Multidimensional Identification of County-Level Shrinkage by Improved Mapping of Urban Entities Based on Time-Series Remote Sensing Data: A Case Study of Yangtze River Delta Urban Agglomerations

Abstract

Although measurements of urban shrinkage in China have received much attention, most have relied on statistical yearbook data based on political–administrative city boundaries, and remote-sensing-based quantification is mainly one-dimensional. This has caused problems in incorporating rural areas and spatiotemporal inconsistencies, as well as an inadequate understanding, which has subsequently resulted in an inaccurate shrinkage identification. This study merely utilized the latest multisensory and time-series remote sensing data, including nighttime light, land use, and population grids, to quantify the spatiotemporal patterns of multidimensional shrinkage based on the county-level urban entity mapping of Yangtze River Delta urban agglomerations (YRD-UAs) from 2003 to 2023. County-level urban entities were acquired from a pioneering mapping effort that utilized city-specific commuting distance and land use maps. The results demonstrated that urban entities in 215 counties grew at a generally slowing pace. The degree of economic, population, and space shrinkage was mainly slight, and the shrinking trajectory was dominated by temporary shrinkage. Most counties experienced population shrinkage in their coastal-oriented distribution, whereas economic shrinkage affected the fewest counties, with the lowest spatial clustering occurring northward. Population shrinkage also displayed the highest spatial autocorrelation, but its agglomeration weakened against space shrinkage clustering. This study concluded that the exclusive utilization of remote sensing products to measure urban-entity-based multidimensional shrinkage reduced the uncertainty associated with rural area inclusion and resulted in satisfactory assessment accuracy. The spatiotemporal patterns of multidimensional shrinkage suggested strengthening ecological land allocation within urban entities across the entire region, implementing polycentric development strategies in the north, as well as enhancing county-level economic governance in the northwest. This study presents a spatiotemporally comparable methodology for quantifying the multidimensional shrinking of county-level urban entities at a large scale and contributes to further optimizing the developments of YRD-UAs.

1. Introduction

As a global phenomenon since the mid-20th century, urban shrinkage occurs when the paradigm of growth is broken and manifests as population loss, economic decline, and eco-environmental deterioration [1,2,3]. China, the world’s largest developing country, is promoting a new urbanization strategy centered on urban agglomerations and metropolitan circles, which exhibits a distinguished urban shrinking issue alongside increasing urbanization [4,5,6]. Urban shrinkage in China has attracted much attention, and the shrinking degree and its process, spatial divergence, and mechanisms have become a topic of research [7,8,9].

The delineation of urban shrinkage encompasses both single-dimensional identification based on population indicators and multidimensional definition integrating two or more indicators from population, economic, social, and spatial dimensions [10,11]. The data sources comprise statistical-yearbook-derived datasets encompassing demographic parameters (permanent population and employment rates) and socioeconomic metrics (gross output value and fixed asset investments), complemented by remote-sensing-based spatial indicators from nighttime light, land use maps, and population products [10,12,13]. Statistical-yearbook-derived and census datasets were traditionally used as true values in shrinking measurements based on political–administrative units, but these data sources were limited by spatiotemporal inconsistencies due to different statistical standards across regions, departments, and time [14,15]. Therefore, it has become a cost-effective and spatiotemporally comparable way to delineate urban shrinkage using satellite remote sensing data [16,17]. The inherent diversity of urban shrinkage patterns poses significant challenges for deriving appropriate indicators from remote sensing data to accurately identify multiscale shrinkage. The emergence of novel sensors and multidimensional remote sensing products presents new opportunities for identifying optimal urban shrinkage indicators. The satellite sensors, such as the Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) and the National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (NPP-VIIRS), provide nighttime light remote sensing data as a common proxy to estimate changes in human activities and the urbanization process [18,19]. Remote-sensing-derived land use maps with spatiotemporal consistency observations have been widely adopted to quantify urbanized land dynamics [6,20,21]. The emerging open global population grids have been extensively used to calculate population loss for shrinkage identification, particularly with the updated LandScan data, which offers the superiority of microscale research [14,22,23]. However, the existing remote-sensing-based literature has mostly focused on a one-dimensional estimation of population loss within political–administrative cities using nighttime light data or population grids [24,25]. It overlooked the multiple manifestations of urban shrinkage and the important role of rural-to-urban migration, especially in the county-level shrinking process, resulting from the single-dimension definition and inclusion of rural areas in the political–administrative units [10,26]. Urban-entity-based shrinkage studies have remained relatively nascent, predominantly employing nighttime light data or impervious surface maps, coupled with uniform regional commuting thresholds, to delineate prefecture-level urban entities for unidimensional shrinkage characterization [26,27,28]. Namely, a critical research gap has persisted in spatiotemporally explicit and efficient quantification of multidimensional shrinking of county-level urban entities in a spatiotemporally comparable manner.

To fill the above research gap, this study aimed to develop a remote-sensing-based methodology for the comprehensive analysis of degree, trajectory, and spatial patterns of multidimensional shrinkage within county-level urban entities. Earning a reputation for a well-developed urban hierarchy, unparalleled economic vitality, and cutting-edge innovation capacity, the Yangtze River Delta urban agglomerations (YRD-UAs) hold strategic primacy in advancing the national modernization blueprint and shaping an all-round globalization paradigm of China. Taking this typical metropolitan region as the study area, the shrinking indices for the economic, population, and space dimensions were calculated using the latest NPP-VIIRS-like nighttime light data, the China Land Cover Dataset (CLCD) land use maps, and the LandScan Global population count grids. The specific goals were to identify county-level urban entities from impervious surfaces in land use maps using city-specific commuting distances; determine spatiotemporal county-level multidimensional shrinkage from remote sensing products; and quantify the spatiotemporal patterns of county-level shrinkage, including trajectory and spatial heterogeneity.

