawl, but it requires a large number of input data as well as a complex calculation process. In addition, the results calculated by the multidimensional index method are often difficult to compare with relevant studies, so they are often applied to small-scale urban sprawl studies (Ewing, Pendall, and Chen 2003; Lopez 2014; Sudhira, Ramachandra, and Jagadish 2004). The single-dimension index method focuses mainly on the dynamic relationship between the population and land use, and this limits their ability to identify different characteristics of urban sprawl. However, the relatively simple calculation process of the single-dimension index method allows them to measure urban sprawl readily and effectively when data are difficult to obtain (Bhatta, Saraswati, and Bandyopadhyay 2010). Moreover, using the same indicator to measure urban sprawl enables comparisons to be made among cities (Lopez 2014). Therefore, when the datasets required by multidimensional index methods are unavailable, it is reasonable to adopt a single-dimensional method to measure urban sprawl at large scales.

In this study, the spatiotemporal dynamic characteristics of China’s urban sprawl are described by a single-dimensional index. Specifically, we used the USI constructed by Fulton et al. (2001) to quantify the extent of urban sprawl. The index has been widely used in the research of urban sprawl (Gao et al. 2016; Guan et al. 2020; Li and Li 2019). This index involves two important variables that are key features of urban sprawl: the growth rate of the urban area and the growth rate of the urban population. Thus, the index can effectively reflect the main characteristics of urban sprawl. The USI is calculated as follows:

(1) $\mathit{US}{I}_{\left(t_1,t_2\right)}=\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UA}}-\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UP}}$(1)

where $\mathit{US}{I}_{\left(t_1,t_2\right)}$is the USI of a city during the$t_1$-$t_2$period and $\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UA}}$and $\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UP}}$ are the annual growth rates of the urban area and population from $t_1$-$t_2$, respectively. The calculation formulas of $\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UA}}$ and $\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UP}}$ are expressed as follows:

(2) $\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UA}}=\left({\left(\frac{U{A}_{t_2}}{\mathit{U}{A}_{t_1}}\right)}^{\frac{1}{t_2-t_1}}-1\right)\times 100$(2)

(3) $\mathrm{\Delta }{R}_{\left(t_1,t_2\right)}^{\mathit{UP}}=\left({\left(\frac{U{P}_{t_2}}{\mathit{U}{P}_{t_1}}\right)}^{\frac{1}{t_2-t_1}}-1\right)\times 100$(3)

where $\mathit{U}{A}_{t_1}$and $\mathit{U}{A}_{t_2}$refer to the urban land areas at times $t_1$ and $t_2$, respectively, and $\mathit{U}{P}_{t_1}$ and $\mathit{U}{P}_{t_2}$ represent the urban populations at times $t_1$ and $t_2$, respectively.

To simplify the statistical analysis, we divided the degree of sprawl samples according to the USI calculation results from 2000 to 2020 and set 0 as the threshold value to identify whether urban sprawl occurred. Cities with sprawl were then divided according to the average value of the national sprawl index. This classification was applied in two periods: 2000 ~ 2010 and 2010 ~ 2020.

3.2. Analyzing the spatiotemporal dynamics of urban sprawl in different regions and sizes

This study divided the study area into four regions, east, central, west, and northeast (Figure 1), by referring to the classification scheme of the National Bureau of Statistics. In the eastern region, there were 86 cities, including 10 provincial-level administrative regions such as Beijing, Tianjin, and Hebei. In the central region, there were 81 cities, including 6 provincial-level administrative regions such as Shanxi, Anhui, and Jiangxi. In the western region, there were 112 cities, including 12 provincial-level administrative regions such as Sichuan, Shaanxi, and Xinjiang. In the northeastern region, there were 36 cities, including 3 provincial-level administrative regions such as Liaoning, Jilin, and Heilongjiang.

To divide the cities by size, we consulted the classification standard issued by the State Council of China in 2014, in which cities were divided into five categories and seven grades (megacities, super cities, large cities A, large cities B, medium cities, small cities A, and small cities B). To simplify this city size classification scheme, we merged the categories and sorted the 315 cities into megacities (>5 million), large cities (1 ~ 5 million), medium-sized cities (0.5 ~ 1 million), and small cities (<0.5 million) based on their urban population in 2020 (Figure 1, Supplementary Table S1). Based on this classification, we calculated and analyzed the spatiotemporal dynamics of urban sprawl in different regions and among different city sizes.

Moran’s I usually has been utilized to quantify spatiotemporal characteristics by identifying spatial correlation and spatial heterogeneity. We utilized a global Moran’s I test to measure spatial autocorrelation of USI. The positive and significant values of global Moran’s I indicated that USI was more clustered, while the opposite indicated more dispersed. The equation for calculating global Moran’s I was set as follows:

(4) $\mathrm{global}\mathrm{Mora}{\mathrm{n}}^{\prime }\mathrm{s}\mathrm{I}=\frac{n}{\sum _{\mathit{i}=1}^n\sum _{\mathit{i}\ne \mathit{j}}^n{W}_{\mathit{ij}}}\times \frac{\sum _{\mathit{i}=1}^n\sum _{\mathit{i}\ne \mathit{j}}^n{W}_{\mathit{ij}}\left(x_i-x^{\ast }\right)\left(x_j-x^{\ast }\right)}{\sum _{\mathit{i}=1}^n{\left(x_i-x^{\ast }\right)}^2}$(4)

where $x_i$ denotes the value of urban sprawl in city $\mathit{i}$, $x^{\ast }$ denotes the mean value of $\mathit{x}$, $\mathit{n}$ stands for the number of cities, and $W_{\mathit{ij}}$ denotes the spatial weight matrix.