2. Materials and Methods

2.1. Study Area and Data Sources

The study area, i.e., YRD-UAs, is located between 115°45′22″ and 122°50′3″E and 27°8′36″ and 34°28′4″N. It spans four provincial-level administrative divisions—Shanghai Municipality, Jiangsu, Zhejiang, and Anhui provinces—in China and covers an area of approximately 2.25 × 10⁵ km² (Figure 1). This megaregion encompasses 27 cities including one municipality and 26 prefecture-level cities forming a contiguous urban network across the eastern coastal zone and the lower Yangtze River Basin (

Table 1. The administrative divisions of YRD-UAs.

Provincial-Level	City (Municipality and Prefecture-Level)	County Number
Shanghai	Shanghai	16
Jiangsu	Nanjing, Wuxi, Changzhou, Suzhou, Nantong, Yancheng, Yangzhou, Zhenjiang, Taizhou	68
Zhejiang	Hangzhou, Ningbo, Wenzhou, Jiaxing, Huzhou, Shaoxing, Jinhua, Zhoushan, Taizhou	75
Anhui	Hefei, Wuhu, Ma’anshan, Tongling, An’qing, Chuzhou, Chizhu, Xuancheng	56

). Since the reform and opening-up in 1978, YRD-UAs have played one of the most important supporting roles in the strategic development of China, whose permanent resident population reached 2.38 × 10² million in 2023 [29]. In terms of the economy, the gross domestic product (GDP) of YRD-UAs reached CNY 2.63 × 10⁴ billion in 2023, accounting for 20.86% of the total national GDP [30]. The ongoing expansion of the urban scale of YRD-UAs is indicated by the increasing growth of impervious surfaces, with an average annual rate of 7.17 × 10² km²/year from 2003 to 2023. At the same time, the imbalance of regional development in YRD-UAs is highlighted, especially in county-level administrative units, where population and economic factors are distributed unevenly [31,32]. A total of 215 county-level administrative units belonging to Shanghai Municipality and 26 prefecture-level cities are adopted as the basic units to quantify spatiotemporally explicit patterns of urban entity shrinkage.

2.2. Data and Methods

To quantify spatiotemporally explicit patterns of county-level shrinkage based on urban entities using multidimensional remote sensing data, the workflow contained three major sections, as presented in Figure 2. Urban entities of county-level administration units were identified from impervious surfaces of land use maps using the commuting monitoring report and the location of local governments. Then, based on the distribution of urban entities, using multisource remote sensing products, corresponding indices for three dimensions were calculated: average nighttime light brightness for economy, the empty city index for population, and the ecological land proportion for space. After that, shrinking indices for four periods for each dimension were constructed to illustrate explicit county-level shrinking variations. Spatiotemporal pattern analysis using shrinking trajectory classification and spatial autocorrelation quantification was therefore conducted. Finally, spatiotemporal patterns of county-level shrinkage in YRD-UAs were comprehensively revealed from a multidimensional comparison and summary based on shrinking degree, trajectory, and clusters. This pioneering methodology improved urban entity mapping for county-level multidimensional shrinkage identification merely based on time-series remote sensing data.

2.2.1. Remote Sensing Products and Preprocessing

The three remote sensing products, adopted between 2003 and 2023, in this study included NPP-VIIRS-like nighttime light data, CLCD land use maps, and the LandScan Global population count raster (

Table 2. The adopted remote sensing products in the years of 2003, 2018, 2013, 2018, and 2023.

Source	Description	Spatial Resolution	Usage
NPP-VIIRS-like	Nighttime light raster	500 m	Average brightness calculation from economic dimension
CLCD	Land use maps	30 m	Urban entity identification, built-up area calculation for empty city index from population dimension, area calculation of ecological land and non-construction land for ecological land proportion from space dimension
LandScan Global	Population raster	1000 m	Total population calculation for empty city index from population dimension

). By effectively advancing both DMSP-OLS and NPP-VIIRS nighttime light sources, the NPP-VIIRS-like dataset has proven its ability to clearly monitor socioeconomic activities within urban areas as well as their long-term dynamics [28,33]. In this study, NPP-VIIRS-like products were downloaded by year and extracted by the study area using ArcGIS 10.2 software for the subsequent calculation of average brightness within the urban entity of each county-level unit.

Based on training samples and Landsat images, using a random forest classifier and Google Earth Engine platform, CLCD land use maps are annually generated with an overall accuracy of 79% and have earned a reputation for accurate land use area estimation [34,35,36]. These land use maps were downloaded according to location and year and then reclassified into three types: built-up (impervious surfaces in CLCD), ecological (forest, shrub, grassland, water, snow and ice, and wetland in CLCD), and other lands (cropland and barren in CLCD). Namely, non-construction land includes ecological and other lands. Then, the area of built-up, ecological, and other lands was calculated separately.

Exploiting census and remotely sensed data in a multivariable machine learning modeling approach, Oak Ridge National Laboratory’s LandScan Global products have been widely employed to map wall-to-wall population accurately in multiscale regions [37,38]. These LandScan Global grids were downloaded by year and extracted by the study area to facilitate the follow-up calculation of total population within the urban entity of each county-level unit.

2.2.2. Identification of Urban Entities

Based on the above preprocessed land use maps, the final boundaries of urban entities of a total of 215 county-level units within YRD-UAs were determined by two steps with ArcGIS 10.2 software [26], as illustrated as Figure S1: (1) the adjacent patches of built-up land were merged into one large patch, and the location of the county-level’s government was set as the center, and (2) the distance between merged patches and the center was calculated, and these patches within a certain radius around the center were kept as urban entities. A certain radius for each county-level unit was determined by the commuting monitoring report for major Chinese cities 2024 released by Baidu Map, as listed in

Table 3. The radius of different cities around local governments based on the commuting monitoring report for major Chinese cities 2024 released by Baidu Map ¹.

City	Radius (km)	Criterion
Shanghai	9.8	The commuting distance listed in the report.
Nanjing	8.8
Wuxi	7.4
Changzhou	7
Suzhou	7.9
Nantong	8.6
Hangzhou	8.1
Ningbo	7.3
Wenzhou	6.6
Shaoxing	7.9
Yancheng, Jiaxing, Jinhua, Taizhou_Z, Hefei	8.3	The average commuting distance of megacities listed in the report.
Yangzhou, Zhenjiang, Taizhou_J, Huzhou, Wuhu, An’qing, Chuzhou	7.7	The average commuting distance of Type I large cities listed in the report.
Zhoushan, Ma’anshan, Tongling, Chizhu, Xuancheng	8	The average commuting distance of Type II large cities listed in the report.

¹ The report can be downloaded from the website: https://jiaotong.baidu.com/reports (accessed on 12 March 2025). Taizhou_Z and Taizhou_J represent Taizhou City of Zhengjiang and Jiangsu provinces, respectively.

[39]. Using these two steps, we separately identified urban entities of 215 county-level units in 2003, 2008, 2013, 2018, and 2023.

2.2.3. Quantification of Multidimensional Shrinking

We quantified county-level shrinkage across economic, population, and space dimensions [10,40]. The average brightness from preprocessed NPP-VIIRS-like products within each acquired county-level urban entity was calculated as a perspective on the economic dimension [41,42]. As for the population dimension, the built-up land area from land use maps was divided by the total population from the LandScan Global dataset within each county-level urban entity, yielding the value of the empty city index [43]. Owing to the disorderly urban construction, the space of YRD-UAs had been under inefficient utilization, and its ecosystem function had been degraded, resulting in deteriorating environmental quality [44]. Thus, the percentage of non-construction land occupied by ecological land within a circular region centered at the local county-level government with a certain radius as listed in Table 3 was computed as a space dimension in this study.

After that, the shrinking index for each dimension was constructed as the following formula [40,45]:

$$S_{i (t_1,t_2)}^k={\displaystyle \frac{x_{\left(i,t_2\right)}^k-x_{\left(i,t_1\right)}^k}{x_{\left(i,t_1\right)}^k}}$$

(1)

where $S_{i (t_1,t_2)}^k$ is the shrinking index of the dimension i (average brightness of nighttime light from economic dimension or empty city index from population dimension or ecological land proportion from space dimension) of a county-level unit k in the period from t₁ to t₂, and $x_{\left(i,t_1\right)}^k$ and $x_{\left(i,t_2\right)}^k$ are the values of the dimension i in the k of years t₁ and t₂, respectively. When $S_{i (t_1,t_2)}^k$ is less than zero, the county-level unit k experiences shrinkage of dimension i, observed in the period from t₁ to t₂. Based on previous studies, the following criteria was used to classify shrinking indices into three degrees of shrinkage: severe ($S_{i (t_1,t_2)}^k$ < −0.3), moderate (−0.3 ≤ $S_{i (t_1,t_2)}^k$ < −0.15), and slight (−0.15 ≤ $S_{i (t_1,t_2)}^k$ < 0) [46].

2.2.4. Spatiotemporal Pattern Analysis

Aiming at determining temporal patterns of county-level shrinkage of YRD-UAs, the trajectory types were identified based on the urban shrinkage trajectory typology proposed by SCiRN [47]. Indeed, the shrinking trajectories of county-level urban entities from 2003 to 2023 for each dimension in YRD-UAs were classified into three types, i.e., continuous, episodic, and temporary (

Table 4. The classification system of shrinking trajectories.

Shrinking Trajectories	Subtype	Characteristic
Continuous shrinkage	/	Shrinking in all four 5-year periods (2003–2023)
Episodic shrinkage	Periodic shrinkage	Shrinking over the period, with a stable or even growing rate in at least one 5-year period (shrinking in 2008–2023 or 2013–2023)
	Discontinued shrinkage	Shrinking over the period, with a stable or even growing rate in at least one 5-year period (shrinking in 2003–2018, 2003–2013 or 2008–2018)
Temporary shrinkage	/	Shrinking in at least one 5-year period or discontinued for two 5-year periods

and Figure 2).

In this study, global Moran’s I, an effective tool for measuring spatial autocorrelation, was calculated based on the location and attribute values of the multidimensional shrinkage using the following formula [48,49]:

$$I_{ (t_1,t_2)}^i={\displaystyle \frac{n}{S}}{\displaystyle \frac{{\sum }_{k=1}^n{\sum }_{q=1}^n{w}_{k,q}{z}_{i (t_1,t_2)}^k{z}_{i (t_1,t_2)}^q}{{\sum }_{k=1}^n{\left[z_{i (t_1,t_2)}^k\right]}^2}}$$

(2)

where $I_{ (t_1,t_2)}^i$ is the global Moran’s I statistics of the dimension i shrinkage in the period from t₁ to t₂; w_k,q is the spatial weight between the county-level unit k and q; $z_{i (t_1,t_2)}^q$ is the deviation of the absolute values of $S_{i (t_1,t_2)}^k$ (dimension i shrinking index of the k between t₁ and t₂) from its mean; $S$ is the aggregate of all the spatial weights; and n is equal to the total number of county-level units, i.e., 215. When $I_{ (t_1,t_2)}^i$ reaches statistical significance with p-values less than 0.05 and is more than zero, the dimension i shrinkage in the period from t₁ to t₂ exhibits a statistically significant positive spatial autocorrelation pattern, indicating clustering. While $I_{ (t_1,t_2)}^i$ is less than zero with a statistical significance, this shrinkage shows a statistically significant negative spatial autocorrelation pattern, indicating dispersion. In addition, $I_{ (t_1,t_2)}^i$ is close to zero with a statistical insignificance, representing a random distribution [50].