To observe the changes in urban sprawl hotspots, we chose the Getis-Ord G_i* index to present the local spatial clustering distributional characteristics of urban sprawl (Ord and Getis 2001). First, we calculated the local spatial correlation index Getis-Ord G_i* of the USI in 2000 ~ 2020, 2000 ~ 2010 and 2010 ~ 2020. Then, the local Getis-Ord G_i* statistics were used to determine whether there were high-value clusters or low-value clusters in the urban sprawl results. We obtained the threshold of 2.58 value based on the high and low scores (through a 99% confidence test, and identified the locations corresponding to this threshold. Finally, we used the hotspot analysis tool in the ArcGIS spatial analysis module for spatial visualization. The formulas are shown as follows:

(5) $G_i^{\ast }=\frac{\sum _j^n{W}_{\mathit{ij}}{X}_j}{\sum _j^n{X}_j}$(5)

(6) $\mathit{Z}=\frac{G_i^{\ast }-\mathit{E}\left(G_i^{\ast }\right)}{\sqrt{\mathit{Var}\left(G_i^{\ast }\right)}}$(6)

where $G_i^{\ast }$ is the agglomeration index of urban unit $\mathit{i}$, $\mathit{Z}$ is the significance of the agglomeration index, $W_{\mathit{ij}}$ is the spatial weight defined by the distance, $X_i$ and$X_j$ are the attribute values of urban units $\mathit{i}$ and $\mathit{j}$, respectively, and $\mathit{E}\left(G_i^{\ast }\right)$ and $\mathit{Var}\left(G_i^{\ast }\right)$ are the mathematical expectation of and variance in $G_i^{\ast }$, respectively.

To investigate whether urban sprawl differences existed among different regions and at different sizes, we used the Kolmogorov-Smirnov (K-S) method to test the normal distribution of the USI in different regions and at different city-size scales. For the samples that met the normal distribution, we used the analysis of variance method. For the others, we used the Kruskal‒Wallis (K-W) rank sum test method to see whether distinctions existed among different regions and different sizes in China. To further gauge and determine the sources of these urban sprawl differences among different regions, we used the Theil index to measure and decompose the regional differences in urban sprawl (Li et al. 2020; Zhang et al. 2021). In addition, we decomposed the overall difference into the differences within and among the four major regions of eastern, central, western and northeastern China. The specific calculation and decomposition methods are expressed as follows:

(7) $\mathit{T}=T_W+T_b$(7)

(8) $T_{\mathit{wj}}=\sum _i\frac{\mathit{U}{\mathit{S}}_{\mathit{ji}}}{\mathit{U}{S}_j}ln\frac{\mathit{U}{\mathit{S}}_{\mathit{ji}}/\mathit{U}{S}_j}{1/n_j}$(8)

(9) $T_w=\sum _j\frac{U{S}_j}{\mathit{US}}{T}_{\mathit{wj}}$(9)

(10) $T_b=\sum _j\frac{U{S}_j}{\mathit{US}}ln\frac{U{S}_j/\mathit{US}}{n_j/\mathit{n}}$(10)

where $\mathit{i}$ represents city $\mathit{i}$, $\mathit{j}$ represents region $\mathit{j}$, $n_j$ represents the number of cities included in region $\mathit{j}$, $\mathit{n}$ represents the number of cities in China, $\mathit{T}$, $T_W$, $T_b$, and $T_{\mathit{wj}}$ represent the Theil index of urban sprawl in China as a whole, within the national region, between regions and within $\mathit{j}$ regions, respectively, and $\mathit{US}$, $\mathit{U}{S}_j$, and$\mathit{U}{S}_{\mathit{ji}}$ represent the USI values of the nation, region $\mathit{j}$, and city $\mathit{i}$ within region $\mathit{j}$, respectively.

3.3. Analyzing the driving force of urban sprawl

The driving mechanism of urban sprawl is a complicated process, and the dominant factors of urban sprawl vary among countries and regions (Osman, Divigalpitiya, and Arima 2016; Wang, Shi, and Zhou 2020; Wu, Li, and Wang 2021). Existing studies have shown that urban sprawl in China is closely related to policy, population, economy, transportation, education and other factors (D. Feng et al. 2019; Tian, Ge, and Li 2017). Accordingly, we selected 16 potential factors (Supplementary Table S2) that influence urban sprawl from five dimensions: national policies, economic development, society, transportation, and education level (Supplementary Table S2). To avoid redundancy and multicollinearity among the 16 potential factors, we used the Pearson correlation to select a total of 10 factors (two factor of each dimension) for Geodetector (Table 2). In addition, we used the change rate of each factor as the independent variable since the dependent variable (USI) is an indicator representing the dynamic process of urban sprawl.

Table 2. Selected driving factors of urban sprawl in China.