Then, an agglomeration and outlier analysis was conducted based on the local Moran’s I calculation [9]:

$$I_{i (t_1,t_2)}^k={\displaystyle \frac{\left|S_{i (t_1,t_2)}^k\right|-\overline{\left|S_{i (t_1,t_2)}\right|}}{\left[\frac{{\sum }_{q=1,q\ne k}^n{(\left|S_{q (t_1,t_2)}^k\right|-\overline{\left|S_{i (t_1,t_2)}\right|})}^2}{n-1}\right]}}{\sum }_{q=1,q\ne k}^n{w}_{k,q}(\left|S_{q (t_1,t_2)}^k\right|-\overline{\left|S_{i (t_1,t_2)}\right|})$$

(3)

where $I_{i (t_1,t_2)}^k$ is the local Moran’s I statistics of the dimension i shrinkage for the county-level unit k in the period from t₁ to t₂; $\left|S_{i (t_1,t_2)}^k\right|$ is the absolute values of $S_{i (t_1,t_2)}^k$ (same as Formulas (1) and (2)); $\overline{\left|S_{i (t_1,t_2)}\right|}$ is the mean of the $\overline{\left|S_{i (t_1,t_2)}\right|}$; and w_k,q and n are the same as Formula (2). To illustrate the results of local Moran’s I with a statistical significance, a local indicator of spatial association (LISA) clustering map was produced to identify the four types of spatial correlation: (1) high–high cluster (HH), representing relatively severe shrinkage aggregation, is classified as a hot spot; (2) low–low cluster (LL), representing relatively slight shrinkage aggregation, is classified as a cold spot; (3) high–low cluster (HL), where the relatively severe shrinkage is surrounded by the relatively slight shrinkage, is classified as an outlier of hot spots; and (4) low–high cluster (LH), where the relatively slight shrinkage is surrounded by the relatively severe shrinkage, is classified as an outlier of cold spots [51].

3. Results

3.1. Spatiotemporal Variations in Urban Entities

The spatiotemporal distribution of a total of 215 county-level units within YRD-UAs is drawn in Figure 3. The urban entities displayed an outstanding expansion with distinct variations. It was concentrated in the northeastern region of YRD-UAs, especially the surrounding area of Shanghai Municipality.

The area of urban entities as well as their changes was summed up by cities and provinces, as shown in Figure 4. Generally, between 2003 and 2023, the urban entities of YRD-UAs grew from 6315.47 to 15,051.21 km² with a slowing growth rate from 9.10 to 2.08%/year. It was demonstrated that Shanghai Municipality played a dominant role in the urban entity expansion of YRD-UAs, which grew from 1094.85 to 1945.40 km². However, the rate of area change in urban entities in Shanghai Municipality was the smallest among the four provincial-level administrative divisions, which rapidly decreased from 7.90 to 0.94%/year, followed by Jiangsu, Zhejiang, and Anhui provinces. Urban entities in Jiangsu and Anhui provinces also exhibited a progressively decelerating increase, especially Wuxi, Suzhou, and Chizhou cities. At the same time, urban entities in Zhejiang Province exhibited a slow growth trajectory from 10.33 to 2.60%/year in the first 15 years and then increased by 2.67%/year. Among the twenty-six prefecture-level cities of three provinces, Suzhou, Hangzhou, Ningbo, Nanjing, Wuxi, and Hefei were the top six cities that obtained the largest area of urban entities. The majority had a decelerating growth, excluding nine cities. The cities of Nantong, Yancheng, Zhenjiang, and Taizhou in Jiangsu Province experienced an accelerating increase in the first 5 years and then grew at a decelerated speed. At the same time, the cities of Wenzhou, Jiaxing, Huzhou, Zhoushan, and Taizhou in Zhejiang Province had the same growth trajectory as the whole Zhejiang Province, as reported above.

3.2. Multidimensional County-Level Shrinking Dynamics

Based on the average brightness of nighttime light within urban entities of YRD-UAs, the county-level shrinking index of the economic dimension is drawn in Figure 5. It was noted that the county-level economic shrinkage changed with some fluctuations. Specifically, the economic shrinkage was initially greatly mitigated in the southern region of YRD-UAs at the beginning, and then intensified in the north, ultimately resulting in a slight alleviation.

The economic shrinking degree was summed up by cities, as shown in Figure 6. The outstanding spatiotemporal variations in county-level economic shrinkage were depicted, especially for the first 10 years. During the study period, the province with the highest number of economically shrinking counties underwent a sequential transition. Initially observed in Zhejiang Province, this distinction subsequently shifted to Jiangsu Province, followed by a transitional period where Anhui Province temporarily held this position, before ultimately reverting back to Jiangsu Province. Shanghai Municipality was the only provincial-level administrative division that exhibited a continuous mitigation of economic shrinkage. The provincial difference was relatively large for the first five years, and then it decreased over the next 10 years, whereas it increased in the recent five years. For the four study periods, the cities registering the greatest number of economic shrinking counties shifted sequentially: Ningbo and Jinhua (Phase I: nine counties), Jiaxing (Phase II: two counties), Xuancheng (Phase III: five counties), and Yancheng and Zhenjiang (Phase IV: four counties).

Based on the area of built-up and total population within the urban entity of YRD-UAs, the county-level shrinking index of population dimension is presented in Figure 7. The population shrinkage was even more severe compared to county-level economic shrinkage. In detail, population shrinkage originally intensified in the north, and then the number of shrinking counties increased to a slighter degree in the east, which eventually slightly rose.

The population shrinkage displayed fewer spatiotemporal variations among cities and provinces compared to economic shrinkage (Figure 8). Moreover, Zhejiang Province emerged as the predominant provincial-level administrative division across all study periods, with the exception of the 2008–2013 interval during which Anhui Province exhibited a dominant status. The interprovincial disparities across the four study periods were relatively even. Both Zhejiang Province and Shanghai Municipality exhibited distinct V-shaped rebound trajectories in 2013. At the same time, Anhui Province demonstrated an inverse path. Jiangsu Province stood out as the sole provincial-level division demonstrating persistent intensification of population shrinkage with a transition from a rising rate to progressive deceleration. Serial transitions of the preponderate cities holding the most population shrinking counties across the four study phases were observed: Taizhou of Zhejiang Province (Phase I: eight counties), Yancheng (Phase II: nine counties), Hangzhou (Phase III: nine counties), and Hangzhou and Wenzhou (Phase IV: nine counties).

The county-level shrinking index of space dimension was acquired from the area calculation of reclassified land use types, as shown in Figure 9. The spatiotemporal distribution of space shrinkage was prominently distinguished from the economic and population shrinking dimensions. The space shrinkage demonstrated intensifying magnitude with a distinct geospatial progression, i.e., initial emergence in western regions, subsequent eastward expansion to central-eastern zones, culminating in pan-regional prevalence, while exhibiting pronounced northern predominance.

The space shrinkage displayed more spatiotemporal variations among cities and provinces compared to population shrinkage but obtained fewer fluctuations than the economic shrinkage (Figure 10). The provincial-level space shrinkage across four study phases revealed rotating leadership transitions, with Zhejiang Province owning the maximum number of space shrinking counties except during two intervals: 2008–2013 (Anhui Province) and 2018–2023 (Jiangsu Province). The interprovincial disparities in space shrinkage demonstrated an intensification of spatial polarization across four study phases. Anhui Province exhibited a unique non-monotonic path in space shrinking dynamics, with its county-level shrinking count demonstrating a distinctive N-shaped fluctuation. This path diverged from the persistent cumulative growth observed in Shanghai Municipality, as well as Zhejiang and Jiangsu provinces. The city with the greatest number of space shrinking counties shifted from Zhoushan (Phase I: one county) to Anqing (Phase II: four counties), subsequently bifurcating into the Hangzhou–Jinhua dual core (Phase III: six counties), before culminating in a polycentric cluster encompassing Yancheng, Wenzhou, Wuhu, and Anqing (Phase IV: eight counties).

The multidimensional county-level shrinkage was compared as listed in

Table 5. The comparison of multidimensional county-level shrinking identification.

Consistency	Dimension	Period	Number of County-Level Units
Three-dimensional consistency	Economic, population and space dimensions	2003–2008	0
		2008–2013	1
		2013–2018	9
		2018–2023	14
Two-dimensional consistency	Economic and population dimensions	2003–2008	52
		2008–2013	8
		2013–2018	17
		2018–2023	6
	Economic and space dimensions	2003–2008	1
		2008–2013	1
		2013–2018	6
		2018–2023	9
	Population and space dimensions	2003–2008	0
		2008–2013	24
		2013–2018	43
		2018–2023	78
None	Economic dimension	2003–2008	66
		2008–2013	4
		2013–2018	12
		2018–2023	0
	Population dimension	2003–2008	33
		2008–2013	87
		2013–2018	70
		2018–2023	38
	Space dimension	2003–2008	0
		2008–2013	9
		2013–2018	23
		2018–2023	53

. A continuous enhancement of counties experiencing simultaneous economic, population, and space shrinkage was discovered, which increased from 0 to 14 during 2003 to 2023. Similarly, the number of counties experiencing economic–space shrinkage increased from one to nine; meanwhile, the number of population–space shrinkage counties rose from 0 to 78. At the same time, counties experiencing concurrent economic and population shrinkage exhibited a distinctive volatility path, which sharply decreased and then increased, ultimately leading to a downward trend. This path was also monitored in counties that have only undergone economic shrinkage. The number of single-dimensional shrinking counties that were subjected to population shrinkage rose before the continuous decline, whereas that of space-shrinking counties progressively grew. Across the four study phases, most counties suffered from single-dimensional shrinkage, except from 2018 to 2023 when two-dimensional shrinkage was dominant.

3.3. Spatiotemporal Patterns of Urban Shrinking in the Yangtze River Delta

The shrinking trajectories of county-level urban entities from 2003 to 2023 for each dimension in YRD-UAs are shown in Figure 11. It delineated that widespread shrinking exposure occurred across counties, with the total experiencing at least one-dimensional shrinkage, although marked by evident heterogeneity in dimensional manifestations and shrinking trajectories. Economic shrinkage affected the fewest counties, followed by space shrinkage, while population shrinkage dominated, impacting 210 counties. As for the economic dimension, a total of 145 counties were subjected to temporary shrinkage, significantly outnumbering those experiencing discontinued or periodic shrinkage. The pronounced compositional heterogeneity of trajectory types of population shrinkage compared to the remaining dimensions was uncovered. Population shrinkage emerged as the sole dimension exhibiting a continued shrinking trajectory with county-level population decline manifesting four distinct temporal patterns: periodic (79 counties), temporary (74 counties), discontinued (43 counties), and continued shrinkages (14 counties). Space shrinkage showed partial parallels with economic shrinkage patterns, dominated by temporary cases of 104 counties, yet diverged in periodic shrinkage, where 65 counties substantially exceeded discontinued cases.

Although three dimensions displayed various shrinking trajectories, 19 counties experienced the same trajectory type of economic, population, and space shrinkages with the majority as temporary shrinkage (

Table 6. The comparison of multidimensional county-level shrinking trajectories.