Dimensions	Code	Indicator
National policies	X1	Public budget expenditure
	X2	Public budget revenue
Economic development	X3	GDP share of secondary industry
	X4	Foreign investment amount
Society	X5	Population density
	X6	Employees of the unit
Transportation	X7	Per capita road area
	X8	Per capita number of buses
Education level	X9	Number of universities
	X10	Number of university students

Geodetector is a statistical method that can reveal drivers by detecting spatial differentiation (Wang et al. 2017). It is mainly divided into four parts (i.e. factor detector; interaction detector; risk detector and ecological detector), and has been widely used in the study of various influencing factors and mechanisms (He et al. 2023; Li and Geng 2023). Geodetector is based on the idea of spatial layered heterogeneity, which divides the study space into sub-regions by variables, and the spatial variance within each sub-region and among different sub-regions are compared to evaluate the determinant power of potential explanatory variables (Luo et al. 2016; Wang, Zhang, and Fu 2016; Wang et al. 2010). It is crucial to determine the optimal scale of spatial layered heterogeneity through spatial data discretization when using Geodetector (Song et al. 2020; Wang et al. 2017). However, traditional Geodetector generally determines spatial data discretization methods and spatial scales based on experiences, and cannot determine the methods objectively and quantitatively. Therefore, we used the newly developed method, optimal parameters-based geographical detector (OPGD) to optimize the discretization of spatial data (Song et al. 2020). The OPGD model selects the optimal combination of discretization methods and break number of spatial strata for each continuous geographical variable, determined through the computation of q values with a factor detector. The parameter combination with the highest q value is then chosen for spatial discretization, reflecting the variable’s utmost significance in spatial stratified heterogeneity.

Specifically, we used the factor detector to analyze the driving force of each factor for urban sprawl, and used the interaction detector to determine the comprehensive driving force of two factors after interaction. To determine the optimal spatial discretization parameters, following the method of Song et al. (2020), we included five classification methods (equal interval; natural breakpoint; quantile classification; geometric interval; and standard deviation classification), and screened the best combinations of spatial discretization parameters with the largest q value from a range of 3 ~ 11 classes (Supplementary Fig. S1-S3). In addition, to further explore the driving forces of urban sprawl in different regions and city sizes, we also use the OPGD model to calculate the optimal spatial discretization method and spatial scale for each region.

Therefore, we adopted the factor detector to estimate the driving forces of each selected factor for urban sprawl as follows (Wang and Xu 2017):

(11) $\mathit{q}=1-\frac{1}{N{\sigma }^2}\sum _{\mathit{h}=1}^L{N}_h{\sigma }_h^2$(11)

where $\mathit{N}$ is the number of sample units in the entire region; $\mathit{L}$ is the number of sub-regions; $N_h$ is the number of sample units in the sub-region; ${\sigma }^2$ is the variance of the urban sprawl index in the entire region; ${\sigma }_h^2$ is the variance of the sub-region. The $\mathit{q}$ value is utilized to evaluate the explanation ability of each driver on urban sprawl. For $\mathit{q}\in \left[0,1\right]$, a larger q value indicates that the driving factor has a stronger explanatory power for urban sprawl.

Then, we adopted the interaction detector to superimpose two driving factors in space and calculate their interactive driving force $\mathit{q}\left(\mathit{x}1{\cap }^{x}2\right)$. The relationship between the two drivers is determined by comparing the factor detector values $\left(\mathit{q}1,\mathit{q}2\right)$ and the interaction detector values $\mathit{q}\left(\mathit{x}1{\cap }^{x}2\right)$. The relationship between the two drivers can be divided into the following categories: nonlinear-weaken, uni-variable weaken, bi-variable enhance, independent and nonlinear-enhance (Supplementary Table S3).

Specifically, we quantified the driving forces of urban sprawl from 2000 to 2020, 2000 to 2010, and 2010 to 2020, respectively, to explore the changes in the driving forces of urban sprawl. Also, we quantified the driving forces of urban sprawl in different regions and for different city sizes from 2000 to 2020 to explore the differences in driving forces of urban sprawl over space and city city-size scales.

4. Results

4.1. Urban sprawl dynamics in China

Overall, a sprawling trend was shown in China from 2000 to 2020, with sprawl existing in most of the 315 cities (Figure 3). The national average USI was 3.04. The total area of the 315 cities in China increased from 2.58 × 10⁴ km² to 9.01 × 10⁴ km², with an average annual growth rate of 6.45%. The total urban population increased from 45.17 × 10⁷ to 88.41 × 10⁷, with an average annual increase of 3.41%. Among the 315 cities, 309 cities experienced urban sprawl (USI >0), accounting for 98.1% of the total number of cities. Moreover, 232 cities (73.7%) experienced sprawl exceeding the national average (USI = 3.04), while only 1.9% of cities did not experience sprawl development. In the past two decades, urban sprawl has exhibited spatial aggregation to some extent. The global Moran’s I index of the USI was 0.37, with a Z score of 18.35; and the p value passed the 99.9% confidence level test (Supplementary Table S4). It indicated that urban sprawl was not distributed randomly; instead, it exhibited both spatial agglomeration and dependence.

Figure 3. The values of urban sprawl index (USI) of 315 cities in China during 2000 ~ 2020.

Overall, the speed of urban sprawl in China has shown a downward trend over the past two decades (Figure 4), with the national average USI dropping from 3.55 to 2.53. From 2000 to 2010, the overall sprawl trend was relatively severe. The average annual growth rate of the built-up areas of 315 cities in China was 7.40%, while the average annual growth rate of the urban population was 3.85%, resulting in a USI value of 3.55. From 2010 to 2020, the average annual growth rate of the built-up area of 315 cities in China was 5.51%, while the average annual growth rate of the urban population was 2.98%. Thus, the USI dropped to 2.53.

Figure 4. Dynamics of urban sprawl measured by urban sprawl index (USI) during the (a) 2000~2010 and (b) 2010~2020 periods.

Meanwhile, the number of sprawling cities gradually increased (Figure 4). During 2000 ~ 2010, 283 cities experienced sprawl, accounting for 89.8% of the national cities analyzed herein. During 2010 ~ 2020, the number of sprawling cities increased to 298, accounting for 94.6% of the analyzed cities. In addition, compared to the previous period, the sprawl phenomena of 13 cities were reversed in 2010 ~ 2020, and the USI was reduced to below 0. However, additional 28 cities experienced sprawling development. Further analysis with Moran’s I index showed that the urban sprawl pattern in China shifted from a concentrated pattern a to dispersed pattern, with the index decreasing from 0.36 during 2000 ~ 2010 to 0.20 during 2010 ~ 2020 (Supplementary Table S4). The p values for both periods passed the 99.9% confidence level test.

4.2. Dynamics of urban sprawl in different regions

Overall, there were salient regional differences in urban sprawl in China over the past two decades (Figure 5a). From 2000 to 2020, the urban sprawl in the western region was more severe than that in other regions, with the average USI reaching 5.33, the highest among the four regions in China. The central and northeastern regions took second and third place, with average USIs of 4.00 and 3.04, respectively. In the eastern region, the urban sprawl was relatively low, with an average USI of only 1.95. Regarding the proportions of cities with varying levels of urban sprawl in different regions, there were salient differences among regions (Figure 5b). Across China, 73.7% of cities exceeded the national average sprawl level (USI = 3.04). In the western region, 89.3% of cities exceeded the national average sprawl level. In the central and northeastern regions, 70.4% and 72.2% of cities exceeded the national average sprawl level, respectively. In the eastern region, this value was 57.0%, substantially lower than the national average.

Figure 5. Variations in urban sprawl among different regions: (a) urban sprawl index in different regions during 2000 ~ 2020 (note: the box plot displays the average, the median, the first and third quartiles of urban sprawl index); (b) the proportion of cities with varying urban sprawl values relative to the national averages.

In the past two decades, urban sprawl has shown different trends in different regions of China. According to the dynamic changes in the USI and the proportions of the cities with varying levels of urban sprawl during 2000 ~ 2010 and 2010 ~ 2020 (Figure 6), the urban sprawl in the four regions presented three trends: acceleration, stability, and slowdown. The northeastern regions experienced an accelerating trend of urban sprawl. The average USI in this region increased from 2.49 to 3.58 (Figure 6a), and the proportion of cities that exceeded the national average level increased from 41.7% to 77.8% (Figure 6b). The urban sprawl in the central region of China has tended to stabilize. The average USIs in the central region during 2000 ~ 2010 and 2010 ~ 2020 were 3.96 and 4.04, respectively (Figure 6a). The number of cities that exceeded the national average urban sprawl level increased from 63.0% to 79.0% (Figure 6b). The USI gradually declined in the eastern and western regions of China. In the past two decades, the average USI in the eastern region dropped from 2.58 to 1.07 (Figure 6a), and the proportion of cities exceeding the national average sprawl level dropped from 62.8% to 40.7% (Figure 6b), indicating that the urban sprawl situation declined significantly. Meanwhile, the average USI in the western region dropped from 6.01 in 2000 ~ 2010 to 4.65 in 2010 ~ 2020 (Figure 6a), and the proportion of cities exceeding the national average sprawl level accounted for 80.4% and 83.9% of all analyzed cities, respectively (Figure 6b).

Figure 6. Variations in urban sprawl among different regions: (a) and (c) average urban sprawl index (USI) in different regions during 2000~2010 and 2010~2020, respectively; (b) and (d) the proportion of cities with varying urban sprawl values relative to the national averages during 2000~2010 and 2010~2020, respectively.

The Kruskal-Wallis test results confirmed that there were significant differences in the USIs among different regions during 2000 ~ 2020 (Supplementary Table S5). The K-W test statistic was 59.38, and the p value was less than 0.001, passing the 99.9% confidence level test. The Bonferroni multiple mean comparison results showed that significant differences existed between any two regions except between the eastern and northeastern regions and between the central and northeastern regions.

According to the analysis of the Theil index, the overall regional urban sprawl differences in China showed a downward trend during 2000 ~ 2020, and the Theil index dropped from 0.38 in 2000 ~ 2010 to 0.31 in 2010 ~ 2020 (Table 3), indicating that urban sprawl in China tended to be evenly distributed spatially. In addition, the overall differences in urban sprawl in China were caused mainly by intraregional differences, which were far larger than interregional differences. From 2000 to 2020, the contribution rate of intraregional differences reached 79.2%, while the contribution rate of interregional differences was only 20.8%. In terms of regions, there were significant distinctions in the regional differences in urban sprawl in China (Table 3). The difference in urban sprawl in the eastern region presented an expanding trend. The Theil index in the eastern region was 0.25 during 2000 ~ 2010 and rose to 0.46 during 2010 ~ 2020. The difference in urban sprawl in the central and western regions gradually decreased. The difference in urban sprawl in the northeastern region remained basically unchanged over the past two decades. The Theil index in the northeastern region was 0.28 during 2000 ~ 2010 and dropped to 0.26 during 2010 ~ 2020.