Consistency	Dimension	Province	Number of County-Level Units
Three-dimensional consistency	Economic, population and space dimensions	Shanghai	2 (all temporary)
		Jiangsu	7 (1 periodic and 6 temporary)
		Zhejiang	3 (all temporary)
		Anhui	7 (all temporary)
Two-dimensional consistency	Economic and population dimensions	Shanghai	2 (all temporary)
		Jiangsu	6 (all temporary)
		Zhejiang	4 (all temporary)
		Anhui	4 (1 periodic and 3 temporary)
	Economic and space dimensions	Shanghai	1 (all temporary)
		Jiangsu	15 (all temporary)
		Zhejiang	17 (all temporary)
		Anhui	16 (1 periodic and 15 temporary)
	Population and space dimensions	Shanghai	4 (2 periodic and 2 temporary)
		Jiangsu	7 (all periodic)
		Zhejiang	6 (1 periodic and 5 temporary)
		Anhui	2 (1 periodic and 1 temporary)
None	Economic, population, or space dimension	Shanghai	1
		Jiangsu	4
		Zhejiang	10
		Anhui	3

). The shrinkage trajectory alignment of two dimensions revealed that shrinking counties predominantly exhibited temporary patterns, except for population-space dyads where periodic shrinkage predominated. For the four provincial-level divisions, the two-dimensional consistency of shrinking trajectories was primary followed by three-dimensional alignment, except Zhejiang Province, which exhibited the poorest three-dimensional synchronization. The two-dimensional consistency of economic and space shrinking trajectories most frequently appeared in Zhejiang, Jiangsu, and Anhui provinces. Shanghai Municipality demonstrated the strongest population–space shrinking trajectory alignment, yet exhibited the weakest economic–space concordance, whereas Anhui Province displayed the inverse pattern. Jiangsu and Zhejiang provinces both showed the weakest consistency in economic–population shrinking trajectory.

The global Moran’s I of multidimensional county-level shrinkage revealed phase-specific spatial autocorrelation patterns across three shrinking dimensions (

Table 7. The statistics of global Moran’s I of multidimensional county-level shrinkage.

Dimension	Period	Moran’s I	p-Value
Economic	2003–2008	0.10	0.011
	2008–2013	−0.03	0.47
	2013–2018	0.00	0.36
	2018–2023	0.00	0.06
Population	2003–2008	0.11	0.005
	2008–2013	0.28	0.000
	2013–2018	0.00	0.13
	2018–2023	0.11	0.011
Space	2003–2008	0.00	0.009
	2008–2013	0.07	0.06
	2013–2018	0.14	0.000
	2018–2023	0.19	0.000

). Population shrinkage demonstrated the highest spatial autocorrelation, followed by space shrinkage, while economic shrinkage exhibited minimal autocorrelation. Temporally, Phase IV (2018–2023) generally showed the highest spatial autocorrelation, surpassing Phase II (2008–2013). Economic shrinkage exhibited statistically significant positive spatial autocorrelation only during the initial phase with the p-value of 0.011 (2003–2008), with less than 5% likelihood that this clustered pattern could be the result of random chance. Positive spatial autocorrelation in population shrinkage was observed in three study phases with temporal intensity variations. The 2008–2013 period exhibited the highest clustering, followed by 2003–2008, while 2018–2023 showed the lowest spatial autocorrelation. Space shrinkage exhibited significantly positive autocorrelation across two study phases, yet demonstrated distinct temporal evolution compared to population shrinkage patterns. Global Moran’s I analysis revealed non-significant spatial clustering during 2008–2013, contrasting sharply with intensifying autocorrelation in subsequent periods: 2013–2018 and 2018–2023.

Significant spatial heterogeneity in shrinking degree patterns across dimensions and temporal phases was delineated by LISA clustering maps (Figure 12). Overall, the population shrinking degree exhibited the highest spatial heterogeneity, followed by space shrinking degree, while the economic shrinking degree demonstrated the lowest spatial dependence. During the 2003–2008 period, the distribution of hot spots (aggregations with relatively severe degrees) and cold spots (aggregations with relatively slight degrees), as well as their outliers, indicated the spatial heterogeneity in county-level economic shrinking degree within Zhejiang and Anhui provinces. The spatial dependence of the county-level economic shrinking degree weakened in the following 15 years. A relatively severe economic shrinkage clustered in Anhui Province during the 2013–2018 period and finally shifted to Jiangsu Province.

As for the county-level population shrinking degree, the hot spot location illustrated the spatial clustering pattern within Zhejiang and Jiangsu provinces, while outliers showed the spatial heterogeneity in Jiangsu and Anhui provinces from 2003 to 2008. After that, spatial heterogeneity was observed widely across YRD-UAs, especially in Zhejiang Province and Shanghai Municipality, with the northward movement of hot spots and relatively slight economic shrinkage clustering in the mid-east regions. Then, the spatial dependence of the county-level population shrinking degree was reduced when relatively severe economic shrinkage aggregated in Anhui Province. Ultimately, relative severe and slight economic shrinkage aggregated in Zhejiang and Anhui provinces, respectively, where spatial heterogeneity was pronounced.

At the beginning, spatial dependence on county-level space shrinkage was weak. Significantly high-value clustering of space shrinkage in Anhui Province was first identified during the 2008–2013 study period. Then, the hot spots moved eastward to Jiangsu Province and Shanghai Municipality. Over the past five years, relatively severe space shrinkage has been concentrated in the north, particularly in Jiangsu and Anhui provinces, while cold spots have clustered in Zhejiang Province. Concurrently, distinct spatial heterogeneity was observed in Jiangsu Province.

4. Discussion

4.1. Spatiotemporal Characteristics of County-Level Shrinking in the Yangtze River Delta

The results revealed outstanding spatiotemporal heterogeneity in county-level shrinkage across YRD-UAs. The spatial distribution of county-level shrinkage was characterized by north–south differentiation and coastal–inland divergence (Figure 5, Figure 6, Figure 7, Figure 8, Figure 9 and Figure 10). Indeed, county-level economic shrinking distribution displayed a north–south distinction, and appeared frequently in the northern region, especially in Jiangsu Province. Conversely, county-level shrinking distribution of population and space manifested the coastal-inland differentiation with the coast-oriented distribution, mainly located in Zhejiang Province. Multidimensional statistics of county-level shrinkage indicated that Yancheng city in Jiangsu Province, as well as Hangzhou and Wenzhou cities in Zhejiang Province, were critically widespread regions requiring targeted attention. Contrary to the aforementioned distributional characteristics, spatial patterns of county-level shrinking severity within YRD-UAs, as determined by global and local spatial autocorrelation analysis, revealed a north–south disparity in economic and space shrinkages, as well as east–west polarization in population shrinkage. Significant spatial dependence on county-level population and space shrinkages was demonstrated. Population shrinkage exhibited heterogeneous spatial clustering, with the highest autocorrelation observed during the study period, suggesting widespread population loss driven by multiple factors. Space shrinkage displayed progressively intensifying dependence with increasingly pronounced clustering effects, following more consistent temporal trends than population shrinkage, indicating dominance by relative singular drivers amenable to policy interventions. Conversely, economic shrinkage revealed minimal spatial dependence, with sporadic clustering confined to limited phases and regions, implying spatially random occurrences are highly responsive to regulatory measures. In particular, economic shrinking severity clusters transitioned from southern to northern regions, with initial slight clusters in Anhui Province later dispersing, while space shrinkage maintained persistent northern severity (Anhui–Jiangsu corridor) alongside emerging slight clusters in Zhejiang Province. Population shrinkage exhibited cyclical migration, with severe clusters shifting east-to-west (Zhejiang/Jiangsu → Anhui) before returning eastward, contrasting slight clusters that emerged in eastern coastal zones (Zhejiang/Shanghai) before concentrating westward in Anhui Province. It was disclosed by integrating county-level shrinking distribution and shrinking degree aggregation that slight economic and severe space shrinkages in Anhui and northern Jiangsu provinces should be given emphasis. Special attention should be paid to the population shrinkage in the western Anhui and eastern Zhejiang provinces as well as overtly slight space shrinkage in southern Zhejiang.

As for temporal variations, county-level shrinkage of YRD-UAs was characterized by declining change rates over time, with the directional polarity (positive/negative) of shrinking rates varying across dimensions and phases. Specifically, economic shrinkage fluctuated greatly, followed by population shrinkage, while space shrinkage continuously increased. Among the four provincial administrative divisions, Jiangsu exhibited the strongest temporal monotonicity in multidimensional shrinkage, marked by persistent increases in shrinking counties across both population and space dimensions. Shanghai underwent sustained decline in economic shrinkage alongside rising space shrinkage, while Zhejiang demonstrated monotonic space shrinkage growth exclusively. Anhui displayed non-monotonic trajectories in all dimensional shrinkages over time. The temporal evolution of multidimensional shrinking patterns exhibited a V-shaped trajectory in spatial autocorrelation and clustering, characterized by initial declines followed by late-phase intensification, with the highest aggregation observed during the 2018–2023 period. The shrinking trajectory was dominated by temporary shrinkage, suggesting high recovery potential. Multidimensional shrinking trajectories necessitated prioritized attention to periodic and continuous shrinkage. Continuous shrinkage (2003–2023) was exclusively observed in the population dimension, spatially concentrated in Zhejiang and Jiangsu provinces, which signified structurally entrenched population decline with limited resilience. The periodic trajectory (2013–2023) of economic shrinkage was concentrated in Anhui, while that of population and space shrinkages demonstrated the broader spatial distributions across the region. This widespread periodic shrinkage of population and space, reflecting sustained recent shrinking pressure, required intensive intervention. Economic shrinkage demonstrated the highest transience, with minimal episodic occurrence geographically confined to Anhui Province, indicating that temporary disturbances are highly amenable to policy intervention and rapid recovery.

In summary, the spatiotemporal characteristics of county-level shrinking in YRD-UAs based on urban entities are primarily characterized by spatial north–south divergence and a polarized east–west configuration, coupled with temporal decay in shrinkage intensity alongside intensified spatial agglomeration. Critical intervention areas included economic shrinkage in northwestern regions, specifically Anhui Province, as well as population and space shrinkages clustered in Jiangsu and Zhejiang provinces.

4.2. Optimal Dimension for Remote-Sensing-Based County-Level Shrinking Monitoring

The aforementioned findings revealed that urban-entity-based county-level shrinkage exhibited dimensional variations despite discernible spatiotemporal characteristics (Table 5, Table 6 and Table 7). Population shrinkage was the most prevalent dimension across county-level urban entities, followed by space shrinkage, with temporal progression showing a sequential shift in dominance from economic to population and ultimately to space shrinking patterns. Temporal comparison showed enhanced convergence in county-level shrinkage across the three dimensions, with principal inconsistencies predominantly resulting from synchronized economic-population shrinkage. Economic shrinkage exhibited dramatic fluctuations over time, whereas space shrinkage demonstrated a monotonically increasing trend. However, space shrinkage showed the strongest heterogeneity among cities and provinces, contrasting with the relative homogeneity observed in population shrinking patterns across administrative units. Throughout the study period, space shrinkage demonstrated the highest representativeness as an identification dimension, followed by economic shrinkage. The most representative dimension transitioned from population to economic before stabilizing as space dominance in the final two phases, while the dimension exhibiting the greatest divergence shifted from economic to population shrinkage during intermediate stages before reverting to economic patterns.

Derived from multisource remote sensing products, this study, based on urban entities, identified more severe county-level shrinkage of YRD-UAs than most previously documented, even though a few relevant studies have been conducted [9,40,52]. However, based on political–administrative units and nighttime light data, Tan et al. (2023) found more economically shrinking counties than this study [53]. This difference was due to the methodological framework employed in this study, which incorporated commuting-derived urban entities and systematically excluded rural areas, thereby enhancing objectivity in reflecting authentic county-level development dynamics [26]. Moreover, previous studies underestimated county-level shrinkage compared to this study, mainly due to their reliance on administrative unit-derived statistical data [40,52]. On the other hand, Zhang et al. (2024) defined county-level cities and measured shrinkage using land use maps and WorldPop products with the help of census data, which revealed that fewer counties shrank in terms of population [9]. In other words, the population shrinkage calculated in this study, which was based only on remote-sensing-based population products, had a certain level of uncertainty. The space shrinkage was initiated in this study as the percentage of non-construction land occupied by ecological land within urban entities was the most representative dimension. The land use inefficiency and suboptimal allocation of ecological spaces constrained county-level development, constituting a recognized systemic challenge within YRD-UAs [54,55].