Table 3. Theil value of the urban sprawl index and the contributions of interregional and intraregional urban sprawl in China at different periods.

Classification	2000~2020		2000~2010		2010~2020
	Theil index	Contribution	Theil index	Contribution	Theil index	Contribution
China	0.24	100%	0.38	100%	0.31	100%
Interregional	0.05	20.8%	0.06	15.8%	0.05	16.1%
Intraregional	0.19	79.2%	0.32	84.2%	0.26	83.9%
East	0.24	—	0.25	—	0.46	—
Central	0.17	—	0.31	—	0.19	—
West	0.21	—	0.44	—	0.16	—
Northeast	0.16	—	0.28	—	0.26	—;

For the hotspots of urban sprawl, these hotspots showed a centralized distribution trend during 2000 ~ 2020, with the spatial distribution pattern of urban sprawl gradually moving westward (Figure 7). Hotspots were located mainly in the western region, including in Sichuan Province, Guizhou Province, Chongqing City, and Xinjiang Uygur Autonomous Region. There were also some hotspots in the eastern and central regions. Most of these hotspots were located around the eastern provinces of Fujian and Zhejiang, as well as in the central provinces of Hunan and Jiangxi. The cold spots of urban sprawl were concentrated in northern China and parts of the northeastern region.

Figure 7. Spatial clusters of urban sprawl detected by Getis-Ord G_i* during the (a) 2000~2020; (b) 2000~2010; and (c) 2010~2020 period.

For the spatial changes in hotspots over time, the hotspots of urban sprawl showed a trend of gradual westward development (Figures 7b, 7c). The hotspots of urban sprawl in 2000 ~ 2010 were mainly located in the south of China. From 2010 to 2020, the hotspots in the eastern and central regions gradually moved westward, and new urban sprawl hotspots emerged in Ningxia, Gansu, Xinjiang, and other regions. At the same time, the hotspots continued to shrink, while the cold spots gradually expanded (Figures 7b, 7c). From 2000 to 2010, the number of cities located in urban sprawl hotspots was 66, and the number of cities located in cold spots was 60. From 2010 to 2020, the number of cities in urban sprawl hotspots dropped to 46, while the number of cities in cold spots rose to 64.

4.3. Dynamics of urban sprawl at different city-size scales

In the past two decades, significant differences in urban sprawl have occurred among different urban sizes in China (Figure 8a). In general, medium-sized cities and small cities had the highest levels of urban sprawl, with USIs of 6.79 and 7.03 from 2000 to 2020, respectively; these USIs were significantly higher than the national average (Figure 8a). The level of urban sprawl in large cities took third place, with a USI of 4.32 from 2000 to 2020, higher than the national average (Figure 8a). The level of urban sprawl in megacities was the lowest, with a USI of only 1.48 from 2000 to 2020; this USI was lower than the national average (Figure 8a).

Figure 8. The average urban sprawl index obtained at different city scales during the (a) 2000~2020; (b) 2000~2010 and 2010~2020 periods.

There were substantial differences in the sprawl trends among cities of different sizes during 2000 ~ 2010 and 2010 ~ 2020 (Figure 8b). The speed of sprawl in megacities slowed substantially between these periods. From 2000 to 2010, the USI of megacities was 2.67, lower than the national average of 3.55. Furthermore, from 2010 to 2020, the USI of megacities decreased to 0.31, far lower than the national average of 2.53 (Figure 8b). The speed of sprawl in large cities was generally in a relatively stablized status. The USIs of large cities were 4.39 in 2000 ~ 2010 and 4.25 in 2010 ~ 2020, higher than the national averages of 3.55 and 2.53 in the same period (Figure 8b). The speed of sprawl in medium-sized cities remained at a relatively high level. The USIs in 2000 ~ 2010 and 2010 ~ 2020 were 6.71 and 6.88, respectively (Figure 8b). The speed of sprawl in small cities increased substantially during the study period. From 2000 to 2010, the USI of small cities was 4.93, while from 2010 to 2020, the USI reached 10.16, much higher than the national average (Figure 8b).

The Kruskal-Wallis rank sum test results showed that there were significant differences in the USIs of different city sizes from 2000 to 2020 (Supplementary Table S6). The K-W test statistic was 61.35, and the p value was less than 0.001, passing the 99.9% confidence level test. The Bonferroni multiple mean comparison results showed that there were significant differences in the degree of sprawl between megacities and large, medium-sized, and small cities. There was also a significant difference in the degree of urban sprawl between large- and medium-sized cities, while there were no significant differences among the other groups.

4.4. The driving forces of urban sprawl

The results showed that national policies, economic development, and transportation were strongly associated with urban sprawl (Figure 9). Their explanatory power (q values) for urban sprawl exceeded 0.150 during 2000 ~ 2020. Among the three dimensions, GDP share of secondary industry (X3), per capita road area (X7), per capita number of buses (X8) exhibited a higher level of association with urban sprawl in China than the other factors, with an explanatory power (q value) greater than 0.200. Meanwhile, society and education levels also had a certain positive association with urban sprawl, with an explanatory power (q value) ranging from 0.047 ~ 0.159. In addition, in the process of rapid urbanization in China in the past two decades, the driving force of transportation conditions for urban sprawl had greatly increased. From 2000 to 2010, the driving force of transportation conditions was the lowest among the five dimensions, with a q value of only 0.073. However, from 2010 to 2020, the q value increased to 0.340, which had a much higher driving effect on urban sprawl than other dimensions.

Figure 9. Driving factors of urban sprawl based on the factor detection method: explanatory powers of driving factors during (a) 2000~2020; (b) 2000~2010 and 2010~2020; and (c) driving forces of urban sprawl from different dimensions.

Over the last two decade, the explanatory power of these selected driving factors for urban sprawl had gradually increased (Figure 9). Among the five dimensions, the driving forces of transportation showed the greatest increase, with the p values of the per capita road area (X7) and the per capita number of buses (X8) increasing from 0.067 and 0.079 during 2000 ~ 2010 to 0.336 and 0.344 during 2010 ~ 2020, respectively. In addition, the q values of public budget expenditure (X1), GDP share of secondary industry (X3), the foreign investment amount (X4), and the number of university students (X10) also increased during the two decades, with an increasement ranging from 0.015 to 0.093. Meanwhile, the q values for public budget revenue (X2), population density (X5), employees per unit (X6), and the number of universities (X9) were gradually declining, but the overall changes were relatively small.

The results of interaction detector showed that there are certain types of interactive forces for urban sprawl in China, and the q value varied between 0.047 and 0.597 (Figure 10). Among the 45 pairs of driving factor for urban sprawl, there were 44 pairs exhibiting non-linear enhancing relation with urban sprawl, while only 1 pair was showing bi-variable enhancing relation. The average q value based on the interaction detector was 0.412, which was 2.7 times the average value for the factor detector. In other words, the geographical spatial pattern of urban sprawl in China was driven by the interaction of multiple factors. Specifically, the top three pair of interactive drivers with the largest q values were public budget expenditure and employees per unit (q = 0.597), employees per unit and the per capita number of buses (q = 0.596), the per capita road area and the per capita number of buses (q = 0.590).

Figure 10. Driving factors of urban sprawl during 2000~2020 based on the interactive detection of geodetector (the value represents the interactive effect of a combinations of factors, e.g. 0.461 for the combination of X₁ and X₅ on urban sprawl).

Among the four regions (Figure 11a), the q values for urban sprawl in eastern and northeastern China were greater than the values in central and western regions. The average q value for urban sprawl for the five dimensions in the eastern and northeastern regions were 0.418 and 0.428, respectively, while the average values in the central and western regions were 0.252 and 0.257, respectively. Similar to the overall situation in China, the eastern and northeastern regions showed a strong association with urban sprawl in terms of transportation, national policies, and economic development (0.395 ~ 0.587). Nevertheless, the association with urban sprawl in the central and western regions were relatively low in these dimensions, especially in terms of economic development (0.235 ~ 0.238) and education level (0.130 ~ 0.144).

Figure 11. The driving forces of urban sprawl in (a) different regions; and (b) different city sizes.

The association between these drivers and urban sprawl also showed substantial differences among cities with varying sizes (Figure 11b). Generally speaking, the q values of these indicators for urban sprawl in megacities, medium-sized cities and small cities was higher than those in large cities. Among them, the average q values in megacities, medium-sized cities and small cities for the five dimensions were 0.425, 0.345 and 0.395, respectively, while the value in big cities was only 0.162. For the five dimensions, we found that urban sprawl in megacities had the strongest association with education level (0.509), economic development (0.441), and society dimensions (0.528). Urban sprawl in medium-sized cities and small cities had a strong association with national policies (0.384 ~ 0.437) and transportation (0.362 ~ 0.507).

5. Discussion

5.1. Comparison of relevant studies

This study combined the latest population and urban land data to analyze the dynamics and driving factors of urban sprawl in China over the past 20 years and compared the urban sprawl characteristics among different sizes and regions. Compared with the findings during 2000–2010, we found that the speed of urban sprawl in China in the last decade has shown a downward trend, with the USI dropping from 3.55 to 2.53. Through this study, we can enrich our understanding on urban sprawl in China, especially for providing up-to-date information on urban sprawl in China during 2010 ~ 2020. It can provide guidance for related international urban sprawl research, especially those cities in Global South facing the challenge of efficient utilization of urban land, reflected in the SDG 11.3.1 (ratio of land consumption rate to population growth rate).

Our results were overall in agreement with previous findings, especially in terms of historical measurement of urban sprawl before 2015. We found that China’s USI over the past two decades was 3.04, which was consistent with a previous research of Li and Li (2019) (USI = 3.16 during 2006–2014). From a dynamic perspective, we found that urban sprawl in China has gradually slowed since 2010, which was in line with Deng et al. (2020). From a regional perspective, we found that urban sprawl in the western region was more severe than that in other regions; and this finding was also consistent with previous results. For example, Gao et al. (2016) found that urban sprawl was more severe in the western region than in the eastern and central regions. Zhang et al. (2021) also found that the urban sprawl speed in the western region being higher than those in the eastern, central, and northeastern regions. Compared with traditional remote sensing image data, Shi et al. (2023) found that nighttime light data have higher reliability in evaluating urban expansion, which is also confirmed by our urban sprawl measurements based on the nighttime light data in China. In addition, the spatial autocorrelation of urban sprawl in our study exhibited a clustering phenomenon, which is in line with previous findings. For example, Torbick and Corbiere (2015) found that agglomeration of urban land often occurs in the urban sprawling process in the coastal areas of the northeastern United States. In terms of driving forces of urban sprawl, national policies and economic development played an important driving role in urban sprawl, which was consistent with the conclusions of Kuang et al. (2016).