A comparative analysis of multidimensional shrinkages and benchmarking against existing studies demonstrated that the novel space dimension incorporating ecological land proportion proposed in this study is recommended for preliminary county-level shrinking screening, followed by confirmation through nighttime light data variations within urban entities. However, the population dimension, quantified via remote-sensing-derived population datasets, necessitated further verification using census data.

4.3. Uncertainty and Management

This study identified urban entities based on land use maps and the average commuting distance of each city to enhance precision in monitoring county-level shrinking dynamics. The uncertainty propagation predominantly originated from data sources and threshold determination in urban entity delineation. The above findings and discussion proved that nighttime light data and land use maps were operational data sources for economic and space dimensions, which reduced uncertainty in county-level shrinking identification. Nevertheless, it may overestimate population shrinkage solely based on remote-sensing-based population products, which can be verified using census data to mitigate uncertainty. Future research should explore spatial-scale matching and geo-registration techniques to mitigate uncertainty arising from spatial-matching errors in multi-source spatiotemporal remote sensing datasets with inconsistent resolutions. Subsequent research should integrate field surveys and geospatial big data to calibrate commuter distances, thereby reducing methodological uncertainties owing to threshold determination in urban entity delineation. The uncertainty can also be reduced in future work by adopting higher spatial-resolution remote sensing products. The current findings indicated that county-level shrinkage at a five-year interval still fluctuates, underscoring the importance of subsequently investigating an optimal temporal interval [11]. The uncertainty quantification and shrinkage validation were essential but difficult to conduct and should be further studied through comparisons with census and statistical data.

The results revealed a recent decline in counties in YRD-UAs experiencing economic or population shrinkage, in contrast to persistent growth in space shrinkage. Although rates of shrinking changes across all three dimensions displayed deceleration, the prevalence of severe population or space shrinkage continued to escalate, with an increasing number of counties suffering from multidimensional shrinkage. This suggests the initial success of current policy interventions (e.g., integrated urban–rural systems), yet the optimization of ecological space configuration should be emphasized. Based on China’s macro-policy pivot toward “supernormal countercyclical adjustments” [56], addressing YRD-specific shrinkage drivers: manufacturing relocation, aging demographics, and fiscal fragmentation [40,57], the following evidence-based policy interventions are proposed to mitigate county-level shrinkage in YRD-UAs. The strategic prioritization should focus on enhancing the coverage ratios of critical ecological land covers (grasslands, shrubs, forests, water, and wetlands) within urban entities. It is suggested that governments convert vacant industrial lands into mixed-use eco-districts, imposing vacancy taxes on underutilized plots to fund the development of green infrastructure. Cropland/bare land within urban entities should be regulated with strict quotas, linking local fiscal transfers to reduction targets. Blue-green corridors (e.g., Qiantang River basin) should be integrated using flood-adaptive wetlands and urban forests, raising ecological land via fiscal incentives for carbon-sink projects. Particularly in the northern area of YRD-UAs, urban planning requires urgent reinforcement through (1) increasing ecological land proportions; (2) implementing polycentric development; and (3) controlling impervious surfaces and cropland within commutable ranges. Population shrinking patterns demonstrated nascent success in regional population rebalancing, evidenced by emerging county-level shrinkage clusters around Shanghai. Transbay rail networks should be accelerated to reshape labor flows, coupling transit nodes with satellite innovation zones. The northwest of YRD-UAs, specifically Anhui Province, requires targeted county-level economic revitalization and proactive measures to counter population depletion.

To sum up, the urban-entity-based multidimensional shrinkage framework employed in this study reduced uncertainty associated with the inclusion of rural areas. The exclusive utilization of remote sensing products demonstrated satisfactory assessment accuracy, with future refinements achievable through using high-spatial-resolution images and integrating census and geospatial big data analytics. Optimization patterns were achieved by YRD-UAs under the current planning policies. These findings recommended strengthening ecological land allocation within urban entities across the entire region, implementing polycentric development strategies in the north, and enhancing county-level economic governance in the northwest.

5. Conclusions

Urban shrinkage, a new trend spreading from developed to developing countries, has displayed unique spatiotemporal patterns in China, characterized by the co-occurrence of urban shrinkage alongside increasing urbanization. Previous studies on urban shrinkage identification have predominantly employed political–administrative boundaries as analytical units and relied on statistical yearbook data, where remote-sensing-based measurements were mainly single-dimensional, such as nighttime light data or population products. It generated problems with incorporating remote administrative rural areas, as well as spatiotemporal inconsistencies in statistical records, and inadequate understanding, which subsequently resulted in inaccurate shrinkage identification. Thus, this study pioneered a remote-sensing-based framework, using nighttime light, population, and land use maps with spatiotemporal pattern analysis, to identify urban entities and multidimensional county-level shrinkage across economic, population, and space dimensions within YRD-UAs—a paradigmatic metropolitan region in China.

The results demonstrated that urban entities in 215 counties of the YRD-UAs grew with a generally slowing speed between 2003 and 2023. The highest percentage of shrinking cities was for slight shrinkage, and the shrinking trajectory was dominated by temporary shrinkage. Most counties experienced population shrinkage in their coast-oriented distribution, whereas economic shrinkage affected the fewest counties, with lowest spatial clustering occurring northward. Population shrinkage also displayed the highest spatial autocorrelation, but its agglomeration weakened against space shrinkage clustering.

The urban-entity-based multidimensional shrinkage framework employed in this study reduced uncertainty associated with including rural areas. It indicated that the exclusive utilization of remote sensing products resulted in satisfactory assessment accuracy with future refinements achievable through using high-spatial-resolution images and integrating census and geospatial big data analytics. To further optimize patterns of YRD-UAs, it is recommended to strengthen ecological land allocation within urban entities across the entire region and to implement polycentric development strategies in the north, as well as enhance county-level economic governance in the northwest.