Our results also add new understanding of urban sprawl in China. Zhuang et al. (2022) simulated future urban land expansion and claimed that urban sprawl in China is still manifesting after 2015, we further revealed that the momentum of urban sprawl gradually slowed down during 2010–2020. In addition, Liu et al. (2018) found that the sprawl of small and medium-sized cities in the western region of China was most severe during 2000 ~ 2010. Our new analysis further revealed that, small and medium-sized cities in the northeast region have also faced severe urban sprawl in recent years. In addition, Li and Li (2019) found that the hotspot areas of urban sprawl during 2006–2014 were concentrated in the Yangtze River Delta, Sichuan Basin, and North China Plain. Our results further showed that the speed of urban sprawl in the Yangtze River Delta and North China Plain gradually slowed during 2010 ~ 2020, while a new hotspot of urban sprawl appeared in the northwest region of China. In addition, our comparison within regions and between regions further showed that intraregional differences in urban sprawl were far greater than the interregional differences. Therefore, we believe that reducing intraregional differences is the key to reducing the spatial differences in urban sprawl and achieving coordinated development within regions.

5.2. The driving mechanism of urban sprawl in China and policy implications

Urban sprawl is a complex process dominated by anthropogenic drivers. Therefore, analyzing the driving forces of urban sprawl is fundamental promoting sustainable urban development (Wang, Shi, and Zhou 2020; Wu, Li, and Wang 2021; Ye et al. 2024). We selected five perspectives, the national policies, economic development, society, and transportation for further discussion and suggestions.

First, national policies play a significant role in the urban sprawl process in China (Feng et al. 2019; Liu et al. 2018). National policies confine urban sprawl under the umbrella of the Territorial Spatial Planning, regulating public budget expenditure and public budget revenue. This study found that the explanatory power of public budget expenditure and public budget revenue for urban sprawl from 2000 to 2020 was greater than 10%, and they were important driving factors for urban sprawl. For cities with varying sizes, we found that the annual urban areas growth rates in small and medium-sized cities reached 9.67% and 9.35%, respectively, far higher than values in large cities (7.66%) obtained for and megacities (5.12%), while the annual urban populations growth rates were quite similar, ranging from (2.56%~3.63%). Therefore, we believe that rapid growth in urban areas of small and medium-sized cities is the main reason for the severe sprawl observed in China, and this as highly associated with the National New Urbanization Plan issued in 2014, which strongly restricted the rapid expansion of super-large cities and large cities while encouraged the development of small and medium-sized cities. Therefore, the Chinese government should pay attention to the important role of the new national level planning for land use in China, the Territorial Spatial Planning, in limiting urban sprawl. In specific, the new plan should scientifically promote the delineation of urban development boundaries, and strictly control the excessive expansion of small and medium-sized cities in the western region (Gao et al. 2023). In addition, local governments should formulate scientific urban planning, rationally adjust the layout of urban spatial development, and prevent the phenomenon of disorderly sprawl (Tong et al. 2017).

Second, urban sprawl is closely related to economic development (Zhang et al. 2022). Although industrial and economic orientation promotes urban development, industrial facilities in suburban areas may lead to urban sprawl. We found that GDP share of secondary industry from 2000 to 2020 was most strongly associated with for urban sprawl, with a q value reaching 0.265. As the development of most cities was industry- and economy-oriented, a large amount of urban land in the suburbs of cities is used for industrial facilities and infrastructures, which is one common phenomenon of urban sprawl (Zhang et al. 2018). However, some studies also argued that the speed of urban sprawl can be controlled and mitigated with the transformation of economic development pattern (Lu et al. 2020). Our study also echoed with this phenomenon. We found that urban sprawl manifested quickly in the developed eastern Yangtze River Delta during 2000 ~ 2010, but gradually weakened during 2010 ~ 2020, which may be related to the transformation of the industrial structure in the eastern region. In other words, it is necessary to focus on the quality of economic growth in the process of controlling urban sprawl to simultaneously upgrade industrial structure and improve urban land efficiency (Fang, Li, and Zhang 2017).

Third, this study found that society factors had a relatively low association with urban sprawl, but urban sprawl in small and medium-sized cities in the northeastern region increased in the past decade. With the continuous loss of population and the reduction of employment opportunities in Northeast China, the mismatch between urban population and urban land area has intensified urban sprawl. It indicated that urban sprawl in the northeastern region in recent years have been related to the phenomenon of urban shrinkage. We found that while the land areas of small and medium-sized cities in the northeastern region grew rapidly, the urban population constantly decreased. From 2010 to 2020, the average annual growth rate of the land area of small and medium-sized cities in the northeastern region was 6.63%, while the average annual growth rate of the corresponding urban population was −1.48% during the same period, and the USI was as high as 8.11. Currently, urban shrinkage caused by the continuous loss of the urban population has gradually become one of the key drivers of urban sprawl in the northeastern region (Chen et al. 2018). In pursuit of a better living environment, some urban residents in northeastern region migrate to the larger cities in this region or other large cities in developed coastal regions in China, which exacerbated the emergence of urban sprawl in this region (Guo et al. 2021; Yang et al. 2021). Therefore, local governments should adjust the total supply of urban lands in a timely manner to promote the coordinated development of urban lands and the population.

Fourth, it is worth noting that the driving forces of transportation on urban sprawl in China substantially increased during the last two decades. Intuitively, the development of transportation facilities and the prevalence of private transportation tools such as automobiles in China reduce the travel costs of residents while promoting the expansion of urban space and have hence become an important factor inducing urban sprawl (Mehriar, Masoumi, and Mohino 2020). In the past two decades, the q values for per capita road area and per capita bus number had increased from 0.067 and 0.079 to 0.336 and 0.344, respectively, which were much higher than other driving factors. Also, the q values were much higher in small (0.36) and medium cities (0.51) than in large cities (0.21) and megacities (0.27). Such finding is consistent with previous conclusions that high-speed rail brought great convenience to residents’ travel, and led to the intensification of urban sprawl (Long, Zheng, and Song 2018; Taotao and Dandan 2018). The government generally tends to build high-speed rail stations in urban suburbs, but this may bring about problems such as commuting and carbon emissions for residents in the city centers and suburbs. Therefore, the government needs to pay attention to land use efficiency when establishing high-speed rail stations in the future, and can rely on existing satellite cities instead of uncontrolled inefficient sprawl (Deng et al. 2020; Hou et al. 2023; Wu, Li, and Wang 2022).

Finally, education level also plays a nonnegligible role in urban sprawl. Local governments usually build university towns on the edge of cities meet to the demand of higher education expansion, and this urban expansion approach may contribute to urban sprawl (Yue, Liu, and Fan 2013). Our study found that the association between urban sprawl and education level was lower in central (q = 0.144) and western regions (q = 0.130), but stronger in eastern (q = 0.300) and northeastern regions (q = 0.332). This was in line with the rapid development of university towns in developed coastal areas of China (Shao et al. 2018). It also provided valuable experience for the development of university towns in western region of China. Since the economic conditions, terrain conditions and vegetation conditions in the west may be worse, more attention should be paid to the ecological and environmental effects of building university towns in the suburbs to avoid causing greater environmental damage.

5.3. Limitations and future perspectives

There are still several limitations in this study, mainly reflected in two aspects. First, due to the lack of reliable large-scale datasets, the urban sprawl measurement in this study only focused on the difference between the urban land growth rate and urban population growth rate, which makes it difficult to measure urban sprawl from a socioeconomic perspective (Zhuang et al. 2022). Urban sprawl is a complex process dominated by anthropogenic drivers. In addition to the 10 driving factors we have selected, it is also influenced by natural environmental factors such as terrain, elevation, and distance from water sources. Therefore, our research may not comprehensively reflect the driving mechanisms of urban sprawl.

In the future, we should comprehensively consider urban form and urban micro-dynamics, and establish a multidimensional assessment method to study the complexity and diversity of urban sprawl (Guo et al. 2023; Song et al. 2018). In addition, long-term and time-series urban sprawl data can be used to explore the spatiotemporal timing between urban area expansion and population growth. Finally, we can consider additional driving factors, including natural environment, population growth, economic development, and government intervention, to comprehensively analyze the driving forces of urban sprawl.

6. Conclusions

In the past two decades, Chinese cities have continued to experience urban sprawl, with an average USI of 3.04. However, the speed of urban sprawl gradually slowed, and the USI decreased from 3.55 in 2000 ~ 2010 to 2.53 in 2010 ~ 2020. Second, there were significant regional differences in urban sprawl. Among the four regions, the urban sprawl situation in the western region was still relatively severe. The USI in the western region was 5.33 from 2000 to 2020, and 89.3% of cities in this region exceeded the national average urban sprawl speed. In addition, the sprawl speeds of small cities increased significantly over the study period, and the trend of urban sprawl became increasingly severe. The USI of small cities increased from 4.93 in 2000 ~ 2010 to 10.16 in 2010 ~ 2020. Finally, national policies, economic development, and transportation were strongly associated with urban sprawl. Their explanatory power (q values) for urban sprawl exceeded 15% during 2000 ~ 2020.

In the future, placed-based strategies should be adopted to address urban sprawl in different regions of China. First, the Chinese government should pay attention to the important role of the Territorial Spatial Planning in limiting urban sprawl and strictly control the excessive expansion of small and medium-sized cities in the western region. Second, local governments should formulate a scientific urban transportation development plan, to control the inefficient sprawl caused by the construction of high-speed rail stations. Finally, more attention should be paid to the ecological and environmental effects of building university towns in the suburbs to avoid causing greater environmental damage.

Acknowledgments

We thank the anonymous reviewers and editors for their insightful and critical comments, which have improved the quality of the manuscript.

Data availability statement

The urban area data used in this study can be accessed through the National Tibetan Plateau Scientific Data Center, which were derived from the Chinese urban built-up area dataset (1992 ~ 2020)V1.0 (https://doi.org/10.11888/HumanNat.tpdc.272851).The socio-economic data was available at the China City Statistical Yearbook (https://data.cnki.net/yearBook/). The urban population data from population census data can be obtained through https://www.stats.gov.cn/sj/pcsj/. The geographic auxiliary data can be accessed through the National Geomatics Center of China (http://bzdt.ch.mnr.gov.cn/). The data that support the findings of this study are available from the corresponding author upon reasonable request.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15481603.2024.2351262

Additional information

Funding

This work was supported by the National Natural Science Foundation of China (grant No. 42361144859); the Beijing Nova Program (Grant No. 20220484163); and Beijing Normal University Tang Scholar.